Upload 172 files

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+training/evidence/plots/rl_chosen_ok_by_method.png filter=lfs diff=lfs merge=lfs -text
+training/reward_curves/reward_curve.png filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,11 @@
+FROM python:3.11-slim
+ENV PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1
+WORKDIR /app
+COPY . .
+RUN pip install --no-cache-dir .
+EXPOSE 7860
+HEALTHCHECK --interval=30s --timeout=5s --start-period=20s --retries=3 \
+  CMD python -c "import urllib.request; urllib.request.urlopen('http://127.0.0.1:7860/health', timeout=3).read()" || exit 1
+CMD ["python", "-m", "uvicorn", "server.app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -1,11 +1,176 @@
 ---
-title: AutoDataLab2.0
-emoji: 🏃
-colorFrom: yellow
-colorTo: green
+title: AutoDataLab Plus Plus
+emoji: 🏢
+colorFrom: indigo
+colorTo: purple
 sdk: docker
+app_port: 7860
 pinned: false
-short_description: OpenEnv Hackathon Round2
 ---
+# AutoDataLab++
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+> **The first OpenEnv environment whose reward explicitly punishes the LLM for not knowing when to stop.**
+> A Chief-of-Staff (CoS) policy must route work across four typed specialists — Data Analyst, Finance, Strategy, HR — and *submit* a complete CEO brief inside a step budget. Skip an expert and the grader penalises the brief; over-consult and the shaped reward penalises the trajectory. Both errors hurt, so the policy has to learn the *capability gap* between knowing facts and knowing when it has enough.
+
+## Headline result (fallback disabled)
+
+![Terminal reward, fallback disabled — base vs trained MLP CoS vs oracle upper bound, 3 hard tasks, 5 seeds](training/evidence/plots/headline_terminal_reward.png)
+
+| Task | Base (naive) | **Trained MLP CoS** (REINFORCE, 600 ep) | Oracle router (upper bound) |
+|---|---|---|---|
+| `hard_brief`   | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.90 |
+| `expert_brief` | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.89 |
+| `crisis_brief` | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.89 |
+
+Numbers are **terminal grader scores in `[0,1]`**, mean over **5 seeds**, with the environment's expert-auto-fill **safety net OFF** (`eval_mode=true`). Two RAG settings shown per cell (off / on). Raw runs in `training/evidence/headline_benchmark.json`. Reproduce with `python3 training/scripts/run_headline_benchmark.py`.
+
+What this says, plainly:
+
+- **Untrained baseline ~0.27.** A naive policy that consults the analyst and submits gets penalised, because it never routes to finance / strategy / HR and the grader sees an incomplete brief.
+- **Our actually-trained MLP CoS ~0.73.** A 2-layer MLP routing policy trained with REINFORCE for 600 episodes (`training/checkpoints/cos_final.pt`) recovers the bulk of the headroom — **+0.46 absolute reward over the naive baseline**, on tasks it was not memorising. This is the headline number for "what we trained."
+- **Oracle router ~0.88.** Handcoded canonical-order policy. *Upper bound*, not a trained model — published so judges can see how much routing headroom remains for a future GRPO/SFT LLM run.
+
+The take-away is the **+0.46 trained-vs-baseline gap**, plus the **~0.15 oracle headroom** that future RL runs can chase. The base→oracle gap (~0.6) is the size of the routing problem; our trained MLP closes ~75 % of it.
+
+## Live demo
+
+- **Office UI (Hugging Face Space):** [https://uchihamadara1816-autodatalab2-0.hf.space/ui/](https://uchihamadara1816-autodatalab2-0.hf.space/ui/)
+- Health endpoint: [https://uchihamadara1816-autodatalab2-0.hf.space/health](https://uchihamadara1816-autodatalab2-0.hf.space/health)
+- Pick a task, toggle RAG, run the four policies side-by-side. The naive baseline submits an incomplete brief; the **MLP trained CoS** and **oracle** rows finish with all four boxes lit.
+
+## Quickstart (local)
+
+```bash
+pip install -e .
+# or: uv sync && uv run server
+python3 -m server.app
+```
+
+| Endpoint | URL |
+| --- | --- |
+| API root | [http://127.0.0.1:7860/](http://127.0.0.1:7860/) |
+| Health | [http://127.0.0.1:7860/health](http://127.0.0.1:7860/health) |
+| **Demo UI** | [http://127.0.0.1:7860/ui/](http://127.0.0.1:7860/ui/) |
+
+Pre-submission checks:
+
+```bash
+python3 validate_submission.py
+openenv validate --verbose
+```
+
+Oracle rollout (3 tasks):
+
+```bash
+python3 inference.py --oracle
+```
+
+### Honest evaluation: turn off the safety net
+
+The environment ships with a *production-mode* fallback that auto-completes any
+required expert the policy forgot, so end-users always see a full brief. For
+**evaluation** you want the opposite: the policy must succeed (or fail) on its
+own.
+
+```bash
+# CLI
+python3 training/scripts/run_headline_benchmark.py    # 5 seeds, fallback OFF
+
+# HTTP
+curl -X POST http://127.0.0.1:7860/reset \
+     -H 'content-type: application/json' \
+     -d '{"task":"hard_brief","use_rag":true,"eval_mode":true}'
+```
+
+When `eval_mode=true`, the env runs with `auto_fill_required=False` and
+`shaping="strict"`. Submitting an incomplete brief is allowed and is reflected
+in the terminal grader score, which is exactly what produces the headline
+trained-vs-baseline gap shown above.
+
+## Deployment
+
+### Environment variables
+
+| Variable | Purpose |
+| -------- | ------- |
+| `API_BASE_URL` | OpenAI-compatible endpoint (default: Hugging Face router) |
+| `API_KEY` or `HF_TOKEN` | For LLM CoS; if unset, `inference.py` uses **oracle** |
+| `MODEL_NAME` | Model id for the CoS LLM path |
+| `AUTODATALAB_PLUS_TASKS` | Comma-separated task ids (default: all three briefs) |
+
+Copy `.env.example` to `.env` and adjust. For Docker / Spaces, set secrets in the platform UI.
+
+### Docker
+
+Build and run (port **7860** matches `openenv.yaml`):
+
+```bash
+docker build -t autodatalab-plus .
+# First build can take several minutes: PyTorch + OpenEnv stack download ~1.5GB+ of wheels.
+docker run --rm -p 7860:7860 autodatalab-plus
+```
+
+Smoke test:
+
+```bash
+curl -s http://127.0.0.1:7860/health
+```
+
+The image includes a **HEALTHCHECK** on `/health`. Training checkpoints under `training/checkpoints/` (if present in the build context) are included so the **trained CoS** policy is available in `/visualize/run` when a checkpoint exists.
+
+### Hugging Face Space (OpenEnv)
+
+- Type: **Docker** or use the `openenv.yaml` `app: server.app:app` + `port: 7860` as documented in the [OpenEnv](https://huggingface.co/docs/hub/en/spaces) flow you use for the hackathon.  
+- Ensure **build context** is this repo; **do not** commit large secrets.  
+- After deploy: hit `/health`, open `/ui/`, run `python3 validate_submission.py` against the **live URL** (adjust `ROOT` or use env if your script supports it).
+
+## Project layout (high level)
+
+- `ceo_brief_env/` — Pydantic models, environment, graders, `tasks/`
+- `inference.py` — oracle / baselines / LLM / trained CoS, `[START]`/`[STEP]`/`[END]` logs
+- `server/app.py` — FastAPI; `/reset`, `/step`, `/state`, `/visualize/run`
+- `training/scripts/` — re-runnable SFT/DPO/RL/Kaggle scripts and notebooks
+- `training/evidence/` — small replayable evidence JSON plus loss/reward plots
+- `training/checkpoints/` — small local MLP CoS artifacts, when present
+- `subenvs/` — analyst + email/HR tools
+
+## Evidence map (what every artefact actually is)
+
+We deliberately label every artefact so a strict reader can tell training from smoke-tests at a glance. Two distinct experiments live in this repo:
+
+**Experiment A — MLP+REINFORCE training (this *is* the trained model in the headline)**
+
+- `training/checkpoints/cos_final.pt` — 2-layer MLP routing policy, trained with REINFORCE for 600 episodes at lr 0.003 (`training/scripts/train_cos_local.py`).
+- `training/reward_curves/before_after.json` — real before/after evaluation under the *production* env (safety net ON): `mean_terminal_before ≈ 0.405 → mean_terminal_after ≈ 0.878` across all 6 tasks. This is the training-curve story, not the headline regime.
+- `training/reward_curves/reward_curve.png`, `training/evidence/plots/loss_curve.png` — per-episode reward and training loss for that REINFORCE run.
+- The same checkpoint, evaluated under the **honest regime** (safety net OFF, 5 seeds, 3 hard tasks), produces the `trained_mlp` row in the headline plot above (~0.73). The two numbers (~0.88 with safety net / ~0.73 without) are both real — they're the same model, evaluated under the production env vs. the honest-eval env.
+- This is the trained model we put behind the `MLP trained CoS` policy in the live demo, and it is the headline trained-model number.
+
+**Experiment B — Qwen2.5-1.5B SFT/DPO/GRPO routing**
+
+- `training/scripts/kaggle_rl_1p5b_methods.py`, `kaggle_run_all_1p5b_experiments.py` — re-runnable RL scripts on Kaggle.
+- `training/evidence/sft/`, `dpo/`, `sft_dpo/`, `grpo_rlvr/` — per-method `evidence.json` rollouts on the 3 hard tasks × RAG on/off. **These are short runs** (≤70 GRPO steps, capped by free-tier compute) and we treat them as smoke tests of the training loop, not as a converged GRPO policy. The `train_metrics.json` in each folder is the raw metric stream from those runs.
+- `training/evidence/plots/rl_*.png`, `policy_rewards_by_method.png`, `terminal_scores_by_method.png` — plots derived from those short runs.
+- We **do not** claim a fully converged GRPO policy from these. The Qwen path is a working pipeline; longer runs are future work.
+
+**Experiment C — Headline benchmark, fallback OFF (the `eval_mode` story)**
+
+- `training/scripts/run_headline_benchmark.py` — 3 hard tasks × 4 policies × 2 RAG × 5 seeds, run with `auto_fill_required=False` and `shaping="strict"`.
+- `training/evidence/headline_benchmark.json` — raw cell-level numbers (mean / std / per-seed runs). Schema `autodatalab-plus.headline_benchmark.v2`.
+- `training/evidence/plots/headline_terminal_reward.png` (the bar chart at the top of this README) — reproduced from that JSON.
+- The four bars are clearly labelled: two **untrained baselines** (`base_naive`, `base_roundrobin`), one **actually-trained model** (`trained_mlp`, the REINFORCE-trained MLP CoS at `training/checkpoints/cos_final.pt`), and one **upper bound** (`oracle_router`, the handcoded canonical-order policy that our SFT/GRPO LLM trajectories imitate when they succeed). We do **not** label the oracle as a trained model anywhere in this README.
+
+If a judge wants the LLM-driven version of Experiment C, point any LoRA adapter at `inference.py` (or load it inside `kaggle_three_llms_text.py`) and re-run with `eval_mode=true`; the same script gives a trained-LLM number that fits between `trained_mlp` and `oracle_router`.
+
+## Honest known gaps
+
+We want to be calibrated. As of submission:
+
+- **GRPO runs are short.** ~70 update steps on a 4-action routing problem is enough to see the loop work end-to-end, not enough to claim a converged RL policy. The shaped-reward signal becomes meaningful at longer horizons; we only had compute for the smoke run.
+- **Per-method evidence rollouts are deterministic** (decoded with low temperature for reproducibility). They demonstrate the routing pattern under each method, not the variance you'd get from a stochastic policy. The headline benchmark uses 5 seeds and reports std bars precisely to put a real spread on the trained-vs-base comparison.
+- **`eval_mode` is opt-in.** Production `/reset` defaults to `eval_mode=false` (auto-fill on) so end-users always see a complete brief in the demo UI; the policy-comparison endpoint `/visualize/run` and the benchmark script default to `eval_mode=true` so terminal scores reflect the policy's own routing competence.
+- **`memory/` corpus is intentionally small** (BM25 over a few company SOPs/policies). RAG gives a steady ~+0.01 terminal reward across all 3 hard tasks (visible in the headline JSON); we report it as a small, citable lift, not a magic boost.
+
+## License
+
+Hackathon / team use per repository owner.

blog.md

@@ -0,0 +1,214 @@
+# AutoDataLab++: Teaching a Chief of Staff to Know When to Stop
+
+**Team:** Peaky Blinders — Sai Kamal Nannuri ([Uchihakamal1816](https://github.com/Uchihakamal1816)) & Ajeet Kumar ([ajeetkartikay](https://github.com/ajeetkartikay))
+**Live demo:** [uchihamadara1816-autodatalab2-0.hf.space/ui/](https://uchihamadara1816-autodatalab2-0.hf.space/ui/)
+**Code:** [github.com/Uchihakamal1816/AutoDataLab-](https://github.com/Uchihakamal1816/AutoDataLab-) (the trailing dash is the actual repo slug)
+
+---
+
+## TL;DR
+
+We built an OpenEnv environment where a **Chief of Staff (CoS)** policy must route work across four specialists — Data Analyst, Finance, Strategy, HR — and **submit a complete CEO brief inside a step budget**. Skipping an expert leaves the brief incomplete and the grader penalises it; over-consulting wastes the step budget and the shaped reward penalises that too. Either error hurts. The policy has to learn the *capability gap* between knowing facts and knowing when it has enough.
+
+We trained a **2-layer MLP routing policy with REINFORCE** (600 episodes, lr 0.003) and benchmarked it against an untrained baseline and a handcoded oracle — fallback off, 5 seeds, 3 hard tasks.
+
+![Headline result: terminal reward, fallback disabled — base vs trained MLP CoS vs oracle upper bound](training/evidence/plots/headline_terminal_reward.png)
+
+| Task | Base (naive) | **Trained MLP CoS** (REINFORCE, 600 ep) | Oracle router (upper bound) |
+|---|---|---|---|
+| `hard_brief`   | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.90 |
+| `expert_brief` | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.89 |
+| `crisis_brief` | 0.27 / 0.32 | **0.73 / 0.74** | 0.88 / 0.89 |
+
+*Cells show RAG-off / RAG-on. Numbers are terminal grader scores in `[0, 1]`, mean over 5 seeds, with the env's expert-auto-fill safety net OFF (`eval_mode=true`). Raw runs in [`training/evidence/headline_benchmark.json`](training/evidence/headline_benchmark.json); reproduce with `python3 training/scripts/run_headline_benchmark.py`.*
+
+The trained MLP closes **~75 %** of the base→oracle gap (+0.46 absolute reward over the naive baseline). The remaining ~0.15 of headroom is what a future GRPO/SFT LLM run can chase. We also wired up a Qwen2.5-1.5B SFT/DPO/GRPO pipeline as Experiment B; we treat it as a working pipeline, not a converged result, and we're explicit about that throughout.
+
+---
+
+## Why this is interesting
+
+Most LLM benchmarks test single-turn reasoning. Real enterprise work isn't like that. A CEO doesn't ask one question; they ask *"look at the data, project the quarter, recommend an action, and draft the memo."* That's a **routing problem**: which expert, in what order, with what info, and — critically — when to stop.
+
+Next-token training rewards plausible-sounding text. **It does not reward stopping.** That's why a prompted-only LLM tends to either spam "consult strategy" or invent a plausible answer instead of asking HR. Our hypothesis is that a *small* policy trained with RL on the routing decision can outperform a much larger one that's just prompted to orchestrate, because the small one is being shaped exactly on the bit that matters.
+
+So we made an environment where that bit *is* the reward — and where ignoring HR doesn't get hidden behind a polished sentence.
+
+---
+
+## The environment
+
+Six tasks, three of them on the critical path:
+
+| Task | Difficulty | Required experts |
+|---|---|---|
+| `easy_brief`    | easy   | analyst, finance, hr |
+| `medium_brief`  | medium | analyst, finance, strategy, hr |
+| `hard_brief`    | hard   | analyst, finance, strategy, hr |
+| `expert_brief`  | hard   | analyst, finance, strategy, hr |
+| `risk_brief`    | hard   | analyst, finance, strategy, hr |
+| `crisis_brief`  | hard   | analyst, finance, strategy, hr |
+
+The CoS has a small discrete action space: `consult(expert)`, `ask(expert, sub_question)` (a focused consult), `summarize`, `submit`, `noop`. Episodes max out at 12–14 steps. The reward is shaped:
+
+- **+0.10** for a productive first consult of a required expert
+- **+0.02** for any other useful action (e.g. summarising once)
+- **−0.05** for repeating a consult, **−0.10** for a third+ consult of the same expert (`shaping="strict"`)
+- **−0.05** for summarising before required experts have reported
+- **−0.02** per step (efficiency pressure)
+- **+ terminal grader score** in `(0.001, 0.999)` on submit
+
+The four specialists are real subenvs:
+
+- **Data Analyst** — wraps `subenvs/autodatalab/analytics.py`: cleaning, KPIs, revenue derivation, data-quality scoring.
+- **Finance** — deterministic projection, variance vs plan, break-even sensitivity.
+- **Strategy** — 3-bullet operating plan with citations to upstream numbers; emits structured `Present / Future` buy-sell-hold-trim guidance on the NVDA / AAPL / JPM watchlist.
+- **HR / Comms** — wraps `subenvs/email/`: drafts memos and scores them on a blended structure + tone + audience rubric.
+
+Each specialist returns a typed `ExpertReport`. The CoS composes them into a final `Brief`. Schemas live in `ceo_brief_env/models.py`.
+
+### The honest-evaluation flag
+
+The env ships with a *production* mode that auto-completes any missing required expert before grading — so end-users always see a full brief in the UI. That's a nice product, but it lies to a researcher: a policy that never thought to consult HR still gets HR's report stapled in.
+
+So `/reset` and `/visualize/run` both accept **`eval_mode=true`**. When set, the env runs with `auto_fill_required=False` and `shaping="strict"`. Submitting an incomplete brief is *allowed*, the grader sees the gap, and the terminal score reflects the policy's own routing competence. **All headline numbers in this post are reported in this regime.** It's also our headline contribution as an environment: we believe this is the first OpenEnv env whose reward explicitly punishes the LLM for not knowing when to stop.
+
+---
+
+## What we trained — Experiment A (the headline number)
+
+The trained model in the headline is **a 2-layer MLP routing policy with REINFORCE**, trained on featurised observations of the env. Code: [`training/scripts/train_cos_local.py`](training/scripts/train_cos_local.py). Checkpoint: [`training/checkpoints/cos_final.pt`](training/checkpoints/cos_final.pt).
+
+- **Input features**: a small fixed-length vector — task one-hot, expert-consulted bits, current step, data-quality score, whether each required expert has reported.
+- **Output**: logits over the discrete action space (consult-each-expert, ask-each-expert, summarize, submit, noop).
+- **Algorithm**: REINFORCE with a small entropy bonus, batched over 600 episodes, lr 0.003.
+- **Training time**: ~5 minutes on CPU.
+
+We chose this as the headline trained model because it's **honest evidence that the env is trainable end-to-end** — the policy improves from random to near-oracle performance under shaped reward. Before-vs-after numbers, evaluated under the *production* env (safety net ON, all 6 tasks):
+
+| Task | Before training | After training |
+|---|---|---|
+| `easy_brief`    | 0.451 | 0.866 |
+| `medium_brief`  | 0.467 | 0.887 |
+| `hard_brief`    | 0.399 | 0.883 |
+| `expert_brief`  | 0.523 | 0.883 |
+| `risk_brief`    | 0.283 | 0.868 |
+| `crisis_brief`  | 0.307 | 0.881 |
+| **mean**        | **0.405** | **0.878** |
+
+(Source: [`training/reward_curves/before_after.json`](training/reward_curves/before_after.json). Curve in [`training/reward_curves/reward_curve.png`](training/reward_curves/reward_curve.png).)
+
+The same checkpoint, evaluated under the **honest regime** (safety net OFF, 5 seeds, 3 hard tasks) lands at **~0.73 terminal reward**. Both numbers are real: same model, two regimes. The 0.88-with-safety-net number tells you *the brief content is correct when the policy gets all four experts on board*; the 0.73-without-safety-net number tells you *the policy gets ~3.7 of the 4 required experts right on its own*. We report the second one as the headline because it's the one that doesn't depend on the env helping.
+
+---
+
+## Experiment B — Qwen2.5-1.5B routing pipeline (working, not converged)
+
+We also built a full HF-TRL pipeline to train **Qwen2.5-1.5B-Instruct** as the CoS using SFT, DPO, and a thin custom GRPO loop on top of HF transformers + PEFT (LoRA).
+
+**Pipeline:** [`training/scripts/kaggle_run_all_1p5b_experiments.py`](training/scripts/kaggle_run_all_1p5b_experiments.py)
+- SFT on demonstration trajectories (`trl.SFTTrainer`)
+- DPO on chosen-vs-rejected routing pairs (`trl.DPOTrainer`)
+- GRPO + RLVR on top of the SFT adapter, with an oracle log-prob anchor (`--sft-anchor`) to prevent collapse
+- Optional GRPO and PPO variants for ablation
+
+**Why we don't put this in the headline.** Free-tier compute capped us at ~70 GRPO update steps. That's enough to verify the loop runs end-to-end and that the SFT + GRPO trajectories converge to the canonical routing pattern; it is **not** enough to claim a converged RL policy. The metric streams ([`training/evidence/rl_training_metrics/`](training/evidence/rl_training_metrics/)) reflect that — reward is flat-ish around zero with high variance, which is what a 70-step run on a 4-action problem should look like.
+
+What the pipeline *did* deliver, evaluated under the production env:
+
+| Method | Routes all 4 required experts? | Needed env fallback? | Terminal score (RAG off / on) |
+|---|---|---|---|
+| DPO-only             | 3 / 4 (HR missing)        | yes, every run | 0.882 / 0.892 |
+| SFT                  | 4 / 4 | no  | 0.882 / 0.892 |
+| SFT + DPO            | 4 / 4 | no  | 0.882 / 0.892 |
+| SFT + GRPO + RLVR    | 4 / 4 | no  | 0.882 / 0.892 |
+
+(Source: per-method `evidence.json` files under [`training/evidence/`](training/evidence/). Decoded at low temperature, so the rollouts are deterministic — they show the policy *route*, not stochastic variance. The headline benchmark uses 5 seeds for that.)
+
+The interesting line is the **DPO-only row**: the policy converges to "consult analyst → finance → strategy → strategy → strategy → strategy" and never picks HR. The env's safety net then auto-completes HR before grading, so the *terminal* score still looks fine — but the *trajectory* tells you the policy hasn't actually solved the routing problem. That gap between "score" and "behaviour" is exactly why we added `eval_mode=true`: under that flag the DPO row would collapse to ~0.27, the same as the naive baseline, and the SFT/GRPO row would stay near 0.88.
+
+If a judge wants the LLM-driven version of the headline benchmark, point any LoRA adapter at [`inference.py`](inference.py) (or load it inside [`training/scripts/kaggle_three_llms_text.py`](training/scripts/kaggle_three_llms_text.py)) and re-run with `eval_mode=true`. The same script that produced our headline plot will slot the LLM into a 5th bar between `trained_mlp` and `oracle_router`.
+
+---
+
+## RAG, and why it's the right fit for this env
+
+The `memory/` package contains a small corpus of company artifacts: SOPs (data quality, finance forecasting, comms style, strategy playbook), policies (compliance constraints), and history (last-quarter review, exemplar memos). The retriever is **BM25 over the markdown corpus** — no neural embeddings. We chose that on purpose:
+
+1. **It externalises domain knowledge from the policy.** The CoS doesn't memorise SOPs; it just learns *when to invoke the expert that knows them*. That keeps the action space small and the RL credit assignment clean.
+2. **It's deterministic.** RL rollouts need reproducibility. Hallucinated specifics ruin reward signal. Pinning experts to retrieved company text makes the brief auditable and the reward consistent across seeds.
+3. **It separates *what* to say from *who* should say it.** The policy learns the latter; the corpus owns the former. New SOPs ship without retraining.
+4. **It gives the policy a measurable, exploitable axis.** From the headline benchmark, RAG-on is **+0.013 terminal reward** for both `trained_mlp` and `oracle_router`, across all 3 hard tasks. The naive baseline gets +0.057 from RAG, but it's still capped at ~0.32 because the brief is incomplete — RAG doesn't fix bad routing.
+
+RAG isn't a magic boost here. It's a steady, citable, ~1 point lift that the policy can learn to prefer when it's available — exactly what you want when the policy is small.
+
+---
+
+## Honest known gaps
+
+We want the calibration to be visible:
+
+- **GRPO runs are short.** ~70 update steps on a 4-action routing problem is enough to see the loop work end-to-end, not enough to claim a converged RL policy. The shaped-reward signal becomes meaningful at longer horizons; we only had compute for the smoke run. We do *not* put GRPO in the headline.
+- **Per-method evidence rollouts are deterministic.** They demonstrate the routing pattern under each training method, not the variance you'd get from a stochastic policy. The headline benchmark uses 5 seeds and reports std bars precisely to put a real spread on the trained-vs-base comparison.
+- **`eval_mode` is opt-in.** Production `/reset` defaults to `eval_mode=false` so the demo UI always shows a complete brief. The policy-comparison endpoint `/visualize/run` and the benchmark script default to `eval_mode=true` so terminal scores reflect the policy's own routing competence. Both modes are documented and reproducible.
+- **`memory/` corpus is intentionally small.** RAG is a working axis, not a knowledge-base experiment. Scaling the corpus is straightforward (drop more markdown into `memory/`); it just isn't part of this submission.
+- **Oracle is an upper bound, not a model.** The handcoded canonical-order policy. We label it that way in the plot, the table, and the JSON schema. It's there so judges can see how much routing headroom remains.
+
+---
+
+## What we'd do with more time
+
+- **Path-quality reward.** The current grader is content-driven; a routing-efficiency component (penalising extra steps even when the brief is complete) would steepen the gradient for RL.
+- **Multi-turn refinement.** Today each expert returns once. Letting the CoS re-query an expert with a refined sub-question opens up real planning behaviour; the `ask(expert, sub_question)` action exists for this and is partially wired up.
+- **Adversarial tasks.** A task where one expert lies (data quality is bad but Analyst doesn't catch it) would force the CoS to learn validation routing.
+- **Longer Qwen RL.** The pipeline is ready; we just need a few hours of free-tier GPU to push GRPO past the smoke-run regime.
+- **Cross-org RAG.** BM25 over a single repo is the simplest case; a per-tenant retriever would let the same env sit in front of any company's SOP corpus.
+
+---
+
+## The stack
+
+- **OpenEnv** — environment spec, validation (`openenv validate` passes on all 6 tasks), deployment.
+- **Qwen2.5-1.5B-Instruct** — base policy model for Experiment B.
+- **Hugging Face TRL** — `SFTTrainer` and `DPOTrainer` for Experiment B; thin custom GRPO/PPO loops on top of HF transformers + PEFT (LoRA) for the RL ablations.
+- **PyTorch** — the MLP routing policy in Experiment A.
+- **FastAPI + uvicorn** — serving (port 7860, matches `openenv.yaml`).
+- **Hugging Face Spaces (Docker)** — deployment.
+- **Pydantic** — typed action / observation / report / brief schemas.
+- **BM25 (rank-bm25)** — RAG retriever. Deterministic, fast, no embeddings.
+
+---
+
+## Try it in 30 seconds
+
+```bash
+# Smoke-test the live deployment
+curl https://uchihamadara1816-autodatalab2-0.hf.space/health
+# {"status":"healthy"}
+
+# Honest-evaluation episode (fallback OFF — the headline regime)
+curl -X POST https://uchihamadara1816-autodatalab2-0.hf.space/reset \
+     -H 'content-type: application/json' \
+     -d '{"task":"hard_brief","use_rag":true,"eval_mode":true}'
+
+# Or open the office UI
+# https://uchihamadara1816-autodatalab2-0.hf.space/ui/
+```
+
+In the UI: pick `crisis_brief`, toggle RAG on, run **`MLP trained CoS`** then **`naive baseline`** side-by-side. The trained MLP lights up all four office boxes and submits cleanly; the naive baseline stops after the analyst and ships an incomplete brief. That's the headline capability gap, visible in 30 seconds.
+
+To reproduce the full headline plot locally:
+
+```bash
+git clone https://github.com/Uchihakamal1816/AutoDataLab-.git autodatalab-plus
+cd autodatalab-plus && pip install -e .
+python3 training/scripts/run_headline_benchmark.py
+# writes training/evidence/headline_benchmark.json
+# writes training/evidence/plots/headline_terminal_reward.{png,svg}
+```
+
+---
+
+## Acknowledgments
+
+Built for the OpenEnv hackathon. Thanks to the OpenEnv team for the spec and the validation tooling, Hugging Face for the Spaces hosting and the TRL stack, and Kaggle for the free GPU minutes that ran our SFT and short GRPO passes.

cache/ablation_results.json

@@ -0,0 +1,22 @@
+{
+  "single": {
+    "easy_brief": 0.5618,
+    "medium_brief": 0.4643,
+    "hard_brief": 0.4544
+  },
+  "roundrobin": {
+    "easy_brief": 0.907,
+    "medium_brief": 0.8534,
+    "hard_brief": 0.8436
+  },
+  "oracle_or_llm": {
+    "easy_brief": 0.907,
+    "medium_brief": 0.8534,
+    "hard_brief": 0.8436
+  },
+  "trained_cos": {
+    "easy_brief": 0.907,
+    "medium_brief": 0.7286,
+    "hard_brief": 0.7187
+  }
+}

ceo_brief_env/__init__.py

@@ -0,0 +1,12 @@
+from .environment import CEOBriefEnvironment, oracle_action_for_observation
+from .models import Brief, CoSAction, CoSObservation, CoSState, ExpertReport
+
+__all__ = [
+    "Brief",
+    "CEOBriefEnvironment",
+    "CoSAction",
+    "CoSObservation",
+    "CoSState",
+    "ExpertReport",
+    "oracle_action_for_observation",
+]

ceo_brief_env/checkpoint_load.py

@@ -0,0 +1,15 @@
+"""Load trusted local PyTorch state-dict checkpoints (CoS policy weights)."""
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any
+
+
+def load_state_dict(ckpt_path: Path, map_location: str | Any = "cpu") -> Any:
+    import torch
+
+    p = Path(ckpt_path)
+    try:
+        return torch.load(p, map_location=map_location, weights_only=True)
+    except TypeError:
+        return torch.load(p, map_location=map_location)

ceo_brief_env/environment.py

@@ -0,0 +1,284 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any, Dict
+from uuid import uuid4
+
+import pandas as pd
+
+from .experts import DataAnalystExpert, FinanceExpert, HRExpert, StrategyExpert
+from .graders import grade_episode, load_metric_ground_truth
+from .models import Brief, CoSAction, CoSObservation, CoSState, ExpertReport, RewardBreakdown
+
+TASK_ROOT = Path(__file__).resolve().parent / 'tasks'
+
+# Per-task order for “single” baselines, oracle, and any policy that should cover required experts.
+REQUIRED_EXPERTS_BY_TASK: dict[str, list[str]] = {
+    'easy_brief': ['analyst', 'finance', 'hr'],
+    'medium_brief': ['analyst', 'finance', 'strategy', 'hr'],
+    'hard_brief': ['analyst', 'finance', 'strategy', 'hr'],
+    'expert_brief': ['analyst', 'finance', 'strategy', 'hr'],
+    'risk_brief': ['analyst', 'finance', 'strategy', 'hr'],
+    'crisis_brief': ['analyst', 'finance', 'strategy', 'hr'],
+}
+
+
+def required_experts_for_task(task_name: str) -> list[str]:
+    return list(REQUIRED_EXPERTS_BY_TASK.get(task_name, ['analyst', 'finance', 'hr']))
+
+
+class CEOBriefEnvironment:
+    def __init__(self, shaping: str = "default", auto_fill_required: bool = True) -> None:
+        """Multi-agent CEO-brief env.
+
+        ``shaping`` controls the dense per-step reward. The terminal grader is
+        unchanged either way; that is what hackathon scoring uses.
+
+        - ``"default"``: legacy per-step rewards. Stable; matches existing
+          trained checkpoints and submitted runs.
+        - ``"strict"``: anti-degenerate shaping for RL training. Adds a
+          repetition penalty, an over-consult penalty, an early-finish bonus
+          when all required experts are covered, and a stronger penalty for
+          summarizing before required experts have reported. Use for new
+          GRPO/REINFORCE runs to discourage "summarize-spam -> submit" lazy
+          policies.
+
+        ``auto_fill_required`` keeps the production/demo environment robust by
+        filling any missing required experts before composing or grading. Turn
+        it off only for policy-evidence runs where we want to observe what the
+        LLM actually routed by itself.
+        """
+        self.analyst = DataAnalystExpert()
+        self.finance = FinanceExpert()
+        self.hr = HRExpert()
+        self.strategy = StrategyExpert()
+        self.use_rag = False
+        self.shaping = shaping if shaping in {"default", "strict"} else "default"
+        self.auto_fill_required = auto_fill_required
+        self.reset()
+
+    def reset(self, task: str = 'easy_brief', episode_id: str | None = None, use_rag: bool = False) -> CoSObservation:
+        self.use_rag = use_rag
+        self.episode_id = episode_id or str(uuid4())
+        self.task_name = task if (TASK_ROOT / task).exists() else 'easy_brief'
+        task_dir = TASK_ROOT / self.task_name
+        self.raw_df = pd.read_csv(task_dir / 'raw.csv')
+        self.gt_metrics = load_metric_ground_truth(str(task_dir / 'ground_truth.csv')) if (task_dir / 'ground_truth.csv').exists() else {}
+        with open(task_dir / 'metadata.json', encoding='utf-8') as f:
+            self.meta = json.load(f)
+        self.step_count = 0
+        self.done = False
+        self.cumulative_reward = 0.0
+        self.expert_reports: Dict[str, ExpertReport] = {}
+        self.current_brief: Brief | None = None
+        self.history: list[str] = []
+        self.last_reward = 0.0
+        self.last_terminal = None
+        self.last_data_quality = 0.0
+        self.last_issues = ['No experts consulted yet.']
+        self._consult_counts: Dict[str, int] = {}
+        self._last_action_key: str | None = None
+        return self._observe(initial=True)
+
+    def state(self) -> CoSState:
+        return CoSState(
+            episode_id=self.episode_id,
+            task_name=self.task_name,
+            step_count=self.step_count,
+            done=self.done,
+            rag_enabled=self.use_rag,
+            consulted_experts=list(self.expert_reports.keys()),
+            expert_reports=self.expert_reports,
+            current_brief=self.current_brief,
+            cumulative_reward=self.cumulative_reward,
+        )
+
+    def _observe(self, initial: bool = False) -> CoSObservation:
+        return CoSObservation(
+            done=self.done,
+            reward=0.0 if initial else self.last_reward,
+            instruction=self.meta['instruction'],
+            history=list(self.history),
+            issues=list(self.last_issues),
+            data_quality_score=self.last_data_quality,
+            task_name=self.task_name,
+            task_difficulty=self.meta['difficulty'],
+            max_steps=int(self.meta.get('max_steps', 12)),
+            step_count=self.step_count,
+            rag_enabled=self.use_rag,
+            consulted_experts=list(self.expert_reports.keys()),
+            expert_reports=self.expert_reports,
+            current_brief=self.current_brief,
+            reward_breakdown=RewardBreakdown(
+                immediate=self.last_reward,
+                cumulative=self.cumulative_reward,
+                terminal_grader=self.last_terminal,
+            ),
+            terminal_grader_score=self.last_terminal,
+        )
+
+    def _compose_brief(self) -> Brief:
+        metrics: Dict[str, Any] = {}
+        recommendations: list[str] = []
+        summary_parts: list[str] = []
+        for expert_id in ('analyst', 'finance'):
+            report = self.expert_reports.get(expert_id)
+            if report:
+                metrics.update(report.metrics)
+                summary_parts.append(report.summary)
+        if 'strategy' in self.expert_reports:
+            recommendations = list(self.expert_reports['strategy'].bullet_points)
+            summary_parts.append(self.expert_reports['strategy'].summary)
+        hr_memo = self.expert_reports['hr'].memo if 'hr' in self.expert_reports and self.expert_reports['hr'].memo else ''
+        summary = ' '.join(summary_parts) if summary_parts else 'No brief drafted yet.'
+        self.current_brief = Brief(
+            summary=summary,
+            metrics=metrics,
+            recommendations=recommendations,
+            hr_memo=hr_memo,
+            consulted_experts=list(self.expert_reports.keys()),
+        )
+        return self.current_brief
+
+    def _run_expert(self, expert_id: str, focused: bool = False) -> ExpertReport:
+        question = self.meta['instruction']
+        if expert_id == 'analyst':
+            report = self.analyst.run(
+                self.task_name, question, self.raw_df, focused=focused, use_rag=self.use_rag
+            )
+            self.last_data_quality = float(report.metrics.get('data_quality_score', 0.0))
+            self.last_issues = report.issues or ['analyst:no material issues']
+            return report
+        if expert_id == 'finance':
+            analyst = self.expert_reports.get('analyst') or self._run_expert('analyst')
+            return self.finance.run(
+                self.task_name,
+                question,
+                self.raw_df,
+                analyst.metrics,
+                self.meta,
+                focused=focused,
+                use_rag=self.use_rag,
+            )
+        if expert_id == 'strategy':
+            analyst = self.expert_reports.get('analyst') or self._run_expert('analyst')
+            finance = self.expert_reports.get('finance') or self._run_expert('finance')
+            return self.strategy.run(
+                self.task_name, self.meta, analyst, finance, focused=focused, use_rag=self.use_rag
+            )
+        if expert_id == 'hr':
+            analyst = self.expert_reports.get('analyst') or self._run_expert('analyst')
+            finance = self.expert_reports.get('finance') or self._run_expert('finance')
+            strategy = self.expert_reports.get('strategy')
+            return self.hr.run(
+                self.task_name, self.meta, analyst, finance, strategy, focused=focused, use_rag=self.use_rag
+            )
+        raise ValueError(f'Unknown expert {expert_id!r}')
+
+    def _ensure_required_experts(self) -> list[str]:
+        """Run any task-required experts that the policy never consulted.
+
+        This guarantees the strategist (and any other required role) always
+        contributes to the brief, so the UI / grader always has their report.
+        Returns the list of expert ids that were auto-filled.
+        """
+        if not self.auto_fill_required:
+            return []
+        auto: list[str] = []
+        for expert_id in required_experts_for_task(self.task_name):
+            if expert_id in self.expert_reports:
+                continue
+            try:
+                self.expert_reports[expert_id] = self._run_expert(expert_id)
+                auto.append(expert_id)
+            except Exception:
+                continue
+        return auto
+
+    def step(self, action: CoSAction) -> CoSObservation:
+        if self.done:
+            return self._observe()
+        self.step_count += 1
+        immediate = -0.02
+        details = action.model_dump(exclude_none=True)
+        action_key = json.dumps(details, sort_keys=True)
+        self.history.append(action_key)
+        strict = self.shaping == 'strict'
+        if strict and self._last_action_key is not None and action_key == self._last_action_key:
+            immediate -= 0.05
+        self._last_action_key = action_key
+        required = list(self.meta.get('required_experts', []))
+        if action.action_type in {'consult', 'ask'}:
+            if not action.expert_id:
+                immediate -= 0.03
+                self.last_issues = ['action_missing_expert']
+            else:
+                prior = action.expert_id in self.expert_reports
+                report = self._run_expert(action.expert_id, focused=action.action_type == 'ask')
+                self.expert_reports[action.expert_id] = report
+                immediate += 0.10 if not prior and action.expert_id in required else 0.02
+                if prior:
+                    immediate -= 0.05
+                if strict:
+                    self._consult_counts[action.expert_id] = self._consult_counts.get(action.expert_id, 0) + 1
+                    if self._consult_counts[action.expert_id] > 2:
+                        immediate -= 0.10
+                self.last_issues = report.issues or [f'{action.expert_id}:ok']
+        elif action.action_type == 'summarize':
+            brief_already_exists = self.current_brief is not None
+            missing_required = [e for e in required if e not in self.expert_reports]
+            self._ensure_required_experts()
+            self._compose_brief()
+            immediate += 0.04 if len(self.expert_reports) >= 2 else -0.02
+            if strict and missing_required:
+                immediate -= 0.05 * len(missing_required)
+            if strict and brief_already_exists:
+                immediate -= 0.08
+            self.last_issues = ['brief_composed']
+        elif action.action_type == 'submit':
+            auto_filled = self._ensure_required_experts()
+            if self.current_brief is None or auto_filled:
+                self._compose_brief()
+            self.done = True
+            self.last_terminal = grade_episode(
+                self.gt_metrics, self.meta, self.current_brief, self.expert_reports, use_rag=self.use_rag
+            )
+            immediate += self.last_terminal
+            if strict and not auto_filled:
+                max_steps = int(self.meta.get('max_steps', 12))
+                steps_saved = max(0, max_steps - self.step_count)
+                if steps_saved > 0 and all(e in self.expert_reports for e in required):
+                    immediate += min(0.10, 0.01 * steps_saved)
+            self.last_issues = ['submitted'] + (
+                [f'auto_consulted:{",".join(auto_filled)}'] if auto_filled else []
+            )
+        else:
+            self.last_issues = ['noop']
+            immediate -= 0.01
+
+        if not self.done and self.step_count >= int(self.meta.get('max_steps', 12)):
+            auto_filled = self._ensure_required_experts()
+            if self.current_brief is None or auto_filled:
+                self._compose_brief()
+            self.done = True
+            self.last_terminal = grade_episode(
+                self.gt_metrics, self.meta, self.current_brief, self.expert_reports, use_rag=self.use_rag
+            )
+            immediate += self.last_terminal
+            self.last_issues = ['forced_termination:max_steps'] + (
+                [f'auto_consulted:{",".join(auto_filled)}'] if auto_filled else []
+            )
+
+        self.last_reward = round(immediate, 4)
+        self.cumulative_reward = round(self.cumulative_reward + self.last_reward, 4)
+        return self._observe()
+
+
+def oracle_action_for_observation(obs: CoSObservation) -> CoSAction:
+    for expert in required_experts_for_task(obs.task_name):
+        if expert not in obs.consulted_experts:
+            return CoSAction(action_type='consult', expert_id=expert)
+    if obs.current_brief is None:
+        return CoSAction(action_type='summarize')
+    return CoSAction(action_type='submit')

ceo_brief_env/experts/__init__.py

@@ -0,0 +1,6 @@
+from .data_analyst import DataAnalystExpert
+from .finance import FinanceExpert
+from .hr import HRExpert
+from .strategy import StrategyExpert
+
+__all__ = ["DataAnalystExpert", "FinanceExpert", "HRExpert", "StrategyExpert"]

