task_id stringlengths 17 84 | title stringlengths 5 62 ⌀ | summary stringlengths 76 379 | category stringclasses 14
values | subdomain stringlengths 6 53 | task_split stringclasses 3
values | task_prompt stringlengths 502 5.84k | agent_must_do listlengths 0 9 | software listlengths 0 9 | input_files listlengths 0 14 | taxonomy dict | source_repo_path stringlengths 23 90 |
|---|---|---|---|---|---|---|---|---|---|---|---|
agriculture_env/crop_rotation_d02 | Crop Rotation Audit for Stable Parcels | Audit French stable agricultural parcels for crop-rotation eligibility and risk, then produce full eligible and flagged GeoPackage outputs with exact schema and layer requirements. | agriculture_env | Precision Agriculture | near-term | You are auditing crop rotation eligibility for stable agricultural parcels in France.
Task directory:
- `base`
Visible inputs:
- Parcel GeoPackage: `base\input\seq1524_d02.gpkg`
- Task brief: `base\input\task_prompt.md`
- Citation note: `base\input\citation.txt`
- Optional Python runtime manifest:
- `base\input\run... | [
"Read the staged crop-rotation brief in `input/task_prompt.md` and apply it exactly.",
"Load the `seq1524_d02` layer from `input/seq1524_d02.gpkg`.",
"Compute the eight required audit columns for every eligible `id_lcp`.",
"Write `output/eligible_units.gpkg` with preserved original columns and geometry plus t... | [
"Python",
"GeoPandas",
"pandas",
"pyogrio",
"shapely",
"pyproj",
"fiona"
] | [
{
"name": "seq1524_d02.gpkg",
"format": "GeoPackage",
"path": "input/seq1524_d02.gpkg",
"description": "Stable agricultural parcel polygons with yearly crop-code columns from 2015 through 2024"
},
{
"name": "task_prompt.md",
"format": "Markdown",
"path": "input/task_prompt.md",
"... | {
"domain_id": "12",
"domain_code": "agriculture_env",
"subdomain_id": "12.2",
"subdomain_code": "precision_ag",
"subdomain_name": "Precision Agriculture"
} | tasks/agriculture_env/crop_rotation_d02 |
agriculture_env/ndvi_zonal_statistics_d02 | Sentinel-2 NDVI Zonal Statistics | Compute a Sentinel-2 NDVI GeoTIFF and polygon-level NDVI zonal statistics for agricultural units with a CRS mismatch between vector and raster inputs. | agriculture_env | Precision Agriculture | full-spectrum | You are given a polygon shapefile of agricultural units and a 6-band Sentinel-2 GeoTIFF.
Task directory:
`base`
Inputs:
- `base\input\rotation_count.shp`
- associated `.dbf`, `.shx`, `.prj`, and `.cpg` files are present in the same directory
- each feature has `id_lcp`, `rot_count`, and polygon geometry
- `base\i... | [
"Use the staged shapefile `input/rotation_count.shp` and Sentinel-2 raster `input/s2_l2a_30m_clip.tif`.",
"Compute NDVI from raster band 4 and band 3 without extra scaling or masking.",
"Write `output/ndvi.tif` as single-band float32 with the same georeferencing as the input raster.",
"Reproject polygons to t... | [
"Python",
"rasterio",
"geopandas"
] | [
{
"name": "rotation_count.shp",
"format": "ESRI Shapefile",
"path": "input/rotation_count.shp",
"description": "Agricultural unit polygons with `id_lcp` and precomputed `rot_count` fields"
},
{
"name": "s2_l2a_30m_clip.tif",
"format": "GeoTIFF",
"path": "input/s2_l2a_30m_clip.tif",
... | {
"domain_id": "12",
"domain_code": "agriculture_env",
"subdomain_id": "12.2",
"subdomain_code": "precision_ag",
"subdomain_name": "Precision Agriculture"
} | tasks/agriculture_env/ndvi_zonal_statistics_d02 |
business_finance/american_option_pricing_ls | American Option Pricing via Longstaff-Schwartz | Implement a three-tier Monte Carlo option-pricing benchmark covering Black-Scholes validation, a single-asset American put solved with Longstaff-Schwartz regression, and a correlated 5-asset basket American put with pathwise Greeks. | business_finance | Quantitative Finance & Trading | full-spectrum | You are working on an Ubuntu coding task about pricing American-style options with Monte Carlo and Longstaff-Schwartz regression.