ceo_brief_env/experts/data_analyst.py

@@ -0,0 +1,84 @@
+from __future__ import annotations
+
+import pandas as pd
+
+from subenvs.autodatalab.analytics import (
+    clean_orders,
+    compute_kpis,
+    compute_revenue_share,
+    data_quality_score,
+    derive_revenue,
+    validate_schema,
+)
+
+from ..models import ExpertReport
+
+_ANALYST_QUERY = (
+    "data quality cleaning duplicates imputation category median "
+    "KPIs total revenue top category data_quality_score"
+)
+
+
+class DataAnalystExpert:
+    expert_id = "analyst"
+
+    def run(
+        self,
+        task_name: str,
+        question: str,
+        raw_df: pd.DataFrame,
+        focused: bool = False,
+        use_rag: bool = False,
+    ) -> ExpertReport:
+        cleaned = clean_orders(raw_df)
+        enriched = derive_revenue(cleaned.df)
+        kpis = compute_kpis(cleaned.df)
+        share = compute_revenue_share(cleaned.df)
+        schema = validate_schema(enriched)
+        top_row = share.iloc[0]
+        summary = (
+            f"Cleaned {len(raw_df)} raw rows into {len(cleaned.df)} trusted rows. "
+            f"Top category is {top_row['Category']} and total revenue is {kpis['total_revenue']:.2f}."
+        )
+        bullets = [
+            f"Removed {cleaned.duplicates_removed} duplicate rows and imputed {cleaned.imputed_prices} missing prices.",
+            f"Data quality score is {data_quality_score(cleaned.df):.2f} with {schema['invalid_date_rows']} invalid date rows.",
+            f"{top_row['Category']} leads revenue at {float(top_row['Revenue']):.2f}.",
+        ]
+        metrics = {
+            "duplicates_removed": cleaned.duplicates_removed,
+            "imputed_prices": cleaned.imputed_prices,
+            "data_quality_score": data_quality_score(cleaned.df),
+            "total_revenue": kpis["total_revenue"],
+            "avg_order_value": kpis["avg_order_value"],
+            "top_category": str(top_row["Category"]),
+            "top_category_revenue": round(float(top_row["Revenue"]), 2),
+        }
+        issues = [f"risk:{name}" for name in schema["risk_flags"]]
+        citations = share["Category"].astype(str).head(3).tolist()
+
+        memory_citations: list[str] = []
+        memory_snippets: list[str] = []
+        if use_rag:
+            from memory import get_retriever
+
+            hits = get_retriever().query(_ANALYST_QUERY, k=2)
+            memory_citations = [h.as_citation() for h in hits]
+            memory_snippets = [h.snippet for h in hits]
+            if hits:
+                summary = summary + f" Grounded in SOP {hits[0].source.split('#')[0]}."
+                bullets.append(
+                    f"Grounded against SOP: {hits[0].source.split('#')[0]} (score {hits[0].score:.2f})."
+                )
+
+        return ExpertReport(
+            expert_id="analyst",
+            title="Data Analyst Report",
+            summary=summary,
+            metrics=metrics,
+            bullet_points=bullets,
+            issues=issues,
+            citations=citations,
+            memory_citations=memory_citations,
+            memory_snippets=memory_snippets,
+        )

ceo_brief_env/experts/finance.py

@@ -0,0 +1,71 @@
+from __future__ import annotations
+
+import pandas as pd
+
+from subenvs.autodatalab.analytics import clean_orders, monthly_revenue
+
+from ..models import ExpertReport
+from .finance_tools import break_even, compute_variance, project_next_quarter
+
+_FINANCE_QUERY = (
+    "finance forecasting next quarter projection confidence band "
+    "variance plan_value favorable unfavorable break-even unit_margin fixed_cost"
+)
+
+
+class FinanceExpert:
+    expert_id = "finance"
+
+    def run(
+        self,
+        task_name: str,
+        question: str,
+        raw_df: pd.DataFrame,
+        analyst_metrics: dict,
+        task_meta: dict,
+        focused: bool = False,
+        use_rag: bool = False,
+    ) -> ExpertReport:
+        cleaned = clean_orders(raw_df).df
+        monthly = monthly_revenue(cleaned)
+        projection = project_next_quarter(monthly)
+        variance = compute_variance(float(analyst_metrics.get("total_revenue", 0.0)), float(task_meta.get("plan_value", 1.0)))
+        be = break_even(float(task_meta.get("fixed_cost", 10000.0)), float(task_meta.get("unit_margin", 500.0)))
+        metrics = {
+            **projection,
+            **variance,
+            **be,
+        }
+        bullets = [
+            f"Next-quarter revenue projection is {metrics['projection_next_quarter']:.2f} with +/- {metrics['confidence_band']:.2f} band.",
+            f"Variance versus plan is {metrics['variance_abs']:.2f} ({metrics['variance_pct']:.2f}%).",
+            f"Break-even sits at {metrics['break_even_units']:.2f} units.",
+        ]
+        memory_citations: list[str] = []
+        memory_snippets: list[str] = []
+        summary = (
+            f"Finance projects next quarter at {metrics['projection_next_quarter']:.2f} and "
+            f"marks performance as {metrics['variance_flag']} plan."
+        )
+        if use_rag:
+            from memory import get_retriever
+
+            hits = get_retriever().query(_FINANCE_QUERY, k=2)
+            memory_citations = [h.as_citation() for h in hits]
+            memory_snippets = [h.snippet for h in hits]
+            if hits:
+                summary = summary + f" Following SOP {hits[0].source.split('#')[0]}."
+                bullets.append(
+                    f"SOP alignment: {hits[0].source.split('#')[0]} (score {hits[0].score:.2f})."
+                )
+
+        return ExpertReport(
+            expert_id="finance",
+            title="Finance Forecast",
+            summary=summary,
+            metrics=metrics,
+            bullet_points=bullets,
+            citations=list(monthly['Month'].astype(str).tail(3)),
+            memory_citations=memory_citations,
+            memory_snippets=memory_snippets,
+        )

ceo_brief_env/experts/finance_tools.py

@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+
+def project_next_quarter(revenue_by_month: pd.DataFrame) -> dict[str, float]:
+    if revenue_by_month.empty:
+        return {"projection_next_quarter": 0.0, "confidence_band": 0.0}
+    y = revenue_by_month["Revenue"].astype(float).to_numpy()
+    x = np.arange(1, len(y) + 1, dtype=float)
+    if len(y) == 1:
+        next_values = np.array([y[-1], y[-1], y[-1]])
+        residual = 0.0
+    else:
+        slope, intercept = np.polyfit(x, y, 1)
+        future_x = np.arange(len(y) + 1, len(y) + 4, dtype=float)
+        next_values = slope * future_x + intercept
+        predicted = slope * x + intercept
+        residual = float(np.std(y - predicted))
+    return {
+        "projection_next_quarter": round(float(next_values.sum()), 2),
+        "confidence_band": round(residual * 3, 2),
+    }
+
+
+def compute_variance(actual: float, plan: float) -> dict[str, float | str]:
+    variance_abs = round(float(actual) - float(plan), 2)
+    variance_pct = round((variance_abs / float(plan)) * 100, 2) if plan else 0.0
+    flag = "ahead" if variance_abs >= 0 else "behind"
+    return {
+        "variance_abs": variance_abs,
+        "variance_pct": variance_pct,
+        "variance_flag": flag,
+    }
+
+
+def break_even(fixed_cost: float, unit_margin: float) -> dict[str, float]:
+    if unit_margin <= 0:
+        return {"break_even_units": 0.0, "break_even_revenue": 0.0}
+    units = float(fixed_cost) / float(unit_margin)
+    return {
+        "break_even_units": round(units, 2),
+        "break_even_revenue": round(units * float(unit_margin), 2),
+    }

ceo_brief_env/experts/hr.py

@@ -0,0 +1,114 @@
+from __future__ import annotations
+
+from ..models import ExpertReport
+from subenvs.email.hr_tools import build_hr_memo, score_memo
+from subenvs.email.graders import grade_response
+
+# Maps CoS brief tasks to the Round 1 email-env task IDs so the right
+# task-specific rubric fires inside grade_response.
+_BRIEF_TO_EMAIL_TASK = {
+    "easy_brief": "task_1",
+    "medium_brief": "task_2",
+    "hard_brief": "task_3",
+    "expert_brief": "task_3",
+}
+
+_HR_QUERY = (
+    "memo style guide audience tone bullet highlights "
+    "call to action professional direct exemplar"
+)
+
+
+def _clamp(score: float) -> float:
+    """Match ceo_brief_env.graders._clamp so every emitted score is in (0.001, 0.999)."""
+    return max(0.001, min(0.999, round(float(score), 4)))
+
+
+class HRExpert:
+    expert_id = "hr"
+
+    def run(
+        self,
+        task_name: str,
+        task_meta: dict,
+        analyst_report: ExpertReport | None,
+        finance_report: ExpertReport | None,
+        strategy_report: ExpertReport | None = None,
+        focused: bool = False,
+        use_rag: bool = False,
+    ) -> ExpertReport:
+        # --- 1. Build the memo from upstream reports -----------------------
+        highlights: list[str] = []
+        if analyst_report:
+            highlights.extend(analyst_report.bullet_points[:2])
+        if finance_report:
+            highlights.extend(finance_report.bullet_points[:2])
+        if strategy_report:
+            highlights.extend(strategy_report.bullet_points[:2])
+
+        memory_citations: list[str] = []
+        memory_snippets: list[str] = []
+        if use_rag:
+            from memory import get_retriever
+
+            hits = get_retriever().query(_HR_QUERY, k=2)
+            memory_citations = [h.as_citation() for h in hits]
+            memory_snippets = [h.snippet for h in hits]
+            if hits:
+                highlights.insert(
+                    0,
+                    f"Memo follows {hits[0].source.split('#')[0]} (style SOP retrieved from company memory).",
+                )
+
+        audience = str(task_meta.get("memo_audience", "team"))
+        title = str(task_meta.get("title", task_name))
+        memo = build_hr_memo(audience, title, highlights)
+
+        # --- 2. Three-part scoring ----------------------------------------
+        structure_score = score_memo(memo, task_meta.get("hr_required_terms", []))
+
+        email_task_id = _BRIEF_TO_EMAIL_TASK.get(task_name, "task_1")
+        tone_score = grade_response(
+            email=task_meta.get("instruction", ""),
+            response=memo,
+            task_id=email_task_id,
+        )
+
+        audience_hit = 1.0 if audience.lower() in memo.lower() else 0.0
+
+        blended = 0.45 * structure_score + 0.45 * tone_score + 0.10 * audience_hit
+        score = _clamp(blended)
+
+        # --- 3. Return a typed ExpertReport -------------------------------
+        issues: list[str] = []
+        if structure_score < 0.4:
+            issues.append("hr:weak_structure_or_missing_required_terms")
+        if tone_score < 0.4:
+            issues.append("hr:weak_professional_tone")
+        if audience_hit == 0.0:
+            issues.append(f"hr:audience_not_referenced:{audience}")
+
+        return ExpertReport(
+            expert_id="hr",
+            title="HR / Communications Memo",
+            summary=(
+                "Drafted the internal memo and scored it on structure, "
+                "professional tone, and audience relevance."
+            ),
+            metrics={
+                "memo_structure_score": _clamp(structure_score),
+                "memo_tone_score": _clamp(tone_score),
+                "audience_reference": audience_hit,
+                "memo_score": score,
+            },
+            bullet_points=[
+                f"Memo addressed to {audience}.",
+                f"Structure score {structure_score:.3f}, tone score {tone_score:.3f}.",
+                "Blended score uses 45% structure, 45% tone, 10% audience bonus.",
+            ],
+            issues=issues,
+            memo=memo,
+            score=score,
+            memory_citations=memory_citations,
+            memory_snippets=memory_snippets,
+        )

ceo_brief_env/experts/strategy.py

@@ -0,0 +1,274 @@
+from __future__ import annotations
+
+from ..models import ExpertReport
+
+_STRATEGY_QUERY = (
+    "strategy playbook portfolio buy sell hold trim NVDA AAPL JPM "
+    "projection variance break-even margin diversification"
+)
+
+# External watchlist for “portfolio” recommendations (narrative layer on top of internal P&L).
+WATCHLIST: tuple[tuple[str, str], ...] = (
+    ("NVDA", "NVIDIA"),
+    ("AAPL", "Apple"),
+    ("JPM", "JPMorgan Chase"),
+)
+
+
+def _stance(
+    key: str,
+    variance_flag: str,
+    projection: float,
+    plan_value: float,
+    top_category_lower: str,
+) -> str:
+    """
+    Return one of: buy more, add, hold, reduce (sell/trim) — deterministic, CoS-graded via brief text.
+    Tech names tilt on Electronics-heavy internal mix; JPM is treated as a defensive/financial anchor.
+    """
+    stress = variance_flag == "behind" or (plan_value > 0 and projection < 0.95 * plan_value)
+    electronic_tilt = "electronic" in top_category_lower or "phone" in top_category_lower or "tablet" in top_category_lower
+
+    if key == "NVDA":
+        if stress:
+            return "reduce (trim high-beta tech beta — lock in Q4 if variance stays behind plan)"
+        if electronic_tilt:
+            return "buy more (reinforce tech / AI exposure; aligns with strong Electronics sell-through vs plan)"
+        return "add (modest size-up while projection holds vs plan_value)"
+
+    if key == "AAPL":
+        if stress:
+            return "hold (keep core quality; avoid size-up while variance is behind plan)"
+        if electronic_tilt:
+            return "add (broaden megacap tech anchor alongside internal Electronics strength)"
+        return "buy more (diversifier vs pure growth; use break-even and cash discipline)"
+
+    if key == "JPM":
+        if stress:
+            return "add (increase defensives / quality financials; reduce portfolio beta vs NVDA + AAPL)"
+        return "hold (bank anchor; rotate only on clearer variance reversion to plan)"
+    return "hold"
+
+
+def _static_horizons(
+    key: str,
+    variance_flag: str,
+    projection: float,
+    plan_value: float,
+    top_category_lower: str,
+) -> tuple[str, str, str, str]:
+    """
+    Fixed policy text: *present* = what the desk would do in the **current** reporting window;
+    *future* = staged intent for the **next 1-2 quarters** (not a forecast of prices).
+    Also returns present/future one-word stance tokens for metrics (buy|sell|hold|trim|none).
+    """
+    stress = variance_flag == "behind" or (plan_value > 0 and projection < 0.95 * plan_value)
+    electronic = (
+        "electronic" in top_category_lower
+        or "phone" in top_category_lower
+        or "tablet" in top_category_lower
+    )
+    if key == "NVDA":
+        if stress:
+            return (
+                "Present: **trim / do not add** (reduce high-beta now while internal variance is behind plan).",
+                "Future: re-open **buy** scales only if next-quarter **projection** clears plan and **variance** flips ahead.",
+                "trim",
+                "buy",
+            )
+        if electronic:
+            return (
+                "Present: **hold** to **buy small** only within risk limits; no forced trade today.",
+                "Future: **add / buy** on a staged plan over the next 1-2 quarters while Electronics strength persists.",
+                "hold",
+                "buy",
+            )
+        return (
+            "Present: **hold**; wait for a cleaner internal read vs plan.",
+            "Future: **add** modestly next quarter if the **projection** path holds.",
+            "hold",
+            "add",
+        )
+    if key == "AAPL":
+        if stress:
+            return (
+                "Present: **hold**; **no new buy** until variance to plan improves.",
+                "Future: **buy** or **add** in the *next* quarter if forecast confidence widens and execution stabilizes.",
+                "hold",
+                "buy",
+            )
+        if electronic:
+            return (
+                "Present: **hold** to **add** a sliver of core quality if account policy allows (optional today).",
+                "Future: **buy / add** as a megacap anchor over the next two quarters (static ladder, not market timing).",
+                "hold",
+                "add",
+            )
+        return (
+            "Present: **hold**; keep dry powder for updates vs plan_value.",
+            "Future: **buy** more defensively in the *next* quarter (quality tilt).",
+            "hold",
+            "buy",
+        )
+    if key == "JPM":
+        if stress:
+            return (
+                "Present: **buy / add** defensives to lower portfolio beta (execute now, static sleeve shift).",
+                "Future: **hold** the defensive sleeve; re-check after **break-even** and plan variance improve.",
+                "buy",
+                "hold",
+            )
+        return (
+            "Present: **hold** the bank anchor; no need to day-trade the sleeve.",
+            "Future: only **sell** down if the operating plan is consistently ahead and you rotate into growth; else **hold**.",
+            "hold",
+            "sell" if (not stress and plan_value and projection > plan_value * 1.02) else "hold",
+        )
+    return (
+        "Present: **hold**.",
+        "Future: **hold**; revisit vs plan next cycle.",
+        "hold",
+        "hold",
+    )
+
+
+def _one_line(
+    sym: str,
+    display: str,
+    stance: str,
+    num_token: str,
+    cat_phrase: str,
+    present: str,
+    future: str,
+) -> str:
+    # Keep one finance keyword for rubric: projection / variance / break-even
+    return (
+        f"{sym} ({display}): {stance} — tie to internal numbers including {num_token} {cat_phrase} "
+        f"and the latest projection/variance read vs plan. {present} {future}"
+    )
+
+
+class StrategyExpert:
+    expert_id = "strategy"
+
+    def run(
+        self,
+        task_name: str,
+        task_meta: dict,
+        analyst_report: ExpertReport,
+        finance_report: ExpertReport,
+        focused: bool = False,
+        use_rag: bool = False,
+    ) -> ExpertReport:
+        top_category = str(analyst_report.metrics.get("top_category", "the best category"))
+        top_cat_lower = top_category.lower()
+        fin = finance_report.metrics
+        projection = float(fin.get("projection_next_quarter", 0.0) or 0.0)
+        var_pct = float(fin.get("variance_pct", 0.0) or 0.0)
+        be_units = fin.get("break_even_units", 0.0)
+        vflag = str(fin.get("variance_flag", "behind"))
+        plan_value = float(task_meta.get("plan_value", 0.0) or 0.0)
+
+        # These keys exist on analyst+finance and flow into `brief.metrics` (used by strategy grader for evidence #).
+        tr = analyst_report.metrics.get("total_revenue", 0.0)
+        tr_s = str(tr)
+        proj_s = str(fin.get("projection_next_quarter", projection))
+        vps = str(fin.get("variance_pct", var_pct))
+        be_s = str(be_units)
+
+        cats = [str(c) for c in (analyst_report.citations or []) if c]
+        if top_category and top_category not in cats:
+            cats.insert(0, top_category)
+        cat_a = cats[0] if cats else top_category
+        cat_b = cats[1] if len(cats) > 1 else (cats[0] if cats else "operations")
+
+        st_nvda = _stance("NVDA", vflag, projection, plan_value, top_cat_lower)
+        st_aapl = _stance("AAPL", vflag, projection, plan_value, top_cat_lower)
+        st_jpm = _stance("JPM", vflag, projection, plan_value, top_cat_lower)
+        pnv, fnv, nv_pr, nv_fu = _static_horizons("NVDA", vflag, projection, plan_value, top_cat_lower)
+        paa, faa, aa_pr, aa_fu = _static_horizons("AAPL", vflag, projection, plan_value, top_cat_lower)
+        pjp, fjp, jp_pr, jp_fu = _static_horizons("JPM", vflag, projection, plan_value, top_cat_lower)
+
+        # Each line embeds a distinct evidence token that appears in `brief.metrics` after analyst+finance merge.
+        line_nvda = _one_line(
+            "NVDA", "NVIDIA", st_nvda, tr_s, f"while top internal category {cat_a!r} is in focus", pnv, fnv
+        )
+        line_aapl = _one_line("AAPL", "Apple", st_aapl, proj_s, f"cross-check against {cat_b!r} demand mix", paa, faa)
+        line_jpm = _one_line("JPM", "JPMorgan", st_jpm, vps, f"and break-even path ~{be_s} units in our operating model", pjp, fjp)
+
+        # Third line: explicit break-even token for grader’s projection|variance|break-even check.
+        if "break-even" not in line_jpm.lower() and be_s != "0":
+            line_jpm = line_jpm + f" (break-even reference {be_s})."
+
+        bullets = [line_nvda, line_aapl, line_jpm]
+
+        memory_citations: list[str] = []
+        memory_snippets: list[str] = []
+        summary = (
+            f"Portfolio call on {', '.join(s[0] for s in WATCHLIST)}: map internal "
+            f"{top_category!r} to actions vs plan, with static **Present** (this cycle) "
+            f"and **Future** (next 1-2Q) buy/sell/hold/trim guidance per line."
+        )
+        if use_rag:
+            from memory import get_retriever
+
+            hits = get_retriever().query(_STRATEGY_QUERY, k=2)
+            memory_citations = [h.as_citation() for h in hits]
+            memory_snippets = [h.snippet for h in hits]
+            if hits:
+                summary = summary + f" Structure follows {hits[0].source.split('#')[0]}."
+
+            # Advanced RAG: lightweight Stooq daily CSV "scrape" (HTTP + fixture fallback).
+            from ..stooq_scrape import DEFAULT_WATCHLIST, scrape_watchlist
+
+            stq = scrape_watchlist(DEFAULT_WATCHLIST)
+            for _sym, cite, snip in stq:
+                memory_citations.append(cite)
+                memory_snippets.append(snip)
+            ext = " | ".join(s for _, __, s in stq)
+            summary = summary + f" External tape (Stooq, scraped in RAG mode): {ext}"
+        else:
+            # Non-RAG: no HTTP scrape — still attach multi-hundred-row local “tape” CSVs for grounding text.
+            from ..stooq_scrape import DEFAULT_WATCHLIST, scrape_watchlist_from_long_csv
+
+            stq = scrape_watchlist_from_long_csv(DEFAULT_WATCHLIST, last_n=5)
+            n0 = stq[0][3] if stq else 0
+            for _sym, cite, snip, _n in stq:
+                memory_citations.append(cite)
+                memory_snippets.append(snip)
+            ext = " | ".join(s[2] for s in stq)
+            summary = (
+                summary
+                + f" External tape (bundled long CSV, ~{n0} trading days per symbol, no network): {ext}"
+            )
+
+        def _action_token(stance: str) -> str:
+            head = stance.split()[0].lower() if stance else "hold"
+            if head in ("reduce", "hold", "add"):
+                return head
+            if head == "buy":
+                return "buy_more"
+            return head
+
+        return ExpertReport(
+            expert_id="strategy",
+            title="Strategy — public equities (watchlist)",
+            summary=summary,
+            metrics={
+                "recommendation_count": len(bullets),
+                "nvda": _action_token(st_nvda),
+                "aapl": _action_token(st_aapl),
+                "jpm": _action_token(st_jpm),
+                "nvda_present": nv_pr,
+                "nvda_future": nv_fu,
+                "aapl_present": aa_pr,
+                "aapl_future": aa_fu,
+                "jpm_present": jp_pr,
+                "jpm_future": jp_fu,
+                "watchlist": "NVDA,AAPL,JPM",
+            },
+            bullet_points=bullets,
+            citations=list(analyst_report.citations[:3]) or [top_category, cat_a, cat_b],
+            memory_citations=memory_citations,
+            memory_snippets=memory_snippets,
+        )

ceo_brief_env/fixtures/stooq/aapl_us.csv

@@ -0,0 +1,4 @@
+Date,Open,High,Low,Close,Volume
+2024-10-28,220.0,224.0,218.0,222.1,40000000
+2024-10-29,222.0,226.0,221.0,224.5,38000000
+2024-10-30,224.0,228.0,223.0,225.0,36000000

ceo_brief_env/fixtures/stooq/jpm_us.csv

@@ -0,0 +1,4 @@
+Date,Open,High,Low,Close,Volume
+2024-10-28,210.0,213.0,208.0,211.0,8000000
+2024-10-29,211.0,215.0,210.0,213.4,7500000
+2024-10-30,213.0,216.0,212.0,214.2,7200000

ceo_brief_env/fixtures/stooq/nvda_us.csv

@@ -0,0 +1,4 @@
+Date,Open,High,Low,Close,Volume
+2024-10-28,130.0,132.5,128.0,131.2,45000000
+2024-10-29,131.0,135.0,130.0,133.8,48000000
+2024-10-30,134.0,136.0,132.0,135.1,42000000

ceo_brief_env/fixtures/stooq_long/aapl_us.csv

@@ -0,0 +1,281 @@
+Date,Open,High,Low,Close,Volume
+2022-06-01,150.0000,153.0614,149.0632,151.6332,40551601
+2022-06-02,151.6332,153.4470,150.4806,151.0050,23116470
+2022-06-03,151.0050,151.4769,148.7328,149.3193,28859720
+2022-06-06,149.3193,149.9026,145.8916,146.4382,43323520
+2022-06-07,146.4382,146.5626,143.9913,144.3422,42037244
+2022-06-08,144.3422,145.6126,142.3213,143.7062,35893004
+2022-06-09,143.7062,144.9616,139.5929,140.8703,35011039
+2022-06-10,140.8703,142.7739,139.3237,142.2634,56022729
+2022-06-13,142.2634,146.3343,141.8659,144.6782,42172092
+2022-06-14,144.6782,148.0800,144.0006,146.6694,22650088
+2022-06-15,146.6694,148.0570,145.4279,147.8713,16333835
+2022-06-16,147.8713,148.7226,147.5242,147.7365,39049284
+2022-06-17,147.7365,148.6139,146.1476,147.9587,57571368
+2022-06-20,147.9587,151.6875,147.9127,151.2431,34077358
+2022-06-21,151.2431,152.5230,147.9049,148.3731,22162397
+2022-06-22,148.3731,149.0914,145.5985,146.8948,38129914
+2022-06-23,146.8948,148.0888,145.4880,145.5672,26041890
+2022-06-24,145.5672,148.5606,145.3902,147.2319,19190460
+2022-06-27,147.2319,148.6599,147.2034,147.4980,36965252
+2022-06-28,147.4980,147.6486,145.6590,145.9587,29720348
+2022-06-29,145.9587,146.1553,144.8364,145.9227,58180230
+2022-06-30,145.9227,147.6848,144.7871,147.5644,46425650
+2022-07-01,147.5644,148.2457,142.8168,143.9439,61606791
+2022-07-04,143.9439,148.4399,142.9268,147.4323,62006500
+2022-07-05,147.4323,148.3331,145.2850,145.3923,53180621
+2022-07-06,145.3923,147.1653,145.0485,146.7944,41138742
+2022-07-07,146.7944,149.2158,146.2247,148.9729,40612341
+2022-07-08,148.9729,150.5540,146.0292,147.3108,61859297
+2022-07-11,147.3108,149.8476,146.8062,148.6325,40226129
+2022-07-12,148.6325,149.4859,148.1869,148.2981,56784517
+2022-07-13,148.2981,149.1137,144.1771,145.8044,40621804
+2022-07-14,145.8044,149.3362,144.9875,148.8708,34204067
+2022-07-15,148.8708,149.5308,146.9185,147.7077,56404746
+2022-07-18,147.7077,151.7883,146.5243,151.6162,41384367
+2022-07-19,151.6162,151.6903,150.4895,151.1458,31216002
+2022-07-20,151.1458,152.2222,149.6456,150.3077,30197143
+2022-07-21,150.3077,152.0704,148.0356,148.7349,62697979
+2022-07-22,148.7349,149.9199,145.3486,146.3221,50074664
+2022-07-25,146.3221,148.1913,145.2905,146.6534,18238939
+2022-07-26,146.6534,149.3839,146.2217,148.2928,51071273
+2022-07-27,148.2928,151.6566,148.0287,151.1539,29878387
+2022-07-28,151.1539,152.9081,151.0229,152.0975,37583425
+2022-07-29,152.0975,153.7670,151.4517,152.7065,53693519
+2022-08-01,152.7065,154.8454,152.0717,154.3667,60415621
+2022-08-02,154.3667,155.1956,151.7603,152.3160,43996373
+2022-08-03,152.3160,155.1567,151.9681,154.9271,19061040
+2022-08-04,154.9271,155.8952,151.1959,152.0881,61384083
+2022-08-05,152.0881,153.0471,151.8897,151.9379,26819323
+2022-08-08,151.9379,152.3824,149.2065,150.6309,36015787
+2022-08-09,150.6309,151.5964,147.6761,147.9533,24932317
+2022-08-10,147.9533,149.1763,146.6099,149.0226,37062747
+2022-08-11,149.0226,150.2280,144.1232,145.8055,51448952
+2022-08-12,145.8055,150.3378,145.3305,148.6623,39341972
+2022-08-15,148.6623,151.5042,146.9376,150.7860,36316767
+2022-08-16,150.7860,150.9581,146.4359,147.1901,16908063
+2022-08-17,147.1901,150.0374,145.5422,148.2680,63361666
+2022-08-18,148.2680,148.5887,146.3941,146.5182,60678872
+2022-08-19,146.5182,146.6456,141.9117,143.0571,61491371
+2022-08-22,143.0571,143.6392,142.9967,143.5306,55336787
+2022-08-23,143.5306,145.7196,141.8356,145.1789,29499185
+2022-08-24,145.1789,146.8622,142.5965,142.6918,33415837
+2022-08-25,142.6918,143.7873,139.6609,140.4849,28098960
+2022-08-26,140.4849,141.8285,138.8273,141.6468,53606023
+2022-08-29,141.6468,142.6598,139.2177,139.6728,43591329
+2022-08-30,139.6728,140.0251,137.6988,138.5144,19437891
+2022-08-31,138.5144,143.6863,138.1364,142.3407,44740851
+2022-09-01,142.3407,142.6256,139.4717,140.1161,50368913
+2022-09-02,140.1161,141.6904,136.7132,137.1267,23883653
+2022-09-05,137.1267,138.7331,136.4428,137.2424,41767984
+2022-09-06,137.2424,141.8819,136.6708,140.9899,25084158
+2022-09-07,140.9899,142.4328,139.9880,141.3622,25292440
+2022-09-08,141.3622,143.0000,136.7804,137.9206,56549399
+2022-09-09,137.9206,141.3430,137.6117,140.6247,62249603
+2022-09-12,140.6247,143.9042,139.5618,143.7906,40018841
+2022-09-13,143.7906,144.7720,139.2849,140.3807,37434662
+2022-09-14,140.3807,141.8738,137.6514,138.7696,29374531
+2022-09-15,138.7696,142.5623,138.1234,141.0231,43526395
+2022-09-16,141.0231,141.2473,136.7909,138.2010,61832825
+2022-09-19,138.2010,141.8212,137.0369,141.5683,34835447
+2022-09-20,141.5683,143.0864,139.4827,139.6427,19899141
+2022-09-21,139.6427,139.8875,138.0773,138.7379,46295064
+2022-09-22,138.7379,140.5963,137.9340,139.0592,22950518
+2022-09-23,139.0592,139.6654,137.9584,139.6264,23627897
+2022-09-26,139.6264,140.4765,138.0684,139.7416,22127554
+2022-09-27,139.7416,140.2926,137.4767,138.3396,34547060
+2022-09-28,138.3396,143.6562,136.7728,142.0769,48139682
+2022-09-29,142.0769,145.4728,141.7326,144.5175,23988391
+2022-09-30,144.5175,145.8481,141.2125,142.3560,17645555
+2022-10-03,142.3560,142.4829,140.1131,141.0005,32394415
+2022-10-04,141.0005,141.7078,140.3177,141.0816,37788029
+2022-10-05,141.0816,143.5555,139.7986,143.4373,44089705
+2022-10-06,143.4373,146.1430,142.9412,145.8696,59548311
+2022-10-07,145.8696,150.1263,145.2110,149.6357,32166577
+2022-10-10,149.6357,151.3380,147.6023,148.4964,36419715
+2022-10-11,148.4964,153.0525,146.8255,151.8950,47181775
+2022-10-12,151.8950,152.0384,150.7532,151.6573,41037869
+2022-10-13,151.6573,154.3835,150.9917,152.8299,42379740
+2022-10-14,152.8299,157.4605,151.3190,155.6863,28648380
+2022-10-17,155.6863,159.8747,155.0975,158.5930,42558530
+2022-10-18,158.5930,162.8619,158.1285,161.2986,39446059
+2022-10-19,161.2986,163.2745,159.8202,162.9390,53912955
+2022-10-20,162.9390,165.7712,162.3205,163.8920,29399026
+2022-10-21,163.8920,169.4780,163.8793,168.2351,36799955
+2022-10-24,168.2351,169.9060,164.9097,166.8574,22559495
+2022-10-25,166.8574,170.4136,166.6272,168.6820,37220744
+2022-10-26,168.6820,169.3992,163.6109,164.7627,50015989
+2022-10-27,164.7627,166.0650,161.9548,163.0381,33504603
+2022-10-28,163.0381,164.9036,161.4943,163.3323,36083122
+2022-10-31,163.3323,166.9905,162.4053,166.1144,26884459
+2022-11-01,166.1144,166.4681,164.4281,166.3540,33759598
+2022-11-02,166.3540,168.0635,164.0036,165.3645,37841852
+2022-11-03,165.3645,170.0003,163.6566,168.7904,28620900
+2022-11-04,168.7904,173.5442,167.9788,171.5928,52491586
+2022-11-07,171.5928,177.4088,170.0348,176.1594,21304671
+2022-11-08,176.1594,177.9010,171.2243,172.8210,23405715
+2022-11-09,172.8210,176.0650,171.0556,175.7622,60639633
+2022-11-10,175.7622,176.9523,174.1573,176.3209,57627183
+2022-11-11,176.3209,177.3192,172.0211,173.3712,52578145
+2022-11-14,173.3712,175.0428,167.9039,169.1718,55506239
+2022-11-15,169.1718,170.0118,164.8528,166.0996,59087464
+2022-11-16,166.0996,167.9726,164.5363,165.5902,59956679
+2022-11-17,165.5902,170.1847,165.1824,170.0223,31119361
+2022-11-18,170.0223,172.0425,168.8116,170.5827,32562829
+2022-11-21,170.5827,171.4699,165.9147,167.9132,25292506
+2022-11-22,167.9132,169.0001,164.9456,165.2998,54162806
+2022-11-23,165.2998,165.5045,164.3623,165.3514,49610913
+2022-11-24,165.3514,166.7788,163.4866,166.3270,21204170
+2022-11-25,166.3270,166.8136,160.6813,162.4023,41236986
+2022-11-28,162.4023,164.7246,161.7036,163.9993,37099895
+2022-11-29,163.9993,165.1690,160.9217,161.5277,61627825
+2022-11-30,161.5277,164.5285,159.7175,164.1563,21913335
+2022-12-01,164.1563,169.0072,162.5169,168.1067,41259288
+2022-12-02,168.1067,168.8293,166.5666,167.7023,41456754
+2022-12-05,167.7023,168.6673,163.6443,164.5951,56312267
+2022-12-06,164.5951,164.9769,161.7618,161.9279,34765115
+2022-12-07,161.9279,162.6246,161.2947,161.4628,60351536
+2022-12-08,161.4628,162.7852,157.4684,157.8839,39307532
+2022-12-09,157.8839,158.8417,157.0807,157.2742,26398357
+2022-12-12,157.2742,158.6024,154.0704,154.1138,48288576
+2022-12-13,154.1138,154.3120,150.9453,152.4740,51802247
+2022-12-14,152.4740,152.9385,148.4520,149.7980,30815538
+2022-12-15,149.7980,152.9901,148.8072,151.9651,30082768
+2022-12-16,151.9651,154.0332,151.8021,152.4305,39214558
+2022-12-19,152.4305,152.6015,151.9140,152.5755,25603644
+2022-12-20,152.5755,155.4452,151.4205,154.1834,35741884
+2022-12-21,154.1834,158.4177,152.8941,156.8650,18562426
+2022-12-22,156.8650,158.4173,154.1391,154.5203,44784454
+2022-12-23,154.5203,158.7765,153.6953,158.1909,33874261
+2022-12-26,158.1909,159.5171,157.7942,158.8448,61676357
+2022-12-27,158.8448,160.6424,157.2460,158.0613,22794627
+2022-12-28,158.0613,158.8446,154.0200,155.8668,37029864
+2022-12-29,155.8668,156.5515,152.1522,152.9085,43839108
+2022-12-30,152.9085,153.4613,150.0348,150.2470,31723429
+2023-01-02,150.2470,151.9073,150.0122,150.4768,22020956
+2023-01-03,150.4768,151.1848,146.1678,147.7686,63386705
+2023-01-04,147.7686,149.3873,147.2179,147.7778,33690213
+2023-01-05,147.7778,148.6890,145.5279,145.9164,56872653
+2023-01-06,145.9164,149.5629,144.7856,148.3077,32013903
+2023-01-09,148.3077,149.4503,146.4922,147.1435,44120255
+2023-01-10,147.1435,148.1802,143.4742,144.9628,47038088
+2023-01-11,144.9628,145.4797,143.0683,143.6195,32144223
+2023-01-12,143.6195,146.4374,142.5767,146.2717,26090165
+2023-01-13,146.2717,148.0524,145.5106,146.7912,25916784
+2023-01-16,146.7912,148.3105,145.8002,147.7372,51160231
+2023-01-17,147.7372,148.6910,145.5677,146.5816,40478412
+2023-01-18,146.5816,146.9642,143.7413,145.2270,24041128
+2023-01-19,145.2270,145.2272,141.0273,142.6215,27644199
+2023-01-20,142.6215,144.9270,141.5365,144.3558,43897042
+2023-01-23,144.3558,145.0180,141.4161,142.9636,48578854
+2023-01-24,142.9636,144.4040,139.7050,140.4072,16160311
+2023-01-25,140.4072,142.3254,139.3981,140.9088,29079710
+2023-01-26,140.9088,141.7393,140.7650,141.5966,18896294
+2023-01-27,141.5966,143.0342,140.9088,142.1154,46020088
+2023-01-30,142.1154,143.8130,139.8781,140.1731,35763875
+2023-01-31,140.1731,141.5010,138.5103,141.3207,19454086
+2023-02-01,141.3207,143.3532,140.5007,143.2395,59089857
+2023-02-02,143.2395,144.7575,141.6480,144.3791,24487809
+2023-02-03,144.3791,145.2654,141.0388,142.1378,34185329
+2023-02-06,142.1378,143.5469,141.7807,142.6275,36967477
+2023-02-07,142.6275,144.1770,139.6226,140.7197,22468977
+2023-02-08,140.7197,141.1762,139.3685,140.5382,63891500
+2023-02-09,140.5382,141.6623,138.6237,139.1115,36777340
+2023-02-10,139.1115,140.2410,138.2406,139.4196,42239771
+2023-02-13,139.4196,141.4074,137.9228,140.0077,26644522
+2023-02-14,140.0077,141.8013,138.4348,140.4361,35038013
+2023-02-15,140.4361,141.6201,137.2364,138.7804,24542043
+2023-02-16,138.7804,139.9166,137.2607,137.4724,29076086
+2023-02-17,137.4724,137.9714,135.7770,136.6441,37826103
+2023-02-20,136.6441,137.8303,133.6829,134.5077,52832008
+2023-02-21,134.5077,135.9924,131.0542,131.7410,49383270
+2023-02-22,131.7410,133.7500,130.2145,132.2229,41407365
+2023-02-23,132.2229,132.2315,131.5787,131.7150,24508352
+2023-02-24,131.7150,134.3708,131.5056,133.8885,48515940
+2023-02-27,133.8885,134.1262,130.6624,131.3146,61579399
+2023-02-28,131.3146,132.1462,129.4757,130.3199,16704423
+2023-03-01,130.3199,130.5609,128.6511,129.0302,30446269
+2023-03-02,129.0302,129.7247,127.9010,128.6080,24186687
+2023-03-03,128.6080,128.6449,125.2151,126.4900,41351451
+2023-03-06,126.4900,127.1243,124.7468,124.7673,57510976
+2023-03-07,124.7673,127.9902,124.2014,127.2919,19367493
+2023-03-08,127.2919,129.9797,126.5960,129.7326,59116507
+2023-03-09,129.7326,132.8077,129.2628,132.6947,50921025
+2023-03-10,132.6947,135.1480,131.7717,134.6063,37811377
+2023-03-13,134.6063,139.0367,133.1152,137.5597,50343278
+2023-03-14,137.5597,140.0697,137.3062,138.8865,46647009
+2023-03-15,138.8865,139.7698,138.4016,139.0210,30616019
+2023-03-16,139.0210,140.4934,137.0606,138.7211,37600461
+2023-03-17,138.7211,140.6040,138.1897,139.2467,63904563
+2023-03-20,139.2467,139.8619,138.6762,139.6637,45892599
+2023-03-21,139.6637,143.8524,139.0000,143.2696,37968751
+2023-03-22,143.2696,143.2799,139.3070,140.6730,19906742
+2023-03-23,140.6730,144.0949,139.2280,142.6671,16501972
+2023-03-24,142.6671,142.9714,142.4706,142.9437,29117645
+2023-03-27,142.9437,143.5775,142.6406,142.8772,37542918
+2023-03-28,142.8772,145.0555,141.6908,144.4807,31027008
+2023-03-29,144.4807,148.1383,143.6584,146.9116,32187951
+2023-03-30,146.9116,147.3035,145.5146,145.5431,40577332
+2023-03-31,145.5431,148.2819,144.6893,147.5319,61103835
+2023-04-03,147.5319,148.2751,144.5336,146.0747,52634786
+2023-04-04,146.0747,146.7849,142.7085,143.6763,28735915
+2023-04-05,143.6763,147.4556,143.6094,146.0349,44362706
+2023-04-06,146.0349,146.9070,142.3872,142.4001,46293742
+2023-04-07,142.4001,142.4878,138.3520,139.5344,18746277
+2023-04-10,139.5344,139.6012,137.7011,138.6622,27155763
+2023-04-11,138.6622,139.8434,137.3353,138.2472,50522230
+2023-04-12,138.2472,141.1127,137.7835,140.7977,24828501
+2023-04-13,140.7977,144.2846,140.4378,143.2484,27139664
+2023-04-14,143.2484,144.0432,140.0562,140.4710,35547093
+2023-04-17,140.4710,142.0960,139.5788,140.9138,38581607
+2023-04-18,140.9138,142.1146,139.1607,139.4430,26649410
+2023-04-19,139.4430,141.1119,137.9747,138.2508,55547535
+2023-04-20,138.2508,142.8210,136.6856,141.2508,49948656
+2023-04-21,141.2508,141.5880,137.5066,138.7865,50452607
+2023-04-24,138.7865,139.6707,137.3383,139.1869,32406891
+2023-04-25,139.1869,141.3401,138.4521,139.8045,47992099
+2023-04-26,139.8045,141.0832,137.6149,139.0651,17261688
+2023-04-27,139.0651,140.6903,137.1073,138.7606,41182408
+2023-04-28,138.7606,139.0076,134.6397,136.1457,46879299
+2023-05-01,136.1457,137.7118,133.6466,133.7489,31610983
+2023-05-02,133.7489,135.4959,132.8496,135.4748,41255116
+2023-05-03,135.4748,137.9852,134.6214,137.3000,39211755
+2023-05-04,137.3000,138.4522,134.8214,135.4000,48234431
+2023-05-05,135.4000,137.3661,134.1352,136.1354,29860333
+2023-05-08,136.1354,140.0433,135.0874,138.7268,30166737
+2023-05-09,138.7268,140.3499,134.7882,135.9078,21143153
+2023-05-10,135.9078,140.2104,135.5300,139.1656,30905890
+2023-05-11,139.1656,143.4398,138.6643,142.9997,45233969
+2023-05-12,142.9997,143.3075,141.4119,142.5113,21881872
+2023-05-15,142.5113,143.5008,137.7018,138.9492,58596133
+2023-05-16,138.9492,139.1475,135.7163,136.0713,22354681
+2023-05-17,136.0713,136.6504,132.6879,133.1196,50790544
+2023-05-18,133.1196,134.0172,129.1524,130.7168,21989903
+2023-05-19,130.7168,131.5026,126.9655,128.2068,31027042
+2023-05-22,128.2068,129.3728,126.1629,126.6792,35036773
+2023-05-23,126.6792,127.6917,126.0449,127.0633,43089801
+2023-05-24,127.0633,131.8342,126.4674,130.2904,19247247
+2023-05-25,130.2904,132.3496,129.5528,131.5643,59721419
+2023-05-26,131.5643,132.8025,127.8454,129.0915,57321755
+2023-05-29,129.0915,132.4151,128.4796,131.4945,39711549
+2023-05-30,131.4945,131.5959,129.1600,129.7038,51989991
+2023-05-31,129.7038,132.7931,129.4506,131.9624,18181473
+2023-06-01,131.9624,133.5361,131.6584,131.9087,26932135
+2023-06-02,131.9087,132.9228,128.3772,129.2774,39127430
+2023-06-05,129.2774,130.0173,124.9391,126.2356,42304905
+2023-06-06,126.2356,129.8914,124.7713,128.7644,17322086
+2023-06-07,128.7644,131.8121,127.4767,131.0285,48283550
+2023-06-08,131.0285,135.8626,129.6838,134.2628,26692862
+2023-06-09,134.2628,135.4602,132.7162,134.5304,46241067
+2023-06-12,134.5304,135.9920,133.8481,134.9338,52545464
+2023-06-13,134.9338,136.0484,130.3476,131.5739,43731435
+2023-06-14,131.5739,133.6187,130.2123,132.1579,37540238
+2023-06-15,132.1579,136.2544,131.0292,135.8573,24346444
+2023-06-16,135.8573,138.3284,135.4662,137.6395,36946771
+2023-06-19,137.6395,138.9008,133.6924,134.7936,57883516
+2023-06-20,134.7936,137.9968,134.5681,137.0318,42430153
+2023-06-21,137.0318,138.4179,134.5249,135.3774,49176819
+2023-06-22,135.3774,137.8429,134.3712,136.6948,28571688
+2023-06-23,136.6948,137.2033,133.7525,134.7591,40173582
+2023-06-26,134.7591,135.4331,131.7575,132.8981,47833938
+2023-06-27,132.8981,134.4429,132.1697,132.6250,39713466