Task directory:
- `base`
Visible input files:
- Problem specification: `base/input/problem_spec.md`
- Python runtime manifest: `base/input/runtime_env/pyproject.toml`
- Python lockfile: `b... | [
"Read the staged problem specification in full before implementing the solver.",
"Use the staged Python wrapper if a task-local NumPy/SciPy runtime is needed.",
"Produce `results.json` and `exercise_boundary_tier2.npy` under the writable output directory.",
"Keep the reported metrics, exercise boundary sample... | [
"Python",
"NumPy",
"SciPy",
"uv"
] | [
{
"name": "problem_spec.md",
"format": "Markdown",
"path": "input/problem_spec.md",
"description": "Full public specification for the three-tier American option pricing benchmark."
},
{
"name": "runtime_env/pyproject.toml",
"format": "TOML",
"path": "input/runtime_env/pyproject.toml"... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.7",
"subdomain_code": "quant_finance",
"subdomain_name": "Quantitative Finance & Trading"
} | tasks/business_finance/american_option_pricing_ls |
business_finance/ar_full_1500 | Annual Report Full Extraction (1500 files) | Download 1500 annual reports from East Money, extract core technical personnel records, and produce one consolidated workbook with the required schema. | business_finance | Accounting & Finance | last-exam | Task Type:
- Full pipeline (download + parse + structure output), with file verification.
Goal:
- Download all reports listed in base\input\file_list.txt from East Money (东方财富).
- Extract core technical personnel data.
- Output one consolidated Excel file.
Input Files:
1) base\input\file_list.txt
- UTF-8 text file... | [
"Read `input/file_list.txt` to determine the 1500 annual-report PDFs in scope",
"Download all listed annual reports from East Money into `output/downloads/`",
"Extract core technical personnel data for each report",
"Produce `final_dataset.xlsx` with the required columns and data rules"
] | [
"Microsoft Edge",
"Python"
] | [
{
"name": "file_list.txt",
"format": "txt",
"path": "input/file_list.txt",
"description": "UTF-8 list of 1500 annual-report PDF filenames"
}
] | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/ar_full_1500 |
business_finance/ar_full_300 | Annual Report Full Extraction (300 files) | Download 300 annual reports from East Money, extract core technical personnel records, and produce one consolidated workbook with the required schema. | business_finance | Accounting & Finance | full-spectrum | Task Type:
- Full pipeline (download + parse + structure output), with file verification.
Goal:
- Download all reports listed in base\input\file_list.txt from East Money (东方财富).
- Extract core technical personnel data.
- Output one consolidated Excel file.
Input Files:
1) base\input\file_list.txt
- UTF-8 text file... | [
"Read the file list from `{task_dir}\\input\\file_list.txt` (300 PDF filenames).",
"Download all 300 annual report PDFs from East Money (dongchaiwang) to `{remote_output_dir}\\downloads`.",
"Parse each PDF to extract core technical personnel data.",
"Output a single consolidated Excel file at `{remote_output_... | [
"Google Chrome",
"Python"
] | [
{
"name": "file_list.txt",
"format": "UTF-8 text",
"path": "input/file_list.txt",
"description": "300 PDF filenames, one per line. Example: 688001_SomeCompany_2023AnnualReport.pdf"
}
] | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/ar_full_300 |
business_finance/basel_operational_risk_bia_cn | Basel Operational Risk BIA Classification | Classify Chinese operational-risk loss events into Basel event categories and business lines, then calculate regulatory capital under the Basic Indicator Approach. | business_finance | Actuarial & Risk Modeling | full-spectrum | You are working on a Linux VM as an operational-risk analyst.