ceo_brief_env/fixtures/stooq_long/jpm_us.csv

@@ -0,0 +1,281 @@
+Date,Open,High,Low,Close,Volume
+2022-06-01,120.0000,124.2809,118.7149,122.8821,4002086
+2022-06-02,122.8821,124.2947,122.1005,123.6658,4450908
+2022-06-03,123.6658,124.3256,121.5094,121.8326,7568316
+2022-06-06,121.8326,121.9580,117.9332,118.9467,7243984
+2022-06-07,118.9467,120.2550,118.4343,119.2048,3752523
+2022-06-08,119.2048,122.0288,118.2636,121.1722,9161517
+2022-06-09,121.1722,124.9185,120.0674,124.3770,6739364
+2022-06-10,124.3770,126.0296,123.9078,125.0385,4023066
+2022-06-13,125.0385,126.1316,124.1523,125.0542,7495534
+2022-06-14,125.0542,126.4611,124.7863,126.1855,6304758
+2022-06-15,126.1855,128.7889,126.0124,128.4873,4146600
+2022-06-16,128.4873,128.9121,124.6773,125.5402,11078366
+2022-06-17,125.5402,126.0971,124.1714,124.6028,6125959
+2022-06-20,124.6028,127.4570,124.2164,126.6228,4228421
+2022-06-21,126.6228,129.8096,125.3638,128.7174,5485053
+2022-06-22,128.7174,130.6454,127.8019,129.2207,12276477
+2022-06-23,129.2207,130.7326,127.0504,127.1357,7195743
+2022-06-24,127.1357,128.3708,123.0912,124.1147,12547862
+2022-06-27,124.1147,127.7451,123.3126,126.3070,12686888
+2022-06-28,126.3070,128.6958,124.8301,128.1634,6819191
+2022-06-29,128.1634,128.2801,125.5664,126.2724,5087791
+2022-06-30,126.2724,130.8390,125.9452,129.3636,4193633
+2022-07-01,129.3636,130.0204,128.5196,129.7562,11003290
+2022-07-04,129.7562,130.4129,126.8662,128.2526,7469048
+2022-07-05,128.2526,129.6260,126.1325,127.3763,11793743
+2022-07-06,127.3763,129.4728,126.3170,128.8348,11166268
+2022-07-07,128.8348,130.3526,128.3784,129.9807,7132317
+2022-07-08,129.9807,132.9299,129.1229,132.0472,11134792
+2022-07-11,132.0472,133.0474,128.9882,129.5050,8569810
+2022-07-12,129.5050,132.3944,128.0262,132.0010,3954323
+2022-07-13,132.0010,133.4952,131.6286,132.6329,8584142
+2022-07-14,132.6329,133.3485,131.4827,132.6261,3237843
+2022-07-15,132.6261,133.8763,131.7370,131.9616,11526951
+2022-07-18,131.9616,133.0402,129.8162,130.0865,7513510
+2022-07-19,130.0865,130.2142,128.6584,129.7477,7809679
+2022-07-20,129.7477,132.0911,128.8822,131.2420,8250816
+2022-07-21,131.2420,132.6575,129.8976,130.0324,11615514
+2022-07-22,130.0324,133.5252,128.8200,132.7323,9664320
+2022-07-25,132.7323,134.1384,132.2232,133.6706,6554049
+2022-07-26,133.6706,135.2100,131.9271,132.4696,6920482
+2022-07-27,132.4696,133.7675,131.3462,132.1035,8516022
+2022-07-28,132.1035,132.5443,130.4501,130.5897,5285683
+2022-07-29,130.5897,131.9253,128.3726,129.1514,4259735
+2022-08-01,129.1514,129.3388,128.2131,129.0899,7444486
+2022-08-02,129.0899,132.1447,127.7719,131.7349,5576984
+2022-08-03,131.7349,136.3317,131.6674,135.3021,10948887
+2022-08-04,135.3021,135.8902,133.6607,133.8962,11239370
+2022-08-05,133.8962,135.0824,130.5671,131.1901,9722647
+2022-08-08,131.1901,131.3133,128.5210,129.9481,9485616
+2022-08-09,129.9481,131.4703,126.4689,127.8680,4952982
+2022-08-10,127.8680,128.8055,124.5924,124.8008,10226144
+2022-08-11,124.8008,125.8627,123.9289,124.7821,10598787
+2022-08-12,124.7821,128.0158,123.8965,127.0963,6337847
+2022-08-15,127.0963,127.2374,123.5386,124.8222,5178459
+2022-08-16,124.8222,125.2966,123.5977,124.5236,8320125
+2022-08-17,124.5236,127.7449,123.9828,127.0799,9451693
+2022-08-18,127.0799,130.3511,126.6044,128.9194,11710175
+2022-08-19,128.9194,133.9631,128.7195,132.4022,6849072
+2022-08-22,132.4022,135.8650,132.3395,134.5235,6992622
+2022-08-23,134.5235,137.5639,133.6597,136.3235,5221318
+2022-08-24,136.3235,136.4193,135.3400,135.8546,7302739
+2022-08-25,135.8546,139.3623,135.1791,138.3837,5560958
+2022-08-26,138.3837,139.4975,136.5028,137.2364,4861988
+2022-08-29,137.2364,140.7480,135.6431,139.7824,8349295
+2022-08-30,139.7824,141.6269,138.7621,140.9039,12020811
+2022-08-31,140.9039,145.8572,140.1764,144.7094,3801734
+2022-09-01,144.7094,148.2287,144.0453,147.5357,3748289
+2022-09-02,147.5357,148.2271,142.4285,144.1142,10309527
+2022-09-05,144.1142,147.0957,143.6423,145.9252,4819077
+2022-09-06,145.9252,147.0478,144.8954,146.9357,8139003
+2022-09-07,146.9357,147.7018,146.8165,147.4870,7743032
+2022-09-08,147.4870,148.6223,144.3994,144.9378,9158184
+2022-09-09,144.9378,148.1476,144.0132,147.0773,5094285
+2022-09-12,147.0773,148.8033,145.6358,146.1067,5929550
+2022-09-13,146.1067,147.2322,144.7203,146.2738,6034334
+2022-09-14,146.2738,150.1214,144.8865,150.0813,8669397
+2022-09-15,150.0813,153.7794,148.8401,153.4974,12293243
+2022-09-16,153.4974,155.2959,151.8011,155.0868,12508518
+2022-09-19,155.0868,156.6263,151.9755,152.1203,3980409
+2022-09-20,152.1203,156.0667,150.3731,154.7300,9266517
+2022-09-21,154.7300,155.5642,153.6663,155.0580,12727597
+2022-09-22,155.0580,158.1401,154.6906,156.4959,10255680
+2022-09-23,156.4959,158.7879,154.9556,158.6530,4890282
+2022-09-26,158.6530,158.7696,155.2696,156.5143,4566056
+2022-09-27,156.5143,160.6951,154.9939,160.6951,10488007
+2022-09-28,160.6951,162.6350,159.1563,161.5263,9309078
+2022-09-29,161.5263,163.0709,158.0075,158.6813,8804160
+2022-09-30,158.6813,160.7491,157.8923,159.0239,11392829
+2022-10-03,159.0239,159.6463,154.4846,155.7378,9074700
+2022-10-04,155.7378,159.1104,154.2349,158.7144,10837955
+2022-10-05,158.7144,158.9807,154.8572,155.9488,12565978
+2022-10-06,155.9488,157.9296,155.0003,157.1227,10971684
+2022-10-07,157.1227,158.5588,156.0041,157.5440,12148113
+2022-10-10,157.5440,161.9027,157.0408,160.2886,8961424
+2022-10-11,160.2886,164.6808,158.4578,163.2738,11986026
+2022-10-12,163.2738,164.2695,159.6082,160.2963,5169308
+2022-10-13,160.2963,162.4385,158.8185,160.5436,4151672
+2022-10-14,160.5436,161.2557,156.9529,158.6379,7611174
+2022-10-17,158.6379,159.6853,155.5503,156.0132,3568049
+2022-10-18,156.0132,158.8804,154.5605,157.0134,10267560
+2022-10-19,157.0134,158.0718,153.1785,154.3583,10487620
+2022-10-20,154.3583,154.4570,152.4377,154.1538,5749175
+2022-10-21,154.1538,157.2039,152.7124,156.3330,6442193
+2022-10-24,156.3330,157.1360,152.3084,153.1908,10808646
+2022-10-25,153.1908,154.4159,151.9499,152.6391,10233544
+2022-10-26,152.6391,154.2511,148.5301,150.2326,3427613
+2022-10-27,150.2326,154.7431,148.9833,153.3117,4216650
+2022-10-28,153.3117,156.8730,152.1595,155.4337,12451105
+2022-10-31,155.4337,160.5604,153.7084,159.3488,4580365
+2022-11-01,159.3488,160.3620,156.8941,157.8844,12347274
+2022-11-02,157.8844,160.4461,156.7073,159.2587,12277537
+2022-11-03,159.2587,160.6718,157.5144,158.6571,7063328
+2022-11-04,158.6571,158.9496,157.5208,157.5414,6425652
+2022-11-07,157.5414,157.6949,155.2001,156.1095,6568579
+2022-11-08,156.1095,161.3657,154.8636,159.8684,11331092
+2022-11-09,159.8684,160.0914,157.0415,158.1989,5383764
+2022-11-10,158.1989,163.3319,156.5167,162.5031,5559174
+2022-11-11,162.5031,163.7258,157.9237,159.2478,5657863
+2022-11-14,159.2478,164.9935,157.9345,163.0890,12620870
+2022-11-15,163.0890,167.8077,161.6596,167.5099,10361298
+2022-11-16,167.5099,170.5118,166.4178,169.8198,8845446
+2022-11-17,169.8198,173.3313,169.0552,173.2243,6222772
+2022-11-18,173.2243,176.4074,173.0961,176.1609,4308118
+2022-11-21,176.1609,177.0808,171.0447,172.8402,6609315
+2022-11-22,172.8402,174.2298,168.7693,169.4458,8683254
+2022-11-23,169.4458,172.8643,167.9547,171.9238,9147793
+2022-11-24,171.9238,176.9916,170.0645,175.3871,10364424
+2022-11-25,175.3871,176.4491,172.0168,173.6258,10732745
+2022-11-28,173.6258,177.7866,171.9078,176.0060,8394012
+2022-11-29,176.0060,177.6741,174.9976,177.5807,10176995
+2022-11-30,177.5807,183.5891,176.6446,181.7390,5404289
+2022-12-01,181.7390,187.1903,181.3212,185.7254,3561854
+2022-12-02,185.7254,191.0383,185.4814,189.7253,12703238
+2022-12-05,189.7253,192.6250,187.8894,190.8872,5656723
+2022-12-06,190.8872,191.7955,189.4843,191.4000,7372043
+2022-12-07,191.4000,194.8994,190.9279,194.6290,6915516
+2022-12-08,194.6290,195.5785,187.5874,189.7795,11510901
+2022-12-09,189.7795,191.6736,186.3412,186.8459,5500504
+2022-12-12,186.8459,189.1539,186.3946,188.4887,4022045
+2022-12-13,188.4887,188.9240,183.5121,185.5424,3602831
+2022-12-14,185.5424,186.6898,183.9387,186.2156,3228953
+2022-12-15,186.2156,187.8640,185.1683,185.2259,7148769
+2022-12-16,185.2259,186.3021,182.4089,183.7954,3967302
+2022-12-19,183.7954,188.2436,182.0926,187.6356,4476017
+2022-12-20,187.6356,188.0717,183.6927,183.8778,5579311
+2022-12-21,183.8778,184.0686,178.7943,180.6408,8273098
+2022-12-22,180.6408,181.9819,178.8839,180.8787,3640717
+2022-12-23,180.8787,182.3247,176.7766,176.7915,8054607
+2022-12-26,176.7915,180.0439,176.7900,178.2270,7058755
+2022-12-27,178.2270,179.4678,176.6575,176.7275,10438646
+2022-12-28,176.7275,182.4666,175.3147,181.2731,11217222
+2022-12-29,181.2731,186.6653,180.2132,184.8478,4041284
+2022-12-30,184.8478,189.9130,183.2930,188.0095,6220570
+2023-01-02,188.0095,194.4669,187.3047,193.0242,10746678
+2023-01-03,193.0242,193.5818,188.4223,190.4916,3729913
+2023-01-04,190.4916,195.4367,189.1840,194.5106,9833389
+2023-01-05,194.5106,199.4775,194.2937,198.9526,8116695
+2023-01-06,198.9526,199.9652,196.0821,196.3418,4064015
+2023-01-09,196.3418,197.4612,195.7137,197.3386,6926543
+2023-01-10,197.3386,199.2463,191.3059,193.3983,3508276
+2023-01-11,193.3983,197.2490,191.7052,196.8335,3782379
+2023-01-12,196.8335,201.9243,196.2216,199.6898,4784633
+2023-01-13,199.6898,201.7502,196.4446,198.3011,4704496
+2023-01-16,198.3011,204.6209,196.0235,202.8219,12537898
+2023-01-17,202.8219,204.5928,199.3959,199.4544,5040959
+2023-01-18,199.4544,200.6992,198.9320,200.1391,6634504
+2023-01-19,200.1391,202.3435,198.6140,200.0090,8538489
+2023-01-20,200.0090,203.0914,198.0809,202.6333,10522656
+2023-01-23,202.6333,207.4270,201.3158,206.3126,3918231
+2023-01-24,206.3126,207.9062,203.8179,206.1681,11802870
+2023-01-25,206.1681,208.0393,202.6176,204.7206,12428558
+2023-01-26,204.7206,207.8051,202.5746,206.9865,3356134
+2023-01-27,206.9865,209.1811,204.7939,208.0376,9163038
+2023-01-30,208.0376,208.4508,204.9598,205.0485,10421504
+2023-01-31,205.0485,209.3681,204.4982,208.4342,8552006
+2023-02-01,208.4342,209.7515,204.5361,205.3829,6263340
+2023-02-02,205.3829,208.3914,204.5660,207.7713,8613593
+2023-02-03,207.7713,213.3322,207.0072,212.9485,5817322
+2023-02-06,212.9485,214.1947,209.1011,211.1799,4939991
+2023-02-07,211.1799,215.0544,209.7304,213.9453,8025563
+2023-02-08,213.9453,215.8977,212.3637,215.4337,7060133
+2023-02-09,215.4337,220.5662,214.3791,219.0005,12675816
+2023-02-10,219.0005,219.0367,215.2850,215.6037,9822228
+2023-02-13,215.6037,217.5064,211.0396,211.2635,6054504
+2023-02-14,211.2635,213.0026,210.1661,212.9494,12793620
+2023-02-15,212.9494,216.8491,210.6280,215.8409,8664424
+2023-02-16,215.8409,220.6278,214.8524,219.7212,9925061
+2023-02-17,219.7212,224.9434,219.3249,223.0892,7245303
+2023-02-20,223.0892,223.7089,220.9659,221.2747,10383140
+2023-02-21,221.2747,222.2141,217.3536,217.4025,9965697
+2023-02-22,217.4025,219.3782,214.9237,219.1728,12751937
+2023-02-23,219.1728,224.5564,217.3224,222.4059,3411422
+2023-02-24,222.4059,223.0380,220.0079,220.0650,9328101
+2023-02-27,220.0650,222.4213,218.0890,221.4589,5449045
+2023-02-28,221.4589,221.7647,218.5993,219.2181,11367526
+2023-03-01,219.2181,221.3707,216.9352,219.4309,9597356
+2023-03-02,219.4309,221.9779,217.9358,218.1454,7165900
+2023-03-03,218.1454,224.7494,217.2197,222.9244,11140838
+2023-03-06,222.9244,228.2700,222.2115,226.3055,9728032
+2023-03-07,226.3055,230.8174,224.9126,228.5140,12121244
+2023-03-08,228.5140,232.1048,226.9279,231.5228,5084609
+2023-03-09,231.5228,234.2228,228.7502,233.3110,6485378
+2023-03-10,233.3110,239.6246,232.2656,237.2448,8854476
+2023-03-13,237.2448,238.3134,233.0059,234.2404,10030876
+2023-03-14,234.2404,235.1465,232.4162,234.6934,8154120
+2023-03-15,234.6934,236.6168,232.2523,235.3583,9481077
+2023-03-16,235.3583,239.1575,234.2236,237.7274,7299903
+2023-03-17,237.7274,241.6003,237.5292,240.3373,5319086
+2023-03-20,240.3373,240.4789,233.9970,235.0101,6826810
+2023-03-21,235.0101,235.5774,234.1631,234.8198,7492339
+2023-03-22,234.8198,235.5550,231.1920,232.5639,10530852
+2023-03-23,232.5639,232.6019,226.2755,227.4984,11784461
+2023-03-24,227.4984,230.4352,224.8428,229.4306,9716599
+2023-03-27,229.4306,236.0940,229.3236,233.7317,5071039
+2023-03-28,233.7317,238.8013,231.2066,237.6856,3883633
+2023-03-29,237.6856,245.3728,236.5122,243.5589,11164148
+2023-03-30,243.5589,245.7301,243.0203,244.3826,10725761
+2023-03-31,244.3826,245.2627,239.9133,240.1672,3351627
+2023-04-03,240.1672,240.2950,236.6047,237.9445,6530639
+2023-04-04,237.9445,239.2691,231.4259,232.2240,5422641
+2023-04-05,232.2240,234.0091,232.0722,233.0788,12137360
+2023-04-06,233.0788,239.9350,231.2347,239.2549,8816291
+2023-04-07,239.2549,240.8002,235.6728,236.4257,5842183
+2023-04-10,236.4257,236.5536,235.5446,236.1966,12644125
+2023-04-11,236.1966,240.8874,233.3716,239.9330,6357119
+2023-04-12,239.9330,245.7064,237.3587,244.7592,7539856
+2023-04-13,244.7592,250.3608,243.6008,249.9502,6508234
+2023-04-14,249.9502,256.6957,249.5435,254.0572,7136646
+2023-04-17,254.0572,254.2579,249.6394,250.0298,3908345
+2023-04-18,250.0298,251.3222,245.0882,247.7778,9458268
+2023-04-19,247.7778,252.5722,246.8212,251.5742,7155279
+2023-04-20,251.5742,251.6345,247.5670,248.3879,12113022
+2023-04-21,248.3879,249.8902,242.1745,242.8504,10520843
+2023-04-24,242.8504,244.3374,235.2054,237.2696,7638689
+2023-04-25,237.2696,239.8005,233.0888,233.1424,12095987
+2023-04-26,233.1424,239.9420,230.7209,237.4236,6749117
+2023-04-27,237.4236,240.7173,235.7570,239.3671,6846539
+2023-04-28,239.3671,245.1273,237.7551,244.2251,3854434
+2023-05-01,244.2251,246.8190,237.8559,238.8783,6795398
+2023-05-02,238.8783,247.5126,238.4327,245.3853,8412506
+2023-05-03,245.3853,247.3145,244.5825,245.5333,5846625
+2023-05-04,245.5333,253.4887,243.9103,251.6885,9840282
+2023-05-05,251.6885,255.9782,249.3238,255.4298,7834270
+2023-05-08,255.4298,256.3997,248.0959,251.0811,11312519
+2023-05-09,251.0811,258.3679,249.1639,256.8252,7004001
+2023-05-10,256.8252,262.0800,255.7233,260.4060,12583052
+2023-05-11,260.4060,260.9360,257.8615,260.8097,4120744
+2023-05-12,260.8097,262.2276,257.5036,258.1593,7513412
+2023-05-15,258.1593,259.8016,255.7068,257.7236,9780284
+2023-05-16,257.7236,258.4569,251.6161,251.9687,6395641
+2023-05-17,251.9687,254.5443,251.7286,251.7713,10978173
+2023-05-18,251.7713,254.0090,245.0351,247.5555,12666689
+2023-05-19,247.5555,249.0845,246.7016,248.9191,7382929
+2023-05-22,248.9191,254.5195,246.3010,251.5951,3805841
+2023-05-23,251.5951,253.5623,246.4069,247.0210,8801940
+2023-05-24,247.0210,249.1306,245.9860,247.5120,11938742
+2023-05-25,247.5120,248.5840,244.6981,245.1738,5981868
+2023-05-26,245.1738,247.0549,240.7116,240.7709,5529867
+2023-05-29,240.7709,247.5672,239.7895,245.0617,10253024
+2023-05-30,245.0617,251.4015,243.2353,249.4553,10436621
+2023-05-31,249.4553,250.4461,246.6829,247.4441,5416825
+2023-06-01,247.4441,249.5945,246.8775,247.9897,4019178
+2023-06-02,247.9897,248.6239,241.6604,242.1140,9162525
+2023-06-05,242.1140,248.0577,240.2431,246.4770,7243465
+2023-06-06,246.4770,248.6870,239.0704,241.2877,4617720
+2023-06-07,241.2877,243.2641,237.2003,239.2261,4774051
+2023-06-08,239.2261,240.3330,236.5938,238.1536,11073328
+2023-06-09,238.1536,239.2282,231.4565,234.0478,4903638
+2023-06-12,234.0478,236.2554,228.6634,230.3883,7769161
+2023-06-13,230.3883,230.5605,228.4559,230.2267,5738777
+2023-06-14,230.2267,232.7928,228.4887,229.8565,6989437
+2023-06-15,229.8565,230.6419,226.7111,228.4049,10077229
+2023-06-16,228.4049,228.4169,223.0283,224.0905,4788911
+2023-06-19,224.0905,226.0346,222.7344,225.5989,7755982
+2023-06-20,225.5989,232.4850,224.8894,231.5953,6440311
+2023-06-21,231.5953,234.5721,230.9235,233.8974,10027660
+2023-06-22,233.8974,234.8155,231.5217,234.6144,4417179
+2023-06-23,234.6144,236.1519,231.1599,232.4335,10551176
+2023-06-26,232.4335,234.2596,227.9579,228.4488,4316188
+2023-06-27,228.4488,233.3523,228.1529,231.1993,3519610

ceo_brief_env/fixtures/stooq_long/nvda_us.csv

@@ -0,0 +1,281 @@
+Date,Open,High,Low,Close,Volume
+2022-06-01,48.0000,48.3224,47.4191,47.9509,47939004
+2022-06-02,47.9509,48.3813,47.8446,48.0427,50714518
+2022-06-03,48.0427,48.9065,47.9884,48.4455,38204075
+2022-06-06,48.4455,48.9162,47.0722,47.4672,22512820
+2022-06-07,47.4672,49.3158,47.0947,48.7514,56933762
+2022-06-08,48.7514,48.7602,47.6356,47.9396,23573066
+2022-06-09,47.9396,48.0788,47.2073,47.2244,47836067
+2022-06-10,47.2244,47.7018,46.8527,47.1464,58417502
+2022-06-13,47.1464,47.5919,46.8876,47.2165,36689773
+2022-06-14,47.2165,49.1126,46.7405,48.5327,62468577
+2022-06-15,48.5327,48.6665,47.9634,48.1304,24213409
+2022-06-16,48.1304,49.1167,47.6414,48.8818,43190811
+2022-06-17,48.8818,50.6517,48.8815,50.1418,32583044
+2022-06-20,50.1418,51.5967,49.5520,51.3074,43845463
+2022-06-21,51.3074,51.6949,49.7541,50.2233,36186535
+2022-06-22,50.2233,50.4237,48.6305,49.1997,65482431
+2022-06-23,49.1997,49.3451,48.2188,48.2773,23593604
+2022-06-24,48.2773,49.2145,47.9533,49.1097,46845492
+2022-06-27,49.1097,49.5411,48.3023,48.3783,58622907
+2022-06-28,48.3783,48.6226,47.3463,47.4676,36187698
+2022-06-29,47.4676,49.1933,47.2944,48.7236,73091906
+2022-06-30,48.7236,48.9541,47.5570,48.0496,58510139
+2022-07-01,48.0496,48.6200,46.9833,47.1039,35496653
+2022-07-04,47.1039,48.0442,46.9364,47.8553,24403913
+2022-07-05,47.8553,48.1899,46.7507,46.8875,56077030
+2022-07-06,46.8875,47.1425,46.1022,46.6390,49023471
+2022-07-07,46.6390,47.3809,46.5367,46.8933,29248119
+2022-07-08,46.8933,48.4495,46.7529,47.9787,31388039
+2022-07-11,47.9787,49.2086,47.8655,48.6595,77008151
+2022-07-12,48.6595,50.0782,48.4134,49.7181,26230380
+2022-07-13,49.7181,50.2925,48.4382,48.5771,62274767
+2022-07-14,48.5771,49.0573,47.6806,48.0243,37606120
+2022-07-15,48.0243,48.4395,47.2312,47.2702,33703780
+2022-07-18,47.2702,47.9756,46.9218,47.4899,36813163
+2022-07-19,47.4899,48.7306,47.4804,48.6116,36151635
+2022-07-20,48.6116,48.6469,48.4420,48.5446,42127122
+2022-07-21,48.5446,48.8802,48.3337,48.8031,73456413
+2022-07-22,48.8031,50.5143,48.3983,50.1192,55066416
+2022-07-25,50.1192,50.1403,49.2284,49.2390,74612747
+2022-07-26,49.2390,50.4131,49.2264,49.8373,58171071
+2022-07-27,49.8373,50.3023,49.6466,49.8651,79961457
+2022-07-28,49.8651,50.1919,48.3857,48.8174,74011752
+2022-07-29,48.8174,49.9221,48.3527,49.5041,74900154
+2022-08-01,49.5041,49.9111,48.6578,49.1896,72266075
+2022-08-02,49.1896,49.6562,48.5392,49.0474,54368449
+2022-08-03,49.0474,49.6727,48.7044,49.4458,56532014
+2022-08-04,49.4458,49.8252,47.8427,48.4198,72787509
+2022-08-05,48.4198,49.3069,47.9927,49.0781,54857172
+2022-08-08,49.0781,49.5718,48.9477,48.9970,65012610
+2022-08-09,48.9970,49.3505,47.5733,47.8494,33813298
+2022-08-10,47.8494,48.7131,47.4966,48.4242,75227856
+2022-08-11,48.4242,48.4308,47.6972,47.8702,60688218
+2022-08-12,47.8702,47.9676,46.6745,47.1874,59599400
+2022-08-15,47.1874,47.6923,46.9281,47.1129,59953943
+2022-08-16,47.1129,47.3565,45.9817,46.4308,74853271
+2022-08-17,46.4308,47.6550,46.1059,47.4362,38989210
+2022-08-18,47.4362,47.7188,46.1246,46.5927,70923218
+2022-08-19,46.5927,47.7220,46.4379,47.1841,30147734
+2022-08-22,47.1841,47.3399,47.0104,47.1315,44839088
+2022-08-23,47.1315,47.7979,46.9531,47.5163,70347104
+2022-08-24,47.5163,49.0665,47.4737,48.8015,21889146
+2022-08-25,48.8015,49.8638,48.3865,49.8390,54234917
+2022-08-26,49.8390,50.3124,49.3980,49.4093,28152869
+2022-08-29,49.4093,49.4240,48.8737,49.3652,34244540
+2022-08-30,49.3652,49.3930,48.1334,48.4996,46788846
+2022-08-31,48.4996,49.2909,48.0297,48.9064,77507639
+2022-09-01,48.9064,49.5763,48.6276,49.4580,30721165
+2022-09-02,49.4580,49.7382,47.8363,48.2498,30745928
+2022-09-05,48.2498,48.4499,47.3405,47.7400,51225376
+2022-09-06,47.7400,48.5377,47.5146,48.1012,67515934
+2022-09-07,48.1012,49.2605,47.5629,49.2090,63342630
+2022-09-08,49.2090,49.4768,47.9549,48.3176,74597921
+2022-09-09,48.3176,48.6474,47.5673,48.0746,67806034
+2022-09-12,48.0746,49.5528,47.6988,49.2786,32293661
+2022-09-13,49.2786,50.4225,48.8992,49.9322,63059717
+2022-09-14,49.9322,50.4714,48.6689,49.2483,78641511
+2022-09-15,49.2483,49.8876,49.0590,49.4187,74599383
+2022-09-16,49.4187,50.6356,49.3696,50.4247,46454071
+2022-09-19,50.4247,51.1015,50.1297,50.6347,21704621
+2022-09-20,50.6347,51.5799,50.1487,51.5403,30373778
+2022-09-21,51.5403,52.0276,51.0807,51.1669,28920533
+2022-09-22,51.1669,51.7338,50.6512,51.2885,61362292
+2022-09-23,51.2885,52.8879,50.7043,52.5771,25161410
+2022-09-26,52.5771,52.9094,51.6990,51.8797,63730592
+2022-09-27,51.8797,52.6677,51.3544,52.3393,53598265
+2022-09-28,52.3393,52.5787,51.3691,51.8955,74031502
+2022-09-29,51.8955,52.4253,50.5762,51.1708,51453556
+2022-09-30,51.1708,51.5696,50.8418,51.4455,50190446
+2022-10-03,51.4455,51.8269,50.8471,51.8096,50961066
+2022-10-04,51.8096,52.3077,51.2657,51.6144,49462426
+2022-10-05,51.6144,52.2555,51.2807,52.2142,44826445
+2022-10-06,52.2142,54.1503,52.0456,53.5569,48389689
+2022-10-07,53.5569,53.8356,52.0637,52.5783,74033328
+2022-10-10,52.5783,52.7920,52.4575,52.5918,57073692
+2022-10-11,52.5918,53.9398,52.1000,53.8561,21367156
+2022-10-12,53.8561,54.0030,52.6263,53.0637,39324934
+2022-10-13,53.0637,53.4584,52.6701,52.7365,63853625
+2022-10-14,52.7365,53.0596,51.6057,51.7613,31863730
+2022-10-17,51.7613,52.1943,51.5279,51.9222,44804120
+2022-10-18,51.9222,52.1807,51.7949,52.0808,57878899
+2022-10-19,52.0808,52.8751,51.5488,52.5412,56702668
+2022-10-20,52.5412,53.7631,52.0741,53.6135,68630494
+2022-10-21,53.6135,55.0460,53.4108,54.8381,75366018
+2022-10-24,54.8381,55.4951,53.5250,54.1012,28035933
+2022-10-25,54.1012,54.5729,53.2685,53.4349,25820952
+2022-10-26,53.4349,54.7313,52.9284,54.4558,27560196
+2022-10-27,54.4558,54.9036,54.2453,54.2569,32056043
+2022-10-28,54.2569,55.4627,53.6264,54.8627,26924614
+2022-10-31,54.8627,55.4616,54.5317,54.9615,61140957
+2022-11-01,54.9615,55.0081,54.0691,54.1381,22246152
+2022-11-02,54.1381,54.7034,53.7686,54.3675,28793289
+2022-11-03,54.3675,54.5006,53.0003,53.5401,79418082
+2022-11-04,53.5401,54.8944,53.5003,54.8318,77089696
+2022-11-07,54.8318,55.3349,54.5888,54.8038,48017592
+2022-11-08,54.8038,55.2139,54.4086,54.9304,20794760
+2022-11-09,54.9304,56.1613,54.8109,55.5981,47237058
+2022-11-10,55.5981,56.6609,55.4679,56.3867,29904599
+2022-11-11,56.3867,56.5193,55.7823,56.5088,68102369
+2022-11-14,56.5088,57.7974,56.0820,57.2066,44270594
+2022-11-15,57.2066,57.9429,56.5320,57.5944,68289284
+2022-11-16,57.5944,58.2242,56.4342,56.9429,66679278
+2022-11-17,56.9429,58.2601,56.3302,57.9779,72790490
+2022-11-18,57.9779,59.2035,57.4455,58.6634,44344730
+2022-11-21,58.6634,59.4939,58.4228,59.4436,48130423
+2022-11-22,59.4436,59.6973,57.5464,57.9909,57441216
+2022-11-23,57.9909,58.6483,56.7969,57.2545,40268936
+2022-11-24,57.2545,58.2204,56.8883,57.8252,43375644
+2022-11-25,57.8252,59.9021,57.3386,59.4439,65704382
+2022-11-28,59.4439,61.0624,59.0050,61.0456,64327692
+2022-11-29,61.0456,61.3398,60.3133,60.3498,31726765
+2022-11-30,60.3498,60.4211,59.8619,60.0427,74338595
+2022-12-01,60.0427,60.6595,59.3458,60.2920,54078710
+2022-12-02,60.2920,62.4390,59.7063,61.9646,24571759
+2022-12-05,61.9646,62.9462,61.9310,62.3778,75809229
+2022-12-06,62.3778,62.7310,61.2226,61.3471,49731254
+2022-12-07,61.3471,61.8440,60.6513,61.8006,45242232
+2022-12-08,61.8006,62.4256,61.5295,61.9810,37145024
+2022-12-09,61.9810,63.0045,61.7701,62.5835,62996106
+2022-12-12,62.5835,62.5940,61.8185,62.0008,22566440
+2022-12-13,62.0008,62.5623,60.6801,60.9654,73852526
+2022-12-14,60.9654,61.8966,60.2420,61.8593,76667440
+2022-12-15,61.8593,62.5315,60.2417,60.5538,48656506
+2022-12-16,60.5538,62.3450,60.1735,62.1632,76236257
+2022-12-19,62.1632,63.3443,61.4331,62.9903,69001762
+2022-12-20,62.9903,63.5183,62.5184,63.4307,47341321
+2022-12-21,63.4307,63.4704,62.1334,62.5296,27459554
+2022-12-22,62.5296,63.0308,62.0928,62.4341,35727764
+2022-12-23,62.4341,63.1160,61.8512,62.7998,51845975
+2022-12-26,62.7998,65.2883,62.2465,64.5504,34304661
+2022-12-27,64.5504,65.2419,62.7302,63.3262,57498585
+2022-12-28,63.3262,63.5325,62.4311,62.9485,53893424
+2022-12-29,62.9485,63.8301,62.3795,63.3488,31388926
+2022-12-30,63.3488,64.0936,61.9131,62.6010,72726584
+2023-01-02,62.6010,62.6466,60.9682,61.1670,45511578
+2023-01-03,61.1670,61.7345,60.7695,61.6587,24349570
+2023-01-04,61.6587,62.1590,60.0006,60.3997,57866325
+2023-01-05,60.3997,60.7465,59.9325,60.0843,40622031
+2023-01-06,60.0843,61.5674,60.0307,60.9541,27187311
+2023-01-09,60.9541,62.5087,60.3918,62.0444,32789879
+2023-01-10,62.0444,62.2365,61.2872,61.8887,42136999
+2023-01-11,61.8887,63.2273,61.6470,62.4855,52916808
+2023-01-12,62.4855,64.0948,62.1649,63.3943,42155630
+2023-01-13,63.3943,64.0601,62.0769,62.1354,24981285
+2023-01-16,62.1354,62.8034,61.6244,62.4401,37932240
+2023-01-17,62.4401,63.5034,62.2803,63.4455,59725896
+2023-01-18,63.4455,63.6768,61.5935,62.1341,61638850
+2023-01-19,62.1341,62.2407,61.2483,61.5124,63558537
+2023-01-20,61.5124,61.5991,60.6485,61.1691,54156473
+2023-01-23,61.1691,63.3241,60.4987,62.6178,46279562
+2023-01-24,62.6178,63.9505,62.3791,63.7175,43975501
+2023-01-25,63.7175,65.9818,63.5276,65.2809,41701652
+2023-01-26,65.2809,65.5655,64.6056,64.9137,43254792
+2023-01-27,64.9137,65.3529,63.3498,63.9616,52433687
+2023-01-30,63.9616,65.6382,63.8261,65.1979,65536841
+2023-01-31,65.1979,66.8368,65.1805,66.6119,50938477
+2023-02-01,66.6119,67.3230,65.8394,66.8676,59072247
+2023-02-02,66.8676,68.0988,66.4289,68.0465,67283772
+2023-02-03,68.0465,68.1131,66.0589,66.6484,73944311
+2023-02-06,66.6484,67.1556,65.1687,65.2814,64747492
+2023-02-07,65.2814,66.0889,65.1087,65.8948,65921181
+2023-02-08,65.8948,66.6752,65.5830,66.0690,40268815
+2023-02-09,66.0690,68.3549,65.6776,67.8078,52240390
+2023-02-10,67.8078,69.2873,67.0633,68.7035,44636331
+2023-02-13,68.7035,70.5545,67.9998,69.9945,68354435
+2023-02-14,69.9945,72.0986,69.1900,71.3379,58416401
+2023-02-15,71.3379,72.1446,70.6511,71.5351,45296655
+2023-02-16,71.5351,71.6606,70.7062,71.3408,79460622
+2023-02-17,71.3408,71.4843,70.8031,70.9771,45496710
+2023-02-20,70.9771,71.8031,70.2513,70.3012,38506676
+2023-02-21,70.3012,70.8479,68.3293,68.9715,30765972
+2023-02-22,68.9715,69.3512,67.0022,67.4751,74557695
+2023-02-23,67.4751,67.5631,65.7723,65.9183,33041860
+2023-02-24,65.9183,66.4857,64.6282,65.0928,33437559
+2023-02-27,65.0928,65.3123,63.9710,64.1037,65450232
+2023-02-28,64.1037,64.5254,62.9369,63.5602,48769467
+2023-03-01,63.5602,64.2370,62.1591,62.8486,40524498
+2023-03-02,62.8486,63.8253,62.4846,63.1006,33260314
+2023-03-03,63.1006,63.8181,61.0960,61.6893,43098593
+2023-03-06,61.6893,62.2550,61.4861,61.8720,79414739
+2023-03-07,61.8720,62.6597,61.8640,62.4815,48358760
+2023-03-08,62.4815,63.0789,61.6179,62.1499,56364393
+2023-03-09,62.1499,62.2669,60.8782,61.1143,35565400
+2023-03-10,61.1143,62.7884,60.6471,62.4019,79587131
+2023-03-13,62.4019,62.8022,61.6117,61.7233,66222663
+2023-03-14,61.7233,62.3317,59.5736,60.1847,69325303
+2023-03-15,60.1847,60.3796,58.4362,58.9430,25830017
+2023-03-16,58.9430,59.3009,58.2672,58.9693,55198007
+2023-03-17,58.9693,60.3938,58.4107,60.1745,45018230
+2023-03-20,60.1745,61.6612,59.5779,61.5939,32504498
+2023-03-21,61.5939,62.2179,61.4775,61.8279,69920547
+2023-03-22,61.8279,62.0585,61.2467,61.3027,38342082
+2023-03-23,61.3027,61.8288,61.0236,61.2881,61225660
+2023-03-24,61.2881,61.5781,59.7664,60.0995,78014498
+2023-03-27,60.0995,61.7205,60.0905,61.2399,42627605
+2023-03-28,61.2399,62.1901,60.8254,62.0134,47485269
+2023-03-29,62.0134,62.7513,59.9171,60.4975,32403806
+2023-03-30,60.4975,61.1728,60.2245,60.9604,52386055
+2023-03-31,60.9604,61.2072,60.1156,60.3998,60001328
+2023-04-03,60.3998,60.5448,59.1867,59.7108,39763881
+2023-04-04,59.7108,61.6221,59.1922,61.2087,39899507
+2023-04-05,61.2087,62.4300,61.1053,62.3610,25663785
+2023-04-06,62.3610,63.5777,62.2264,63.0418,44136741
+2023-04-07,63.0418,64.7192,62.9737,64.2621,33594978
+2023-04-10,64.2621,64.3578,62.8821,63.1906,24360950
+2023-04-11,63.1906,65.0256,62.8027,64.6941,58834852
+2023-04-12,64.6941,66.3535,64.3944,65.7061,39884630
+2023-04-13,65.7061,65.7183,65.1838,65.4987,61992841
+2023-04-14,65.4987,67.9075,64.9798,67.2700,56516241
+2023-04-17,67.2700,67.5372,65.3847,65.6544,59043339
+2023-04-18,65.6544,65.9519,63.9893,64.3828,67313257
+2023-04-19,64.3828,66.0703,64.2605,65.5890,65990762
+2023-04-20,65.5890,67.5297,65.3115,67.0883,50024936
+2023-04-21,67.0883,67.6627,65.1346,65.9152,50962181
+2023-04-24,65.9152,67.4359,65.7592,66.7665,76693427
+2023-04-25,66.7665,67.4763,66.6998,67.3164,34675961
+2023-04-26,67.3164,68.2565,67.0505,67.6903,75743602
+2023-04-27,67.6903,68.0664,67.1937,67.2939,78414699
+2023-04-28,67.2939,68.0239,65.6642,66.0952,53643678
+2023-05-01,66.0952,66.6164,65.3739,66.4057,79527871
+2023-05-02,66.4057,68.1109,66.3080,67.6228,69479116
+2023-05-03,67.6228,68.3507,66.7730,66.9665,54415063
+2023-05-04,66.9665,68.3928,66.5263,68.2409,24576281
+2023-05-05,68.2409,68.3883,65.8613,66.6507,76366678
+2023-05-08,66.6507,67.5586,66.1138,67.3101,64260744
+2023-05-09,67.3101,67.7882,66.3427,66.9888,20980949
+2023-05-10,66.9888,67.3651,65.9092,66.0225,43182666
+2023-05-11,66.0225,66.5031,65.6107,66.3648,35819491
+2023-05-12,66.3648,66.9695,65.8538,66.7037,22271791
+2023-05-15,66.7037,67.5253,65.9358,67.4082,56005598
+2023-05-16,67.4082,67.7410,66.8973,67.4019,61358523
+2023-05-17,67.4019,69.0420,67.0089,68.4247,79676891
+2023-05-18,68.4247,70.4697,68.0889,69.7540,46037392
+2023-05-19,69.7540,70.8640,69.3137,70.1025,51500069
+2023-05-22,70.1025,70.8632,69.6866,69.9559,23280846
+2023-05-23,69.9559,71.6630,69.3414,70.8972,55851297
+2023-05-24,70.8972,72.2131,70.3923,71.9500,24187987
+2023-05-25,71.9500,72.3359,70.1998,70.6258,43803458
+2023-05-26,70.6258,71.2147,68.6186,69.0547,34348857
+2023-05-29,69.0547,69.3824,68.2556,68.4493,24106379
+2023-05-30,68.4493,70.8564,67.9025,70.0441,71989491
+2023-05-31,70.0441,70.7209,69.6715,69.8574,64797363
+2023-06-01,69.8574,70.9706,69.7748,70.2095,67541445
+2023-06-02,70.2095,70.6624,68.1285,68.9149,20034594
+2023-06-05,68.9149,69.1624,67.8177,68.0830,23758054
+2023-06-06,68.0830,70.2921,67.7584,69.6109,41389863
+2023-06-07,69.6109,70.0693,69.5849,70.0308,73910730
+2023-06-08,70.0308,70.4496,68.6446,69.4226,78636742
+2023-06-09,69.4226,69.5981,69.1766,69.4261,75358941
+2023-06-12,69.4261,71.1568,68.7280,70.9903,41249240
+2023-06-13,70.9903,71.1397,70.2409,70.9933,79724244
+2023-06-14,70.9933,71.5320,69.8177,69.9785,72810169
+2023-06-15,69.9785,70.0676,67.8187,68.4145,38779274
+2023-06-16,68.4145,70.6988,67.8290,69.9686,28092788
+2023-06-19,69.9686,71.5970,69.5949,70.8002,24739373
+2023-06-20,70.8002,71.0612,69.2722,69.8679,31787909
+2023-06-21,69.8679,70.0653,68.2429,68.7915,67429400
+2023-06-22,68.7915,69.3380,67.7036,68.4299,55394024
+2023-06-23,68.4299,68.6770,66.7985,67.5497,60022008
+2023-06-26,67.5497,68.0684,66.7799,66.8520,78981678
+2023-06-27,66.8520,67.2759,66.3158,66.7409,22720502