## Task Directory
`base`
## Visible Inputs
- Task prompt: `base/input/TASK_PROMPT.md`
- Raw Chinese loss events: `base/input/operational_risk_events.xlsx`
- Basel loss event categories: `base/input/loss_event_categories.xlsx`
- Basel business line mapping:... | [
"Read all 60 Chinese operational-risk event descriptions.",
"Assign one Basel loss event category code from `EL1` through `EL7` to each event.",
"Assign one Basel business line code from `BL1` through `BL8` to each event.",
"Calculate operational-risk regulatory capital using the Basel Basic Indicator Approac... | [
"LibreOffice Calc",
"Python",
"openpyxl"
] | [
{
"name": "TASK_PROMPT.md",
"format": "md",
"path": "input/TASK_PROMPT.md",
"description": "Solve-time task instructions and output contract."
},
{
"name": "operational_risk_events.xlsx",
"format": "xlsx",
"path": "input/operational_risk_events.xlsx",
"description": "Chinese oper... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.2",
"subdomain_code": "actuarial_risk",
"subdomain_name": "Actuarial & Risk Modeling"
} | tasks/business_finance/basel_operational_risk_bia_cn |
business_finance/bpmn_category_governance_restructuring_l3 | Flowable BPMN Category Governance Restructuring L3 | Redesign Z Global's Maternal & Baby category governance workflow in Flowable BPMN 6.5.0 for a compound organizational disruption, then produce the required BPMN, compliance, scenario-result, and design-decision artifacts. | business_finance | Enterprise Analytics & Planning | last-exam | You are working on a Linux VM with Docker, Docker Compose, Python, and curl available.
## Your Task
Redesign Z Global's Maternal & Baby category governance workflow in Flowable BPMN 6.5.0 so it handles a compound organizational disruption:
1. the M&B category owner vacancy immediately before a promotion cycle
2. tempo... | [
"Read the staged task prompt, starter BPMN, business rules, org hierarchy, scenarios, and stakeholder artifacts.",
"Modify the BPMN to handle the L3 compound governance disruption while preserving anchored original elements and IDs.",
"Use process definition key `zMBCategoryGovernance_modified_L3` and keep new ... | [
"Flowable 6.5.0",
"Docker",
"Docker Compose",
"Python",
"curl"
] | [
{
"name": "task_prompt.md",
"format": "Markdown",
"path": "input/task_prompt.md",
"description": "Agent-visible task contract"
},
{
"name": "starter_project/original_process.bpmn20.xml",
"format": "BPMN XML",
"path": "input/starter_project/original_process.bpmn20.xml",
"descripti... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.3",
"subdomain_code": "enterprise_analytics",
"subdomain_name": "Enterprise Analytics & Planning"
} | tasks/business_finance/bpmn_category_governance_restructuring_l3 |
business_finance/bpmn_supply_disruption_l3 | Flowable BPMN Supply Disruption L3 | Redesign LY Juice Company's monthly production scheduling workflow in Flowable BPMN 6.5.0 so it handles a compound supply-shortage and quality-hold disruption, then produce the required BPMN and compliance output bundle. | business_finance | Enterprise Analytics & Planning | near-term | You are working on a Linux VM with Docker, Docker Compose, Python, and curl available.
## Your Task
Redesign LY Juice Company's monthly production scheduling workflow in Flowable BPMN 6.5.0 so it handles a compound disruption:
1. raw material supply shortage
2. incoming material quality hold
## Visible Inputs
- task ... | [
"Read the staged task prompt and starter-project materials under `input/`.",
"Modify the BPMN to handle the L3 compound disruption while preserving anchored original elements and IDs.",
"Use process definition key `monthlyProductionScheduling_modified_L3` and keep new manual work as `userTask` nodes.",
"Start... | [
"Flowable 6.5.0",
"Docker",
"Docker Compose",
"Python"
] | [
{
"name": "task_prompt.md",
"format": "Markdown",
"path": "input/task_prompt.md",
"description": "Agent-visible task contract derived from the raw submission prompt with evaluator-only sections removed"
},
{
"name": "starter_project/original_process.bpmn20.xml",
"format": "BPMN XML",
... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.3",
"subdomain_code": "enterprise_analytics",
"subdomain_name": "Enterprise Analytics & Planning"
} | tasks/business_finance/bpmn_supply_disruption_l3 |
business_finance/digital_marketing_ab_test_analysis_1 | Digital Marketing A/B Test Analysis | Design and analyze a marketing A/B test with exclusion handling, stratified assignment checks, two-proportion inference, BH correction, and a final ship/hold recommendation. | business_finance | Sales & Marketing | near-term | You are working on a Linux VM to complete a digital marketing A/B test analysis.