ceo_brief_env/graders.py

@@ -0,0 +1,158 @@
+from __future__ import annotations
+
+from typing import Dict, Iterable, List
+
+import pandas as pd
+
+from .models import Brief, ExpertReport
+
+_SCORE_MIN = 0.001
+_SCORE_MAX = 0.999
+
+
+# RAG+ mode: Stooq watchlist citations (see ``ceo_brief_env.stooq_scrape``).
+STOOQ_CITATION_PREFIX = "stooq:"
+STOOQ_CITATION_SUFFIXES = frozenset({"nvda.us", "aapl.us", "jpm.us"})
+
+
+def _citation_grounds(retriever_sources: set[str], c: str) -> bool:
+    if not isinstance(c, str) or not c:
+        return False
+    if c.startswith("memory:") and c[len("memory:") :] in retriever_sources:
+        return True
+    if c.startswith(STOOQ_CITATION_PREFIX):
+        rest = c[len(STOOQ_CITATION_PREFIX) :]
+        return rest in STOOQ_CITATION_SUFFIXES
+    return False
+
+
+def grounding_score(reports: Dict[str, ExpertReport]) -> float:
+    """Fraction of specialists with a resolvable ``memory:`` or ``stooq:`` citation (RAG on only)."""
+    if not reports:
+        return 0.0
+    try:
+        from memory import get_retriever
+    except Exception:
+        known_sources: set[str] = set()
+    else:
+        known_sources = set(get_retriever().sources())
+    grounded = 0
+    for report in reports.values():
+        cites = getattr(report, "memory_citations", []) or []
+        if any(_citation_grounds(known_sources, c) for c in cites):
+            grounded += 1
+    return grounded / max(len(reports), 1)
+
+
+def _clamp(score: float) -> float:
+    return max(_SCORE_MIN, min(_SCORE_MAX, round(float(score), 4)))
+
+
+def load_metric_ground_truth(path: str) -> Dict[str, str]:
+    df = pd.read_csv(path)
+    return {str(row["metric"]): str(row["value"]) for _, row in df.iterrows()}
+
+
+def _try_float(value: object) -> float | None:
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def metric_match_score(expected: Dict[str, str], actual: Dict[str, object]) -> float:
+    if not expected:
+        return 1.0
+    matched = 0.0
+    for key, expected_value in expected.items():
+        actual_value = actual.get(key)
+        ev = _try_float(expected_value)
+        av = _try_float(actual_value)
+        if ev is not None and av is not None:
+            tolerance = max(0.5, abs(ev) * 0.02)
+            if abs(ev - av) <= tolerance:
+                matched += 1.0
+        elif str(actual_value) == str(expected_value):
+            matched += 1.0
+    return matched / max(len(expected), 1)
+
+
+def strategy_rubric_score(recommendations: List[str], evidence_numbers: Iterable[str], categories: Iterable[str]) -> float:
+    if not recommendations:
+        return 0.001
+    score = 0.0
+    if len(recommendations) == 3:
+        score += 0.25
+    lowered = [r.lower() for r in recommendations]
+    nums = [str(n) for n in evidence_numbers]
+    if nums and all(any(n in rec for n in nums) for rec in lowered[: min(3, len(lowered))]):
+        score += 0.25
+    cat_hits = 0
+    for cat in set(str(c).lower() for c in categories if c):
+        if any(cat in rec for rec in lowered):
+            cat_hits += 1
+    if cat_hits >= 2:
+        score += 0.25
+    if any("projection" in rec or "variance" in rec or "break-even" in rec for rec in lowered):
+        score += 0.25
+    return _clamp(score)
+
+
+def grade_episode(
+    expected_metrics: Dict[str, str],
+    task_meta: Dict[str, object],
+    brief: Brief,
+    reports: Dict[str, ExpertReport],
+    use_rag: bool = False,
+) -> float:
+    required_experts = list(task_meta.get("required_experts", []))
+    coverage = 0.0
+    if required_experts:
+        covered = sum(1 for expert in required_experts if expert in brief.consulted_experts)
+        coverage = covered / len(required_experts)
+
+    metric_score = metric_match_score(expected_metrics, brief.metrics)
+    hr_score = reports.get("hr").score if reports.get("hr") and reports.get("hr").score is not None else 0.001
+    analyst_cats = []
+    if reports.get("analyst"):
+        analyst_cats = [c for c in reports["analyst"].citations if c]
+    evidence_numbers = [str(v) for v in brief.metrics.values()]
+    strategy_needed = "strategy" in required_experts
+    strategy_score = 1.0 if not strategy_needed else strategy_rubric_score(brief.recommendations, evidence_numbers, analyst_cats)
+
+    if use_rag:
+        ground_score = grounding_score(reports)
+        weights = {
+            "coverage": 0.2,
+            "metrics": 0.40,
+            "hr": 0.15,
+            "strategy": 0.15,
+            "grounding": 0.10,
+        }
+        if not strategy_needed:
+            weights["metrics"] += weights["strategy"]
+            weights["strategy"] = 0.0
+        total = (
+            weights["coverage"] * coverage
+            + weights["metrics"] * metric_score
+            + weights["hr"] * hr_score
+            + weights["strategy"] * strategy_score
+            + weights["grounding"] * ground_score
+        )
+    else:
+        weights = {
+            "coverage": 0.2,
+            "metrics": 0.5,
+            "hr": 0.15,
+            "strategy": 0.15,
+        }
+        if not strategy_needed:
+            weights["metrics"] += weights["strategy"]
+            weights["strategy"] = 0.0
+        total = (
+            weights["coverage"] * coverage
+            + weights["metrics"] * metric_score
+            + weights["hr"] * hr_score
+            + weights["strategy"] * strategy_score
+        )
+    return _clamp(total)

ceo_brief_env/models.py

@@ -0,0 +1,84 @@
+from __future__ import annotations
+
+from typing import Any, Dict, List, Literal, Optional
+
+from pydantic import BaseModel, ConfigDict, Field
+
+
+class RewardBreakdown(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    immediate: float
+    cumulative: float
+    terminal_grader: Optional[float] = None
+
+
+class CoSAction(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+
+    action_type: Literal["consult", "ask", "summarize", "submit", "noop"]
+    expert_id: Optional[Literal["analyst", "finance", "hr", "strategy"]] = None
+    sub_question_id: Optional[str] = None
+    notes: Optional[str] = None
+
+
+class ExpertReport(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    expert_id: Literal["analyst", "finance", "hr", "strategy"]
+    title: str
+    summary: str
+    metrics: Dict[str, float | int | str] = Field(default_factory=dict)
+    bullet_points: List[str] = Field(default_factory=list)
+    issues: List[str] = Field(default_factory=list)
+    citations: List[str] = Field(default_factory=list)
+    memory_citations: List[str] = Field(default_factory=list)
+    memory_snippets: List[str] = Field(default_factory=list)
+    memo: Optional[str] = None
+    score: Optional[float] = None
+
+
+class Brief(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    summary: str
+    metrics: Dict[str, float | int | str] = Field(default_factory=dict)
+    recommendations: List[str] = Field(default_factory=list)
+    hr_memo: str = ""
+    consulted_experts: List[str] = Field(default_factory=list)
+
+
+class CoSObservation(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    done: bool
+    reward: float = 0.0
+    instruction: str
+    history: List[str] = Field(default_factory=list)
+    issues: List[str] = Field(default_factory=list)
+    data_quality_score: float = 0.0
+    task_name: str
+    task_difficulty: str
+    rag_enabled: bool = False
+    max_steps: int = 12
+    step_count: int = 0
+    available_experts: List[str] = Field(default_factory=lambda: ["analyst", "finance", "hr", "strategy"])
+    consulted_experts: List[str] = Field(default_factory=list)
+    expert_reports: Dict[str, ExpertReport] = Field(default_factory=dict)
+    current_brief: Optional[Brief] = None
+    reward_breakdown: Optional[RewardBreakdown] = None
+    terminal_grader_score: Optional[float] = None
+
+
+class CoSState(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    episode_id: str
+    task_name: str
+    step_count: int = 0
+    done: bool = False
+    rag_enabled: bool = False
+    consulted_experts: List[str] = Field(default_factory=list)
+    expert_reports: Dict[str, ExpertReport] = Field(default_factory=dict)
+    current_brief: Optional[Brief] = None
+    cumulative_reward: float = 0.0

ceo_brief_env/stooq_scrape.py

@@ -0,0 +1,139 @@
+"""
+Lightweight Stooq daily OHLCV fetch (HTTP GET of CSV) for RAG-advanced mode.
+
+Not financial advice. For simulation / grounding only. Falls back to bundled fixtures
+when offline or on error so CI and judges stay reproducible.
+"""
+from __future__ import annotations
+
+import csv
+import io
+from pathlib import Path
+from typing import List, Tuple
+from urllib.error import HTTPError, URLError
+from urllib.request import Request, urlopen
+
+# Must match grader `STOOQ_CITATION_SUFFIXES`.
+DEFAULT_WATCHLIST: tuple[str, ...] = ("nvda.us", "aapl.us", "jpm.us")
+
+_STOOQ_DAILY = "https://stooq.com/q/d/l/?s={symbol}&i=d"
+_USER_AGENT = "AutoDataLab-Plus/0.1 (research; +https://github.com/)"
+_TIMEOUT_SEC = 8.0
+
+_FIXTURES = Path(__file__).resolve().parent / "fixtures" / "stooq"
+# Bundled multi-hundred-row daily history (Stooq-shaped) for Strategy when RAG is off
+# (no network; used as “enterprise tape” context).
+_LONG_FIXTURES = Path(__file__).resolve().parent / "fixtures" / "stooq_long"
+
+
+def _parse_csv_tail(text: str, symbol: str, last_n: int = 3) -> str:
+    text = text.strip()
+    if not text or "Date" not in text:
+        return f"{symbol}: (no data)"
+    lines = [ln for ln in text.splitlines() if ln.strip()]
+    if len(lines) < 2:
+        return f"{symbol}: (no data)"
+    rdr = csv.reader(io.StringIO(text))
+    rows: List[list[str]] = list(rdr)
+    if not rows:
+        return f"{symbol}: (no data)"
+    header, *data = rows
+    if not data:
+        return f"{symbol}: (no data)"
+    tail = data[-last_n:]
+    parts = [f"Stooq {symbol} daily:"]
+    for row in tail:
+        if not row or row[0] == "No data":
+            continue
+        date = row[0]
+        try:
+            close = row[4] if len(row) > 4 else row[-1]
+        except IndexError:
+            close = "—"
+        parts.append(f"  {date} close≈{close}")
+    return " ".join(parts) if len(parts) > 1 else f"{symbol}: (unparseable)"
+
+
+def _read_fixture(symbol: str) -> str | None:
+    path = _FIXTURES / f"{symbol.replace('.', '_')}.csv"
+    if not path.is_file():
+        return None
+    try:
+        return path.read_text(encoding="utf-8", errors="replace")
+    except OSError:
+        return None
+
+
+def _looks_like_stooq_csv(text: str) -> bool:
+    t = text.lstrip("\ufeff").strip()
+    if not t or "Date" not in t.splitlines()[0]:
+        return False
+    if "<html" in t.lower() or "<!doctype" in t.lower():
+        return False
+    return True
+
+
+def fetch_stooq_daily_csv(symbol: str) -> str:
+    """Return raw CSV text from network or local fixture."""
+    url = _STOOQ_DAILY.format(symbol=symbol.lower())
+    req = Request(url, headers={"User-Agent": _USER_AGENT})
+    try:
+        with urlopen(req, timeout=_TIMEOUT_SEC) as resp:
+            raw = resp.read().decode("utf-8", errors="replace")
+    except (HTTPError, URLError, OSError, TimeoutError):
+        raw = ""
+    if not _looks_like_stooq_csv(raw):
+        fix = _read_fixture(symbol)
+        if fix is not None:
+            return fix
+        return "Date,Open,High,Low,Close,Volume\n"
+    return raw
+
+
+def read_long_fixture_csv(symbol: str) -> str:
+    """Read the bundled long daily CSV for ``symbol`` (e.g. ``nvda.us``)."""
+    path = _LONG_FIXTURES / f"{symbol.replace('.', '_')}.csv"
+    if not path.is_file():
+        return "Date,Open,High,Low,Close,Volume\n"
+    try:
+        return path.read_text(encoding="utf-8", errors="replace")
+    except OSError:
+        return "Date,Open,High,Low,Close,Volume\n"
+
+
+def scrape_watchlist_from_long_csv(
+    symbols: tuple[str, ...] = DEFAULT_WATCHLIST,
+    last_n: int = 5,
+) -> List[Tuple[str, str, str, int]]:
+    """
+    Like :func:`scrape_watchlist` but only reads local long CSVs (no HTTP).
+    Returns ``(stooq_symbol, citation, snippet, row_count_excl_header)``.
+    Citations use the same ``stooq:`` prefix so graders stay consistent if RAG is on elsewhere.
+    """
+    out: list[tuple[str, str, str, int]] = []
+    for sym in symbols:
+        raw = read_long_fixture_csv(sym)
+        rows = list(csv.reader(io.StringIO(raw)))
+        n_data = max(0, len(rows) - 1)
+        snip = _parse_csv_tail(raw, sym, last_n=last_n)
+        cite = f"stooq:{sym}"
+        out.append((sym, cite, snip, n_data))
+    return out
+
+
+def scrape_watchlist(
+    symbols: tuple[str, ...] = DEFAULT_WATCHLIST,
+) -> List[Tuple[str, str, str]]:
+    """
+    For each symbol, fetch Stooq daily history and return
+    (stooq_symbol, citation, snippet) for RAG + grounding.
+
+    Citation format: ``stooq:nvda.us`` (used by ``graders.grounding_score``).
+    """
+    out: list[tuple[str, str, str]] = []
+    for sym in symbols:
+        raw = fetch_stooq_daily_csv(sym)
+        snip = _parse_csv_tail(raw, sym)
+        cite = f"stooq:{sym}"
+        out.append((sym, cite, snip))
+    return out

ceo_brief_env/tasks/_build_ground_truth.py

@@ -0,0 +1,52 @@
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+import pandas as pd
+
+REPO = Path(__file__).resolve().parents[2]
+if str(REPO) not in sys.path:
+    sys.path.insert(0, str(REPO))
+
+from ceo_brief_env.experts.data_analyst import DataAnalystExpert
+from ceo_brief_env.experts.finance import FinanceExpert
+from ceo_brief_env.experts.hr import HRExpert
+from ceo_brief_env.experts.strategy import StrategyExpert
+
+ROOT = Path(__file__).resolve().parent
+
+
+def build_task(task_name: str) -> None:
+    task_dir = ROOT / task_name
+    raw_df = pd.read_csv(task_dir / 'raw.csv')
+    meta = json.loads((task_dir / 'metadata.json').read_text(encoding='utf-8'))
+    analyst = DataAnalystExpert().run(task_name, meta['instruction'], raw_df)
+    finance = FinanceExpert().run(task_name, meta['instruction'], raw_df, analyst.metrics, meta)
+    strategy = None
+    if 'strategy' in meta.get('required_experts', []):
+        strategy = StrategyExpert().run(task_name, meta, analyst, finance)
+    hr = HRExpert().run(task_name, meta, analyst, finance, strategy)
+    metrics = {
+        'data_quality_score': analyst.metrics['data_quality_score'],
+        'total_revenue': analyst.metrics['total_revenue'],
+        'top_category': analyst.metrics['top_category'],
+        'projection_next_quarter': finance.metrics['projection_next_quarter'],
+        'variance_pct': finance.metrics['variance_pct'],
+        'break_even_units': finance.metrics['break_even_units'],
+        'memo_score': hr.metrics['memo_score'],
+    }
+    rows = [{'metric': k, 'value': v} for k, v in metrics.items()]
+    pd.DataFrame(rows).to_csv(task_dir / 'ground_truth.csv', index=False)
+
+
+def main() -> int:
+    for task_dir in ROOT.iterdir():
+        if task_dir.is_dir() and task_dir.name.endswith('_brief'):
+            build_task(task_dir.name)
+    return 0
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())

ceo_brief_env/tasks/_widen_raw_to_400.py

@@ -0,0 +1,84 @@
+"""
+Expand each */raw.csv to 400 columns: 8 business columns (used by the analyst
+pipeline) + 392 wide-table \"enterprise context\" fields (unused by KPI math).
+
+Re-run:  python3 ceo_brief_env/tasks/_widen_raw_to_400.py
+Then:   python3 ceo_brief_env/tasks/_build_ground_truth.py
+"""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+ROOT = Path(__file__).resolve().parent
+N_COLS = 400
+CORE = [
+    "OrderID",
+    "CustomerID",
+    "ExpiryDays",
+    "Product",
+    "Category",
+    "Price",
+    "Quantity",
+    "OrderDate",
+]
+N_EXTRA = N_COLS - len(CORE)  # 392
+
+
+def _wide_frame(df: pd.DataFrame, task_seed: int) -> pd.DataFrame:
+    """Append dim_0001..dim_0392 with reproducible pseudo-enterprise noise."""
+    n = len(df)
+    rng = np.random.default_rng(task_seed)
+    cols: dict[str, np.ndarray] = {}
+    for j in range(N_EXTRA):
+        col = f"dim_{j+1:04d}"
+        kind = j % 5
+        if kind == 0:
+            cols[col] = rng.integers(0, 2, size=n)
+        elif kind == 1:
+            cols[col] = rng.integers(1, 500, size=n)
+        elif kind == 2:
+            cols[col] = np.round(rng.random(n) * 100.0, 4)
+        elif kind == 3:
+            cols[col] = rng.integers(2018, 2025, size=n)
+        else:
+            cols[col] = (rng.integers(0, 1_000_000, size=n) // 1000).astype(np.int64)
+    extra = pd.DataFrame(cols, index=df.index)
+    return pd.concat([df.reset_index(drop=True), extra], axis=1)
+
+
+def main() -> int:
+    for task_dir in sorted(ROOT.iterdir()):
+        if not task_dir.is_dir() or not str(task_dir.name).endswith("_brief"):
+            continue
+        p = task_dir / "raw.csv"
+        if not p.exists():
+            continue
+        task_name = task_dir.name
+        seed = int.from_bytes(task_name.encode(), "big") % (2**31)
+        df = pd.read_csv(p)
+        # Keep only core columns if re-run on already-wide file
+        if len(df.columns) > len(CORE):
+            core_present = [c for c in CORE if c in df.columns]
+            df = df[core_present]
+        missing = [c for c in CORE if c not in df.columns]
+        if missing:
+            print(f"skip {task_name}: missing columns {missing}", file=sys.stderr)
+            continue
+        df = df[[c for c in CORE if c in df.columns]]
+        assert len(df.columns) == len(CORE)
+        wide = _wide_frame(df, seed)
+        if len(wide.columns) != N_COLS:
+            print(f"column count {len(wide.columns)} != {N_COLS}", file=sys.stderr)
+            return 1
+        wide.to_csv(p, index=False)
+        print(f"wrote {p}  rows={len(wide)}  cols={len(wide.columns)}", flush=True)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

ceo_brief_env/tasks/crisis_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,0.9167
+total_revenue,282600.0
+top_category,Electronics
+projection_next_quarter,-70800.0
+variance_pct,101.86
+break_even_units,26.67
+memo_score,0.6805

ceo_brief_env/tasks/crisis_brief/metadata.json

@@ -0,0 +1,22 @@
+{
+  "title": "Crisis brief — mid-quarter shock with hiring freeze",
+  "difficulty": "hard",
+  "max_steps": 14,
+  "instruction": "A mid-quarter shock has put the operating plan well behind expectation. Run the full office: tight data quality read, finance variance and break-even sensitivity, defensive strategy on NVDA/AAPL/JPM, and an HR memo that names the variance, the projection path, and the strategy stance. Be conservative: prefer holds/trims unless evidence supports otherwise.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "strategy",
+    "hr"
+  ],
+  "memo_audience": "CEO, CFO, COO",
+  "hr_required_terms": [
+    "variance",
+    "projection",
+    "strategy",
+    "break-even"
+  ],
+  "plan_value": 140000,
+  "fixed_cost": 48000,
+  "unit_margin": 1800
+}

ceo_brief_env/tasks/crisis_brief/raw.csv

@@ -0,0 +1,13 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+601,C300,30,Workstation,Electronics,98000.0,1.0,2024-01-07,1,29,11.2991,2024,570,1,50,81.1799,2023,469,1,398,35.0699,2024,482,1,34,76.8981,2020,913,1,301,49.0841,2019,311,0,404,87.1558,2018,700,1,490,47.5491,2023,213,0,403,17.7827,2022,882,1,40,2.8595,2021,337,0,108,35.8678,2018,173,0,187,11.7391,2021,984,0,218,76.582,2023,619,0,207,7.1317,2020,75,0,480,9.2095,2023,602,0,104,2.6198,2023,945,0,400,24.3749,2022,413,0,123,72.9151,2024,757,0,70,77.7896,2023,787,1,484,79.7165,2023,877,0,111,68.4621,2019,984,1,488,44.3529,2022,647,1,344,36.5666,2024,436,0,199,45.5387,2021,749,1,434,64.7971,2021,535,1,60,22.1585,2019,596,0,267,46.6497,2019,27,1,154,78.7417,2023,205,1,329,54.2622,2021,31,0,73,31.2542,2018,847,1,235,92.122,2021,555,1,246,95.9976,2021,737,0,104,83.7357,2018,583,1,268,13.6289,2021,742,1,52,16.7124,2018,243,1,302,10.2863,2019,510,0,93,92.223,2024,860,1,243,98.1271,2022,827,0,485,18.6065,2021,861,1,269,0.2961,2018,751,0,344,29.387,2021,145,1,200,98.955,2024,112,1,437,34.2847,2020,429,1,485,64.8237,2020,551,1,231,63.1177,2019,631,1,241,7.9994,2018,846,1,373,18.1263,2022,878,0,224,3.7138,2019,504,1,315,80.1548,2023,167,0,246,18.9959,2023,274,1,84,31.0602,2022,863,0,139,89.972,2022,567,1,215,84.2751,2018,465,1,74,77.668,2020,998,0,364,51.6699,2018,107,0,253,18.1034,2022,884,1,114,57.335,2023,856,0,340,78.0612,2022,507,1,128,86.0943,2018,240,0,343,93.8413,2019,537,1,389,3.1262,2024,377,1,393,68.1358,2022,293,0,229,52.2671,2020,601,1,81,71.0462,2023,610,0,257,71.6916,2018,764,1,6,83.0219,2019,739,0,364,49.1545,2018,521,1,146,38.7768,2023,903,1,86,48.2355,2023,478,0,10,60.0645,2021,558,0,49,51.997,2023,544,0,30,68.7592,2018,131,1,101,91.4483,2023,878,0,486,88.0248,2021,599,1,255,67.7937,2019,92,1,42,6.0513,2023,403,0,336,92.4016,2019,585,0,199,71.6468,2018,0,0,31,46.115,2023,963,0,385
+602,C301,45,Phone,Electronics,42000.0,1.0,2024-01-19,0,47,42.714,2022,388,1,439,23.88,2023,709,0,229,99.5758,2019,551,1,397,39.5286,2019,210,1,49,0.5169,2023,86,1,120,23.3109,2020,477,1,128,71.8955,2023,922,1,296,46.6067,2021,11,0,22,44.3949,2021,69,1,10,56.5891,2018,529,1,82,54.0554,2020,484,1,372,41.1074,2020,447,0,384,4.2352,2021,352,0,61,85.8191,2019,861,0,183,94.536,2019,198,1,163,47.4782,2021,688,0,249,32.0634,2022,119,0,394,66.1202,2024,356,1,105,91.4205,2024,602,0,75,56.426,2021,618,1,349,68.7099,2018,689,0,148,38.807,2020,35,1,98,46.647,2020,100,1,36,39.199,2024,662,0,98,45.8999,2024,419,0,18,2.7541,2019,886,1,187,69.9959,2018,15,0,451,78.0545,2019,906,1,231,4.3398,2024,734,0,203,44.1864,2024,758,1,479,31.4024,2018,251,1,115,24.5504,2024,925,1,256,34.494,2018,428,0,372,91.0575,2020,52,1,206,31.6637,2023,825,0,94,7.1261,2024,36,1,200,31.7797,2020,354,1,313,14.6798,2020,931,1,269,46.5135,2022,450,1,331,60.4155,2018,377,1,357,79.7807,2021,595,0,451,90.5237,2020,100,0,87,20.6699,2023,620,1,395,6.0762,2024,697,0,420,18.3574,2024,507,0,459,46.4478,2023,809,0,316,62.3586,2018,186,0,99,55.6374,2020,818,1,38,56.7895,2021,882,0,342,43.1359,2019,70,0,285,50.5273,2018,918,1,277,15.4151,2022,129,0,394,1.632,2018,584,0,66,10.4551,2022,999,0,370,97.1911,2021,931,1,356,50.9879,2019,139,1,114,6.1589,2019,250,1,34,29.8372,2021,410,0,387,13.5284,2020,138,0,43,98.7983,2023,622,0,184,0.75,2019,991,1,291,57.1036,2021,762,1,325,66.3414,2023,951,0,386,20.4953,2021,55,0,66,20.6337,2019,412,0,214,83.608,2020,167,0,12,93.8023,2023,18,0,70,5.5315,2024,942,1,350,61.1375,2018,304,0,329,46.1645,2021,183,0,111,58.3231,2021,29,0,215,14.455,2020,219,0,329,19.4221,2019,323,1,443,40.7827,2023,42,1,170,15.199,2021,580,1,273,79.2681,2020,378,1,398,80.2092,2021,740,1,185,98.3139,2021,444,0,220
+603,C302,60,Headphones,Electronics,7900.0,1.0,2024-02-02,1,162,89.2765,2022,442,0,473,74.9081,2023,335,1,53,70.9068,2019,583,1,108,8.2629,2019,982,0,383,18.4114,2020,843,1,269,15.2606,2024,638,0,478,50.1232,2021,625,1,55,12.2749,2023,556,1,104,94.3254,2021,461,0,436,62.1067,2021,88,0,453,63.8216,2021,431,0,282,79.5521,2023,542,1,278,14.6176,2018,143,1,256,33.4572,2018,470,0,421,16.3202,2023,160,0,52,97.5301,2021,651,1,328,88.0274,2021,719,1,250,0.2501,2018,666,1,349,35.597,2023,523,1,279,97.0134,2020,98,1,382,92.4716,2021,862,1,115,5.2908,2023,60,1,387,74.2714,2022,621,0,154,41.3644,2023,135,1,284,92.996,2018,978,1,198,38.6482,2019,982,0,218,86.1945,2023,768,1,220,37.7807,2020,263,1,351,66.6189,2018,339,1,423,66.4048,2024,821,1,41,41.3513,2018,221,1,96,88.633,2019,968,1,499,36.1479,2020,724,1,381,49.5031,2024,555,1,67,75.8066,2022,844,0,256,69.888,2022,694,1,204,68.7238,2022,154,1,222,3.2801,2018,395,0,377,3.889,2023,220,1,21,68.6911,2024,151,0,162,74.0095,2018,778,1,135,61.5566,2022,203,0,388,42.4485,2019,850,1,297,53.4918,2019,249,1,221,6.7502,2020,525,1,192,66.525,2024,639,1,1,55.1035,2022,819,0,168,86.86,2024,436,1,456,42.8609,2020,562,0,496,24.9022,2019,960,0,289,68.7651,2020,366,1,166,39.3911,2018,80,0,260,89.0157,2024,202,0,194,42.2193,2018,396,1,165,97.1287,2019,866,1,220,54.5223,2022,826,1,461,29.1912,2019,737,0,31,27.4888,2021,369,0,498,66.5258,2018,523,1,332,46.2714,2024,938,0,67,74.6614,2020,79,1,333,69.2029,2018,545,1,124,46.8531,2022,896,0,13,80.1894,2019,962,1,149,72.5293,2023,210,0,399,26.3104,2019,959,0,395,71.4436,2020,350,0,81,28.6851,2024,461,0,478,80.0292,2020,402,0,495,37.5534,2021,718,0,343,46.0989,2019,367,0,399,13.4429,2019,806,1,439,53.7388,2018,702,1,196,72.4649,2021,393,1,45,10.256,2019,295,1,167,71.4386,2020,179,0,43,24.2341,2022,798,0,156,57.7549,2018,19,0,452
+604,C303,30,Camera,Electronics,41000.0,1.0,2024-02-13,1,265,71.8898,2018,336,1,98,88.4471,2023,961,1,456,94.725,2023,907,1,367,93.0463,2022,787,1,404,4.7322,2022,214,0,317,30.0124,2022,595,0,19,59.5029,2023,721,1,86,5.7787,2023,30,1,14,80.266,2022,826,0,96,37.1809,2022,274,0,268,84.2711,2019,411,1,404,43.2429,2020,485,1,219,34.2227,2019,267,0,138,78.0429,2018,746,1,142,44.134,2022,993,0,235,7.0395,2020,258,1,129,64.9228,2024,182,1,342,44.0091,2023,197,1,362,62.7104,2020,224,0,353,56.5413,2020,226,0,461,38.0537,2020,80,1,136,64.1519,2021,398,0,132,89.0276,2020,713,0,215,68.9897,2022,966,1,290,62.9747,2021,134,1,246,60.1625,2023,165,1,284,19.5553,2020,127,1,77,9.4002,2024,379,1,172,2.0021,2018,313,0,187,44.2648,2018,440,1,148,65.3153,2023,961,0,96,88.6686,2018,948,1,427,0.4257,2020,36,1,193,97.9875,2019,805,0,279,83.0474,2018,552,0,468,16.2863,2022,929,0,488,34.4021,2023,504,1,181,84.9543,2024,52,0,40,79.5751,2020,262,1,190,13.2315,2023,831,0,404,9.3724,2022,654,1,86,13.9261,2018,312,0,300,85.8559,2019,510,1,355,77.0651,2019,935,1,260,84.5444,2020,314,1,212,90.4273,2019,376,1,137,9.7238,2020,722,1,74,52.4767,2019,181,1,219,54.3066,2021,716,1,402,4.1763,2018,126,0,373,5.7149,2018,65,0,358,59.4006,2019,65,0,448,36.744,2018,298,0,57,6.3114,2018,671,0,263,2.1276,2023,142,0,397,48.5795,2022,48,0,313,17.4253,2022,660,1,414,6.2271,2024,791,0,289,73.2198,2023,823,0,56,14.5921,2021,672,1,389,41.8046,2019,708,1,466,19.7705,2021,489,0,479,81.2941,2019,341,0,135,66.541,2022,497,0,277,39.9045,2024,732,0,89,9.289,2018,141,1,327,50.7167,2024,759,1,61,58.6488,2018,551,1,26,46.5216,2020,788,1,219,63.8497,2019,620,1,225,99.5888,2021,941,1,172,56.8707,2024,495,0,424,74.1335,2024,905,0,179,77.6533,2018,433,0,311,31.1968,2021,914,1,170,27.8721,2020,735,0,364,11.9546,2023,913,0,397,44.7967,2022,966,1,16
+605,C304,30,Tablet,Electronics,30500.0,,2024-02-25,0,326,6.9968,2021,611,1,415,32.0166,2024,554,0,320,25.1378,2021,647,1,100,54.2937,2019,997,1,18,54.8351,2018,690,0,171,88.5219,2024,133,1,265,10.2271,2022,218,1,260,11.0745,2020,761,1,417,2.3196,2022,50,1,197,55.1762,2023,69,1,85,19.8423,2020,859,0,414,86.7129,2023,466,0,355,62.7176,2020,522,1,235,43.842,2018,817,1,92,44.3295,2023,571,1,24,39.7175,2018,961,1,103,1.8432,2020,333,0,199,80.7418,2021,581,0,164,30.9832,2022,646,1,298,0.1635,2020,249,1,273,69.1898,2018,51,0,122,21.2284,2021,401,1,88,94.0439,2024,959,1,219,74.1111,2022,645,1,231,92.3911,2022,828,1,205,88.9204,2020,747,1,424,61.6248,2023,635,0,388,29.5128,2018,162,0,356,83.6259,2024,817,0,114,85.6201,2021,133,0,461,11.9734,2023,836,1,189,36.818,2020,655,1,153,83.7223,2018,32,0,161,63.3442,2023,308,1,107,58.7303,2021,731,1,288,44.5737,2023,601,1,75,10.9557,2023,846,1,273,46.9717,2019,996,0,430,12.0217,2021,692,0,6,32.952,2018,871,1,33,9.3636,2020,757,0,313,3.2618,2020,363,1,51,93.964,2018,444,0,111,28.0697,2021,585,1,465,94.2951,2020,725,1,345,45.4705,2019,200,1,407,20.1806,2023,922,0,130,45.487,2023,156,1,64,29.3621,2021,361,0,496,39.7893,2018,607,0,191,24.8653,2021,749,0,422,7.1596,2024,349,0,132,24.5381,2018,4,1,391,24.3354,2024,873,1,131,36.8383,2023,235,1,434,9.7381,2018,86,0,122,40.1331,2018,400,1,248,98.9428,2021,836,1,105,81.3327,2020,844,0,275,48.61,2022,587,0,272,16.9499,2021,274,1,148,63.7686,2023,86,0,408,43.9928,2018,271,0,97,80.2777,2018,832,0,289,24.0179,2024,598,1,204,97.1628,2019,486,0,275,73.7156,2020,708,0,256,17.6641,2024,473,0,459,51.7645,2022,156,0,71,19.7144,2020,459,0,231,33.6317,2018,507,0,494,44.7828,2019,63,0,113,69.0979,2022,154,1,405,53.1931,2018,966,0,45,6.5847,2020,650,0,325,35.3675,2020,674,0,413,63.8429,2024,519,1,12,24.9617,2021,230,1,449
+606,C305,45,Serum,Beauty,1300.0,2.0,2024-03-07,1,34,45.6125,2020,830,0,41,90.2784,2019,48,1,474,89.2955,2021,735,0,25,24.3007,2024,45,0,130,31.8463,2024,549,1,468,33.7572,2021,919,1,294,36.4716,2018,110,1,388,86.6995,2019,713,0,411,62.6685,2021,512,1,345,80.2094,2020,889,1,236,39.1209,2023,557,0,256,7.3639,2019,371,0,248,98.0003,2018,710,1,241,99.422,2023,199,1,169,72.3343,2021,125,1,209,45.1824,2022,19,0,499,79.1581,2024,540,1,267,94.8339,2019,536,0,234,40.2764,2024,260,1,202,54.6238,2022,542,1,96,67.1621,2021,983,1,466,68.2438,2023,463,0,129,81.5038,2022,909,1,496,47.6105,2018,396,0,141,19.3653,2021,198,0,5,0.4034,2024,778,0,158,38.7775,2024,61,0,148,56.7554,2022,883,0,147,86.1987,2019,821,0,400,15.8295,2022,831,0,67,4.6801,2023,583,0,160,26.8546,2022,322,1,294,32.9468,2020,82,1,180,62.9521,2022,716,1,250,96.6011,2023,456,0,189,12.0439,2021,492,0,17,78.8695,2020,303,0,142,11.3961,2020,290,0,404,62.6016,2022,72,1,5,83.4673,2020,278,1,228,61.3194,2019,546,1,418,17.7742,2024,644,0,331,14.4934,2018,934,1,220,65.5122,2020,428,0,21,71.4861,2023,835,1,43,3.1781,2018,227,1,323,67.1978,2022,807,1,429,75.0757,2023,622,1,383,10.9511,2018,236,0,59,30.7807,2023,459,0,397,93.1844,2023,140,0,448,91.817,2020,764,1,95,42.9711,2019,90,1,161,69.223,2018,741,1,317,80.4519,2023,126,0,326,32.5647,2024,337,1,498,74.0223,2022,522,1,111,43.5206,2018,507,0,156,1.336,2019,185,0,224,64.7166,2022,833,1,250,52.3901,2022,924,1,113,68.1389,2021,13,0,280,4.7899,2019,174,0,178,91.6959,2021,336,1,382,14.2737,2024,437,0,434,88.7794,2024,186,1,277,96.3444,2022,697,1,53,53.584,2024,272,0,195,88.535,2019,108,0,277,3.8178,2023,673,1,239,53.2048,2018,287,1,450,25.8803,2022,820,0,306,28.9628,2020,124,1,234,57.2559,2018,813,0,13,84.3891,2019,428,0,311,36.2306,2019,364,1,67,75.298,2021,140,1,436,57.1664,2023,785,1,26
+607,C306,60,Desk,Home,12500.0,1.0,2024-03-18,0,345,53.8047,2022,804,0,253,33.0549,2019,44,1,79,22.5355,2020,355,1,377,62.2929,2021,422,0,70,80.0632,2022,545,1,173,50.0556,2024,2,0,97,1.6601,2022,665,1,187,54.262,2023,951,0,426,52.6644,2021,551,1,311,97.2634,2021,138,1,224,19.2569,2022,147,0,72,80.7577,2022,918,1,104,8.4668,2018,122,1,49,63.8585,2021,100,1,466,23.732,2024,895,0,494,23.6486,2020,198,1,197,77.2568,2018,326,0,226,65.3289,2023,647,1,82,86.3155,2024,292,1,151,66.6084,2022,959,0,378,45.3949,2018,424,1,246,94.8786,2018,859,0,464,86.5524,2023,793,1,206,85.1208,2022,92,0,353,91.7173,2019,565,1,345,6.7008,2020,499,0,211,73.9902,2022,359,1,417,98.8436,2023,592,1,408,54.5499,2019,90,0,245,91.0177,2024,274,1,158,27.5686,2023,279,1,106,70.9682,2023,997,0,299,72.2503,2024,815,1,230,43.0567,2018,277,0,121,30.1199,2020,429,0,31,93.3484,2023,575,1,62,53.1013,2020,599,0,467,79.9392,2024,511,1,273,2.2151,2021,545,0,352,12.463,2023,755,0,441,55.8703,2019,461,1,414,83.714,2019,676,0,90,87.0907,2019,315,0,262,63.1309,2018,772,0,336,45.0808,2022,402,1,130,32.6755,2024,152,1,31,27.8969,2020,398,0,367,66.4586,2022,256,1,189,69.9446,2024,687,0,197,44.6948,2022,529,0,470,13.8614,2024,914,0,139,9.179,2023,536,0,291,96.6608,2021,85,1,348,54.4408,2020,650,0,368,14.9704,2021,841,1,39,98.6302,2020,59,0,23,30.4665,2018,532,1,404,24.1201,2022,200,1,105,7.258,2023,259,0,22,33.6908,2021,940,1,167,24.4553,2020,650,0,108,51.1963,2019,735,1,237,49.5187,2023,417,0,490,31.9363,2019,499,0,79,18.7186,2018,612,0,379,7.7791,2020,505,0,392,61.4556,2018,785,0,399,77.1285,2019,34,0,400,55.4015,2021,779,0,29,89.2132,2020,497,0,430,4.7329,2021,775,0,297,40.4557,2023,170,1,158,48.8236,2022,690,1,278,2.814,2024,435,0,39,70.3749,2019,567,1,247,34.3636,2020,721,0,28,6.5223,2023,506,1,371,49.2449,2019,579,0,268
+608,C307,30,Monitor,Electronics,18500.0,1.0,2024-04-04,0,28,81.0418,2019,329,1,498,26.2027,2018,266,1,423,70.1049,2021,140,0,209,6.8436,2024,969,0,278,41.0682,2020,121,1,167,40.8495,2024,996,1,117,28.2872,2019,114,0,477,41.7888,2022,110,1,43,12.8289,2018,497,0,453,39.3874,2023,626,0,480,88.3355,2019,648,0,168,9.4579,2021,614,0,158,56.0907,2019,58,0,194,53.0077,2022,737,0,395,55.5247,2021,901,0,30,17.7533,2024,457,1,312,83.9569,2020,866,0,211,10.5889,2019,757,1,177,73.1204,2022,589,0,384,62.1249,2019,684,1,314,34.0848,2024,892,0,441,10.3404,2021,255,0,70,97.9444,2021,419,1,327,14.9061,2024,860,1,163,46.0826,2021,995,0,437,38.0631,2021,273,1,339,98.7845,2021,311,1,191,46.2244,2020,78,0,403,41.9937,2019,354,0,121,80.8868,2020,191,1,381,87.1074,2024,532,1,465,71.488,2024,110,0,92,42.7386,2024,468,0,137,67.7976,2019,342,0,350,6.3912,2024,600,1,318,8.9182,2021,918,1,294,4.7646,2021,632,0,382,30.6837,2020,735,0,424,24.1036,2021,572,1,474,19.2626,2020,713,0,1,16.7838,2020,696,0,495,31.3454,2019,162,1,121,63.3282,2019,941,1,187,71.3009,2023,527,0,418,47.3461,2024,547,0,324,47.9481,2023,238,1,488,20.9736,2018,479,0,52,49.8454,2019,825,0,57,72.1301,2021,202,0,12,30.3385,2021,104,1,290,58.5497,2022,119,1,238,94.3822,2022,465,1,4,56.6079,2023,410,0,266,52.7364,2018,568,1,19,24.1914,2018,955,1,138,41.3301,2021,190,0,337,5.0866,2021,304,1,346,58.6776,2024,503,1,27,61.9477,2024,180,1,213,7.9824,2018,910,1,353,1.6168,2023,270,1,317,4.7318,2021,323,1,471,1.5135,2019,993,0,399,43.0371,2022,487,1,211,7.0348,2023,433,0,63,79.4095,2021,310,1,128,41.357,2024,578,1,265,66.0135,2019,585,1,134,97.1917,2021,526,0,10,53.2297,2024,761,0,204,0.6791,2020,372,1,318,85.7888,2021,715,1,457,46.2532,2024,846,1,288,28.4242,2023,302,1,236,98.0117,2022,149,1,368,74.8703,2018,502,0,195,34.3001,2018,203,1,17,61.6395,2022,715,0,336
+609,C308,75,Shoes,Fashion,3200.0,2.0,2024-04-15,0,175,47.8909,2021,202,1,200,87.079,2021,457,0,477,63.5543,2019,250,0,425,50.7921,2018,251,1,140,18.531,2024,463,1,62,23.7837,2018,866,1,283,63.5852,2018,346,0,182,51.6689,2018,303,0,315,5.8712,2021,37,0,275,47.6738,2019,98,1,213,33.3327,2019,840,1,258,1.7105,2018,346,0,196,1.4748,2018,290,0,397,69.0669,2022,932,0,147,88.4798,2019,777,0,261,97.003,2024,819,1,268,50.5681,2024,515,0,456,81.5716,2022,624,0,299,69.8387,2024,791,0,231,16.0837,2020,943,1,426,0.4444,2022,977,1,309,48.7493,2019,879,1,104,30.4319,2020,653,0,378,76.6094,2019,137,1,68,61.5046,2022,25,0,480,67.6713,2020,494,1,17,96.8785,2022,736,0,167,72.2481,2022,747,0,498,66.0362,2024,618,1,192,64.6547,2022,58,1,122,73.3119,2018,725,0,70,56.1347,2023,588,1,377,81.9548,2018,119,1,308,7.2491,2018,303,0,93,91.2426,2019,819,0,154,51.0652,2023,120,1,249,35.6257,2018,166,0,469,22.5172,2023,953,1,289,52.5254,2023,495,1,418,19.1392,2024,877,1,372,48.4309,2023,263,0,244,21.549,2023,153,0,270,26.1433,2018,20,1,212,12.3694,2018,538,1,476,41.3417,2023,658,0,317,82.8539,2022,555,1,26,25.5726,2018,893,0,329,85.4137,2018,856,1,190,3.9267,2021,915,0,476,93.9418,2018,928,1,233,52.0063,2023,243,0,466,84.561,2018,312,0,56,3.6933,2022,743,0,214,31.2558,2024,343,0,104,68.1593,2024,567,0,18,4.8924,2018,727,0,233,84.0284,2022,730,0,41,14.0718,2018,756,0,485,33.3656,2024,425,0,175,89.6337,2023,281,1,387,46.8855,2020,63,1,406,3.8396,2024,384,0,484,10.7994,2019,985,1,433,41.9599,2019,150,0,71,11.2811,2019,934,1,191,56.3049,2023,50,0,10,10.3323,2019,995,0,291,67.7881,2018,356,0,402,41.1253,2023,493,1,93,28.884,2020,351,1,222,23.0337,2021,753,1,312,39.0545,2019,831,0,107,1.498,2023,185,0,298,71.996,2023,885,1,333,32.0281,2023,799,1,358,69.2373,2021,22,1,29,27.2678,2021,261,1,292,85.948,2024,673,1,317
+610,C309,90,Keyboard,Electronics,3700.0,1.0,2024-04-28,1,377,36.3951,2018,559,1,82,88.6105,2022,874,1,356,61.5928,2022,58,1,56,76.6528,2023,771,0,356,91.2929,2018,223,0,426,84.918,2024,918,1,233,12.7016,2020,679,0,455,59.6133,2024,379,1,454,3.9599,2021,530,1,169,2.2761,2022,76,1,165,84.3403,2023,243,1,206,64.0305,2022,256,1,353,86.6067,2022,983,1,166,35.3545,2022,958,1,397,94.8278,2021,416,1,85,2.8694,2021,868,0,38,76.396,2018,799,0,475,25.3447,2021,481,1,421,65.0742,2023,942,0,460,94.3213,2021,11,0,36,73.0891,2023,953,1,60,22.1366,2024,155,1,347,19.9297,2022,160,1,287,22.6341,2022,546,0,156,87.2321,2022,748,1,202,13.4762,2022,15,0,379,5.5991,2018,277,1,448,60.948,2022,921,1,64,61.4745,2019,738,1,126,7.2064,2020,270,0,323,87.3784,2021,86,1,364,43.203,2021,64,0,449,87.2666,2024,589,0,316,34.6785,2021,652,1,393,66.5856,2021,660,1,384,5.7473,2021,113,0,98,79.2803,2024,555,0,359,83.4134,2023,904,0,495,88.8027,2022,42,0,453,93.4389,2021,151,0,86,76.0442,2018,263,0,103,2.7304,2019,174,0,270,68.4925,2019,557,1,167,97.1451,2019,467,0,199,46.1039,2020,254,1,128,68.0284,2021,95,1,298,88.7513,2023,117,1,422,27.1523,2021,293,0,424,80.4072,2019,749,0,46,81.709,2021,768,1,68,48.6141,2020,392,0,193,12.8098,2022,389,1,45,67.9927,2020,687,1,56,65.9521,2024,991,0,121,62.7859,2020,449,1,261,87.688,2019,541,1,105,61.8234,2023,43,1,396,14.6636,2022,920,1,195,24.6699,2024,632,0,417,76.7522,2018,422,1,18,91.2658,2020,635,0,124,7.7524,2019,914,0,398,67.0776,2019,103,0,89,99.2569,2019,549,1,338,74.9316,2021,92,0,272,50.1049,2021,77,0,156,44.5596,2023,692,0,167,48.6705,2020,763,0,94,15.8735,2019,349,1,431,3.4488,2024,660,0,405,82.7042,2019,264,0,428,49.8136,2019,338,0,209,57.9633,2020,606,0,158,80.1876,2024,664,0,100,67.1782,2023,196,1,283,47.4361,2018,716,1,320,45.8388,2023,777,1,290,7.3426,2022,771,0,383
+611,C310,30,Phone,Electronics,42000.0,1.0,2024-05-09,0,272,91.1024,2023,609,1,389,53.5688,2023,754,0,416,28.2974,2024,983,1,40,63.3961,2018,861,1,123,53.4032,2021,504,1,432,96.1176,2018,66,1,288,31.6162,2021,260,0,215,6.1848,2018,581,0,111,89.3017,2021,303,1,410,9.7654,2019,690,1,173,29.1723,2023,359,0,188,15.7017,2018,921,1,375,66.0324,2023,979,0,413,2.0443,2023,967,0,28,22.7686,2020,7,1,38,23.1003,2021,747,1,183,0.4943,2018,628,1,393,58.1929,2023,122,1,498,62.4852,2024,727,1,154,10.8294,2023,69,1,264,97.5929,2023,529,0,61,65.6891,2024,910,0,430,64.2018,2019,51,0,178,28.9301,2022,235,1,158,73.1798,2022,689,0,147,31.5309,2024,63,0,403,68.8166,2023,559,0,332,30.0548,2024,825,1,263,78.507,2018,386,0,205,28.4976,2024,903,0,264,33.5015,2023,495,1,247,55.1445,2019,210,0,436,9.3946,2021,608,1,153,78.919,2022,109,0,102,41.3952,2022,275,0,171,14.1166,2022,145,1,308,11.8544,2024,988,0,364,38.205,2021,731,1,112,86.5166,2022,909,1,161,87.3357,2021,70,0,273,5.3497,2021,218,0,306,93.1111,2022,992,1,477,16.9701,2024,16,1,379,83.157,2023,749,0,188,76.4299,2023,454,0,187,91.9783,2018,797,0,60,44.6919,2023,154,1,227,74.7193,2019,256,1,357,15.4229,2023,108,0,237,36.0042,2019,371,0,380,31.3895,2019,590,1,230,63.7532,2023,113,1,351,43.8288,2024,122,0,229,74.4885,2021,395,1,354,30.1564,2023,818,0,383,51.1065,2019,787,1,451,20.0141,2021,155,0,482,59.0639,2022,448,0,33,69.7741,2021,814,0,434,55.3449,2018,436,0,413,70.5357,2021,115,1,209,27.0294,2019,984,0,157,1.5446,2020,227,0,346,87.2463,2020,627,1,74,11.84,2018,581,0,322,78.6573,2022,390,0,374,99.8867,2021,353,1,152,5.6584,2022,318,0,327,54.731,2024,195,1,447,86.7438,2021,743,0,15,93.1257,2023,865,1,340,98.0048,2020,700,1,446,54.2752,2024,895,1,283,71.2797,2019,845,0,268,35.0642,2018,3,1,383,22.6124,2018,791,0,159,4.0251,2020,506,0,230,67.2233,2023,534,0,304
+612,C311,45,Headphones,Electronics,8000.0,1.0,2024-05-21,0,114,25.1746,2023,350,1,378,60.6678,2019,185,1,435,95.9221,2022,987,0,103,84.1202,2020,586,1,291,37.9491,2021,930,0,451,75.0386,2021,670,0,173,40.845,2024,345,0,388,41.7529,2021,501,1,205,75.9259,2020,286,1,355,68.8055,2019,585,0,194,46.8454,2023,597,1,322,61.7538,2019,539,1,418,82.3195,2019,140,0,8,65.9815,2020,825,0,404,89.5488,2021,831,1,66,19.5934,2021,657,1,198,75.8448,2018,793,0,55,75.7624,2024,145,0,195,28.4838,2020,727,1,209,94.1845,2018,838,1,183,71.9382,2022,198,0,498,69.5826,2020,384,1,451,92.0282,2022,46,0,112,74.1567,2020,673,0,41,61.211,2019,848,0,354,88.0356,2021,841,0,357,36.4826,2019,455,1,250,30.0309,2018,261,0,286,60.0315,2024,29,0,265,16.4333,2024,260,1,455,23.8735,2019,943,0,332,18.7975,2024,878,0,48,12.5775,2019,637,1,492,30.32,2024,124,0,347,9.34,2019,998,1,34,22.063,2018,831,0,421,52.4875,2023,513,1,219,49.6862,2018,76,0,234,82.4235,2019,538,1,104,46.2018,2023,264,1,66,6.036,2021,244,1,24,6.4716,2024,994,0,13,23.9393,2021,900,0,360,20.2078,2018,5,1,117,97.2884,2018,648,1,80,68.9284,2024,44,1,80,17.6141,2020,765,0,233,76.8382,2021,986,1,393,14.1893,2021,856,0,357,68.0392,2024,21,1,278,93.1926,2021,9,0,91,19.7164,2018,162,0,487,52.228,2018,748,1,387,39.743,2021,495,1,477,9.6685,2022,117,0,257,89.2016,2021,596,1,181,10.9361,2018,118,0,447,26.3534,2020,808,0,358,83.8325,2021,50,1,423,46.3884,2023,592,1,60,82.5943,2018,265,0,179,17.1842,2021,820,0,59,1.235,2024,342,1,321,4.6736,2022,677,0,356,31.3905,2022,701,0,11,6.0762,2024,947,1,147,91.0698,2024,97,0,401,4.2701,2021,994,0,110,78.5132,2022,435,0,108,98.0623,2022,980,0,394,25.1927,2022,579,1,154,21.9261,2020,328,1,107,66.8068,2020,584,1,85,26.7067,2021,684,1,44,32.0077,2023,29,0,128,41.4427,2022,141,1,56,42.6543,2019,232,0,41,63.3698,2019,961,1,226

ceo_brief_env/tasks/easy_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,1.0
+total_revenue,58998.58
+top_category,Electronics
+projection_next_quarter,102114.97
+variance_pct,126.92
+break_even_units,14.12
+memo_score,0.725

ceo_brief_env/tasks/easy_brief/metadata.json

@@ -0,0 +1,20 @@
+{
+  "title": "Revenue sanity check",
+  "difficulty": "easy",
+  "max_steps": 8,
+  "instruction": "Is our raw sales file trustworthy, what's next quarter's revenue looking like, and send the team a one-paragraph update.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "hr"
+  ],
+  "memo_audience": "team",
+  "hr_required_terms": [
+    "revenue",
+    "projection",
+    "team"
+  ],
+  "plan_value": 26000,
+  "fixed_cost": 12000,
+  "unit_margin": 850
+}

ceo_brief_env/tasks/easy_brief/raw.csv

@@ -0,0 +1,9 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+101,C001,25,Perfume,Beauty,,3,2024-01-05,1,151,6.6543,2022,742,1,264,83.0928,2022,831,1,381,87.079,2021,554,1,124,94.9615,2021,99,1,141,73.5733,2018,199,1,103,50.7921,2018,997,1,402,25.8798,2020,169,0,464,54.928,2023,341,0,291,23.7837,2018,133,1,295,58.8683,2021,318,1,7,11.0141,2021,520,1,204,51.6689,2018,761,1,92,82.1888,2018,407,1,266,51.2716,2020,393,1,286,47.6738,2019,69,0,319,47.193,2019,452,1,119,55.7923,2022,828,0,201,1.7105,2018,466,0,208,49.6581,2023,377,0,18,71.041,2018,470,1,250,69.0669,2022,817,0,144,33.6834,2024,906,1,83,12.5833,2020,47,1,234,97.003,2024,961,0,335,99.9777,2023,781,1,455,54.0921,2019,397,1,25,81.5716,2022,581,1,90,46.7452,2023,371,1,492,26.0389,2019,596,0,477,16.0837,2020,249,1,254,19.1717,2018,155,1,353,98.3351,2022,243,0,299,48.7493,2019,401,0,374,25.796,2020,396,0,169,90.9689,2019,438,1,474,76.6094,2019,645,1,255,28.0711,2019,692,0,294,19.8954,2020,410,1,356,67.6713,2020,747,1,380,31.4986,2023,392,1,343,6.1121,2018,776,0,139,72.2481,2022,162,0,66,81.6074,2022,419,1,441,9.0394,2023,226,0,221,64.6547,2023,133,1,487,31.6536,2024,871,0,20,27.961,2023,377,0,443,56.1347,2023,655,1,294,59.8245,2020,427,0,1,81.5397,2018,320,0,489,7.2491,2022,308,1,12,24.1107,2019,63,0,124,42.9194,2023,575,1,82,51.0652,2018,601,1,479,12.3134,2019,47,1,28,59.9933,2023,547,1,424,22.5172,2020,996,1,399,54.6272,2024,241,1,381,54.5808,2022,11,1,67,19.1392,2019,871,1,293,88.2664,2018,167,0,379,46.1159,2023,626,0,70,21.549,2020,363,1,164,17.8916,2018,633,0,25,31.5676,2021,220,0,385,12.3694,2022,585,1,413,67.2601,2024,473,1,367,40.2784,2023,690,1,452,82.8539,2021,200,0,314,6.051,2023,209,0,3,39.8109,2024,260,1,262,85.4137,2021,156,0,467,37.8447,2018,721,0,250,68.7795,2020,993,1,21,93.9418,2020,607,0,475,94.184,2020,585,1,391,91.4982,2023,844,0,297,84.561,2024,349,1,58
+102,C002,45,Headphones,Electronics,2950.0,2,2024-01-12,0,329,5.4567,2018,478,1,5,32.9736,2022,82,1,160,88.6105,2022,48,0,390,84.7199,2019,635,0,333,14.0566,2023,48,1,271,76.6528,2023,45,1,54,55.6728,2021,185,1,64,12.1544,2020,937,1,442,84.918,2024,919,1,400,23.3943,2018,635,1,155,11.4967,2018,776,0,56,59.6133,2024,713,0,390,8.5232,2018,58,0,249,49.7135,2023,690,0,276,2.2761,2022,889,1,483,96.1917,2019,333,0,374,64.8193,2021,512,1,433,64.0305,2022,371,0,404,31.5316,2022,14,1,326,5.8568,2023,482,0,219,35.3545,2022,199,0,471,79.117,2021,884,0,271,90.1492,2022,418,0,199,2.8694,2021,19,0,104,62.4036,2018,505,1,71,86.6347,2020,533,1,403,25.3447,2021,536,1,333,35.387,2020,698,1,366,58.9308,2018,404,1,1,94.3213,2021,542,1,28,62.8638,2022,4,0,418,89.2162,2021,931,0,106,22.1366,2024,463,1,384,13.851,2024,304,0,319,41.9916,2023,992,1,370,22.6341,2022,396,0,102,32.5837,2024,615,1,327,99.5718,2021,9,0,444,13.4762,2022,778,1,25,67.9196,2022,968,0,11,31.1943,2024,295,1,148,60.948,2022,883,1,431,99.6608,2019,530,1,106,61.8301,2024,800,0,112,7.2064,2020,831,1,225,24.3231,2021,713,0,242,72.5311,2022,318,0,349,43.203,2021,322,1,67,75.3554,2021,448,0,489,11.9982,2020,359,0,175,34.6785,2021,716,1,387,18.461,2021,364,1,226,81.9647,2023,378,1,490,5.7473,2021,492,1,244,49.7177,2023,59,0,494,16.627,2021,283,1,340,83.4134,2023,290,1,211,57.8393,2021,511,1,314,49.505,2019,9,0,446,93.4389,2021,278,1,223,74.3825,2019,570,0,41,26.3068,2018,836,1,254,2.7304,2019,644,0,80,54.031,2023,840,1,137,2.0184,2024,439,1,333,97.1451,2019,428,0,22,95.3483,2020,263,1,173,65.8071,2022,84,0,48,68.0284,2021,227,0,268,5.0386,2018,957,0,380,89.3062,2022,859,0,376,27.1523,2021,622,1,104,38.0364,2023,443,1,103,91.5052,2021,118,0,18,81.709,2021,459,0,362,46.5466,2024,465,0,203,24.3056,2018,896,1,9,12.8098,2022,764,0,456
+103,C003,60,Jeans,Fashion,1580.0,1,2024-02-02,1,54,75.3739,2019,160,1,377,55.9925,2022,506,1,34,53.5688,2023,44,1,3,71.1985,2024,465,0,472,5.8679,2024,754,1,347,63.3961,2018,422,0,313,71.2884,2021,214,0,223,22.366,2023,346,1,235,96.1176,2018,2,0,371,46.6825,2019,821,1,227,67.9788,2018,373,1,483,6.1848,2018,951,1,160,90.8118,2023,569,0,218,53.0183,2023,623,0,413,9.7654,2019,138,0,359,33.0132,2023,515,0,419,24.3553,2019,143,1,450,15.7017,2018,918,1,458,70.7117,2018,186,0,390,98.3891,2020,97,0,145,2.0443,2023,100,0,34,79.535,2019,137,1,179,41.6355,2023,988,1,397,23.1003,2021,198,1,472,7.4488,2022,524,0,3,79.9477,2024,451,1,412,58.1929,2023,647,1,435,84.3114,2024,856,1,475,94.2933,2022,301,1,285,10.8294,2023,959,1,381,7.0746,2019,637,0,393,95.3235,2018,492,1,220,65.6891,2024,859,1,86,69.4908,2020,657,1,132,16.0842,2020,410,0,26,28.9301,2022,92,1,410,31.1881,2020,106,0,317,74.8216,2020,689,1,450,31.5309,2024,499,0,203,75.7573,2023,174,1,422,27.7785,2020,834,0,121,30.0548,2024,592,1,444,52.5845,2023,660,0,39,38.6077,2018,490,1,428,28.4976,2024,274,1,139,52.7624,2018,733,0,415,49.5948,2021,211,1,184,55.1445,2019,997,1,200,87.3053,2023,819,0,273,60.8819,2022,459,0,317,78.919,2022,277,1,42,20.4355,2022,912,1,337,27.5169,2018,62,1,223,14.1166,2022,575,1,332,61.7015,2018,356,1,453,98.8982,2021,935,0,235,38.205,2021,511,0,475,22.3604,2018,525,1,286,90.9225,2023,704,0,165,87.3357,2021,755,0,104,54.5322,2018,484,0,247,21.832,2021,829,0,17,93.1111,2022,676,0,380,95.4845,2024,261,0,435,1.6942,2024,525,1,141,83.157,2023,772,1,178,37.5522,2024,413,0,374,45.437,2018,259,0,227,91.9783,2018,152,1,26,11.8771,2019,255,1,385,15.475,2022,733,1,227,74.7193,2019,256,1,364,71.5253,2020,39,0,361,10.8205,2019,393,0,199,36.0042,2019,529,0,101,76.1046,2019,520,0,139,59.0909,2022,276,1,36,63.7532,2023,536,0,264
+103,C003,60,Jeans,Fashion,1580.0,1,2024-02-02,1,123,22.6578,2021,363,0,423,35.0098,2021,997,0,451,60.6678,2019,266,1,445,87.0686,2024,615,1,320,98.715,2020,418,0,122,84.1202,2020,969,1,323,58.2904,2020,912,1,233,93.0111,2019,332,0,169,75.0386,2021,996,0,59,34.5087,2020,127,1,447,34.547,2019,955,0,433,41.7529,2021,110,1,459,41.0459,2022,39,1,200,28.6362,2022,906,1,401,68.8055,2019,626,0,24,38.7036,2020,843,0,423,59.7603,2023,334,0,37,61.7538,2019,614,1,325,83.5863,2024,866,0,111,14.0785,2020,388,1,497,65.9815,2020,737,1,4,80.8125,2023,948,1,267,83.1843,2024,59,0,226,19.5934,2021,457,1,263,39.6326,2023,763,1,34,79.3618,2020,421,1,474,75.7624,2024,757,1,229,38.9362,2020,650,0,180,72.7054,2024,768,1,273,94.1845,2018,684,0,487,36.492,2022,730,0,356,19.8507,2019,883,0,341,69.5826,2020,255,0,480,90.219,2023,199,0,340,4.6488,2018,653,0,238,74.1567,2020,860,1,118,8.1789,2019,872,1,101,84.8583,2021,875,0,3,88.0356,2021,273,1,2,71.3824,2020,55,0,73,45.5204,2022,380,1,284,30.0309,2018,78,1,36,13.8745,2023,949,0,485,65.8275,2019,241,0,169,16.4333,2024,191,1,497,45.6478,2019,736,1,104,4.8678,2018,930,0,142,18.7975,2024,110,1,237,51.7928,2022,94,0,119,65.3943,2023,273,0,264,30.32,2024,342,0,244,15.4404,2019,37,0,77,41.7061,2022,635,0,390,22.063,2018,918,0,194,61.8054,2021,89,1,413,36.4303,2023,765,0,287,49.6862,2018,735,0,4,0.7348,2023,503,0,95,18.4396,2023,949,0,384,46.2018,2023,713,0,320,91.2568,2019,11,1,283,66.0704,2022,991,1,33,6.4716,2024,162,0,25,36.8087,2022,351,0,244,85.5086,2021,374,0,13,20.2078,2018,527,1,361,1.2975,2019,595,0,65,98.3333,2022,647,0,42,68.9284,2024,238,1,426,27.2059,2023,579,0,211,73.7256,2019,102,1,142,76.8382,2021,825,1,375,82.9701,2022,261,1,259,52.2804,2019,23,0,84,68.0392,2024,104,0,431,58.2422,2023,826,0,219,77.0009,2019,475,0,429,19.7164,2018,465,0,174
+104,C004,90,Lamp,Home,870.0,2,2024-02-14,0,29,11.2991,2020,907,0,285,87.6315,2018,399,0,36,74.5718,2021,457,0,398,35.0699,2020,139,0,241,90.9415,2018,850,1,267,28.2215,2024,251,1,301,49.0841,2021,962,0,156,96.7308,2023,123,1,244,32.835,2022,866,1,490,47.5491,2021,338,1,107,93.7307,2023,363,1,384,49.8643,2024,303,1,40,2.8595,2021,699,1,169,85.1374,2019,549,0,498,5.3971,2019,98,1,187,11.7391,2023,650,0,492,84.6579,2021,516,1,487,30.7963,2022,346,0,207,7.1317,2022,582,0,38,23.1429,2024,794,0,306,28.3732,2022,932,1,104,2.6198,2023,579,1,472,81.2493,2023,522,0,453,45.9313,2020,819,1,123,72.9151,2019,102,0,378,34.7917,2018,913,1,442,96.3588,2023,624,0,484,79.7165,2021,713,1,438,12.6743,2019,462,1,108,52.8901,2024,943,1,488,44.3529,2024,877,0,324,48.6673,2022,618,1,5,29.3695,2021,879,1,199,45.5387,2018,1,1,374,81.9871,2024,756,0,370,86.3754,2021,137,1,60,22.1585,2021,722,1,298,19.3957,2021,960,1,388,20.3585,2018,494,1,154,78.7417,2020,900,1,103,25.009,2022,333,0,279,25.424,2018,747,0,73,31.2542,2024,614,1,423,96.7819,2021,383,0,79,98.1138,2021,58,0,246,95.9976,2018,873,1,368,60.569,2019,138,0,135,20.5493,2022,588,1,268,13.6289,2020,872,0,371,57.1197,2018,615,1,315,89.8031,2019,303,1,302,10.2863,2024,665,1,255,31.6239,2019,307,0,61,65.3571,2024,120,1,243,98.1271,2023,792,1,414,93.0922,2024,938,0,57,8.9959,2024,953,0,269,0.2961,2022,888,1,376,75.9751,2022,837,1,417,88.9471,2019,877,1,200,98.955,2022,760,0,57,93.1981,2024,488,1,89,58.4507,2021,153,1,485,64.8237,2019,684,0,276,87.6392,2021,423,0,327,28.4076,2022,538,1,241,7.9994,2023,461,0,423,79.3825,2023,634,0,16,93.095,2024,555,1,224,3.7138,2022,887,1,252,33.5353,2022,658,1,375,85.9028,2019,856,1,246,18.9959,2018,804,0,137,5.1409,2019,953,0,154,22.7695,2024,928,0,139,89.972,2024,486,1,283,77.6769,2021,932,0,459,11.0417,2021,312,0,74
+105,C005,120,Racket,Sports,1730.0,1,2024-03-03,1,47,42.714,2021,911,0,194,57.2446,2024,162,1,165,76.493,2022,874,1,229,99.5758,2019,282,0,276,56.4798,2023,111,0,311,63.1296,2019,771,0,49,0.5169,2023,534,1,44,34.2597,2019,852,0,250,57.3504,2021,918,1,128,71.8955,2018,316,0,461,55.3715,2022,910,0,271,79.0221,2018,379,0,22,44.3949,2021,893,1,35,31.8996,2018,337,1,486,59.7847,2021,76,1,82,54.0554,2019,291,1,242,62.3806,2023,412,1,403,37.3206,2021,256,0,384,4.2352,2018,660,0,176,47.3081,2018,331,0,319,8.0667,2024,958,1,183,94.536,2019,227,0,99,47.319,2020,169,1,119,41.1509,2022,868,0,249,32.0634,2023,4,1,60,84.5566,2023,951,0,326,73.7287,2020,481,0,105,91.4205,2024,624,1,301,16.1198,2019,921,1,333,33.077,2022,11,1,349,68.7099,2021,975,0,345,61.8571,2020,119,0,474,56.0511,2018,155,0,98,46.647,2024,642,1,51,1.7474,2018,573,0,425,70.3823,2022,546,0,98,45.8999,2024,731,1,210,68.6213,2018,403,0,34,84.2282,2024,15,0,187,69.9959,2023,688,1,8,62.8688,2024,896,0,494,95.9372,2024,921,0,231,4.3398,2023,63,0,367,26.1181,2020,252,0,80,43.092,2023,270,0,479,31.4024,2022,457,1,126,73.8316,2019,728,1,468,33.6178,2024,64,1,256,34.494,2023,713,0,214,48.3446,2023,633,0,280,75.8613,2018,652,1,206,31.6637,2021,708,1,413,92.4943,2019,769,0,29,79.2744,2018,113,0,200,31.7797,2018,680,0,177,55.4834,2022,717,1,17,20.6443,2024,904,0,269,46.5135,2018,333,1,225,0.314,2022,907,1,325,0.9522,2020,151,1,357,79.7807,2020,812,0,298,17.8222,2024,204,0,242,31.6463,2018,174,0,87,20.6699,2024,898,0,310,3.1975,2023,333,0,37,45.6321,2022,467,0,420,18.3574,2023,723,1,254,75.6098,2024,255,1,105,18.3792,2023,95,1,316,62.3586,2024,468,1,93,30.1892,2019,844,1,80,74.2783,2023,293,1,38,56.7895,2021,910,0,441,0.8273,2022,90,0,494,8.0484,2018,768,0,285,50.5273,2022,783,1,459,46.6083,2021,385,1,355,86.8212,2018,389,1,394
+106,C006,75,Phone,Electronics,28810.0,1,2024-03-10,0,162,89.2765,2024,438,1,221,35.1768,2024,778,1,342,18.6732,2020,754,1,53,70.9068,2020,818,0,292,5.8639,2019,78,1,480,95.2648,2024,861,0,383,18.4114,2023,8,1,421,51.4506,2021,865,1,79,54.1033,2022,66,0,478,50.1232,2023,436,1,313,66.2604,2018,429,1,179,16.5286,2021,581,0,104,94.3254,2019,787,1,231,59.6337,2024,819,0,460,42.525,2018,690,1,453,63.8216,2020,186,1,216,40.9548,2021,374,0,153,27.5346,2021,921,1,278,14.6176,2023,667,0,72,68.6175,2021,826,1,359,77.9182,2021,967,1,421,16.3202,2024,200,1,81,64.5495,2018,74,1,359,63.922,2022,747,1,328,88.0274,2023,890,0,360,73.209,2021,787,0,159,14.6466,2022,122,1,349,35.597,2023,950,1,262,95.6643,2021,307,0,460,62.1779,2018,69,0,382,92.4716,2022,210,0,431,69.2204,2019,121,0,305,78.1939,2018,910,0,387,74.2714,2023,187,1,311,87.2693,2020,355,0,55,4.8737,2018,235,0,284,92.996,2021,840,0,489,40.2421,2020,293,1,412,87.0848,2024,63,0,218,86.1945,2018,257,1,384,0.2459,2021,664,0,401,66.6517,2019,825,1,351,66.6189,2021,785,0,170,3.9129,2023,410,1,455,90.6137,2023,903,1,41,41.3513,2019,335,1,111,54.8415,2019,494,1,355,79.7123,2024,210,0,499,36.1479,2023,93,1,362,50.6429,2023,306,0,215,11.6132,2021,109,0,67,75.8066,2020,413,0,422,46.094,2021,340,1,466,73.9891,2022,145,1,204,68.7238,2021,118,0,77,17.7402,2021,728,0,399,96.0293,2020,731,1,377,3.889,2018,865,1,110,8.7815,2018,320,1,63,81.4889,2019,70,0,162,74.0095,2021,53,0,389,93.5618,2019,612,1,418,19.4281,2019,992,0,388,42.4485,2024,169,0,425,24.8221,2022,759,1,316,10.4367,2019,749,0,221,6.7502,2021,764,1,263,94.7722,2020,373,1,164,93.4439,2022,797,1,1,55.1035,2019,446,0,410,30.0502,2020,453,1,332,61.2768,2021,256,1,456,42.8609,2022,154,1,281,28.924,2024,474,0,224,59.5252,2024,371,0,289,68.7651,2018,313,1,183,25.7745,2020,459,0,46,76.2477,2018,113,0,260
+107,C007,30,Moisturiser,Beauty,650.0,2,2024-03-18,0,265,71.8898,2023,251,0,168,87.2184,2019,756,1,131,1.8023,2024,185,1,456,94.725,2022,959,1,453,4.1705,2023,205,1,35,92.5176,2023,586,0,404,4.7322,2020,379,0,107,90.9802,2022,903,0,204,86.6962,2022,670,1,19,59.5029,2019,408,0,361,1.7606,2019,776,0,209,70.6459,2018,501,1,14,80.266,2018,759,0,413,26.015,2019,710,0,197,75.7025,2019,585,0,268,84.2711,2024,468,0,206,19.0746,2023,644,0,48,44.6833,2021,539,0,219,34.2227,2021,823,0,134,29.2405,2019,15,1,265,63.2993,2023,825,0,142,44.134,2021,895,0,496,16.6216,2021,130,0,89,89.5891,2019,657,1,129,64.9228,2023,758,0,92,43.5028,2022,108,0,53,18.1607,2019,145,1,362,62.7104,2023,284,1,113,13.6536,2022,417,1,311,24.5513,2019,838,1,461,38.0537,2020,719,1,41,35.6929,2019,996,1,52,48.7308,2020,384,0,132,89.0276,2024,920,0,357,80.2398,2021,223,1,75,87.2403,2024,673,1,290,62.9747,2021,612,1,68,36.7122,2021,708,1,190,46.964,2019,841,1,284,19.5553,2024,364,0,64,96.597,2019,499,0,231,38.9561,2020,261,0,172,2.0021,2023,13,0,157,89.725,2020,529,0,203,68.7313,2021,260,1,148,65.3153,2021,208,1,480,24.5924,2019,663,1,371,79.0403,2024,878,0,427,0.4257,2019,938,1,19,70.2096,2020,985,0,275,1.6965,2023,124,0,279,83.0474,2018,230,1,276,4.4048,2024,66,0,127,81.5582,2024,831,1,488,34.4021,2019,304,1,252,1.434,2020,438,0,426,77.7966,2018,76,0,40,79.5751,2024,934,1,131,85.4413,2020,207,1,232,87.7352,2023,264,0,404,9.3724,2020,946,0,327,4.3843,2019,46,0,163,72.1793,2020,994,1,300,85.8559,2019,983,0,255,68.2368,2022,719,0,174,10.4976,2024,5,0,260,84.5444,2022,964,1,157,30.1674,2020,158,0,348,68.2181,2020,44,1,137,9.7238,2018,985,0,361,39.1171,2019,466,0,217,9.5306,2019,986,0,219,54.3066,2021,860,0,358,32.1899,2023,713,1,153,2.0633,2018,21,1,373,5.7149,2022,546,1,33,11.4459,2023,182,0,466,5.5255,2018,162,1,448

ceo_brief_env/tasks/expert_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,1.0
+total_revenue,292540.0
+top_category,Electronics
+projection_next_quarter,121332.0
+variance_pct,232.43
+break_even_units,16.0
+memo_score,0.6955

ceo_brief_env/tasks/expert_brief/metadata.json

@@ -0,0 +1,21 @@
+{
+  "title": "Full expert stack — data, finance, market strategy, comms",
+  "difficulty": "hard",
+  "max_steps": 14,
+  "instruction": "Run the full office: data quality and KPIs on Q3, finance forecast and variance vs plan, strategy view on the NVDA/AAPL/JPM watchlist aligned to our numbers, and a crisp executive memo. When advanced memory (RAG) is on, tie strategy to SOPs and external tape; otherwise internal metrics only.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "strategy",
+    "hr"
+  ],
+  "memo_audience": "CFO and strategy leads",
+  "hr_required_terms": [
+    "forecast",
+    "variance",
+    "strategy"
+  ],
+  "plan_value": 88000,
+  "fixed_cost": 32000,
+  "unit_margin": 2000
+}