Visible task files:
- `base/input/experiment_brief.md`
- `base/input/historical_metrics.csv`
- `base/input/eligible_population.csv`
- `base/input/exclusion_rules.yaml`
- `base/input/active_experiment_customers.csv`
- `base/input/experimen... | [] | [
"Python 3.12",
"scipy",
"statsmodels",
"pandas",
"GrowthBook"
] | [] | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.6",
"subdomain_code": "sales_marketing",
"subdomain_name": "Sales & Marketing"
} | tasks/business_finance/digital_marketing_ab_test_analysis_1 |
business_finance/digital_marketing_audience_segmentation_1 | Digital Marketing Audience Segmentation | Segment customer profiles into a high-transaction email-disengaged cohort for re-engagement via SMS and push, applying governance rules and overlap analysis. | business_finance | Sales & Marketing | full-spectrum | You are a digital marketing analyst performing audience segmentation for a re-engagement campaign.
## Task Directory
`base`
## Your Task
1. Read the segmentation brief at `base/input/segmentation_brief.md` and the governance policies at `base/input/governance_policies.yaml`. Use `base/input/data_dictionary.tsv` for f... | [
"Filter unified_profiles.parquet to high-transaction email-disengaged customers per the brief.",
"Apply governance suppression (active support tickets) and channel opt-in/out rules.",
"Strip geo/demographic PII columns (age, gender, city, state) from the output roster.",
"Compute per-customer SMS, push, and a... | [
"Python",
"pandas",
"pyarrow"
] | [
{
"name": "unified_profiles.parquet",
"format": "Parquet",
"path": "input/unified_profiles.parquet",
"description": "~10,000 customer profiles with demographics, purchase, and engagement fields."
},
{
"name": "segmentation_brief.md",
"format": "Markdown",
"path": "input/segmentation_... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.6",
"subdomain_code": "sales_marketing",
"subdomain_name": "Sales & Marketing"
} | tasks/business_finance/digital_marketing_audience_segmentation_1 |
business_finance/equity_research_summary | Equity Research Summary | Build a company equity-research workbook from staged filings and market sources, then produce the required summary outputs. | business_finance | Enterprise Analytics & Planning | full-spectrum | You are an equity research analyst preparing a one-page Tesla workbook in LibreOffice Calc.
## Variant
`tsla_fy2023`: Single Tesla equity-research workbook using live Yahoo pages plus fixed FY2023 SEC values.
## Runtime Entry Points
- Read the staged instructions: `tsla_fy2023\input\instruction.md`
- Read the allowed... | [
"Open Yahoo Finance for Tesla (`TSLA`), visit the quote page and the Statistics page, and collect the live market-data and valuation fields listed in `input/instruction.md`.",
"Open SEC EDGAR for Tesla and locate the Tesla FY2023 annual 10-K financial statements needed for the workbook.",
"Launch LibreOffice Ca... | [
"Chrome"
] | [
{
"name": "instruction.md",
"format": "Markdown",
"path": "input/instruction.md",
"description": "Sanitized runtime brief with required sections, output filename, and source workflow"
},
{
"name": "source_urls.txt",
"format": "UTF-8 text",
"path": "input/source_urls.txt",
"descri... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.3",
"subdomain_code": "enterprise_analytics",
"subdomain_name": "Enterprise Analytics & Planning"
} | tasks/business_finance/equity_research_summary |
business_finance/ff5_public_reconstruction | FF5 Public Reconstruction | Reconstruct the monthly Fama-French five-factor series from 2015-01 onward using only the staged allowlisted public sources, then save the required factor CSV. | business_finance | Enterprise Analytics & Planning | last-exam | You are working on a Linux finance benchmark task.