ceo_brief_env/tasks/expert_brief/raw.csv

@@ -0,0 +1,12 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+401,C130,30,Monitor,Electronics,18500.0,2,2024-01-08,1,123,22.6578,2024,152,1,176,8.2049,2022,745,0,93,89.055,2021,615,0,368,5.8639,2023,622,0,198,58.6001,2019,47,0,489,54.928,2021,871,0,375,52.3668,2022,233,1,318,14.2836,2024,592,1,433,41.7529,2020,935,1,471,5.8712,2021,851,0,355,80.2094,2019,626,1,236,63.8216,2021,484,0,428,64.4314,2019,446,1,104,49.6581,2023,500,0,71,29.7861,2021,353,1,100,61.4805,2018,237,1,267,83.1843,2021,70,0,319,1.982,2022,729,0,379,6.6713,2019,421,0,408,14.6466,2023,208,1,201,28.4838,2023,136,0,485,16.0837,2018,532,1,183,67.1621,2022,892,1,466,5.2908,2023,35,0,480,90.219,2018,473,1,436,99.2768,2018,863,1,336,23.5451,2019,872,0,100,40.2421,2022,67,0,243,84.1431,2018,195,0,451,6.1121,2021,542,1,150,9.926,2021,816,1,210,92.2315,2019,470,1,428,28.4976,2019,874,0,157,87.1074,2019,48,0,332,26.8546,2020,110,0,294,36.1479,2020,428,1,426,5.526,2020,235,0,290,24.1107,2019,745,1,499,18.568,2018,494,1,246,81.428,2024,8,1,413,36.4303,2024,363,0,146,51.1389,2024,170,1,467,75.1767,2022,949,1,51,81.4889,2018,714,1,133,6.036,2018,304,1,115,21.549,2020,450,1,13,14.4934,2024,941,0,220,53.4918,2018,697,1,361,1.2975,2020,981,0,325,25.9336,2022,720,1,23,44.7849,2018,579,0,403,30.0502,2020,158,1,223,14.4828,2019,707,0,58,68.7795,2020,10,0,168,86.336,2019,580,1,233,87.7534,2021,554,0,429,19.7164,2022,866,1,220,3.6933,2021,381,1,387,69.223,2019,568,0,317,97.1287,2024,931,1,222,80.8092,2023,568,0,488,4.5514,2019,254,1,60,25.6277,2021,240,1,254,64.4252,2024,224,1,250,93.6697,2023,689,1,317,94.0421,2022,52,1,380,29.3441,2021,633,1,50,18.9402,2023,651,1,452,1.235,2018,300,1,72,41.9599,2021,473,0,356,14.2737,2019,433,0,434,26.3104,2023,167,1,279,80.3184,2023,864,1,258,79.838,2021,999,0,497,1.914,2018,547,1,54,40.4241,2020,23,1,343,13.3863,2023,807,0,22,77.511,2021,872,0,445
+402,C131,45,Serum,Beauty,1100.0,3,2024-01-19,0,29,11.2991,2024,527,0,151,99.7281,2019,762,1,294,45.7429,2019,139,1,178,4.1705,2020,960,0,456,64.8436,2022,335,1,308,12.1544,2022,534,1,11,47.7808,2023,566,1,411,25.1362,2021,109,0,384,49.8643,2020,182,0,373,3.9599,2023,495,0,460,97.2634,2022,98,0,224,84.2711,2019,431,1,218,76.582,2019,110,1,432,31.5316,2021,26,0,41,12.2157,2021,324,1,51,0.0719,2024,893,0,472,81.2493,2019,467,0,208,45.7343,2019,233,1,472,11.9559,2020,913,0,445,18.1607,2022,698,1,238,90.2845,2024,235,1,115,94.3213,2019,579,1,134,45.3949,2023,977,0,246,64.1519,2021,60,1,199,45.5387,2022,336,0,283,65.3956,2019,790,1,389,19.4642,2024,722,1,282,36.7122,2018,824,0,14,85.2095,2023,789,0,310,31.1943,2019,328,1,232,90.6755,2018,805,1,265,28.0504,2024,405,0,169,16.4333,2018,755,0,107,73.3119,2019,605,0,28,70.9682,2021,588,0,299,0.4257,2020,724,0,52,16.7124,2022,16,0,26,18.461,2018,598,1,209,51.1396,2024,57,0,287,95.8226,2019,531,0,414,93.0922,2023,849,1,480,95.3307,2018,2,0,412,22.0019,2019,837,1,96,87.7352,2018,323,0,306,5.78,2019,43,1,308,2.7304,2020,26,0,184,87.0907,2019,20,1,262,77.0651,2019,249,0,241,7.9994,2023,734,1,473,63.4024,2022,930,1,79,0.0412,2018,887,0,199,39.1171,2024,664,0,84,75.8617,2023,543,0,489,91.5052,2023,310,1,340,96.0673,2023,465,1,260,78.3053,2020,332,1,459,11.0417,2020,115,0,445,67.9927,2020,53,0,50,54.4408,2024,343,1,368,2.1276,2023,866,1,114,57.335,2023,45,0,160,46.6561,2021,159,0,350,6.1672,2020,146,1,100,47.6325,2024,72,0,268,88.481,2023,145,0,274,28.1618,2023,681,0,413,7.9943,2018,813,1,20,91.0577,2021,246,1,24,70.9632,2018,726,1,401,99.2569,2021,706,0,106,18.7186,2021,934,0,379,9.289,2018,959,0,146,38.7768,2024,199,1,143,58.217,2019,926,1,3,95.6922,2018,158,1,389,75.0932,2022,892,0,272,35.3021,2023,995,1,250,75.3618,2020,914,1,110
+403,C132,60,Workstation,Electronics,95000.0,1,2024-02-12,1,47,42.714,2021,9,0,436,16.2917,2024,764,0,167,91.2197,2024,282,0,71,85.4235,2024,68,0,106,65.9199,2019,548,1,211,22.366,2020,233,0,164,59.5803,2023,466,0,64,30.9181,2022,170,1,271,79.0221,2022,780,1,401,89.3017,2018,318,1,179,39.3874,2019,76,1,480,19.8423,2020,411,0,372,41.1074,2020,652,0,39,70.7117,2020,232,0,116,27.5355,2020,20,1,369,94.3381,2023,555,1,99,47.319,2018,397,1,448,86.8745,2019,320,1,315,18.2912,2024,951,0,127,44.5498,2019,723,1,431,42.7441,2023,88,0,283,10.8294,2019,386,1,222,34.0848,2023,953,1,441,21.2284,2021,398,1,98,46.647,2019,637,1,401,57.3556,2021,703,0,77,58.0077,2020,731,1,497,53.7273,2024,380,0,443,51.1824,2023,616,1,223,27.7785,2023,780,1,127,37.9729,2024,996,1,330,88.18,2021,845,0,79,98.1138,2019,974,1,207,87.3784,2024,442,0,159,71.488,2019,64,1,92,83.7223,2018,36,1,372,91.0575,2023,538,0,71,20.4355,2018,26,1,54,57.5166,2021,511,1,459,66.3906,2020,168,1,177,55.4834,2023,679,1,199,73.164,2024,303,0,327,69.2046,2023,907,0,346,44.691,2024,809,1,189,30.8422,2019,74,0,89,93.1111,2020,470,1,150,63.3282,2024,557,1,187,28.0697,2021,935,1,420,18.3574,2023,345,1,153,37.3078,2021,188,1,73,81.5536,2019,468,0,240,69.2473,2021,434,1,409,76.8679,2023,741,1,224,10.8205,2024,462,0,91,60.7351,2019,134,0,413,27.8491,2021,717,1,355,86.8212,2018,913,0,471,43.8288,2018,430,0,48,52.7364,2021,991,0,19,36.8383,2023,142,1,356,50.9879,2021,201,1,236,90.2847,2018,605,1,308,49.5281,2024,566,0,252,15.1072,2018,62,0,70,35.1858,2024,407,1,2,43.6596,2021,603,0,122,27.4913,2021,247,0,390,26.6006,2019,959,1,445,8.5122,2019,445,1,58,87.2463,2023,57,0,216,7.0348,2018,92,1,63,97.1628,2019,141,0,12,93.8023,2024,722,1,386,30.3609,2022,16,0,451,91.9243,2021,676,1,263,67.0188,2021,992,0,92,12.9781,2018,669,1,179,86.5065,2024,646,0,191
+404,C133,30,Shoes,Fashion,3200.0,2,2024-02-20,1,162,89.2765,2019,755,1,299,75.6653,2024,992,1,253,94.9615,2024,818,1,125,30.0631,2018,74,1,491,24.1479,2022,535,1,464,93.0111,2024,906,0,447,91.9658,2018,575,0,169,89.4187,2022,520,0,179,16.5286,2021,320,1,2,75.9259,2021,600,0,226,47.6738,2019,690,1,213,39.1209,2023,859,1,282,79.5521,2022,2,0,208,83.5863,2018,397,0,352,48.2215,2024,754,1,466,0.6968,2019,906,1,81,64.5495,2018,794,0,450,65.7579,2024,775,1,218,54.0921,2020,787,1,292,96.516,2021,327,1,394,96.9474,2023,629,1,477,94.1845,2021,737,1,297,0.4444,2022,529,1,309,68.2438,2023,401,0,387,74.2714,2024,262,0,114,22.3563,2023,698,1,342,28.0711,2019,840,1,13,44.1694,2020,465,1,491,2.6137,2022,839,1,343,45.5204,2020,45,0,174,88.324,2024,127,0,474,7.8104,2019,373,1,80,43.092,2024,449,0,132,33.5015,2023,738,0,418,56.1347,2024,210,0,377,32.9468,2020,32,1,381,49.5031,2018,594,1,12,15.4404,2020,558,1,158,6.21,2023,141,1,61,48.6048,2023,47,1,77,17.7402,2021,469,1,389,95.5585,2023,465,1,384,54.5808,2023,320,1,467,14.5694,2023,65,0,279,17.0002,2019,498,0,70,6.4716,2022,645,1,324,26.1433,2021,16,1,212,65.5122,2020,585,1,221,6.7502,2024,747,0,151,69.7436,2018,183,0,256,6.051,2023,446,1,446,62.9852,2023,498,0,218,52.2401,2023,293,1,250,52.2804,2018,431,0,114,52.9922,2024,761,1,243,56.7069,2018,844,0,46,76.2477,2019,528,0,184,52.228,2018,673,0,258,31.2558,2021,395,0,104,80.4519,2023,235,0,220,54.5223,2021,147,1,18,69.3036,2020,138,1,215,80.7792,2022,590,1,378,68.7228,2023,619,1,262,16.4702,2023,486,0,422,68.9975,2021,7,1,181,65.0755,2018,418,0,39,60.1489,2018,816,0,248,75.8667,2020,238,0,333,4.6736,2020,294,1,415,11.2811,2022,581,0,191,88.7794,2024,486,1,395,71.4436,2019,490,1,41,82.8742,2023,930,1,55,80.0123,2024,547,0,247,9.7451,2020,532,0,359,3.4564,2023,336,1,280,65.8915,2024,144,1,72
+405,C134,90,Headphones,Electronics,8000.0,2,2024-03-05,0,265,71.8898,2024,846,1,292,81.1799,2020,186,1,388,84.7199,2020,959,1,30,79.3659,2022,507,0,393,4.2389,2023,318,0,64,96.7308,2020,152,0,287,99.6766,2024,345,0,158,92.2447,2023,776,1,209,70.6459,2023,918,0,12,42.8658,2021,596,0,283,2.2761,2019,585,1,165,19.2569,2022,557,0,404,43.2429,2023,262,1,404,7.1317,2019,91,1,237,38.8495,2022,659,1,479,36.5177,2022,884,1,496,16.6216,2023,451,0,216,3.0805,2020,880,1,469,86.6347,2018,108,1,291,35.6642,2023,467,0,365,63.0225,2019,886,0,1,52.8901,2024,881,1,343,73.0891,2022,198,0,60,94.8786,2018,463,1,132,89.0276,2021,680,0,418,64.7971,2019,48,1,495,32.5837,2021,612,1,374,3.6041,2023,676,0,83,87.8541,2020,387,0,11,25.009,2023,377,1,479,7.8106,2018,525,0,307,84.7469,2021,226,0,455,90.6137,2019,277,1,326,23.8735,2022,478,0,155,43.203,2022,878,0,449,72.2503,2024,82,0,193,97.9875,2020,915,0,387,10.2863,2024,748,1,220,30.7576,2018,895,1,57,40.8812,2022,59,1,252,1.434,2021,573,0,412,39.3946,2021,608,0,163,49.505,2022,207,0,203,15.1097,2018,456,1,155,75.8981,2022,677,1,254,58.4507,2022,29,0,2,68.4925,2021,900,1,167,63.1309,2018,428,1,260,84.5444,2024,128,0,289,18.1263,2023,134,0,265,5.0386,2022,985,1,59,33.5866,2024,346,1,91,93.5597,2023,588,0,103,5.1409,2020,248,0,41,92.8819,2019,555,0,391,36.6055,2019,896,0,466,5.5255,2024,347,1,493,89.4587,2018,593,1,293,65.9521,2018,495,1,121,14.9704,2021,126,1,397,48.5795,2024,429,1,422,78.0612,2023,605,0,492,8.0854,2021,91,1,460,99.7837,2022,340,0,411,25.0686,2024,425,1,20,20.1205,2020,877,0,140,6.3381,2021,357,1,441,54.5731,2021,560,1,296,15.0743,2020,880,1,316,21.7833,2021,873,1,232,74.9316,2023,701,0,272,7.7791,2020,186,0,327,50.7167,2024,910,0,36,48.2355,2022,922,0,286,80.3627,2024,788,1,175,99.0117,2020,932,0,310,96.5513,2018,762,1,98,45.1042,2022,928,1,499
+406,C135,75,Desk,Home,12000.0,1,2024-03-18,1,326,6.9968,2022,570,0,312,23.88,2024,167,0,4,71.1985,2019,872,1,491,73.419,2024,626,1,498,10.633,2021,483,0,223,34.2597,2020,932,0,495,91.8562,2019,825,0,218,72.1691,2018,373,0,453,91.4978,2019,78,0,358,64.0517,2021,792,1,101,9.7654,2021,141,0,173,88.3355,2019,147,0,414,86.7129,2020,619,1,458,4.2352,2024,96,1,416,33.1595,2022,751,0,483,28.4169,2021,137,0,286,61.6459,2019,907,0,236,80.2235,2022,654,1,179,79.9477,2023,608,1,279,19.7535,2023,163,0,186,73.1922,2018,220,0,285,33.077,2022,135,0,298,97.5929,2022,747,0,61,10.3404,2021,859,0,88,94.0439,2020,749,0,152,39.199,2019,656,0,498,31.1881,2024,192,0,344,49.1334,2022,831,1,373,4.9071,2023,338,0,422,62.8688,2020,97,1,5,74.7235,2019,572,1,32,33.9786,2018,800,0,203,68.7313,2022,994,1,228,4.1149,2019,548,0,123,55.1445,2024,662,0,436,42.7386,2024,815,1,161,63.3442,2020,243,0,42,31.6637,2018,376,0,462,34.0854,2018,220,0,73,14.9649,2018,356,0,423,44.2479,2023,113,0,227,47.9252,2023,38,1,274,90.9225,2022,512,1,436,87.1975,2019,882,1,84,49.3552,2018,874,0,17,31.6463,2019,212,1,104,16.9701,2022,855,0,379,71.3009,2023,772,0,465,94.2951,2024,846,1,24,46.4478,2018,934,0,496,11.8771,2024,176,0,78,26.0189,2020,854,1,79,72.8541,2024,109,1,361,0.8273,2018,249,1,41,37.164,2021,582,1,157,74.9996,2023,276,1,471,71.9424,2019,146,1,123,13.7357,2020,677,0,53,74.4885,2024,462,1,354,24.1914,2018,841,0,434,9.7381,2019,856,1,275,6.1589,2018,81,1,411,96.4072,2022,263,0,224,20.8499,2021,333,0,422,40.6694,2021,647,1,57,23.0667,2022,746,0,276,11.5317,2022,423,0,135,8.6481,2023,474,1,8,32.7241,2019,513,1,401,2.8736,2021,165,1,103,11.84,2020,246,0,322,79.4095,2021,505,1,275,73.7156,2023,903,0,413,5.5315,2022,267,0,145,65.4113,2019,785,1,98,14.0724,2022,288,1,230,43.8475,2024,532,0,410,35.295,2022,134,1,126
+407,C136,30,Tablet,Electronics,,1,2024-04-02,0,34,45.6125,2022,388,1,269,74.9081,2022,18,0,404,87.0686,2020,174,0,493,4.8469,2019,766,1,23,64.5625,2018,800,0,233,51.4506,2018,300,1,270,67.0545,2024,475,0,204,11.0141,2021,955,1,258,52.671,2023,43,1,313,55.7647,2018,260,1,310,68.8055,2018,635,0,194,33.3327,2019,648,0,256,7.3639,2023,447,0,325,14.6176,2022,182,0,343,1.5572,2023,283,1,413,33.6834,2024,948,1,63,91.395,2023,236,1,280,36.4405,2021,649,1,473,79.3618,2021,777,1,379,53.6179,2019,353,1,349,35.8857,2022,148,0,273,62.1779,2024,888,1,462,71.9382,2024,486,1,498,48.7493,2019,255,0,129,81.5038,2022,100,1,14,41.3644,2018,872,0,100,8.1789,2021,982,1,424,0.9585,2022,134,0,389,0.3585,2022,739,1,73,0.2459,2020,94,0,333,82.5791,2021,138,1,392,81.782,2019,490,0,404,85.7758,2023,491,1,60,77.4271,2021,782,0,439,18.7975,2024,879,1,48,81.9548,2018,468,1,180,62.9521,2024,52,1,244,75.8066,2022,881,0,118,97.3721,2024,956,1,416,12.3134,2020,89,0,152,97.0545,2019,32,0,15,83.0262,2022,413,1,316,18.4396,2018,284,1,220,75.5864,2022,0,1,285,11.2657,2024,903,0,33,19.4281,2019,718,0,482,23.9393,2023,669,0,360,12.3694,2018,527,0,21,71.4861,2020,507,1,338,66.525,2022,827,0,311,27.2059,2019,112,1,383,73.3665,2021,862,1,311,49.1481,2019,564,0,259,28.924,2024,647,1,373,17.3233,2022,458,1,273,91.4982,2018,475,0,74,46.5692,2020,787,0,34,54.3874,2023,129,1,295,39.743,2020,493,1,477,68.1593,2024,955,1,326,32.5647,2022,139,0,396,29.1912,2022,462,1,438,11.486,2019,434,0,404,20.8995,2023,780,1,93,77.9134,2018,191,0,396,66.4161,2023,832,0,329,40.323,2022,27,1,297,73.5384,2020,942,0,85,61.0868,2020,772,0,323,28.0289,2019,872,0,167,31.3905,2019,407,1,11,56.3049,2023,310,0,277,96.3444,2020,18,1,59,28.6851,2021,160,0,204,43.7276,2021,82,1,218,5.7777,2022,409,0,337,5.8318,2022,195,0,341,28.0491,2022,44,0,323
+408,C137,60,Racket,Sports,2100.0,1,2024-04-14,0,345,53.8047,2018,442,1,50,88.4471,2021,483,0,308,35.0699,2022,238,1,440,41.8442,2021,228,1,212,9.6227,2018,767,1,156,90.9802,2023,581,1,430,51.8984,2018,749,0,419,11.4967,2018,363,0,297,1.1168,2023,208,0,220,12.3311,2021,389,1,138,5.3971,2019,291,0,363,84.3403,2023,840,0,72,80.7577,2020,542,0,207,34.2227,2022,36,0,417,9.2095,2021,89,1,198,79.117,2021,579,0,447,32.4947,2021,719,0,207,60.0091,2018,825,1,37,34.7917,2022,650,1,408,75.7903,2022,598,1,428,60.2616,2019,558,1,108,24.5513,2018,507,1,465,10.6033,2018,552,0,223,22.1366,2024,879,0,464,86.5524,2020,621,1,434,68.9897,2020,175,1,33,22.1585,2021,56,0,176,87.5468,2024,256,0,250,37.4572,2022,87,0,103,96.597,2024,277,1,411,24.1213,2022,66,1,7,9.0394,2024,241,1,318,55.5301,2023,958,0,125,82.5649,2023,245,1,443,20.5493,2024,525,1,259,87.2666,2024,119,0,230,43.0567,2019,555,1,302,83.0474,2022,248,1,231,92.223,2019,653,0,317,49.7177,2023,792,1,300,44.3207,2019,799,0,430,79.8385,2024,795,1,259,75.9751,2020,293,1,231,87.7181,2021,743,1,242,77.878,2022,269,1,89,72.1793,2019,328,0,212,17.8062,2021,876,1,76,97.1451,2019,538,0,336,45.0808,2020,525,0,373,90.4273,2022,682,0,399,3.7138,2018,687,0,368,65.8513,2023,271,0,128,91.3319,2018,699,1,137,32.1899,2018,41,1,298,44.8005,2023,899,1,466,24.3056,2018,932,0,422,8.0967,2018,519,0,215,67.2145,2022,6,1,211,11.5208,2022,857,1,291,62.7859,2020,567,1,39,98.6302,2022,826,1,340,17.4253,2022,702,0,356,86.0943,2022,492,0,139,97.1239,2022,246,1,130,66.4168,2023,336,1,189,76.3234,2018,418,1,189,78.9622,2024,522,0,86,38.4808,2018,969,1,54,89.6083,2018,24,0,458,15.6253,2021,360,0,362,83.8866,2023,68,0,3,50.1049,2021,50,1,392,61.4556,2024,350,0,86,58.6488,2020,13,1,30,60.0645,2018,381,0,131,18.4742,2023,34,1,249,68.669,2020,47,1,66,82.3491,2024,568,1,422
+409,C138,30,Phone,Electronics,42000.0,1,2024-04-30,0,28,81.0418,2021,336,0,439,32.0166,2020,670,0,494,99.5758,2018,826,1,454,11.1707,2022,633,0,484,80.8464,2022,410,0,44,38.6163,2024,885,1,433,13.4064,2021,718,1,410,67.9788,2018,910,0,298,3.0805,2023,26,1,263,49.855,2020,731,1,186,59.7847,2018,202,1,59,29.1723,2023,243,1,168,9.4579,2023,485,1,384,62.7176,2022,652,0,146,85.8191,2024,80,0,75,79.535,2019,227,0,450,46.9443,2018,177,0,344,20.7373,2020,18,0,204,84.5566,2020,661,0,481,48.1941,2024,843,1,325,22.4724,2024,706,1,333,55.3777,2018,54,1,190,29.8997,2018,618,0,183,65.6891,2024,155,1,70,97.9444,2024,713,0,36,74.1111,2018,633,0,328,45.8999,2022,510,1,97,40.3378,2024,315,0,137,56.7215,2023,987,1,8,98.7706,2022,295,0,195,63.0756,2020,312,1,300,61.8301,2024,383,0,323,82.1514,2023,82,0,24,3.8978,2018,696,0,319,33.6178,2022,566,1,424,9.3946,2021,589,1,137,67.7976,2023,805,1,206,58.7303,2020,451,1,335,7.1261,2018,683,1,467,61.7015,2018,680,0,316,54.7042,2019,852,0,465,95.1119,2019,528,1,255,0.314,2018,79,0,30,7.0625,2019,170,0,6,75.7246,2019,171,0,242,61.0947,2024,158,1,203,9.6113,2021,914,1,392,83.157,2023,467,1,418,47.3461,2020,314,1,459,45.4705,2024,514,1,296,62.3586,2019,339,1,42,45.3853,2021,442,0,412,12.8116,2021,555,0,441,17.6442,2021,34,1,232,37.5141,2020,83,0,45,59.0909,2022,385,1,214,34.9449,2019,896,1,395,99.9074,2018,997,0,329,92.5319,2024,142,0,25,30.1564,2023,449,0,138,41.3301,2018,48,1,114,40.1331,2019,804,1,473,29.8372,2022,878,0,23,6.5054,2023,661,1,90,55.0495,2024,765,1,311,45.664,2023,516,1,392,59.4806,2020,441,1,243,98.4585,2020,795,1,317,27.115,2019,268,1,21,34.08,2022,716,0,160,89.1934,2022,341,1,89,78.6573,2022,77,0,128,41.357,2020,759,0,70,17.6641,2018,410,0,491,61.1375,2024,402,0,16,89.4472,2019,242,0,380,46.1075,2022,603,0,15,95.6966,2022,831,0,234
+409,C138,30,Phone,Electronics,42000.0,1,2024-04-30,1,175,47.8909,2020,611,1,473,90.2784,2022,833,1,124,70.9068,2022,488,1,468,20.5395,2019,619,1,126,25.8798,2019,169,1,421,28.3051,2024,470,1,17,97.1131,2022,306,0,266,34.547,2019,429,0,51,71.3708,2019,834,0,135,40.0189,2020,759,0,286,42.525,2024,720,1,338,46.8454,2023,359,1,258,1.7105,2019,466,1,278,98.0003,2023,780,1,11,33.4572,2022,116,1,205,80.8125,2023,200,1,389,41.8102,2024,384,0,325,52.6488,2020,643,0,455,73.209,2021,513,0,19,14.5411,2021,997,0,356,26.0389,2024,596,1,460,12.6803,2023,763,0,59,43.8155,2024,229,0,431,69.5826,2020,910,1,104,30.4319,2022,959,0,154,47.6105,2023,714,1,255,92.996,2022,576,0,460,70.8354,2018,788,1,247,31.4986,2024,392,0,384,70.3137,2021,241,0,294,18.2568,2022,136,1,445,38.6077,2018,252,0,230,13.3165,2021,294,1,311,83.0984,2021,540,1,443,79.7123,2020,799,1,69,12.5775,2019,608,0,308,7.2491,2022,308,0,67,96.6011,2019,674,1,22,69.888,2020,792,1,381,61.8054,2021,118,1,83,93.5701,2023,306,1,198,53.8888,2021,120,0,381,8.7815,2018,604,0,444,50.7793,2022,545,0,380,46.1159,2022,626,1,418,42.9566,2024,760,1,429,16.2267,2024,31,1,315,20.2078,2018,749,0,476,41.3417,2023,725,1,192,3.1781,2020,21,0,314,55.1035,2024,855,1,168,83.1635,2019,747,0,428,37.8447,2022,721,1,281,16.685,2022,397,1,11,4.7573,2023,505,1,164,77.0009,2019,459,0,64,53.6161,2024,262,1,483,99.8908,2023,445,0,32,29.809,2023,774,1,91,9.6685,2022,818,1,18,4.8924,2024,86,0,461,74.0223,2018,492,0,147,27.4888,2022,544,1,103,23.2578,2022,697,1,213,4.3644,2019,79,1,469,78.6619,2018,169,0,144,37.4449,2019,571,1,60,55.8521,2021,314,0,335,41.7154,2021,193,1,160,76.4496,2022,776,1,200,11.0284,2021,545,0,246,6.0762,2024,390,0,10,10.3323,2022,708,1,81,53.584,2019,596,1,188,80.0292,2024,702,1,43,78.7573,2022,867,1,176,86.0708,2021,6,0,184,20.2184,2020,447,1,107
+410,C139,45,Keyboard,Electronics,3500.0,4,2024-05-06,0,377,36.3951,2022,830,1,98,33.0549,2023,197,1,390,94.725,2021,478,1,282,76.8981,2021,841,0,386,55.6728,2024,185,0,107,23.8932,2019,337,0,457,59.5041,2019,501,0,420,93.7307,2023,776,1,31,11.0989,2019,821,1,65,6.9612,2018,690,1,276,75.7025,2018,779,0,270,33.2918,2024,597,0,206,64.0305,2022,371,0,219,8.4668,2020,220,1,361,78.0429,2019,779,1,267,2.6198,2023,895,0,208,5.9954,2022,970,0,129,49.7094,2020,791,1,71,43.5028,2021,2,0,368,83.3729,2024,389,1,312,58.9308,2022,404,1,311,61.8415,2021,975,1,346,92.081,2020,692,0,194,29.3695,2021,384,0,347,19.9297,2023,909,1,215,85.1208,2021,367,0,102,62.9747,2018,501,1,343,46.6497,2023,880,1,8,67.9196,2022,968,1,64,90.3261,2018,567,0,304,96.4461,2020,43,1,70,65.8275,2019,410,0,36,27.4072,2020,922,0,138,73.7158,2024,113,0,349,79.0403,2024,982,0,125,29.0303,2023,637,1,316,34.6785,2018,716,1,279,30.1199,2018,153,0,379,16.2863,2021,547,1,97,98.1271,2023,304,1,278,93.8561,2018,673,0,26,75.4492,2018,749,1,314,85.4413,2022,607,0,388,38.7945,2020,131,0,406,26.3068,2024,836,1,163,20.3665,2024,49,1,373,4.82,2020,604,0,232,28.4076,2022,5,0,199,46.1039,2022,835,1,212,32.6755,2022,859,1,268,9.7238,2019,639,1,449,80.1548,2023,662,0,147,38.0364,2021,443,0,358,99.3657,2022,168,0,235,95.0801,2024,500,1,63,77.6769,2021,182,0,46,31.2831,2020,189,0,73,20.2355,2020,942,0,338,93.598,2022,876,0,132,15.3047,2024,117,0,261,87.688,2020,337,0,313,30.4665,2018,10,0,330,6.2271,2018,58,1,189,93.8413,2018,118,0,316,35.0588,2021,4,1,336,13.3286,2022,854,1,210,97.9624,2019,600,0,263,55.3138,2019,116,0,469,98.5379,2018,356,0,355,96.276,2021,465,0,255,14.8676,2022,414,0,111,24.2169,2021,947,0,156,44.5596,2018,697,0,61,77.1285,2018,508,0,127,46.5216,2020,249,0,437,51.997,2018,184,1,330,44.3134,2021,962,0,470,79.9028,2023,401,1,75

ceo_brief_env/tasks/hard_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,1.0
+total_revenue,153877.78
+top_category,Electronics
+projection_next_quarter,-7672.0
+variance_pct,102.47
+break_even_units,16.67
+memo_score,0.6955

ceo_brief_env/tasks/hard_brief/metadata.json

@@ -0,0 +1,21 @@
+{
+  "title": "Board-ready summary with risks",
+  "difficulty": "hard",
+  "max_steps": 12,
+  "instruction": "Prepare a board-ready revenue summary with risk flags, Q4 forecast with scenarios, strategic recommendation, and a CFO-facing memo.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "strategy",
+    "hr"
+  ],
+  "memo_audience": "CFO",
+  "hr_required_terms": [
+    "board",
+    "forecast",
+    "CFO"
+  ],
+  "plan_value": 76000,
+  "fixed_cost": 30000,
+  "unit_margin": 1800
+}

ceo_brief_env/tasks/hard_brief/raw.csv

@@ -0,0 +1,11 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+301,C020,45,Camera,Electronics,42500.0,1,2024-01-06,1,54,75.3739,2023,699,1,305,58.4566,2021,684,1,417,87.4468,2022,305,0,409,66.5334,2023,214,1,347,63.3961,2024,76,1,123,54.8351,2018,843,0,400,85.2639,2020,895,1,34,80.0559,2020,813,1,227,67.9788,2021,887,1,200,91.4978,2023,208,1,358,89.3017,2018,495,0,410,55.1762,2023,88,0,359,33.0132,2021,480,1,180,53.3545,2022,305,1,2,25.6145,2022,616,1,334,65.2375,2023,511,1,145,2.0443,2023,408,0,28,44.3295,2023,160,0,387,16.9309,2020,430,0,374,20.7373,2023,746,1,3,79.9477,2020,650,1,279,44.5498,2019,698,1,431,62.4852,2022,690,0,154,0.1635,2020,98,1,381,7.0746,2019,444,1,265,55.9868,2023,610,1,492,15.5929,2019,962,0,94,63.7886,2022,307,1,26,28.9301,2022,684,0,158,92.3911,2022,978,0,172,40.3378,2021,272,0,32,4.9071,2023,87,1,422,27.7785,2023,328,1,114,60.8744,2024,462,1,110,61.4745,2020,183,0,126,44.2648,2018,758,0,225,24.3231,2021,70,0,44,24.5924,2020,937,0,268,58.8934,2023,849,1,357,0.108,2022,745,1,175,34.6785,2023,524,0,393,83.0474,2018,825,1,379,30.7576,2019,956,0,57,95.8226,2019,8,0,494,16.627,2020,734,1,356,77.7966,2018,539,0,264,88.8027,2019,635,1,453,13.2315,2023,377,0,223,74.3825,2020,434,1,132,4.3843,2023,483,1,102,15.3093,2022,299,1,449,4.82,2024,791,1,333,97.1451,2024,403,1,199,84.5444,2020,507,1,289,63.4024,2022,720,1,48,62.7412,2022,922,1,380,89.3062,2019,222,0,221,9.5306,2023,74,1,371,80.4072,2023,267,0,46,4.1763,2018,70,0,362,46.5466,2022,88,0,196,11.4459,2018,710,1,341,31.2831,2024,313,0,458,67.2145,2021,130,1,338,65.9521,2022,493,1,121,2.1276,2023,931,0,377,3.4745,2018,152,0,271,3.5433,2022,596,0,402,73.0773,2023,257,1,283,12.4307,2020,774,1,306,24.6699,2023,2,1,417,14.5921,2021,622,1,498,77.5319,2021,244,0,318,97.9624,2021,239,1,105,38.4808,2024,860,1,469,15.0743,2018,772,1,316,99.2569,2021,473,0,338,39.9045,2024,412,1,249
+302,C021,30,Moisturiser,Beauty,720.0,2,2024-01-20,0,123,22.6578,2021,116,0,175,53.7339,2023,262,1,99,18.5698,2024,995,0,479,64.0496,2018,734,1,122,84.1202,2018,659,0,291,31.8463,2024,214,1,142,90.3383,2019,573,0,335,11.7731,2021,282,1,447,34.547,2023,466,1,209,52.671,2019,26,0,313,75.9259,2021,318,0,355,80.2094,2020,274,1,24,38.7036,2020,260,0,299,85.6753,2023,94,0,131,53.9089,2021,42,0,411,22.098,2022,275,0,497,65.9815,2019,534,0,404,72.3343,2021,993,1,457,13.0438,2024,411,1,329,52.6488,2024,839,1,34,79.3618,2021,661,1,379,96.516,2021,723,1,201,28.4838,2019,161,0,209,54.6238,2022,226,1,487,36.492,2018,298,1,100,23.3502,2018,103,0,162,38.4617,2019,466,1,460,68.0748,2024,430,1,238,74.1567,2020,991,0,41,19.3653,2021,134,0,221,70.8354,2022,842,1,420,0.3585,2024,987,0,73,45.5204,2020,780,1,150,9.926,2019,702,0,418,78.507,2023,967,0,205,85.6201,2021,821,1,139,52.7624,2024,41,1,248,54.0101,2019,709,0,82,21.0916,2020,136,0,47,54.5093,2021,762,0,317,78.919,2020,664,1,102,58.7303,2021,844,0,34,34.0854,2018,653,1,73,66.3906,2020,531,1,453,98.8982,2021,186,0,191,45.4332,2021,754,0,60,86.5166,2022,759,1,161,32.952,2018,151,1,104,54.5322,2019,57,0,109,49.8538,2024,837,0,305,99.2013,2018,648,0,85,87.1028,2018,594,0,141,83.157,2022,611,0,188,94.2951,2020,525,1,24,37.3078,2021,930,1,399,5.0934,2022,278,0,385,15.475,2019,801,1,373,18.6287,2023,913,1,147,15.4229,2020,51,0,237,39.7893,2018,960,0,101,76.1046,2022,327,1,295,58.3205,2019,91,0,359,11.3296,2019,776,0,219,99.9074,2022,388,0,122,74.4885,2024,857,1,354,36.8383,2023,866,1,166,76.7114,2020,587,1,394,55.0105,2018,304,0,246,15.5821,2021,860,0,295,68.3735,2024,997,0,406,69.7741,2023,884,1,434,48.61,2022,938,1,228,82.6348,2018,361,0,58,56.6793,2021,511,0,133,98.4585,2023,710,0,8,32.7241,2019,24,0,401,87.2463,2023,706,0,74,24.0179,2024,210,0,115
+303,C022,90,Sofa,Home,18800.0,1,2024-02-10,1,29,11.2991,2019,471,0,369,87.9216,2022,963,1,235,33.4647,2019,777,0,436,55.1259,2024,199,1,267,28.2215,2020,673,1,393,80.0632,2022,690,1,404,87.1558,2023,990,1,464,25.4842,2024,318,1,107,93.7307,2020,866,0,329,1.1168,2019,834,1,220,42.8658,2021,600,1,460,97.2634,2021,69,0,187,11.7391,2021,408,0,103,74.3899,2020,590,1,310,20.2072,2019,208,1,163,35.2033,2023,470,0,306,28.3732,2018,748,1,124,23.732,2024,571,0,400,24.3749,2019,96,1,102,49.7094,2022,781,0,378,34.7917,2021,513,0,408,35.6642,2023,327,1,238,90.2845,2020,873,1,221,66.6084,2022,249,1,488,44.3529,2022,101,1,373,29.6182,2020,527,1,218,90.6092,2024,99,1,224,10.0628,2021,438,0,370,86.3754,2018,996,1,10,91.7173,2019,828,1,267,46.6497,2023,332,1,462,37.4572,2022,392,1,103,25.009,2023,45,1,232,90.6755,2018,342,0,365,60.0315,2023,465,0,265,15.8295,2022,440,1,497,45.6478,2022,815,0,471,20.6832,2024,743,1,499,66.244,2022,248,1,469,74.2975,2022,385,0,264,30.32,2023,96,0,347,96.6011,2023,552,0,293,97.3721,2020,683,0,416,48.6048,2023,168,0,413,36.4303,2019,266,0,91,86.1215,2018,78,0,374,82.4235,2018,771,1,104,83.4673,2020,831,1,320,91.2568,2021,589,1,122,87.4121,2019,325,0,432,97.5049,2022,2,1,491,55.161,2022,711,1,13,20.2078,2020,527,1,117,71.4861,2023,314,0,338,69.7436,2018,188,1,23,44.7849,2020,98,0,211,73.7256,2024,910,0,331,16.7001,2021,438,1,294,14.1893,2019,720,1,357,30.7807,2023,126,0,431,58.2422,2021,125,1,5,42.9786,2022,933,1,9,88.284,2019,759,1,175,99.8908,2022,114,0,187,39.743,2023,142,1,477,80.4519,2023,142,1,222,80.8092,2019,614,0,298,68.0945,2021,431,0,6,69.9958,2019,474,1,46,24.0007,2018,577,1,345,83.8325,2019,466,0,423,64.7166,2022,672,0,263,66.9145,2018,279,1,133,45.8577,2022,823,0,233,55.8521,2021,343,1,498,61.0868,2021,268,0,323,4.6736,2020,57,0,356,14.2737,2024,732,0,84
+304,C023,75,Racket,Sports,1820.0,2,2024-02-14,1,47,42.714,2020,312,0,438,19.4635,2024,870,1,354,77.6051,2023,709,1,88,90.7177,2018,48,0,311,63.1296,2024,241,1,245,41.0682,2020,549,0,120,23.3109,2023,541,1,259,3.6265,2022,635,1,461,55.3715,2024,122,0,249,3.0805,2024,821,0,263,64.0517,2021,596,0,179,39.3874,2023,889,0,82,54.0554,2023,846,0,423,80.7724,2021,17,0,224,52.9675,2020,146,1,250,26.7396,2020,482,1,319,8.0667,2023,614,0,14,55.5247,2021,125,0,163,47.4782,2018,639,0,16,25.6543,2022,505,1,60,84.5566,2020,2,0,481,19.7535,2023,467,1,431,42.7441,2022,956,1,342,62.1249,2019,542,0,349,68.7099,2023,438,0,243,27.0909,2022,487,1,18,28.0433,2018,742,1,165,71.3714,2023,992,0,425,70.3823,2024,199,1,111,46.0826,2021,198,1,18,2.7541,2018,870,0,426,56.7215,2019,968,0,8,62.8688,2020,377,0,127,37.9729,2024,713,0,431,13.9674,2018,261,1,232,91.0177,2024,133,1,246,95.9976,2023,774,0,25,19.1692,2024,714,0,292,52.555,2022,344,1,63,72.434,2021,320,0,315,89.8031,2019,517,0,78,30.1199,2020,731,0,93,92.223,2021,792,1,317,40.8812,2022,47,1,414,93.0922,2024,146,1,248,39.563,2023,861,0,313,76.7891,2021,3,0,256,12.463,2023,871,0,200,98.955,2021,308,1,330,26.9515,2022,313,1,215,2.6928,2019,206,1,120,85.0633,2019,220,1,327,28.4076,2023,573,0,7,45.0808,2022,725,1,373,18.1263,2023,183,1,79,0.0412,2021,209,1,252,33.5353,2023,556,1,384,43.6743,2021,128,0,55,33.4484,2022,8,1,322,44.6948,2022,607,0,139,89.972,2021,518,0,385,33.2375,2020,943,0,233,39.8259,2022,16,1,261,20.2355,2022,782,1,346,11.5208,2022,774,0,291,14.9704,2021,235,1,114,57.335,2018,107,1,22,92.3611,2020,50,0,254,43.0767,2023,298,0,132,36.9529,2022,208,1,7,6.3694,2019,351,1,468,33.6908,2021,587,0,393,68.1358,2021,551,0,202,66.5348,2022,47,0,301,55.3138,2021,663,0,7,89.6083,2018,193,1,458,21.7833,2021,294,0,106,18.7186,2018,598,0,364
+305,C024,60,Laptop,Electronics,61500.0,1,2024-03-09,0,162,89.2765,2024,683,1,17,8.2049,2023,296,1,168,80.8305,2024,824,0,119,73.5733,2018,754,1,480,95.2648,2019,289,0,55,18.531,2024,545,0,269,15.2606,2024,860,0,66,58.8683,2021,821,1,313,66.2604,2020,928,1,406,71.3708,2019,853,1,135,55.7647,2018,792,0,226,47.6738,2019,138,1,453,63.8216,2022,581,1,162,51.2769,2020,78,1,271,20.7859,2020,586,0,14,71.041,2018,97,0,359,77.9182,2024,799,0,2,88.4798,2019,895,1,52,97.5301,2024,416,1,463,99.9777,2023,524,0,360,73.209,2024,398,0,19,53.6179,2019,163,0,394,96.9474,2023,694,1,473,16.0837,2020,959,1,382,92.4716,2024,100,0,276,93.1951,2022,23,0,31,43.0362,2019,287,0,346,90.9689,2019,410,0,55,4.8737,2018,64,0,239,61.5046,2022,565,1,198,38.6482,2019,350,1,91,31.4986,2023,174,1,384,0.2459,2020,97,1,174,88.324,2021,293,0,363,92.2315,2020,465,0,236,80.8868,2020,831,1,479,31.4024,2023,765,1,284,37.7957,2022,980,0,462,79.9134,2019,241,0,26,3.2444,2018,359,0,280,75.8613,2020,580,1,410,6.3912,2024,456,0,94,7.1261,2021,547,1,467,14.9649,2018,59,1,177,55.4834,2023,784,1,195,99.6502,2020,808,0,62,54.8141,2022,762,0,142,19.2626,2020,278,0,357,79.7807,2018,145,1,1,62.6761,2019,327,0,51,64.5489,2019,965,0,369,44.4957,2023,439,0,37,45.6321,2021,259,1,211,47.3461,2024,835,1,459,46.4478,2018,134,0,73,81.5536,2021,957,0,93,30.1892,2018,813,0,144,15.6662,2018,766,0,282,48.2179,2019,3,1,6,30.3385,2021,459,0,285,50.5273,2023,765,0,317,84.4094,2019,146,0,65,24.2266,2018,439,0,16,0.4916,2018,322,1,342,92.5319,2021,876,1,25,24.1914,2018,126,0,356,50.9879,2018,978,0,455,24.2767,2024,404,1,126,82.2161,2022,197,1,217,83.6948,2023,311,1,113,31.8553,2019,114,0,234,7.9824,2018,833,1,184,0.75,2024,657,1,336,29.6457,2024,98,1,381,24.3893,2024,709,1,380,27.115,2020,356,1,21,2.8736,2021,873,1,216,7.0348,2023,437,0,214
+306,C025,30,Jeans,Fashion,1490.0,3,2024-03-20,1,265,71.8898,2024,152,1,286,99.7281,2020,886,0,480,98.8885,2022,947,1,413,14.0566,2023,418,0,35,92.5176,2023,42,0,3,91.2929,2018,121,1,317,30.0124,2024,866,1,67,23.3943,2018,127,0,361,1.7606,2024,57,1,395,11.0989,2023,85,1,65,12.3311,2021,260,1,283,2.2761,2022,626,1,268,84.2711,2020,178,1,312,33.4871,2024,640,0,243,7.6375,2022,342,0,117,5.8568,2023,388,0,265,63.2993,2024,0,0,472,94.8278,2021,901,1,235,7.0395,2019,895,0,401,62.4036,2018,763,1,92,43.5028,2024,440,1,368,75.7903,2022,353,0,365,63.0225,2024,136,0,282,94.3213,2021,684,1,461,38.0537,2024,920,1,309,88.3726,2019,221,1,199,23.1782,2022,890,0,164,41.9916,2023,653,0,75,87.2403,2021,656,0,230,87.2321,2022,995,0,246,60.1625,2020,469,1,256,67.9196,2022,55,1,64,96.597,2024,94,1,479,7.8106,2018,816,0,273,28.0504,2022,39,0,460,64.6547,2022,274,0,41,41.3513,2024,825,1,303,21.1781,2024,561,0,484,21.2811,2023,361,0,145,81.5397,2022,459,1,215,11.6132,2020,50,0,52,91.2426,2019,429,1,256,69.888,2018,739,1,381,12.3134,2020,356,1,77,17.7402,2024,32,0,45,51.1389,2023,546,0,12,51.7834,2021,87,0,147,19.1392,2024,755,1,162,74.0095,2023,170,1,466,82.929,2018,215,0,102,21.2847,2021,424,1,89,31.5676,2022,525,0,316,10.4367,2021,187,1,40,41.3417,2023,402,1,192,66.525,2022,934,0,256,6.051,2018,255,1,410,30.0502,2022,868,0,429,25.6013,2024,378,1,350,97.96,2024,289,1,155,93.9418,2018,529,0,289,68.7651,2021,877,1,388,27.6949,2022,845,1,41,38.2101,2018,890,1,447,8.5159,2018,695,0,272,29.809,2020,36,0,91,68.1593,2024,841,1,220,54.5223,2024,297,0,428,4.5514,2019,840,0,369,71.3366,2018,274,1,246,20.0072,2021,208,0,37,78.3636,2020,164,1,16,89.6337,2023,940,0,67,74.6614,2024,377,0,395,21.632,2019,38,1,273,53.4317,2024,468,0,355,41.7154,2023,981,0,160,28.0289,2019,165,1,415,11.2811,2019,612,1,399
+307,C026,30,Phone,Electronics,,1,2024-04-15,0,326,6.9968,2021,527,1,396,16.2917,2018,626,0,277,78.0478,2023,470,0,244,5.8679,2024,850,1,38,77.5723,2024,805,0,299,53.4032,2021,463,1,171,88.5219,2020,32,0,377,46.6825,2019,338,1,109,2.3744,2021,408,1,206,37.943,2020,630,1,204,49.855,2020,389,1,101,9.7654,2019,98,1,85,19.8423,2019,238,0,167,41.2778,2024,250,1,233,80.8563,2023,853,0,199,98.3891,2020,794,1,492,62.4195,2024,287,1,377,22.7686,2020,777,0,24,39.7175,2019,901,0,465,7.4488,2022,102,0,167,35.8765,2023,50,1,224,48.1941,2024,598,1,186,73.1922,2023,235,0,332,10.8294,2023,943,0,273,69.1898,2019,706,0,115,11.9538,2021,403,0,201,76.832,2024,387,1,481,16.0842,2020,756,0,282,63.3782,2019,510,1,39,73.1798,2022,25,1,205,88.9204,2023,325,0,411,75.7573,2023,900,1,318,98.7706,2022,277,0,5,74.7235,2024,805,1,422,88.18,2023,264,1,54,7.2064,2020,191,0,148,65.3153,2021,949,1,221,13.8518,2022,911,1,474,58.7717,2024,649,0,193,11.9982,2020,273,0,275,1.6965,2022,141,1,126,66.5856,2021,600,0,468,16.2863,2021,486,0,97,49.7177,2023,89,0,252,1.434,2023,363,1,440,95.3307,2020,849,1,482,76.2162,2018,831,0,40,93.4389,2021,713,1,404,9.3724,2023,320,1,306,48.8411,2021,837,1,157,32.8304,2021,405,1,72,2.0184,2024,374,0,174,10.4976,2021,459,0,107,46.1039,2020,547,1,212,90.4273,2022,827,0,265,5.0386,2018,579,1,361,39.1171,2023,420,0,130,85.626,2019,114,1,278,49.9748,2023,460,0,231,81.709,2021,104,0,373,5.7149,2023,574,0,194,93.2419,2018,427,0,33,11.5342,2018,942,0,19,74.3725,2024,532,1,52,93.598,2024,837,1,132,62.7859,2020,955,0,397,48.5795,2023,568,0,209,46.6561,2023,94,1,330,29.321,2023,333,1,439,75.6471,2024,52,1,31,95.8228,2022,990,0,405,76.7522,2018,910,1,389,41.8046,2018,783,1,462,81.3103,2018,781,1,245,39.1732,2023,828,0,110,98.5379,2021,797,0,355,15.6253,2021,872,1,232,74.9316,2021,433,0,89
+308,C027,365,Yoga Mat,Sports,840.0,2,2024-04-25,0,34,45.6125,2019,9,0,176,75.6653,2024,535,0,25,89.055,2024,251,1,239,98.715,2020,111,1,254,39.5214,2022,106,1,92,37.9491,2021,223,1,468,33.7572,2023,914,1,485,34.5087,2020,316,0,55,26.6298,2023,110,1,83,50.186,2024,908,1,30,40.0189,2020,731,1,310,68.8055,2019,76,0,236,39.1209,2021,748,1,205,64.4314,2018,157,0,186,65.1142,2023,627,1,46,14.0785,2020,331,1,345,32.4528,2023,943,1,82,89.5488,2021,416,0,209,45.1824,2019,431,0,182,39.6326,2023,4,0,270,35.4614,2018,807,0,74,14.5411,2021,843,1,349,35.8857,2023,88,1,485,94.1845,2018,11,1,96,67.1621,2023,835,1,346,99.6091,2022,656,0,232,96.0584,2024,940,0,304,4.6488,2018,573,1,383,36.7511,2022,202,0,465,61.211,2019,748,0,5,0.4034,2024,664,0,14,71.3824,2020,688,0,31,70.3137,2021,295,1,333,82.5791,2024,996,1,210,7.8104,2020,897,0,221,28.4976,2024,58,1,461,11.9734,2020,38,0,369,49.4185,2020,432,1,328,39.9847,2024,4,1,51,60.8819,2022,615,0,339,59.4373,2018,23,0,159,41.3952,2022,819,0,288,44.5737,2018,815,1,448,61.7015,2018,792,0,423,44.2479,2023,849,0,104,73.164,2024,578,0,121,57.1823,2024,854,1,302,87.3357,2021,877,0,33,9.3636,2018,758,0,89,61.2364,2018,27,0,182,76.0114,2024,858,0,48,1.6942,2024,423,1,356,72.7303,2023,826,1,92,76.4299,2023,658,1,345,45.4705,2024,682,0,496,11.8771,2019,887,0,461,69.2473,2021,524,1,371,25.6699,2023,380,0,360,72.1913,2018,321,0,216,36.0042,2019,928,0,191,24.8653,2020,783,1,233,45.9097,2018,128,1,175,52.8222,2022,367,0,69,12.2785,2021,427,0,264,51.3836,2020,363,1,485,30.1564,2023,567,1,434,9.7381,2021,45,0,93,90.2847,2020,618,0,200,44.5822,2024,62,0,272,44.8256,2019,971,1,310,49.3915,2023,250,0,494,55.3449,2018,281,1,272,16.9499,2022,288,1,297,41.8737,2022,77,0,44,35.3019,2020,812,0,43,22.7301,2019,866,1,215,34.08,2022,360,0,103,11.84,2018,934,0,204
+309,C028,120,Lamp,Home,910.0,3,2024-05-01,0,345,53.8047,2024,755,0,151,81.1799,2018,311,0,23,45.7429,2019,971,0,144,90.9415,2018,78,0,313,21.0576,2022,625,0,481,81.3885,2020,504,1,173,50.0556,2018,9,1,145,47.5491,2021,436,0,333,59.201,2024,966,0,403,15.1981,2018,222,0,285,6.9612,2018,759,1,138,5.3971,2019,690,1,224,19.2569,2021,839,1,140,76.582,2024,296,0,459,76.7963,2022,108,1,49,23.1429,2024,826,1,134,86.1328,2020,67,0,295,21.8408,2024,7,0,494,23.6486,2022,472,1,355,72.9151,2019,890,1,163,78.8243,2022,824,1,401,83.3729,2024,997,1,428,60.2616,2019,629,0,115,52.8901,2024,69,1,378,45.3949,2022,787,1,340,36.5666,2018,416,0,429,19.5549,2022,368,0,376,81.9871,2024,355,1,142,66.2994,2023,821,0,450,98.2593,2022,689,0,345,6.7008,2019,975,0,215,78.7417,2020,257,1,180,90.3261,2018,241,0,411,24.1213,2018,127,1,265,84.7469,2018,861,0,53,16.4333,2024,270,1,67,4.6801,2019,483,1,239,5.571,2019,400,1,161,53.6205,2018,550,0,213,65.3943,2023,633,0,248,24.3403,2018,774,1,389,9.34,2019,660,1,189,12.0439,2024,553,1,92,61.8054,2021,680,1,152,97.0545,2021,679,1,146,95.5585,2018,991,0,256,75.1767,2021,208,1,304,46.2018,2023,151,0,228,61.3194,2024,81,1,300,35.7494,2018,91,0,322,97.0763,2018,746,0,52,85.5086,2021,333,0,180,63.1792,2018,43,1,15,97.2884,2018,254,0,43,3.1781,2020,514,0,311,27.2059,2023,468,0,403,62.9852,2021,752,1,377,14.4828,2023,847,1,222,27.4199,2023,430,0,124,68.0392,2024,768,1,397,93.1844,2021,405,0,12,13.3758,2024,91,1,382,14.6875,2024,986,0,95,39.0819,2024,111,1,149,10.7066,2021,26,0,296,9.6685,2022,449,1,326,32.5647,2021,201,0,488,69.3036,2022,584,0,261,25.6277,2024,363,0,30,27.8107,2021,390,0,171,53.7022,2018,406,0,94,46.3884,2023,422,1,250,52.3901,2023,418,0,368,42.3946,2018,882,1,7,16.1292,2018,229,1,123,32.778,2024,176,0,181,76.4496,2022,716,0,167,31.3905,2022,92,0,434
+309,C028,120,Lamp,Home,910.0,3,2024-05-01,1,28,81.0418,2022,846,1,436,23.88,2020,606,1,133,91.2197,2020,892,0,280,56.4798,2023,205,1,383,78.7518,2022,645,1,24,65.9768,2018,930,0,167,40.8495,2020,523,1,297,71.8955,2018,408,0,58,17.0828,2019,866,0,353,82.0915,2018,410,0,20,82.6782,2023,690,0,186,59.7847,2021,585,1,480,88.3355,2020,847,0,96,41.1074,2023,617,1,307,43.8865,2018,980,1,92,47.3081,2018,15,0,177,74.63,2019,6,1,221,66.6641,2019,831,1,30,17.7533,2021,559,1,300,32.0634,2023,758,0,433,68.3869,2023,948,1,91,63.8775,2019,389,0,325,22.4724,2018,886,1,283,33.077,2022,838,0,314,34.0848,2018,711,1,179,38.807,2024,695,1,128,26.27,2021,815,1,237,1.7474,2018,223,1,113,96.6986,2019,235,0,315,51.0998,2020,848,1,437,38.0631,2019,485,0,439,69.9959,2023,364,0,156,15.3276,2021,567,0,195,63.0756,2019,525,0,330,33.9786,2022,29,0,332,98.1138,2021,903,1,158,27.5686,2019,830,1,274,83.7357,2019,551,1,498,99.93,2021,837,1,456,57.1197,2018,306,1,37,55.5488,2023,83,0,379,59.8664,2021,275,0,31,93.3484,2019,709,0,407,98.1271,2023,118,0,300,44.3207,2021,469,1,480,39.3946,2021,223,1,263,22.0019,2019,766,1,343,88.9471,2019,70,1,441,55.8703,2018,493,0,467,34.2847,2019,931,1,338,77.6915,2020,939,1,81,87.6392,2021,759,1,62,24.9775,2024,355,1,34,32.1206,2023,454,0,130,32.6755,2022,21,1,399,3.7138,2022,446,0,199,33.5866,2023,454,0,306,75.8617,2021,715,1,20,56.2796,2021,176,1,125,22.7695,2024,371,1,470,13.8614,2024,278,0,129,84.2751,2024,637,0,268,51.9326,2020,245,1,272,5.3236,2023,457,0,156,39.6928,2021,597,0,485,15.3047,2024,818,0,39,98.6302,2024,147,1,160,78.0612,2023,200,1,266,6.1672,2018,989,0,340,20.5161,2022,65,1,167,52.3368,2024,406,0,231,85.7274,2020,436,1,167,24.4553,2023,489,0,187,52.2671,2022,270,1,367,24.6744,2023,439,1,379,32.4349,2021,691,1,210,96.276,2021,776,0,362,83.8866,2023,581,1,379