Read these staged files first:
- `base/input/agent_prompt.md`
- `base/input/README.md`
- `base/input/manifest.json`
Task-local entry points:
- Browser launcher: `base/software/browser`
- Python wrapper: `base/software/python`
- UV wrapper: `base/software/uv`
If you ... | [
"Read `input/agent_prompt.md`, `input/README.md`, and `input/manifest.json` before starting.",
"Stay inside the staged public-data allowlist while reading the paper, gathering data, and building the factor pipeline.",
"Optionally install task-facing Python dependencies from `input/runtime_env/` with the staged ... | [
"Google Chrome",
"Python",
"uv"
] | [
{
"name": "agent_prompt.md",
"format": "Markdown",
"path": "input/agent_prompt.md",
"description": "Primary agent-visible benchmark brief, including the allowed public sources and output contract."
},
{
"name": "README.md",
"format": "Markdown",
"path": "input/README.md",
"descri... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.3",
"subdomain_code": "enterprise_analytics",
"subdomain_name": "Enterprise Analytics & Planning"
} | tasks/business_finance/ff5_public_reconstruction |
business_finance/financial_stmt_reconstruction_aapl_fy2024 | Apple FY2024 Balance Sheet Reconstruction | Extract Apple Inc.'s FY2024 Consolidated Balance Sheet from its Form 10-K and return a structured JSON report with exact USD-million values. | business_finance | Accounting & Finance | near-term | You are working on a Linux VM as a financial statement extraction analyst.
## Task Directory
`base`
## Visible Inputs
- Task prompt: `base/input/TASK_PROMPT.md`
- Apple FY2024 10-K PDF: `base/input/aapl-2024-10k.pdf`
- SEC EDGAR HTML: `base/input/aapl-20240928.htm`
- Task metadata: `base/input/task.json`
- Output sch... | [
"Locate the Consolidated Balance Sheets in the staged Apple FY2024 Form 10-K materials.",
"Extract the September 28, 2024 value for every required field in the schema.",
"Write `output/balance_sheet.json` without modifying staged input files."
] | [
"Python",
"pdftotext",
"grep"
] | [
{
"name": "aapl-2024-10k.pdf",
"format": "pdf",
"path": "input/aapl-2024-10k.pdf",
"description": "Cached Apple FY2024 Form 10-K PDF."
},
{
"name": "aapl-20240928.htm",
"format": "html",
"path": "input/aapl-20240928.htm",
"description": "SEC EDGAR HTML filing for the same 10-K."
... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/financial_stmt_reconstruction_aapl_fy2024 |
business_finance/internal_employee_agent_instance_1 | Internal Employee Agent Conversation Replay | Read staged HR and IT support conversations, then produce one deterministic `results.json` that captures the correct per-turn responses and tool usage across all sessions. | business_finance | HR & Project Management | full-spectrum | You are working on a Linux VM.
## Your Task
Act as the company's internal HR / IT assistant across all staged conversation sessions and write the final structured artifact to `base/output/results.json`.
## Visible Inputs
- agent rules: `base/input/agent_rules.md`
- HR knowledge base: `base/input/hr_knowledge_base.md`... | [
"Read the staged rules, knowledge bases, conversation sessions, and deterministic web-search grounding.",
"Process each test id in `queries.json` as an independent multi-turn session while preserving turn order and within-session memory.",
"Write exactly one `results.json` file under `output/`.",
"Use the exa... | [
"Python"
] | [
{
"name": "agent_rules.md",
"format": "Markdown",
"path": "input/agent_rules.md",
"description": "Canonical tool names and operational rules for the HR / IT assistant."
},
{
"name": "hr_knowledge_base.md",
"format": "Markdown",
"path": "input/hr_knowledge_base.md",
"description":... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.5",
"subdomain_code": "hr_pm",
"subdomain_name": "HR & Project Management"
} | tasks/business_finance/internal_employee_agent_instance_1 |
business_finance/legal_ma_consistency_audit_01 | M&A Filing Consistency Audit | Review four Chinese regulatory filings about the same share-transfer transaction and write an English audit report that identifies supported inconsistencies with Chinese evidence and location markers. | business_finance | Litigation Support & Discovery | last-exam | You are reviewing four Chinese regulatory filings about the same share-transfer transaction.
## Task
Read the staged materials, cross-check the filings for factual inconsistencies, and write one
English audit report.