ceo_brief_env/tasks/medium_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,1.0
+total_revenue,76562.5
+top_category,Electronics
+projection_next_quarter,123243.75
+variance_pct,82.29
+break_even_units,15.0
+memo_score,0.725

ceo_brief_env/tasks/medium_brief/metadata.json

@@ -0,0 +1,21 @@
+{
+  "title": "Category growth and Q4 forecast",
+  "difficulty": "medium",
+  "max_steps": 10,
+  "instruction": "Which category drove Q3 growth, what is the Q4 forecast plus variance versus plan, what should we do about it, and send the leads a heads-up.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "strategy",
+    "hr"
+  ],
+  "memo_audience": "department leads",
+  "hr_required_terms": [
+    "forecast",
+    "variance",
+    "leads"
+  ],
+  "plan_value": 42000,
+  "fixed_cost": 18000,
+  "unit_margin": 1200
+}

ceo_brief_env/tasks/medium_brief/raw.csv

@@ -0,0 +1,10 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+201,C010,30,Serum,Beauty,1250.0,2,2024-01-04,1,329,5.4567,2020,911,1,168,87.6315,2021,238,0,268,1.8023,2022,780,0,497,22.5355,2020,907,1,427,41.8442,2021,633,1,393,64.8436,2018,5,0,267,42.2249,2020,386,0,117,50.0556,2024,595,1,400,23.3943,2021,316,1,361,93.7307,2020,466,1,31,70.6459,2023,780,1,222,52.6644,2021,826,0,365,90.6417,2024,97,1,137,63.5336,2021,540,1,146,17.8335,2019,533,0,206,80.7577,2022,485,0,404,31.5316,2022,660,0,134,23.1429,2018,858,1,11,63.2993,2020,943,1,472,23.732,2024,993,0,308,5.9954,2024,231,0,129,3.0805,2020,320,0,3,40.8466,2019,358,0,330,65.3289,2023,197,1,333,35.387,2021,624,1,113,12.6743,2020,564,0,55,24.5513,2021,54,1,343,45.3949,2018,80,1,280,88.3726,2018,656,0,199,90.6092,2024,466,0,321,47.3513,2024,836,0,196,85.1208,2022,966,1,102,32.5837,2021,731,0,68,19.3957,2022,27,1,158,46.964,2024,49,0,350,73.9902,2022,127,1,493,38.0783,2021,300,1,190,24.1213,2020,312,1,307,28.0504,2018,897,1,460,15.8295,2022,821,0,487,31.6536,2019,873,1,111,4.8678,2020,837,0,216,79.7123,2024,587,1,69,32.9468,2020,724,1,351,45.9584,2021,346,1,278,79.2483,2021,102,0,333,37.6627,2024,44,0,461,12.0439,2021,694,1,479,12.3134,2021,792,0,77,36.4303,2019,186,1,417,96.0293,2018,798,1,2,62.6016,2022,220,1,427,70.4983,2023,934,0,76,50.7793,2021,989,0,380,17.0002,2019,43,1,172,17.7742,2024,203,1,164,17.8916,2022,684,0,425,85.5086,2021,631,0,485,10.4367,2018,826,0,40,71.4861,2023,525,1,151,25.9336,2018,680,0,319,15.6732,2022,37,1,443,41.8112,2023,391,1,400,75.0757,2023,436,1,467,37.8447,2022,804,0,281,52.2804,2018,310,0,408,59.5252,2018,950,1,449,93.1844,2023,366,0,58,27.6949,2024,128,0,41,88.284,2019,776,0,340,54.3874,2023,6,0,258,69.223,2018,396,0,182,73.65,2019,627,1,433,85.9166,2022,573,1,438,29.7823,2022,35,1,390,74.0223,2022,737,0,147,80.7792,2022,146,0,185,27.8107,2020,938,1,124
+202,C011,45,Headphones,Electronics,3100.0,2,2024-01-11,0,54,75.3739,2021,438,0,306,57.2446,2024,471,0,156,67.0035,2020,5,1,388,70.1049,2021,647,1,430,11.1707,2022,619,1,498,65.9199,2021,598,1,5,12.7113,2024,800,0,453,40.8495,2024,133,1,371,46.6825,2018,436,0,109,55.3715,2024,866,1,200,91.4978,2019,320,0,283,12.8289,2018,50,1,379,33.769,2024,761,1,35,20.2785,2024,570,0,94,74.8463,2023,249,0,330,9.4579,2021,466,0,458,70.7117,2018,667,0,261,47.3081,2020,354,1,377,62.4195,2019,67,1,377,55.5247,2021,571,1,387,16.9309,2019,224,1,480,80.2235,2022,775,1,445,14.0649,2020,891,0,257,10.5889,2019,581,1,435,84.3114,2024,950,1,323,16.1198,2023,665,1,104,55.3777,2021,763,1,298,34.0848,2024,51,0,65,11.9538,2024,416,0,201,28.0433,2018,99,1,94,63.7886,2023,302,1,27,14.9061,2024,645,0,410,31.1881,2024,840,1,414,68.6213,2024,993,0,394,66.4764,2023,404,1,391,98.7845,2021,635,0,209,89.6202,2024,227,0,82,63.0756,2023,136,1,32,88.18,2022,861,0,54,91.0177,2024,440,1,225,24.3231,2018,457,1,480,60.569,2019,309,0,200,79.0403,2022,132,1,125,72.2503,2024,36,1,146,61.5763,2022,671,0,403,66.4277,2019,251,1,354,24.8371,2020,758,0,21,93.3484,2023,929,1,244,49.7177,2023,680,0,252,93.0922,2024,266,0,356,77.7966,2024,421,1,152,2.2151,2021,262,0,104,83.7163,2023,406,1,415,38.7945,2024,147,1,406,75.8981,2022,74,1,402,83.714,2019,312,1,80,54.031,2019,898,0,255,87.6392,2021,463,1,347,10.4976,2024,43,1,107,45.0808,2022,314,1,289,63.4024,2018,603,1,188,51.1038,2018,636,1,234,0.4813,2023,660,1,455,66.4586,2022,181,0,104,38.0364,2018,910,1,358,5.1409,2020,462,0,459,2.0633,2019,725,1,42,13.8614,2024,65,1,28,93.2419,2024,91,1,33,39.8259,2022,759,0,458,67.2145,2020,997,1,293,54.4408,2020,671,0,14,20.7163,2023,503,1,72,59.7554,2018,665,0,420,4.589,2018,845,0,290,30.4665,2018,660,1,330,8.0854,2021,566,1,396,20.5161,2019,892,0,331
+203,C012,90,Shoes,Fashion,2050.0,2,2024-02-08,1,123,22.6578,2024,251,1,415,35.1768,2020,749,1,303,19.7217,2021,890,1,354,63.5543,2019,735,1,151,20.5395,2024,841,0,23,24.1479,2019,184,1,190,46.541,2023,283,0,77,23.7837,2018,919,0,59,34.5087,2023,408,0,55,66.2604,2020,122,0,209,52.671,2023,918,1,471,5.8712,2021,512,0,345,71.0826,2024,688,0,444,77.9019,2020,638,0,234,84.7583,2020,856,1,397,1.7105,2018,371,1,325,83.5863,2023,823,0,355,68.6175,2023,334,1,330,32.4528,2020,6,0,82,88.4798,2019,125,0,457,13.0438,2023,195,1,10,36.4405,2021,880,1,379,6.6713,2022,354,1,2,81.5716,2022,536,1,229,38.9362,2023,284,1,130,95.6643,2021,970,1,470,12.6803,2022,975,1,462,0.4444,2022,983,1,117,99.6091,2022,695,0,232,43.0362,2019,742,1,460,68.0748,2020,27,1,207,76.6094,2019,396,1,118,8.1789,2021,612,1,100,40.2421,2024,386,1,440,48.5996,2019,3,0,431,96.8785,2022,61,1,351,49.965,2023,300,1,441,18.2568,2023,43,1,392,7.8104,2019,29,0,221,80.8868,2020,133,1,139,52.7624,2022,335,1,418,73.8316,2024,245,1,276,16.2488,2024,399,0,173,42.7386,2024,32,0,341,30.6011,2021,789,0,154,51.7819,2022,316,1,207,67.4651,2018,67,1,36,8.9182,2021,601,1,332,61.7015,2018,118,1,423,55.4834,2023,146,0,191,45.4332,2022,951,1,233,24.1036,2021,692,1,383,32.0905,2022,873,0,436,86.197,2018,797,0,27,49.3552,2018,498,1,452,31.3454,2019,363,0,380,95.4845,2024,169,0,223,3.1975,2020,60,0,415,72.7303,2020,355,1,92,47.3461,2024,725,0,24,37.3078,2019,919,1,100,99.2602,2024,623,1,224,77.1028,2018,692,0,278,49.8454,2019,156,1,364,71.5253,2021,154,0,181,0.8273,2018,431,1,180,23.1721,2021,200,1,253,58.5497,2022,749,1,292,45.9097,2022,637,1,175,24.2266,2018,16,1,219,99.9074,2020,445,1,53,52.7364,2018,873,0,299,70.8662,2023,301,0,118,94.227,2019,509,1,256,14.7357,2024,550,1,31,5.0866,2021,400,1,223,96.4072,2022,590,1,418,96.7052,2018,135,0,349
+204,C013,60,Rug,Home,2300.0,1,2024-02-18,1,29,11.2991,2023,883,0,402,87.2184,2019,607,0,359,70.9287,2023,796,1,412,61.5928,2022,355,1,34,76.8981,2018,418,1,212,4.2389,2022,962,0,88,8.6193,2020,808,1,466,84.918,2024,2,0,490,47.5491,2019,838,0,333,1.7606,2024,928,0,329,1.1168,2023,78,1,373,3.9599,2021,551,0,108,35.8678,2020,118,1,70,29.5505,2020,755,1,270,48.473,2022,434,0,153,64.0305,2022,918,1,207,7.1317,2021,326,1,62,29.2405,2020,528,1,375,86.1328,2021,207,1,295,94.8278,2021,895,0,400,24.3749,2019,703,0,100,60.0091,2018,654,1,472,11.9559,2023,139,1,199,25.3447,2021,647,1,484,79.7165,2023,800,1,147,13.6536,2024,228,1,121,61.8415,2020,976,0,465,73.0891,2023,424,1,344,36.5666,2018,863,1,429,23.1782,2022,287,1,224,10.0628,2018,869,1,38,22.6341,2022,92,1,60,22.1585,2021,192,1,282,36.7122,2020,741,0,133,88.652,2021,307,1,348,5.5991,2018,359,1,329,54.2622,2021,463,1,296,96.4461,2024,621,1,7,84.7469,2024,254,0,53,64.6547,2022,831,1,497,45.6478,2019,208,0,292,54.8415,2018,877,1,249,58.3708,2021,473,1,121,81.9548,2018,82,1,426,5.526,2021,24,0,358,5.0407,2019,777,0,115,51.0319,2020,587,0,223,51.0652,2023,492,1,194,61.8054,2021,304,0,152,17.7402,2024,784,1,91,86.1215,2020,7,1,304,52.5254,2023,72,1,369,51.2647,2018,439,1,139,5.8726,2023,854,1,473,11.2657,2024,677,0,115,21.549,2023,644,0,25,36.8087,2024,983,0,467,24.8221,2020,34,1,157,63.1792,2022,722,1,15,41.3417,2023,835,1,338,69.7436,2020,294,1,114,79.9118,2018,347,0,492,50.421,2020,250,1,407,85.4137,2018,622,1,375,82.9701,2022,860,0,118,28.924,2024,248,0,168,86.336,2021,862,0,250,52.0063,2023,140,1,73,13.3758,2018,911,0,382,38.2101,2018,439,1,175,99.8908,2018,942,0,295,31.2558,2024,741,1,397,58.2419,2019,3,1,63,65.113,2018,303,1,243,85.6043,2023,792,0,252,84.0284,2022,522,0,213,11.486,2019,91,0,254,64.4252,2024,731,0,60
+205,C014,75,Cycle,Sports,14500.0,1,2024-03-05,0,47,42.714,2023,677,1,165,58.4566,2023,884,1,373,96.1176,2018,457,1,473,28.2974,2024,140,1,397,39.5286,2018,282,0,484,10.633,2019,47,0,407,21.4044,2018,238,1,150,96.1176,2018,996,1,128,71.8955,2020,820,0,58,2.3744,2021,57,1,249,3.0805,2023,43,0,401,89.3017,2021,497,1,10,56.5891,2019,53,0,313,96.6426,2020,842,0,167,41.1131,2020,743,0,216,15.7017,2018,614,0,384,4.2352,2020,500,1,30,72.2705,2023,780,1,142,74.63,2023,365,0,221,22.7686,2020,901,1,163,47.4782,2020,459,0,229,20.7373,2020,649,1,315,18.2912,2019,788,1,220,58.1929,2023,757,0,105,91.4205,2020,902,1,295,8.8267,2020,565,1,264,22.6258,2023,698,1,190,97.5929,2023,892,0,148,38.807,2021,293,1,128,76.832,2024,890,0,165,71.3714,2022,71,0,345,28.9301,2022,860,1,98,45.8999,2022,982,1,497,53.7273,2020,601,0,102,16.5588,2023,374,1,98,68.8166,2023,311,0,451,78.0545,2023,254,1,39,13.1254,2018,666,1,300,33.9786,2021,470,0,332,7.2064,2020,274,1,246,95.9976,2021,238,0,140,24.5924,2023,886,1,94,96.8568,2021,536,1,181,87.2666,2024,815,1,52,16.7124,2021,303,1,139,2.3419,2023,758,0,47,84.4644,2021,185,1,349,5.7473,2021,575,0,243,98.1271,2019,524,1,300,1.434,2023,32,1,248,39.563,2018,538,1,20,88.8027,2022,545,1,344,29.387,2021,462,1,378,58.596,2019,740,0,31,77.878,2022,874,0,308,2.7304,2019,676,0,485,64.8237,2019,239,0,158,68.2368,2020,534,0,101,24.9775,2021,482,0,34,46.1039,2020,402,1,373,18.1263,2022,689,1,77,62.7412,2022,551,0,88,81.9682,2024,629,1,434,27.1523,2021,256,1,246,18.9959,2021,141,0,344,32.1899,2018,249,0,340,96.0673,2020,277,0,344,48.6141,2020,914,0,215,84.2751,2024,197,1,268,11.5342,2018,890,0,261,20.2355,2021,727,0,211,65.9521,2024,650,0,253,18.1034,2022,96,0,420,27.0103,2021,545,0,446,82.6798,2020,680,0,146,61.8234,2023,532,1,128,86.0943,2022,263,1,100,47.6325,2020,665,0,419
+206,C015,75,Phone,Electronics,30100.0,1,2024-03-17,1,162,89.2765,2021,699,1,102,53.7339,2023,762,1,381,4.861,2023,104,1,235,95.9221,2022,250,0,108,8.2629,2021,191,0,126,64.5625,2022,335,0,193,54.928,2021,537,1,291,75.0386,2021,866,1,478,50.1232,2022,531,1,173,26.6298,2023,408,1,406,71.3708,2019,208,0,2,75.9259,2020,37,1,436,62.1067,2021,432,0,49,4.7471,2020,247,0,240,55.7923,2020,563,1,201,61.7538,2019,346,0,278,14.6176,2018,26,0,145,29.7861,2018,500,1,45,19.9693,2022,843,1,434,89.5488,2021,777,1,52,97.5301,2024,430,1,409,52.6488,2020,825,1,218,54.0921,2023,500,0,25,75.7624,2024,624,0,349,35.597,2022,842,0,396,88.638,2020,954,0,180,54.2351,2024,764,0,59,71.9382,2022,977,1,115,5.2908,2021,790,0,439,96.0584,2024,387,0,346,90.9689,2024,307,0,474,74.1567,2020,137,0,284,92.996,2022,56,0,13,44.1694,2022,712,1,136,77.8202,2018,435,1,394,36.4826,2019,736,0,220,37.7807,2023,348,0,239,88.8664,2021,683,0,445,81.782,2019,405,0,25,28.4976,2024,191,0,479,31.4024,2024,70,1,266,54.0101,2021,638,1,458,65.57,2023,512,1,325,9.3946,2021,468,0,372,91.0575,2022,420,1,171,8.3443,2019,277,0,373,73.1571,2024,187,0,117,14.1166,2022,918,1,200,31.7797,2018,286,1,316,44.2479,2023,363,0,195,99.6502,2021,539,0,207,86.5166,2022,572,0,331,60.4155,2019,58,0,356,20.655,2021,716,0,217,75.7246,2019,903,1,89,93.1111,2022,162,1,87,20.6699,2022,737,1,470,88.7305,2023,126,1,479,58.5555,2022,839,0,153,76.4299,2023,547,1,459,46.4478,2021,720,0,119,5.0934,2018,538,0,56,92.2164,2020,196,0,210,74.7193,2019,825,1,38,56.7895,2023,537,1,101,17.6442,2021,647,0,91,60.7351,2020,570,1,278,31.3895,2019,119,0,277,15.4151,2024,706,1,233,52.8222,2022,942,1,16,0.4916,2022,130,0,329,74.4885,2021,568,1,370,97.1911,2019,510,1,478,33.2489,2023,313,0,76,8.6017,2019,226,1,455,20.0141,2021,304,1,34,29.8372,2022,434,1,252,15.1072,2018,459,1,206
+207,C016,30,Tablet,Electronics,,1,2024-04-03,0,265,71.8898,2019,116,1,280,87.9216,2023,320,0,382,26.6246,2022,912,0,126,17.4748,2021,58,1,367,93.0463,2022,631,1,386,9.6227,2023,548,0,330,12.1544,2023,634,0,442,32.835,2022,918,0,19,59.5029,2023,840,0,340,59.201,2024,110,1,395,11.0989,2019,26,1,12,42.8658,2020,530,1,96,37.1809,2021,597,0,39,16.3902,2023,198,1,130,64.8193,2019,807,0,433,30.7963,2022,256,1,219,34.2227,2019,232,0,491,12.2157,2024,438,0,41,73.7527,2023,284,0,222,21.8408,2024,416,0,235,7.0395,2019,411,0,434,49.7094,2020,18,1,469,86.6347,2018,683,1,403,96.3588,2023,481,1,362,62.7104,2023,427,1,471,14.8864,2021,1,0,166,68.4557,2018,922,0,346,10.6033,2022,953,1,136,64.1519,2018,560,0,78,19.5549,2022,940,1,164,41.9916,2021,430,1,370,86.3754,2021,546,0,290,62.9747,2018,510,1,374,3.6041,2024,889,1,421,27.3496,2021,567,0,308,7.7162,2019,277,1,77,9.4002,2020,959,0,373,14.6274,2022,20,1,70,9.0394,2021,845,1,221,16.4333,2024,58,0,41,41.3513,2022,41,0,362,20.6832,2022,886,1,103,99.787,2024,999,0,3,12.5775,2019,119,1,381,49.5031,2021,898,1,152,60.3482,2022,830,0,299,42.9194,2019,511,0,82,22.063,2018,120,0,204,68.7238,2018,631,1,83,97.0545,2021,849,1,45,51.1389,2018,754,0,398,82.4235,2019,495,1,21,68.6911,2018,889,1,438,40.0348,2018,93,1,29,46.1159,2022,269,0,70,6.4716,2024,153,0,388,42.4485,2023,178,0,11,46.2245,2023,770,0,142,77.2047,2019,442,0,422,97.2884,2018,658,1,192,66.525,2018,930,1,278,44.7849,2019,97,1,343,39.8109,2024,335,0,262,76.8382,2021,856,1,456,42.8609,2021,334,1,457,16.685,2022,41,1,114,52.9922,2021,578,1,29,93.1926,2021,243,0,166,39.3911,2019,110,0,157,14.6875,2024,367,0,447,8.5159,2022,388,1,32,39.743,2021,343,1,165,97.1287,2019,153,1,284,22.6679,2021,485,0,293,5.9869,2024,923,0,87,10.9361,2018,730,1,31,27.4888,2022,492,0,378,68.7228,2023,732,0,32
+208,C017,60,Perfume,Beauty,2400.0,2,2024-04-12,0,326,6.9968,2020,471,0,305,19.4635,2023,66,1,496,87.4468,2024,640,0,485,82.6297,2021,983,0,100,54.2937,2023,231,0,430,80.8464,2021,535,1,12,22.366,2024,341,1,235,57.3504,2021,66,1,265,10.2271,2018,602,1,130,17.0828,2019,966,1,206,37.943,2024,834,0,358,64.0517,2018,303,0,197,55.1762,2018,747,1,345,53.6794,2018,853,1,204,24.3553,2019,828,1,450,37.3206,2021,921,0,355,62.7176,2024,397,0,489,27.5355,2018,289,0,59,95.8878,2018,183,0,109,66.6641,2019,7,0,24,39.7175,2018,96,1,374,25.6543,2023,643,1,179,79.9477,2021,397,1,412,73.7287,2020,122,1,164,30.9832,2022,653,1,363,70.6665,2022,571,1,197,1.1548,2024,546,0,121,29.8997,2022,529,0,122,21.2284,2019,456,0,455,26.27,2021,368,0,481,16.0842,2023,438,1,26,70.3823,2022,235,1,231,92.3911,2024,576,0,344,49.1334,2021,901,0,166,1.5027,2024,848,1,419,31.0963,2018,559,1,388,29.5128,2022,9,1,460,70.2901,2018,219,1,40,61.8301,2018,373,0,112,98.1138,2021,270,1,148,65.3153,2023,815,1,44,19.1692,2018,697,1,3,58.8934,2019,854,1,275,29.0303,2023,589,1,193,97.9875,2018,223,0,326,13.3591,2019,433,1,14,81.9647,2021,937,1,490,65.3571,2024,113,1,488,34.4021,2020,781,1,278,44.3207,2021,679,0,440,95.3307,2023,78,0,264,76.7891,2023,42,0,190,13.2315,2019,776,0,76,32.3114,2024,266,0,295,26.3068,2024,171,0,254,58.4507,2021,174,0,300,85.8559,2019,143,1,279,59.4612,2019,665,0,135,53.8899,2019,519,0,186,32.1206,2023,254,1,212,90.4273,2023,188,1,48,0.0412,2019,85,0,170,89.3062,2020,148,1,376,85.9028,2019,293,1,219,54.3066,2018,549,0,375,99.3657,2022,34,1,41,92.8819,2018,746,1,470,32.7056,2021,392,0,358,59.4006,2023,249,0,195,51.9326,2020,986,0,19,74.3725,2022,114,0,338,11.5208,2018,991,0,263,2.1276,2022,793,1,225,44.0696,2023,904,0,307,72.72,2020,626,1,400,15.9934,2021,43,1,414,6.2271,2018,878,0,460,99.7837,2019,612,0,381
+208,C017,60,Perfume,Beauty,2400.0,2,2024-04-12,0,34,45.6125,2024,312,0,175,8.2049,2023,902,1,94,18.5698,2019,949,0,446,47.8692,2022,987,0,25,24.3007,2019,952,1,293,25.8798,2018,318,1,489,93.0111,2023,937,1,169,54.1033,2022,670,1,294,36.4716,2020,142,1,173,77.6098,2024,866,0,83,50.186,2019,821,1,313,55.7647,2021,286,0,345,80.2094,2024,425,0,293,47.193,2022,391,1,423,59.7603,2022,512,1,37,27.5346,2021,539,0,248,98.0003,2022,91,0,71,48.2215,2019,994,0,389,82.5785,2020,336,0,361,43.8958,2019,831,0,209,45.1824,2024,639,1,329,99.9777,2019,791,0,473,79.3618,2019,533,1,474,14.6466,2022,145,0,234,40.2764,2022,969,1,363,40.4386,2024,546,1,311,83.8329,2021,191,1,336,43.8155,2024,198,1,466,68.2438,2020,781,0,192,25.796,2024,815,0,304,4.6488,2019,992,0,238,4.8737,2018,673,1,141,19.3653,2023,501,1,424,0.9585,2022,4,0,435,84.1431,2023,314,1,194,90.3502,2023,455,0,148,56.7554,2019,666,0,413,71.3292,2018,836,1,461,38.6077,2019,226,0,428,43.092,2023,903,1,461,11.9734,2023,774,0,248,37.7957,2020,368,1,168,21.0916,2019,304,1,418,10.0847,2020,608,0,161,63.3442,2024,758,0,55,21.3317,2023,587,0,279,27.5169,2022,575,1,223,79.2744,2018,145,1,75,10.9557,2018,848,0,494,54.7042,2023,469,1,104,73.164,2020,861,0,60,54.8141,2021,909,1,6,32.952,2022,9,0,36,6.5099,2020,93,0,154,21.832,2020,626,1,17,31.6463,2018,992,1,51,93.964,2020,96,1,9,22.0469,2023,280,1,228,74.9672,2021,930,0,471,39.8195,2021,454,0,345,45.4705,2018,183,1,399,81.5536,2018,201,1,428,15.475,2020,260,0,227,74.2783,2023,256,0,64,29.3621,2019,482,1,54,99.3205,2018,397,1,41,37.164,2024,382,0,125,51.8648,2024,590,1,422,7.1596,2020,868,0,57,26.298,2020,245,0,69,12.2785,2018,782,0,122,92.5319,2024,395,1,131,36.8383,2020,829,1,409,86.8219,2018,97,0,54,78.7951,2021,242,0,201,13.8689,2019,155,1,248,98.9428,2019,544,0,224,20.8499,2022,813,1,159

ceo_brief_env/tasks/risk_brief/ground_truth.csv

@@ -0,0 +1,8 @@
+metric,value
+data_quality_score,0.8333
+total_revenue,271309.09
+top_category,Electronics
+projection_next_quarter,-64505.45
+variance_pct,117.05
+break_even_units,25.88
+memo_score,0.703

ceo_brief_env/tasks/risk_brief/metadata.json

@@ -0,0 +1,22 @@
+{
+  "title": "Board risk brief — data quality, compliance exposure, and capital allocation",
+  "difficulty": "hard",
+  "max_steps": 14,
+  "instruction": "Prepare a board-level risk brief after an internal audit flagged data-quality gaps, delivery exposure, and uncertainty in the operating plan. Run the full office: quantify data issues, finance downside and break-even sensitivity, strategy stance on NVDA/AAPL/JPM under risk controls, and an HR/comms memo that names variance, projection, strategy, and risk governance. The final brief must be suitable for audit committee review.",
+  "required_experts": [
+    "analyst",
+    "finance",
+    "strategy",
+    "hr"
+  ],
+  "memo_audience": "Board audit committee and CFO",
+  "hr_required_terms": [
+    "variance",
+    "projection",
+    "strategy",
+    "risk"
+  ],
+  "plan_value": 125000,
+  "fixed_cost": 44000,
+  "unit_margin": 1700
+}

ceo_brief_env/tasks/risk_brief/raw.csv

@@ -0,0 +1,13 @@
+OrderID,CustomerID,ExpiryDays,Product,Category,Price,Quantity,OrderDate,dim_0001,dim_0002,dim_0003,dim_0004,dim_0005,dim_0006,dim_0007,dim_0008,dim_0009,dim_0010,dim_0011,dim_0012,dim_0013,dim_0014,dim_0015,dim_0016,dim_0017,dim_0018,dim_0019,dim_0020,dim_0021,dim_0022,dim_0023,dim_0024,dim_0025,dim_0026,dim_0027,dim_0028,dim_0029,dim_0030,dim_0031,dim_0032,dim_0033,dim_0034,dim_0035,dim_0036,dim_0037,dim_0038,dim_0039,dim_0040,dim_0041,dim_0042,dim_0043,dim_0044,dim_0045,dim_0046,dim_0047,dim_0048,dim_0049,dim_0050,dim_0051,dim_0052,dim_0053,dim_0054,dim_0055,dim_0056,dim_0057,dim_0058,dim_0059,dim_0060,dim_0061,dim_0062,dim_0063,dim_0064,dim_0065,dim_0066,dim_0067,dim_0068,dim_0069,dim_0070,dim_0071,dim_0072,dim_0073,dim_0074,dim_0075,dim_0076,dim_0077,dim_0078,dim_0079,dim_0080,dim_0081,dim_0082,dim_0083,dim_0084,dim_0085,dim_0086,dim_0087,dim_0088,dim_0089,dim_0090,dim_0091,dim_0092,dim_0093,dim_0094,dim_0095,dim_0096,dim_0097,dim_0098,dim_0099,dim_0100,dim_0101,dim_0102,dim_0103,dim_0104,dim_0105,dim_0106,dim_0107,dim_0108,dim_0109,dim_0110,dim_0111,dim_0112,dim_0113,dim_0114,dim_0115,dim_0116,dim_0117,dim_0118,dim_0119,dim_0120,dim_0121,dim_0122,dim_0123,dim_0124,dim_0125,dim_0126,dim_0127,dim_0128,dim_0129,dim_0130,dim_0131,dim_0132,dim_0133,dim_0134,dim_0135,dim_0136,dim_0137,dim_0138,dim_0139,dim_0140,dim_0141,dim_0142,dim_0143,dim_0144,dim_0145,dim_0146,dim_0147,dim_0148,dim_0149,dim_0150,dim_0151,dim_0152,dim_0153,dim_0154,dim_0155,dim_0156,dim_0157,dim_0158,dim_0159,dim_0160,dim_0161,dim_0162,dim_0163,dim_0164,dim_0165,dim_0166,dim_0167,dim_0168,dim_0169,dim_0170,dim_0171,dim_0172,dim_0173,dim_0174,dim_0175,dim_0176,dim_0177,dim_0178,dim_0179,dim_0180,dim_0181,dim_0182,dim_0183,dim_0184,dim_0185,dim_0186,dim_0187,dim_0188,dim_0189,dim_0190,dim_0191,dim_0192,dim_0193,dim_0194,dim_0195,dim_0196,dim_0197,dim_0198,dim_0199,dim_0200,dim_0201,dim_0202,dim_0203,dim_0204,dim_0205,dim_0206,dim_0207,dim_0208,dim_0209,dim_0210,dim_0211,dim_0212,dim_0213,dim_0214,dim_0215,dim_0216,dim_0217,dim_0218,dim_0219,dim_0220,dim_0221,dim_0222,dim_0223,dim_0224,dim_0225,dim_0226,dim_0227,dim_0228,dim_0229,dim_0230,dim_0231,dim_0232,dim_0233,dim_0234,dim_0235,dim_0236,dim_0237,dim_0238,dim_0239,dim_0240,dim_0241,dim_0242,dim_0243,dim_0244,dim_0245,dim_0246,dim_0247,dim_0248,dim_0249,dim_0250,dim_0251,dim_0252,dim_0253,dim_0254,dim_0255,dim_0256,dim_0257,dim_0258,dim_0259,dim_0260,dim_0261,dim_0262,dim_0263,dim_0264,dim_0265,dim_0266,dim_0267,dim_0268,dim_0269,dim_0270,dim_0271,dim_0272,dim_0273,dim_0274,dim_0275,dim_0276,dim_0277,dim_0278,dim_0279,dim_0280,dim_0281,dim_0282,dim_0283,dim_0284,dim_0285,dim_0286,dim_0287,dim_0288,dim_0289,dim_0290,dim_0291,dim_0292,dim_0293,dim_0294,dim_0295,dim_0296,dim_0297,dim_0298,dim_0299,dim_0300,dim_0301,dim_0302,dim_0303,dim_0304,dim_0305,dim_0306,dim_0307,dim_0308,dim_0309,dim_0310,dim_0311,dim_0312,dim_0313,dim_0314,dim_0315,dim_0316,dim_0317,dim_0318,dim_0319,dim_0320,dim_0321,dim_0322,dim_0323,dim_0324,dim_0325,dim_0326,dim_0327,dim_0328,dim_0329,dim_0330,dim_0331,dim_0332,dim_0333,dim_0334,dim_0335,dim_0336,dim_0337,dim_0338,dim_0339,dim_0340,dim_0341,dim_0342,dim_0343,dim_0344,dim_0345,dim_0346,dim_0347,dim_0348,dim_0349,dim_0350,dim_0351,dim_0352,dim_0353,dim_0354,dim_0355,dim_0356,dim_0357,dim_0358,dim_0359,dim_0360,dim_0361,dim_0362,dim_0363,dim_0364,dim_0365,dim_0366,dim_0367,dim_0368,dim_0369,dim_0370,dim_0371,dim_0372,dim_0373,dim_0374,dim_0375,dim_0376,dim_0377,dim_0378,dim_0379,dim_0380,dim_0381,dim_0382,dim_0383,dim_0384,dim_0385,dim_0386,dim_0387,dim_0388,dim_0389,dim_0390,dim_0391,dim_0392
+701,C400,30,Workstation,Electronics,96000.0,1.0,2024-01-05,1,29,11.2991,2024,570,1,50,81.1799,2023,469,1,398,35.0699,2024,482,1,34,76.8981,2020,913,1,301,49.0841,2019,311,0,404,87.1558,2018,700,1,490,47.5491,2023,213,0,403,17.7827,2022,882,1,40,2.8595,2021,337,0,108,35.8678,2018,173,0,187,11.7391,2021,984,0,218,76.582,2023,619,0,207,7.1317,2020,75,0,480,9.2095,2023,602,0,104,2.6198,2023,945,0,400,24.3749,2022,413,0,123,72.9151,2024,757,0,70,77.7896,2023,787,1,484,79.7165,2023,877,0,111,68.4621,2019,984,1,488,44.3529,2022,647,1,344,36.5666,2024,436,0,199,45.5387,2021,749,1,434,64.7971,2021,535,1,60,22.1585,2019,596,0,267,46.6497,2019,27,1,154,78.7417,2023,205,1,329,54.2622,2021,31,0,73,31.2542,2018,847,1,235,92.122,2021,555,1,246,95.9976,2021,737,0,104,83.7357,2018,583,1,268,13.6289,2021,742,1,52,16.7124,2018,243,1,302,10.2863,2019,510,0,93,92.223,2024,860,1,243,98.1271,2022,827,0,485,18.6065,2021,861,1,269,0.2961,2018,751,0,344,29.387,2021,145,1,200,98.955,2024,112,1,437,34.2847,2020,429,1,485,64.8237,2020,551,1,231,63.1177,2019,631,1,241,7.9994,2018,846,1,373,18.1263,2022,878,0,224,3.7138,2019,504,1,315,80.1548,2023,167,0,246,18.9959,2023,274,1,84,31.0602,2022,863,0,139,89.972,2022,567,1,215,84.2751,2018,465,1,74,77.668,2020,998,0,364,51.6699,2018,107,0,253,18.1034,2022,884,1,114,57.335,2023,856,0,340,78.0612,2022,507,1,128,86.0943,2018,240,0,343,93.8413,2019,537,1,389,3.1262,2024,377,1,393,68.1358,2022,293,0,229,52.2671,2020,601,1,81,71.0462,2023,610,0,257,71.6916,2018,764,1,6,83.0219,2019,739,0,364,49.1545,2018,521,1,146,38.7768,2023,903,1,86,48.2355,2023,478,0,10,60.0645,2021,558,0,49,51.997,2023,544,0,30,68.7592,2018,131,1,101,91.4483,2023,878,0,486,88.0248,2021,599,1,255,67.7937,2019,92,1,42,6.0513,2023,403,0,336,92.4016,2019,585,0,199,71.6468,2018,0,0,31,46.115,2023,963,0,385
+702,C401,45,Phone,Electronics,43000.0,,2024-01-17,0,47,42.714,2022,388,1,439,23.88,2023,709,0,229,99.5758,2019,551,1,397,39.5286,2019,210,1,49,0.5169,2023,86,1,120,23.3109,2020,477,1,128,71.8955,2023,922,1,296,46.6067,2021,11,0,22,44.3949,2021,69,1,10,56.5891,2018,529,1,82,54.0554,2020,484,1,372,41.1074,2020,447,0,384,4.2352,2021,352,0,61,85.8191,2019,861,0,183,94.536,2019,198,1,163,47.4782,2021,688,0,249,32.0634,2022,119,0,394,66.1202,2024,356,1,105,91.4205,2024,602,0,75,56.426,2021,618,1,349,68.7099,2018,689,0,148,38.807,2020,35,1,98,46.647,2020,100,1,36,39.199,2024,662,0,98,45.8999,2024,419,0,18,2.7541,2019,886,1,187,69.9959,2018,15,0,451,78.0545,2019,906,1,231,4.3398,2024,734,0,203,44.1864,2024,758,1,479,31.4024,2018,251,1,115,24.5504,2024,925,1,256,34.494,2018,428,0,372,91.0575,2020,52,1,206,31.6637,2023,825,0,94,7.1261,2024,36,1,200,31.7797,2020,354,1,313,14.6798,2020,931,1,269,46.5135,2022,450,1,331,60.4155,2018,377,1,357,79.7807,2021,595,0,451,90.5237,2020,100,0,87,20.6699,2023,620,1,395,6.0762,2024,697,0,420,18.3574,2024,507,0,459,46.4478,2023,809,0,316,62.3586,2018,186,0,99,55.6374,2020,818,1,38,56.7895,2021,882,0,342,43.1359,2019,70,0,285,50.5273,2018,918,1,277,15.4151,2022,129,0,394,1.632,2018,584,0,66,10.4551,2022,999,0,370,97.1911,2021,931,1,356,50.9879,2019,139,1,114,6.1589,2019,250,1,34,29.8372,2021,410,0,387,13.5284,2020,138,0,43,98.7983,2023,622,0,184,0.75,2019,991,1,291,57.1036,2021,762,1,325,66.3414,2023,951,0,386,20.4953,2021,55,0,66,20.6337,2019,412,0,214,83.608,2020,167,0,12,93.8023,2023,18,0,70,5.5315,2024,942,1,350,61.1375,2018,304,0,329,46.1645,2021,183,0,111,58.3231,2021,29,0,215,14.455,2020,219,0,329,19.4221,2019,323,1,443,40.7827,2023,42,1,170,15.199,2021,580,1,273,79.2681,2020,378,1,398,80.2092,2021,740,1,185,98.3139,2021,444,0,220
+703,C402,15,Headphones,Electronics,8200.0,2.0,2024-01-26,1,162,89.2765,2022,442,0,473,74.9081,2023,335,1,53,70.9068,2019,583,1,108,8.2629,2019,982,0,383,18.4114,2020,843,1,269,15.2606,2024,638,0,478,50.1232,2021,625,1,55,12.2749,2023,556,1,104,94.3254,2021,461,0,436,62.1067,2021,88,0,453,63.8216,2021,431,0,282,79.5521,2023,542,1,278,14.6176,2018,143,1,256,33.4572,2018,470,0,421,16.3202,2023,160,0,52,97.5301,2021,651,1,328,88.0274,2021,719,1,250,0.2501,2018,666,1,349,35.597,2023,523,1,279,97.0134,2020,98,1,382,92.4716,2021,862,1,115,5.2908,2023,60,1,387,74.2714,2022,621,0,154,41.3644,2023,135,1,284,92.996,2018,978,1,198,38.6482,2019,982,0,218,86.1945,2023,768,1,220,37.7807,2020,263,1,351,66.6189,2018,339,1,423,66.4048,2024,821,1,41,41.3513,2018,221,1,96,88.633,2019,968,1,499,36.1479,2020,724,1,381,49.5031,2024,555,1,67,75.8066,2022,844,0,256,69.888,2022,694,1,204,68.7238,2022,154,1,222,3.2801,2018,395,0,377,3.889,2023,220,1,21,68.6911,2024,151,0,162,74.0095,2018,778,1,135,61.5566,2022,203,0,388,42.4485,2019,850,1,297,53.4918,2019,249,1,221,6.7502,2020,525,1,192,66.525,2024,639,1,1,55.1035,2022,819,0,168,86.86,2024,436,1,456,42.8609,2020,562,0,496,24.9022,2019,960,0,289,68.7651,2020,366,1,166,39.3911,2018,80,0,260,89.0157,2024,202,0,194,42.2193,2018,396,1,165,97.1287,2019,866,1,220,54.5223,2022,826,1,461,29.1912,2019,737,0,31,27.4888,2021,369,0,498,66.5258,2018,523,1,332,46.2714,2024,938,0,67,74.6614,2020,79,1,333,69.2029,2018,545,1,124,46.8531,2022,896,0,13,80.1894,2019,962,1,149,72.5293,2023,210,0,399,26.3104,2019,959,0,395,71.4436,2020,350,0,81,28.6851,2024,461,0,478,80.0292,2020,402,0,495,37.5534,2021,718,0,343,46.0989,2019,367,0,399,13.4429,2019,806,1,439,53.7388,2018,702,1,196,72.4649,2021,393,1,45,10.256,2019,295,1,167,71.4386,2020,179,0,43,24.2341,2022,798,0,156,57.7549,2018,19,0,452
+704,C403,60,Tablet,Electronics,,1.0,2024-02-09,1,265,71.8898,2018,336,1,98,88.4471,2023,961,1,456,94.725,2023,907,1,367,93.0463,2022,787,1,404,4.7322,2022,214,0,317,30.0124,2022,595,0,19,59.5029,2023,721,1,86,5.7787,2023,30,1,14,80.266,2022,826,0,96,37.1809,2022,274,0,268,84.2711,2019,411,1,404,43.2429,2020,485,1,219,34.2227,2019,267,0,138,78.0429,2018,746,1,142,44.134,2022,993,0,235,7.0395,2020,258,1,129,64.9228,2024,182,1,342,44.0091,2023,197,1,362,62.7104,2020,224,0,353,56.5413,2020,226,0,461,38.0537,2020,80,1,136,64.1519,2021,398,0,132,89.0276,2020,713,0,215,68.9897,2022,966,1,290,62.9747,2021,134,1,246,60.1625,2023,165,1,284,19.5553,2020,127,1,77,9.4002,2024,379,1,172,2.0021,2018,313,0,187,44.2648,2018,440,1,148,65.3153,2023,961,0,96,88.6686,2018,948,1,427,0.4257,2020,36,1,193,97.9875,2019,805,0,279,83.0474,2018,552,0,468,16.2863,2022,929,0,488,34.4021,2023,504,1,181,84.9543,2024,52,0,40,79.5751,2020,262,1,190,13.2315,2023,831,0,404,9.3724,2022,654,1,86,13.9261,2018,312,0,300,85.8559,2019,510,1,355,77.0651,2019,935,1,260,84.5444,2020,314,1,212,90.4273,2019,376,1,137,9.7238,2020,722,1,74,52.4767,2019,181,1,219,54.3066,2021,716,1,402,4.1763,2018,126,0,373,5.7149,2018,65,0,358,59.4006,2019,65,0,448,36.744,2018,298,0,57,6.3114,2018,671,0,263,2.1276,2023,142,0,397,48.5795,2022,48,0,313,17.4253,2022,660,1,414,6.2271,2024,791,0,289,73.2198,2023,823,0,56,14.5921,2021,672,1,389,41.8046,2019,708,1,466,19.7705,2021,489,0,479,81.2941,2019,341,0,135,66.541,2022,497,0,277,39.9045,2024,732,0,89,9.289,2018,141,1,327,50.7167,2024,759,1,61,58.6488,2018,551,1,26,46.5216,2020,788,1,219,63.8497,2019,620,1,225,99.5888,2021,941,1,172,56.8707,2024,495,0,424,74.1335,2024,905,0,179,77.6533,2018,433,0,311,31.1968,2021,914,1,170,27.8721,2020,735,0,364,11.9546,2023,913,0,397,44.7967,2022,966,1,16
+705,C404,30,Desk,Home,11800.0,1.0,2024-02-19,0,326,6.9968,2021,611,1,415,32.0166,2024,554,0,320,25.1378,2021,647,1,100,54.2937,2019,997,1,18,54.8351,2018,690,0,171,88.5219,2024,133,1,265,10.2271,2022,218,1,260,11.0745,2020,761,1,417,2.3196,2022,50,1,197,55.1762,2023,69,1,85,19.8423,2020,859,0,414,86.7129,2023,466,0,355,62.7176,2020,522,1,235,43.842,2018,817,1,92,44.3295,2023,571,1,24,39.7175,2018,961,1,103,1.8432,2020,333,0,199,80.7418,2021,581,0,164,30.9832,2022,646,1,298,0.1635,2020,249,1,273,69.1898,2018,51,0,122,21.2284,2021,401,1,88,94.0439,2024,959,1,219,74.1111,2022,645,1,231,92.3911,2022,828,1,205,88.9204,2020,747,1,424,61.6248,2023,635,0,388,29.5128,2018,162,0,356,83.6259,2024,817,0,114,85.6201,2021,133,0,461,11.9734,2023,836,1,189,36.818,2020,655,1,153,83.7223,2018,32,0,161,63.3442,2023,308,1,107,58.7303,2021,731,1,288,44.5737,2023,601,1,75,10.9557,2023,846,1,273,46.9717,2019,996,0,430,12.0217,2021,692,0,6,32.952,2018,871,1,33,9.3636,2020,757,0,313,3.2618,2020,363,1,51,93.964,2018,444,0,111,28.0697,2021,585,1,465,94.2951,2020,725,1,345,45.4705,2019,200,1,407,20.1806,2023,922,0,130,45.487,2023,156,1,64,29.3621,2021,361,0,496,39.7893,2018,607,0,191,24.8653,2021,749,0,422,7.1596,2024,349,0,132,24.5381,2018,4,1,391,24.3354,2024,873,1,131,36.8383,2023,235,1,434,9.7381,2018,86,0,122,40.1331,2018,400,1,248,98.9428,2021,836,1,105,81.3327,2020,844,0,275,48.61,2022,587,0,272,16.9499,2021,274,1,148,63.7686,2023,86,0,408,43.9928,2018,271,0,97,80.2777,2018,832,0,289,24.0179,2024,598,1,204,97.1628,2019,486,0,275,73.7156,2020,708,0,256,17.6641,2024,473,0,459,51.7645,2022,156,0,71,19.7144,2020,459,0,231,33.6317,2018,507,0,494,44.7828,2019,63,0,113,69.0979,2022,154,1,405,53.1931,2018,966,0,45,6.5847,2020,650,0,325,35.3675,2020,674,0,413,63.8429,2024,519,1,12,24.9617,2021,230,1,449
+706,C405,45,Camera,Electronics,40500.0,1.0,2024-03-04,1,34,45.6125,2020,830,0,41,90.2784,2019,48,1,474,89.2955,2021,735,0,25,24.3007,2024,45,0,130,31.8463,2024,549,1,468,33.7572,2021,919,1,294,36.4716,2018,110,1,388,86.6995,2019,713,0,411,62.6685,2021,512,1,345,80.2094,2020,889,1,236,39.1209,2023,557,0,256,7.3639,2019,371,0,248,98.0003,2018,710,1,241,99.422,2023,199,1,169,72.3343,2021,125,1,209,45.1824,2022,19,0,499,79.1581,2024,540,1,267,94.8339,2019,536,0,234,40.2764,2024,260,1,202,54.6238,2022,542,1,96,67.1621,2021,983,1,466,68.2438,2023,463,0,129,81.5038,2022,909,1,496,47.6105,2018,396,0,141,19.3653,2021,198,0,5,0.4034,2024,778,0,158,38.7775,2024,61,0,148,56.7554,2022,883,0,147,86.1987,2019,821,0,400,15.8295,2022,831,0,67,4.6801,2023,583,0,160,26.8546,2022,322,1,294,32.9468,2020,82,1,180,62.9521,2022,716,1,250,96.6011,2023,456,0,189,12.0439,2021,492,0,17,78.8695,2020,303,0,142,11.3961,2020,290,0,404,62.6016,2022,72,1,5,83.4673,2020,278,1,228,61.3194,2019,546,1,418,17.7742,2024,644,0,331,14.4934,2018,934,1,220,65.5122,2020,428,0,21,71.4861,2023,835,1,43,3.1781,2018,227,1,323,67.1978,2022,807,1,429,75.0757,2023,622,1,383,10.9511,2018,236,0,59,30.7807,2023,459,0,397,93.1844,2023,140,0,448,91.817,2020,764,1,95,42.9711,2019,90,1,161,69.223,2018,741,1,317,80.4519,2023,126,0,326,32.5647,2024,337,1,498,74.0223,2022,522,1,111,43.5206,2018,507,0,156,1.336,2019,185,0,224,64.7166,2022,833,1,250,52.3901,2022,924,1,113,68.1389,2021,13,0,280,4.7899,2019,174,0,178,91.6959,2021,336,1,382,14.2737,2024,437,0,434,88.7794,2024,186,1,277,96.3444,2022,697,1,53,53.584,2024,272,0,195,88.535,2019,108,0,277,3.8178,2023,673,1,239,53.2048,2018,287,1,450,25.8803,2022,820,0,306,28.9628,2020,124,1,234,57.2559,2018,813,0,13,84.3891,2019,428,0,311,36.2306,2019,364,1,67,75.298,2021,140,1,436,57.1664,2023,785,1,26
+707,C406,90,Serum,Beauty,1200.0,3.0,2024-03-18,0,345,53.8047,2022,804,0,253,33.0549,2019,44,1,79,22.5355,2020,355,1,377,62.2929,2021,422,0,70,80.0632,2022,545,1,173,50.0556,2024,2,0,97,1.6601,2022,665,1,187,54.262,2023,951,0,426,52.6644,2021,551,1,311,97.2634,2021,138,1,224,19.2569,2022,147,0,72,80.7577,2022,918,1,104,8.4668,2018,122,1,49,63.8585,2021,100,1,466,23.732,2024,895,0,494,23.6486,2020,198,1,197,77.2568,2018,326,0,226,65.3289,2023,647,1,82,86.3155,2024,292,1,151,66.6084,2022,959,0,378,45.3949,2018,424,1,246,94.8786,2018,859,0,464,86.5524,2023,793,1,206,85.1208,2022,92,0,353,91.7173,2019,565,1,345,6.7008,2020,499,0,211,73.9902,2022,359,1,417,98.8436,2023,592,1,408,54.5499,2019,90,0,245,91.0177,2024,274,1,158,27.5686,2023,279,1,106,70.9682,2023,997,0,299,72.2503,2024,815,1,230,43.0567,2018,277,0,121,30.1199,2020,429,0,31,93.3484,2023,575,1,62,53.1013,2020,599,0,467,79.9392,2024,511,1,273,2.2151,2021,545,0,352,12.463,2023,755,0,441,55.8703,2019,461,1,414,83.714,2019,676,0,90,87.0907,2019,315,0,262,63.1309,2018,772,0,336,45.0808,2022,402,1,130,32.6755,2024,152,1,31,27.8969,2020,398,0,367,66.4586,2022,256,1,189,69.9446,2024,687,0,197,44.6948,2022,529,0,470,13.8614,2024,914,0,139,9.179,2023,536,0,291,96.6608,2021,85,1,348,54.4408,2020,650,0,368,14.9704,2021,841,1,39,98.6302,2020,59,0,23,30.4665,2018,532,1,404,24.1201,2022,200,1,105,7.258,2023,259,0,22,33.6908,2021,940,1,167,24.4553,2020,650,0,108,51.1963,2019,735,1,237,49.5187,2023,417,0,490,31.9363,2019,499,0,79,18.7186,2018,612,0,379,7.7791,2020,505,0,392,61.4556,2018,785,0,399,77.1285,2019,34,0,400,55.4015,2021,779,0,29,89.2132,2020,497,0,430,4.7329,2021,775,0,297,40.4557,2023,170,1,158,48.8236,2022,690,1,278,2.814,2024,435,0,39,70.3749,2019,567,1,247,34.3636,2020,721,0,28,6.5223,2023,506,1,371,49.2449,2019,579,0,268
+708,C407,30,Shoes,Fashion,3300.0,2.0,2024-03-29,0,28,81.0418,2019,329,1,498,26.2027,2018,266,1,423,70.1049,2021,140,0,209,6.8436,2024,969,0,278,41.0682,2020,121,1,167,40.8495,2024,996,1,117,28.2872,2019,114,0,477,41.7888,2022,110,1,43,12.8289,2018,497,0,453,39.3874,2023,626,0,480,88.3355,2019,648,0,168,9.4579,2021,614,0,158,56.0907,2019,58,0,194,53.0077,2022,737,0,395,55.5247,2021,901,0,30,17.7533,2024,457,1,312,83.9569,2020,866,0,211,10.5889,2019,757,1,177,73.1204,2022,589,0,384,62.1249,2019,684,1,314,34.0848,2024,892,0,441,10.3404,2021,255,0,70,97.9444,2021,419,1,327,14.9061,2024,860,1,163,46.0826,2021,995,0,437,38.0631,2021,273,1,339,98.7845,2021,311,1,191,46.2244,2020,78,0,403,41.9937,2019,354,0,121,80.8868,2020,191,1,381,87.1074,2024,532,1,465,71.488,2024,110,0,92,42.7386,2024,468,0,137,67.7976,2019,342,0,350,6.3912,2024,600,1,318,8.9182,2021,918,1,294,4.7646,2021,632,0,382,30.6837,2020,735,0,424,24.1036,2021,572,1,474,19.2626,2020,713,0,1,16.7838,2020,696,0,495,31.3454,2019,162,1,121,63.3282,2019,941,1,187,71.3009,2023,527,0,418,47.3461,2024,547,0,324,47.9481,2023,238,1,488,20.9736,2018,479,0,52,49.8454,2019,825,0,57,72.1301,2021,202,0,12,30.3385,2021,104,1,290,58.5497,2022,119,1,238,94.3822,2022,465,1,4,56.6079,2023,410,0,266,52.7364,2018,568,1,19,24.1914,2018,955,1,138,41.3301,2021,190,0,337,5.0866,2021,304,1,346,58.6776,2024,503,1,27,61.9477,2024,180,1,213,7.9824,2018,910,1,353,1.6168,2023,270,1,317,4.7318,2021,323,1,471,1.5135,2019,993,0,399,43.0371,2022,487,1,211,7.0348,2023,433,0,63,79.4095,2021,310,1,128,41.357,2024,578,1,265,66.0135,2019,585,1,134,97.1917,2021,526,0,10,53.2297,2024,761,0,204,0.6791,2020,372,1,318,85.7888,2021,715,1,457,46.2532,2024,846,1,288,28.4242,2023,302,1,236,98.0117,2022,149,1,368,74.8703,2018,502,0,195,34.3001,2018,203,1,17,61.6395,2022,715,0,336
+709,C408,60,Monitor,Electronics,19000.0,1.0,2024-04-08,0,175,47.8909,2021,202,1,200,87.079,2021,457,0,477,63.5543,2019,250,0,425,50.7921,2018,251,1,140,18.531,2024,463,1,62,23.7837,2018,866,1,283,63.5852,2018,346,0,182,51.6689,2018,303,0,315,5.8712,2021,37,0,275,47.6738,2019,98,1,213,33.3327,2019,840,1,258,1.7105,2018,346,0,196,1.4748,2018,290,0,397,69.0669,2022,932,0,147,88.4798,2019,777,0,261,97.003,2024,819,1,268,50.5681,2024,515,0,456,81.5716,2022,624,0,299,69.8387,2024,791,0,231,16.0837,2020,943,1,426,0.4444,2022,977,1,309,48.7493,2019,879,1,104,30.4319,2020,653,0,378,76.6094,2019,137,1,68,61.5046,2022,25,0,480,67.6713,2020,494,1,17,96.8785,2022,736,0,167,72.2481,2022,747,0,498,66.0362,2024,618,1,192,64.6547,2022,58,1,122,73.3119,2018,725,0,70,56.1347,2023,588,1,377,81.9548,2018,119,1,308,7.2491,2018,303,0,93,91.2426,2019,819,0,154,51.0652,2023,120,1,249,35.6257,2018,166,0,469,22.5172,2023,953,1,289,52.5254,2023,495,1,418,19.1392,2024,877,1,372,48.4309,2023,263,0,244,21.549,2023,153,0,270,26.1433,2018,20,1,212,12.3694,2018,538,1,476,41.3417,2023,658,0,317,82.8539,2022,555,1,26,25.5726,2018,893,0,329,85.4137,2018,856,1,190,3.9267,2021,915,0,476,93.9418,2018,928,1,233,52.0063,2023,243,0,466,84.561,2018,312,0,56,3.6933,2022,743,0,214,31.2558,2024,343,0,104,68.1593,2024,567,0,18,4.8924,2018,727,0,233,84.0284,2022,730,0,41,14.0718,2018,756,0,485,33.3656,2024,425,0,175,89.6337,2023,281,1,387,46.8855,2020,63,1,406,3.8396,2024,384,0,484,10.7994,2019,985,1,433,41.9599,2019,150,0,71,11.2811,2019,934,1,191,56.3049,2023,50,0,10,10.3323,2019,995,0,291,67.7881,2018,356,0,402,41.1253,2023,493,1,93,28.884,2020,351,1,222,23.0337,2021,753,1,312,39.0545,2019,831,0,107,1.498,2023,185,0,298,71.996,2023,885,1,333,32.0281,2023,799,1,358,69.2373,2021,22,1,29,27.2678,2021,261,1,292,85.948,2024,673,1,317
+710,C409,30,Keyboard,Electronics,3600.0,3.0,2024-04-17,1,377,36.3951,2018,559,1,82,88.6105,2022,874,1,356,61.5928,2022,58,1,56,76.6528,2023,771,0,356,91.2929,2018,223,0,426,84.918,2024,918,1,233,12.7016,2020,679,0,455,59.6133,2024,379,1,454,3.9599,2021,530,1,169,2.2761,2022,76,1,165,84.3403,2023,243,1,206,64.0305,2022,256,1,353,86.6067,2022,983,1,166,35.3545,2022,958,1,397,94.8278,2021,416,1,85,2.8694,2021,868,0,38,76.396,2018,799,0,475,25.3447,2021,481,1,421,65.0742,2023,942,0,460,94.3213,2021,11,0,36,73.0891,2023,953,1,60,22.1366,2024,155,1,347,19.9297,2022,160,1,287,22.6341,2022,546,0,156,87.2321,2022,748,1,202,13.4762,2022,15,0,379,5.5991,2018,277,1,448,60.948,2022,921,1,64,61.4745,2019,738,1,126,7.2064,2020,270,0,323,87.3784,2021,86,1,364,43.203,2021,64,0,449,87.2666,2024,589,0,316,34.6785,2021,652,1,393,66.5856,2021,660,1,384,5.7473,2021,113,0,98,79.2803,2024,555,0,359,83.4134,2023,904,0,495,88.8027,2022,42,0,453,93.4389,2021,151,0,86,76.0442,2018,263,0,103,2.7304,2019,174,0,270,68.4925,2019,557,1,167,97.1451,2019,467,0,199,46.1039,2020,254,1,128,68.0284,2021,95,1,298,88.7513,2023,117,1,422,27.1523,2021,293,0,424,80.4072,2019,749,0,46,81.709,2021,768,1,68,48.6141,2020,392,0,193,12.8098,2022,389,1,45,67.9927,2020,687,1,56,65.9521,2024,991,0,121,62.7859,2020,449,1,261,87.688,2019,541,1,105,61.8234,2023,43,1,396,14.6636,2022,920,1,195,24.6699,2024,632,0,417,76.7522,2018,422,1,18,91.2658,2020,635,0,124,7.7524,2019,914,0,398,67.0776,2019,103,0,89,99.2569,2019,549,1,338,74.9316,2021,92,0,272,50.1049,2021,77,0,156,44.5596,2023,692,0,167,48.6705,2020,763,0,94,15.8735,2019,349,1,431,3.4488,2024,660,0,405,82.7042,2019,264,0,428,49.8136,2019,338,0,209,57.9633,2020,606,0,158,80.1876,2024,664,0,100,67.1782,2023,196,1,283,47.4361,2018,716,1,320,45.8388,2023,777,1,290,7.3426,2022,771,0,383
+711,C410,45,Phone,Electronics,42000.0,1.0,2024-04-29,0,272,91.1024,2023,609,1,389,53.5688,2023,754,0,416,28.2974,2024,983,1,40,63.3961,2018,861,1,123,53.4032,2021,504,1,432,96.1176,2018,66,1,288,31.6162,2021,260,0,215,6.1848,2018,581,0,111,89.3017,2021,303,1,410,9.7654,2019,690,1,173,29.1723,2023,359,0,188,15.7017,2018,921,1,375,66.0324,2023,979,0,413,2.0443,2023,967,0,28,22.7686,2020,7,1,38,23.1003,2021,747,1,183,0.4943,2018,628,1,393,58.1929,2023,122,1,498,62.4852,2024,727,1,154,10.8294,2023,69,1,264,97.5929,2023,529,0,61,65.6891,2024,910,0,430,64.2018,2019,51,0,178,28.9301,2022,235,1,158,73.1798,2022,689,0,147,31.5309,2024,63,0,403,68.8166,2023,559,0,332,30.0548,2024,825,1,263,78.507,2018,386,0,205,28.4976,2024,903,0,264,33.5015,2023,495,1,247,55.1445,2019,210,0,436,9.3946,2021,608,1,153,78.919,2022,109,0,102,41.3952,2022,275,0,171,14.1166,2022,145,1,308,11.8544,2024,988,0,364,38.205,2021,731,1,112,86.5166,2022,909,1,161,87.3357,2021,70,0,273,5.3497,2021,218,0,306,93.1111,2022,992,1,477,16.9701,2024,16,1,379,83.157,2023,749,0,188,76.4299,2023,454,0,187,91.9783,2018,797,0,60,44.6919,2023,154,1,227,74.7193,2019,256,1,357,15.4229,2023,108,0,237,36.0042,2019,371,0,380,31.3895,2019,590,1,230,63.7532,2023,113,1,351,43.8288,2024,122,0,229,74.4885,2021,395,1,354,30.1564,2023,818,0,383,51.1065,2019,787,1,451,20.0141,2021,155,0,482,59.0639,2022,448,0,33,69.7741,2021,814,0,434,55.3449,2018,436,0,413,70.5357,2021,115,1,209,27.0294,2019,984,0,157,1.5446,2020,227,0,346,87.2463,2020,627,1,74,11.84,2018,581,0,322,78.6573,2022,390,0,374,99.8867,2021,353,1,152,5.6584,2022,318,0,327,54.731,2024,195,1,447,86.7438,2021,743,0,15,93.1257,2023,865,1,340,98.0048,2020,700,1,446,54.2752,2024,895,1,283,71.2797,2019,845,0,268,35.0642,2018,3,1,383,22.6124,2018,791,0,159,4.0251,2020,506,0,230,67.2233,2023,534,0,304
+712,C411,75,Racket,Sports,2100.0,,2024-05-10,0,114,25.1746,2023,350,1,378,60.6678,2019,185,1,435,95.9221,2022,987,0,103,84.1202,2020,586,1,291,37.9491,2021,930,0,451,75.0386,2021,670,0,173,40.845,2024,345,0,388,41.7529,2021,501,1,205,75.9259,2020,286,1,355,68.8055,2019,585,0,194,46.8454,2023,597,1,322,61.7538,2019,539,1,418,82.3195,2019,140,0,8,65.9815,2020,825,0,404,89.5488,2021,831,1,66,19.5934,2021,657,1,198,75.8448,2018,793,0,55,75.7624,2024,145,0,195,28.4838,2020,727,1,209,94.1845,2018,838,1,183,71.9382,2022,198,0,498,69.5826,2020,384,1,451,92.0282,2022,46,0,112,74.1567,2020,673,0,41,61.211,2019,848,0,354,88.0356,2021,841,0,357,36.4826,2019,455,1,250,30.0309,2018,261,0,286,60.0315,2024,29,0,265,16.4333,2024,260,1,455,23.8735,2019,943,0,332,18.7975,2024,878,0,48,12.5775,2019,637,1,492,30.32,2024,124,0,347,9.34,2019,998,1,34,22.063,2018,831,0,421,52.4875,2023,513,1,219,49.6862,2018,76,0,234,82.4235,2019,538,1,104,46.2018,2023,264,1,66,6.036,2021,244,1,24,6.4716,2024,994,0,13,23.9393,2021,900,0,360,20.2078,2018,5,1,117,97.2884,2018,648,1,80,68.9284,2024,44,1,80,17.6141,2020,765,0,233,76.8382,2021,986,1,393,14.1893,2021,856,0,357,68.0392,2024,21,1,278,93.1926,2021,9,0,91,19.7164,2018,162,0,487,52.228,2018,748,1,387,39.743,2021,495,1,477,9.6685,2022,117,0,257,89.2016,2021,596,1,181,10.9361,2018,118,0,447,26.3534,2020,808,0,358,83.8325,2021,50,1,423,46.3884,2023,592,1,60,82.5943,2018,265,0,179,17.1842,2021,820,0,59,1.235,2024,342,1,321,4.6736,2022,677,0,356,31.3905,2022,701,0,11,6.0762,2024,947,1,147,91.0698,2024,97,0,401,4.2701,2021,994,0,110,78.5132,2022,435,0,108,98.0623,2022,980,0,394,25.1927,2022,579,1,154,21.9261,2020,328,1,107,66.8068,2020,584,1,85,26.7067,2021,684,1,44,32.0077,2023,29,0,128,41.4427,2022,141,1,56,42.6543,2019,232,0,41,63.3698,2019,961,1,226

inference.py

@@ -0,0 +1,365 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import os
+import sys
+from pathlib import Path
+from typing import Any, Callable, Optional
+
+API_BASE_URL = os.getenv('API_BASE_URL', 'https://router.huggingface.co/v1')
+MODEL_NAME = os.getenv('MODEL_NAME', 'Qwen/Qwen2.5-72B-Instruct')
+API_KEY = os.getenv('API_KEY') or os.getenv('HF_TOKEN') or ''
+BENCHMARK = 'autodatalab_plus'
+TASKS = [t.strip() for t in os.getenv('AUTODATALAB_PLUS_TASKS', 'easy_brief,medium_brief,hard_brief,expert_brief').split(',') if t.strip()]
+
+_LOG_SCORE_MIN = 0.001
+_LOG_SCORE_MAX = 0.999
+
+REPO = Path(__file__).resolve().parent
+if str(REPO) not in sys.path:
+    sys.path.insert(0, str(REPO))
+
+from ceo_brief_env.environment import CEOBriefEnvironment, oracle_action_for_observation, required_experts_for_task
+from ceo_brief_env.models import CoSAction
+
+
+def _bool_str(v: bool) -> str:
+    return str(bool(v)).lower()
+
+
+def log_start(task: str, env: str, model: str) -> None:
+    print(f'[START] task={task} env={env} model={model}', flush=True)
+
+
+def _clamp_log_reward(r: float) -> float:
+    return max(_LOG_SCORE_MIN, min(_LOG_SCORE_MAX, float(r)))
+
+
+def log_step(step: int, action: str, reward: float, done: bool, error: Optional[str]) -> None:
+    err = error if error else 'null'
+    r = _clamp_log_reward(reward)
+    print(f'[STEP] step={step} action={action} reward={r:.2f} done={_bool_str(done)} error={err}', flush=True)
+
+
+def log_end(success: bool, steps: int, score: float, rewards: list[float]) -> None:
+    s = _clamp_log_reward(score)
+    rewards_str = ','.join(f'{_clamp_log_reward(r):.2f}' for r in rewards)
+    print(f'[END] success={_bool_str(success)} steps={steps} score={s:.2f} rewards={rewards_str}', flush=True)
+
+
+def _action_str(action: CoSAction) -> str:
+    return json.dumps(action.model_dump(exclude_none=True), separators=(',', ':'), sort_keys=True)
+
+
+def _single_baseline(obs) -> CoSAction:
+    for e in required_experts_for_task(obs.task_name):
+        if e not in obs.consulted_experts:
+            return CoSAction(action_type='consult', expert_id=e)
+    if obs.current_brief is None:
+        return CoSAction(action_type='summarize')
+    return CoSAction(action_type='submit')
+
+
+def _roundrobin_baseline(obs) -> CoSAction:
+    for expert in ['analyst', 'finance', 'strategy', 'hr']:
+        if expert not in obs.consulted_experts:
+            return CoSAction(action_type='consult', expert_id=expert)
+    if obs.current_brief is None:
+        return CoSAction(action_type='summarize')
+    return CoSAction(action_type='submit')
+
+
+_TRAINED_POLICY: dict = {"model": None, "load_status": None, "load_warned": False}
+
+
+def _trained_action(obs) -> CoSAction:
+    import numpy as np
+    import torch
+
+    from training.train_cos_local import (
+        ACTIONS,
+        PolicyNet,
+        featurize,
+        load_policy_state_dict_from_file,
+    )
+
+    if _TRAINED_POLICY["model"] is None:
+        import sys
+
+        ckpt = REPO / "training" / "checkpoints" / "cos_final.pt"
+        if not ckpt.exists():
+            ckpt = REPO / "training" / "checkpoints" / "cos_ckpt0.pt"
+        model = PolicyNet()
+        status = "random_init"
+        if ckpt.exists():
+            try:
+                status = load_policy_state_dict_from_file(model, ckpt)
+            except (OSError, RuntimeError, KeyError) as e:
+                print(
+                    f"[inference] warning: could not load {ckpt} ({e}); using random init.",
+                    file=sys.stderr,
+                )
+                model = PolicyNet()
+                status = f"load_failed: {e}"
+        else:
+            print(
+                "[inference] warning: no checkpoint; using random PolicyNet. "
+                "Run: python3 training/train_cos_local.py",
+                file=sys.stderr,
+            )
+        if status != "ok" and "padded" in status and not _TRAINED_POLICY.get("load_warned"):
+            print(
+                f"[inference] loaded checkpoint with compat: {status}. "
+                "Re-run training for best results: python3 training/train_cos_local.py",
+                file=sys.stderr,
+            )
+            _TRAINED_POLICY["load_warned"] = True
+        model.eval()
+        _TRAINED_POLICY["model"] = model
+        _TRAINED_POLICY["load_status"] = status
+    model = _TRAINED_POLICY["model"]
+    feats = torch.from_numpy(featurize(obs)).unsqueeze(0)
+    with torch.no_grad():
+        logits = model(feats)
+    idx = int(torch.argmax(logits, dim=-1).item())
+    return ACTIONS[idx]
+
+
+def _cache_path(task_name: str, step_count: int, prompt: str) -> Path:
+    prompt_hash = hashlib.sha256(prompt.encode('utf-8')).hexdigest()[:16]
+    cache_dir = REPO / 'cache'
+    cache_dir.mkdir(exist_ok=True)
+    return cache_dir / f'{task_name}_step{step_count}_{prompt_hash}.json'
+
+
+def _llm_action(obs) -> CoSAction:
+    from openai import OpenAI
+
+    client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
+    prompt = (
+        'You are the Chief of Staff in a multi-agent company simulation. '
+
+        f'Task: {obs.task_name}. Instruction: {obs.instruction}. '
+
+        f'Consulted experts: {obs.consulted_experts}. Current issues: {obs.issues}. '
+
+        'Return exactly one JSON object with keys action_type, expert_id, sub_question_id, notes. '
+
+        'Valid experts: analyst, finance, hr, strategy. Valid actions: consult, ask, summarize, submit, noop.'
+    )
+    cache_path = _cache_path(obs.task_name, obs.step_count, prompt)
+    if cache_path.exists():
+        try:
+            payload = json.loads(cache_path.read_text(encoding='utf-8'))
+        except json.JSONDecodeError:
+            payload = {}
+        if not payload or 'action_type' not in payload:
+            payload = {'action_type': 'noop'}
+        return CoSAction.model_validate(payload)
+    completion = client.chat.completions.create(
+        model=MODEL_NAME,
+        temperature=0.1,
+        messages=[{'role': 'system', 'content': 'Respond with strict JSON only.'}, {'role': 'user', 'content': prompt}],
+    )
+    text = completion.choices[0].message.content or '{}'
+    start = text.find('{')
+    end = text.rfind('}')
+    if start != -1 and end != -1 and end >= start:
+        try:
+            payload = json.loads(text[start : end + 1])
+        except json.JSONDecodeError:
+            payload = {}
+    else:
+        payload = {}
+    if not payload or 'action_type' not in payload:
+        payload = {'action_type': 'noop'}
+    cache_path.write_text(json.dumps(payload, indent=2), encoding='utf-8')
+    return CoSAction.model_validate(payload)
+
+
+def run_episode_collect(
+    task: str,
+    picker: Callable,
+    label: str,
+    use_rag: bool = False,
+    quiet: bool = False,
+) -> dict[str, Any]:
+    """
+    Run one episode and return structured data (for --export). Same dynamics as `run_episode`.
+    """
+    env = CEOBriefEnvironment()
+    obs = env.reset(task=task, use_rag=use_rag)
+    rewards: list[float] = []
+    trace: list[dict[str, Any]] = []
+    steps = 0
+    score = 0.001
+    success = False
+    err: Optional[str] = None
+    if not quiet:
+        log_start(task=task, env=f'{BENCHMARK} rag={use_rag}', model=label)
+    try:
+        while not obs.done and steps < obs.max_steps:
+            steps += 1
+            action = picker(obs)
+            obs = env.step(action)
+            r = float(obs.reward)
+            rewards.append(r)
+            if not quiet:
+                log_step(steps, _action_str(action), r, bool(obs.done), None)
+            trace.append(
+                {
+                    "step": steps,
+                    "action": action.model_dump(exclude_none=True),
+                    "reward": round(r, 4),
+                    "done": bool(obs.done),
+                    "consulted_experts": list(obs.consulted_experts),
+                }
+            )
+        score = float(obs.terminal_grader_score or 0.001)
+        score = max(0.001, min(0.999, score))
+        success = score >= 0.5
+    except Exception as exc:  # noqa: BLE001
+        err = str(exc).replace('\n', ' ')
+        if not quiet:
+            log_step(max(steps, 1), 'exception', _LOG_SCORE_MIN, True, err)
+    finally:
+        if not quiet:
+            log_end(success, steps, score, rewards)
+    return {
+        "task": task,
+        "policy_label": label,
+        "use_rag": use_rag,
+        "success": success,
+        "steps": steps,
+        "terminal_score": round(score, 4),
+        "cumulative_reward": round(sum(rewards), 4),
+        "step_rewards": [round(x, 4) for x in rewards],
+        "trace": trace,
+        "error": err,
+        "final_instruction": obs.instruction,
+        "task_difficulty": obs.task_difficulty,
+        "max_steps": obs.max_steps,
+        "consulted_experts": list(obs.consulted_experts),
+        "current_brief": obs.current_brief.model_dump() if obs.current_brief is not None else None,
+        "expert_reports": {k: v.model_dump() for k, v in obs.expert_reports.items()},
+    }
+
+
+def run_episode(task: str, picker: Callable, label: str, use_rag: bool = False) -> float:
+    out = run_episode_collect(task, picker, label, use_rag=use_rag, quiet=False)
+    return float(out["terminal_score"])
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--oracle', action='store_true')
+    parser.add_argument('--baseline', choices=['single', 'roundrobin'])
+    parser.add_argument('--trained', action='store_true',
+                        help='use locally trained CoS policy (training/checkpoints/cos_final.pt)')
+    parser.add_argument('--task')
+    parser.add_argument('--ablation', action='store_true')
+    parser.add_argument(
+        '--rag',
+        action='store_true',
+        help='enable organizational memory (RAG) in experts and grounding in the grader (default: off, team parity)',
+    )
+    parser.add_argument(
+        '--export',
+        metavar='DIR',
+        help='save one JSON per task (plus summary.json) under DIR; uses all tasks from env unless --task is set',
+    )
+    args = parser.parse_args()
+    use_rag = bool(args.rag)
+
+    if args.trained:
+        picker = _trained_action
+        label = 'trained-cos'
+    elif args.baseline == 'single':
+        picker = _single_baseline
+        label = 'single-baseline'
+    elif args.baseline == 'roundrobin':
+        picker = _roundrobin_baseline
+        label = 'roundrobin-baseline'
+    elif args.oracle or not API_KEY:
+        picker = oracle_action_for_observation
+        label = 'oracle'
+    else:
+        picker = _llm_action
+        label = MODEL_NAME
+
+    tasks = [args.task] if args.task else TASKS
+
+    if args.export:
+        out_dir = Path(args.export).resolve()
+        out_dir.mkdir(parents=True, exist_ok=True)
+        by_task: dict[str, dict[str, Any]] = {}
+        for task in tasks:
+            row = run_episode_collect(task, picker, label, use_rag=use_rag, quiet=True)
+            by_task[task] = row
+            (out_dir / f'{task}.json').write_text(
+                json.dumps(row, indent=2, default=str), encoding='utf-8'
+            )
+            print(
+                f'[export] {task} terminal={row["terminal_score"]} steps={row["steps"]} success={row["success"]}',
+                flush=True,
+            )
+        mean_term = float(sum(t['terminal_score'] for t in by_task.values()) / max(len(by_task), 1))
+        summary: dict[str, Any] = {
+            'policy_label': label,
+            'use_rag': use_rag,
+            'checkpoint_load': _TRAINED_POLICY.get('load_status') if label == 'trained-cos' else None,
+            'tasks': {
+                t: {
+                    'terminal_score': by_task[t]['terminal_score'],
+                    'success': by_task[t]['success'],
+                    'steps': by_task[t]['steps'],
+                    'cumulative_reward': by_task[t]['cumulative_reward'],
+                }
+                for t in by_task
+            },
+            'mean_terminal': round(mean_term, 4),
+            'export_dir': str(out_dir),
+        }
+        (out_dir / 'summary.json').write_text(json.dumps(summary, indent=2), encoding='utf-8')
+        print(f'[export] wrote {len(tasks) + 1} files to {out_dir}', flush=True)
+        return 0
+    else:
+        results = {}
+        for task in tasks:
+            results[task] = run_episode(task, picker, label, use_rag=use_rag)
+
+    if args.ablation and not args.export:
+        ablations: dict[str, dict[str, float]] = {}
+        if picker is _single_baseline:
+            ablations['single'] = dict(results)
+        else:
+            ablations['single'] = {
+                task: run_episode(task, _single_baseline, 'single-baseline', use_rag=use_rag) for task in tasks
+            }
+        if picker is _roundrobin_baseline:
+            ablations['roundrobin'] = dict(results)
+        else:
+            ablations['roundrobin'] = {
+                task: run_episode(task, _roundrobin_baseline, 'roundrobin-baseline', use_rag=use_rag) for task in tasks
+            }
+        ablations['oracle_or_llm'] = dict(results)
+        trained_ckpt = REPO / 'training' / 'checkpoints' / 'cos_final.pt'
+        if trained_ckpt.exists():
+            if picker is _trained_action:
+                ablations['trained_cos'] = dict(results)
+            else:
+                ablations['trained_cos'] = {
+                    task: run_episode(task, _trained_action, 'trained-cos', use_rag=use_rag) for task in tasks
+                }
+        cache_dir = REPO / 'cache'
+        cache_dir.mkdir(exist_ok=True)
+        (cache_dir / 'ablation_results.json').write_text(json.dumps(ablations, indent=2), encoding='utf-8')
+
+    return 0
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())

memory/__init__.py

@@ -0,0 +1,29 @@
+"""Organizational memory (RAG pillar) for AutoDataLab Solo.
+
+Exposes a module-level retriever keyed at ``memory/`` so every specialist
+and the grader can share the same corpus index without passing it through
+every expert signature.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+
+from .retriever import MemoryHit, Retriever
+
+_CORPUS_DIR = Path(__file__).resolve().parent
+_retriever: Retriever | None = None
+
+
+def get_retriever() -> Retriever:
+    """Return the process-wide retriever, building it lazily on first use."""
+    global _retriever
+    if _retriever is None:
+        _retriever = Retriever(_CORPUS_DIR)
+    return _retriever
+
+
+def corpus_dir() -> Path:
+    return _CORPUS_DIR
+
+
+__all__ = ["MemoryHit", "Retriever", "get_retriever", "corpus_dir"]

memory/history/exemplar_memo.md

@@ -0,0 +1,12 @@
+# EXEMPLAR-MEMO: Canonical HR Memo from Last Quarter
+
+Audience: Ops leads.
+
+Hook: Electronics is still our revenue anchor and needs a defend-first posture this week.
+
+Highlights:
+- Analyst flagged Electronics as the top category at 38 percent of total revenue.
+- Finance projection for next quarter is plus 6.2 percent with a plus-or-minus 15 percent band.
+- Variance versus plan is plus 4.1 percent, so we are favorable but not safe.
+
+Call to action: Ops leads, lock the Electronics staffing plan by Thursday and align promotion windows to the break-even unit target.

memory/history/last_quarter_review.md

@@ -0,0 +1,15 @@
+# HIST-Q1: Last Quarter Executive Review
+
+## Headline numbers
+- Top revenue category: Electronics.
+- Data quality score on ingest: 0.82.
+- Variance versus plan: +4.1 percent (favorable).
+- Break-even pressure: moderate; staffing tightened in weeks 10 and 11.
+
+## Lessons recorded for future briefs
+1. Always include a break-even reference when promotions are being considered.
+2. HR memos landed better with the Ops team when the first line named them explicitly.
+3. Strategy recommendations that skipped the variance signal were rated lower by the CEO.
+
+## Precedent for this quarter
+The CEO expects every brief to cite `variance_pct` and `projection_next_quarter` somewhere in the Strategy and HR layers.

memory/policies/compliance_policy.md

@@ -0,0 +1,13 @@
+# POLICY-COMP-01: Company Compliance Policy
+
+## Publication gate
+A CEO brief cannot be published when any of the following is true:
+1. `data_quality_score` is below 0.70.
+2. Strategy recommendations do not cite at least one Analyst metric and one Finance metric.
+3. HR memo does not name its intended audience.
+
+## Grounding rule
+Every specialist report in the final brief must cite at least one source from `memory/`. This ensures that reasoning is grounded in our SOPs, historical precedents, and compliance policies rather than improvised at rollout time.
+
+## Audit
+Citations are programmatically verified against the live `memory/` index. A citation that does not resolve to a real chunk is treated as hallucinated and drops the grounding reward for that specialist.

memory/retriever.py

@@ -0,0 +1,157 @@
+"""Zero-dependency TF-IDF retriever over the company-memory corpus.
+
+Design choices
+--------------
+* Pure Python (re + math) so it runs inside OpenEnv/HF Space containers
+  without pulling sentence-transformers / faiss / langchain.
+* Each markdown file is split into paragraph-level chunks, indexed with
+  token-frequency + inverse-document-frequency, and queried with cosine
+  over sparse TF-IDF vectors.
+* The retriever is the *single source of grounding truth* consumed by
+  both the specialists (at rollout time) and the grader (at scoring
+  time), so the reward becomes verifiable instead of fuzzy.
+"""
+from __future__ import annotations
+
+import math
+import re
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Dict, Iterable, List, Sequence, Tuple
+
+_WORD_RE = re.compile(r"[a-zA-Z][a-zA-Z0-9_-]{1,}")
+
+
+def _tokenize(text: str) -> List[str]:
+    return [tok.lower() for tok in _WORD_RE.findall(text or "")]
+
+
+@dataclass
+class MemoryHit:
+    source: str
+    snippet: str
+    score: float
+
+    def as_citation(self) -> str:
+        """Stable string form used in ExpertReport.citations and briefs."""
+        return f"memory:{self.source}"
+
+
+class Retriever:
+    """Lightweight TF-IDF retriever over a directory of markdown files."""
+
+    def __init__(self, corpus_dir: Path) -> None:
+        self.corpus_dir = Path(corpus_dir)
+        self._docs: List[Tuple[str, str]] = []
+        self._vocab: Dict[str, int] = {}
+        self._tf: List[Dict[int, float]] = []
+        self._df: Dict[int, int] = {}
+        self._norms: List[float] = []
+        self._load()
+        self._build_index()
+
+    # -- indexing ----------------------------------------------------------
+
+    def _load(self) -> None:
+        for path in sorted(self.corpus_dir.rglob("*.md")):
+            rel = str(path.relative_to(self.corpus_dir)).replace("\\", "/")
+            text = path.read_text(encoding="utf-8")
+            chunks = [chunk.strip() for chunk in re.split(r"\n\s*\n", text) if chunk.strip()]
+            for i, chunk in enumerate(chunks):
+                self._docs.append((f"{rel}#chunk{i}", chunk))
+
+    def _build_index(self) -> None:
+        for _, text in self._docs:
+            tokens = _tokenize(text)
+            tf_doc: Dict[int, float] = {}
+            for tok in tokens:
+                if tok not in self._vocab:
+                    self._vocab[tok] = len(self._vocab)
+                idx = self._vocab[tok]
+                tf_doc[idx] = tf_doc.get(idx, 0.0) + 1.0
+            self._tf.append(tf_doc)
+            for idx in tf_doc:
+                self._df[idx] = self._df.get(idx, 0) + 1
+        self._num_docs = len(self._docs)
+        # cache L2 norms of tf-idf vectors for cosine similarity.
+        self._norms = []
+        for tf_doc in self._tf:
+            sq = 0.0
+            for idx, tf in tf_doc.items():
+                idf = math.log((self._num_docs + 1) / (self._df.get(idx, 1) + 1)) + 1.0
+                sq += (tf * idf) ** 2
+            self._norms.append(math.sqrt(sq) or 1.0)
+
+    # -- public API --------------------------------------------------------
+
+    def query(self, text: str, k: int = 3) -> List[MemoryHit]:
+        if not text or not self._docs:
+            return []
+        tokens = _tokenize(text)
+        if not tokens:
+            return []
+        q_tf: Dict[int, float] = {}
+        for tok in tokens:
+            if tok in self._vocab:
+                idx = self._vocab[tok]
+                q_tf[idx] = q_tf.get(idx, 0.0) + 1.0
+        if not q_tf:
+            return []
+        # query norm
+        q_sq = 0.0
+        for idx, tf in q_tf.items():
+            idf = math.log((self._num_docs + 1) / (self._df.get(idx, 1) + 1)) + 1.0
+            q_sq += (tf * idf) ** 2
+        q_norm = math.sqrt(q_sq) or 1.0
+
+        scores: List[Tuple[float, int]] = []
+        for doc_idx, tf_doc in enumerate(self._tf):
+            dot = 0.0
+            for idx, qt in q_tf.items():
+                tfd = tf_doc.get(idx)
+                if tfd is None:
+                    continue
+                idf = math.log((self._num_docs + 1) / (self._df.get(idx, 1) + 1)) + 1.0
+                dot += (qt * idf) * (tfd * idf)
+            if dot <= 0.0:
+                continue
+            cosine = dot / (q_norm * self._norms[doc_idx])
+            scores.append((cosine, doc_idx))
+        scores.sort(reverse=True)
+        hits: List[MemoryHit] = []
+        for score, doc_idx in scores[: max(k, 0)]:
+            source, chunk = self._docs[doc_idx]
+            snippet = chunk.replace("\n", " ").strip()
+            if len(snippet) > 280:
+                snippet = snippet[:277] + "..."
+            hits.append(MemoryHit(source=source, snippet=snippet, score=round(float(score), 4)))
+        return hits
+
+    def sources(self) -> List[str]:
+        return [src for src, _ in self._docs]
+
+    def has_source(self, source: str) -> bool:
+        return any(src == source for src, _ in self._docs)
+
+    def contains_any(self, text: str, sources: Sequence[str]) -> bool:
+        """Return True if any of ``sources`` appears verbatim in ``text``.
+
+        Used by the grader to verify citations are grounded in the corpus
+        rather than hallucinated strings.
+        """
+        if not text:
+            return False
+        return any(src and src in text for src in sources)
+
+    def count_grounded_citations(self, citations: Iterable[str]) -> int:
+        known = set(self.sources())
+        n = 0
+        for citation in citations or []:
+            if not isinstance(citation, str) or not citation.startswith("memory:"):
+                continue
+            if citation[len("memory:"):] in known:
+                n += 1
+        return n
+
+
+__all__ = ["MemoryHit", "Retriever"]