## Visible Inputs
- Task root: `base`
- Task brief: `base/input/task_brief.md`
- Document manifest: `... | [
"Read the staged task brief, document manifest, source PDFs, and extracted-text mirrors.",
"Cross-check the four filings for material factual inconsistencies.",
"Write exactly one English Markdown audit report to `output/audit_report.md`.",
"For every finding, include the original Chinese evidence text and a ... | [
"software/python"
] | [
{
"name": "task_brief.md",
"format": "Markdown",
"path": "input/task_brief.md",
"description": "Agent-visible task instructions and output contract."
},
{
"name": "document_manifest.json",
"format": "JSON",
"path": "input/document_manifest.json",
"description": "Maps the canonica... | {
"domain_id": "7",
"domain_code": "legal",
"subdomain_id": "7.1",
"subdomain_code": "litigation",
"subdomain_name": "Litigation Support & Discovery"
} | tasks/business_finance/legal_ma_consistency_audit_01 |
business_finance/llm_ecosystem_privacy_audit_realdata_1 | GPT Actions Privacy & Policy Audit | Audit 2,253 real GPT Actions against OpenAI's published GPT Actions Usage Policy and compute cross-domain data-exposure amplification across shared backend domains on a Linux VM. | business_finance | Compliance & Regulatory | full-spectrum | You are working on a Linux VM to perform a compliance audit of GPT Actions against OpenAI's published GPT Actions Usage Policy. All inputs are staged locally on this VM. No network access is required or permitted.
## Input Files (read-only)
- Pre-classified Action data: `base/input/pp_action_data_entities.json`
- Data... | [
"Parse `pp_action_data_entities.json`, `taxonomy.csv`, `openai_gpt_actions_usage_policy.md`, and `gpt_action_metadata.json` with only Python stdlib modules.",
"Apply the Policy-to-Taxonomy Mapping Guide verbatim to flag CRITICAL and HIGH taxonomy types per Action field.",
"Write `output/policy_violations.json` ... | [
"Python (`software/python`)"
] | [
{
"name": "pp_action_data_entities.json",
"format": "JSON",
"path": "input/pp_action_data_entities.json",
"description": "Pre-classified data collection profiles for 2,253 GPT Actions, keyed by `\"<backend_domain>, <api_name>\"`. Each Action lists its data fields with an assigned taxonomy `data_type... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.4",
"subdomain_code": "compliance",
"subdomain_name": "Compliance & Regulatory"
} | tasks/business_finance/llm_ecosystem_privacy_audit_realdata_1 |
business_finance/metabase_bi_dashboard_01 | Metabase BI Dashboard | Build a business analytics dashboard in Metabase using a pre-loaded SQLite database, then export key metrics as a structured JSON report. | business_finance | Enterprise Analytics & Planning | near-term | You are a business analyst building a BI dashboard in Metabase.
## Environment
- Metabase is available on this Windows machine. To start it, run the launcher script:
`base\software\launch_metabase.bat`
Then wait for the server to become ready and open http://localhost:3000 in the browser.
- Login credentials are i... | [
"Launch Metabase via the launcher script and wait for the server to start",
"Log in at http://localhost:3000 with the provided credentials",
"Build a dashboard with 6 charts and a heading text card in the Metabase GUI",
"Query the Business Analytics SQLite database to extract all required metrics",
"Save da... | [
"Metabase 0.54.3",
"Microsoft Edge"
] | [
{
"name": "Task requirements",
"format": "md",
"path": "input/requirements.md",
"description": "Detailed dashboard requirements specifying 6 charts and a heading text card."
},
{
"name": "Output schema",
"format": "json",
"path": "input/output_schema.json",
"description": "JSON S... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.3",
"subdomain_code": "enterprise_analytics",
"subdomain_name": "Enterprise Analytics & Planning"
} | tasks/business_finance/metabase_bi_dashboard_01 |
business_finance/odoo | Odoo Supply-Chain End-To-End Workflow | Execute a full Odoo ERP supply-chain workflow — purchases, landed cost, subcontracting, two-level manufacturing, delivery, invoicing, and a returns / credit-note flow — and leave the final business state inside the variant's PostgreSQL database while saving evidence screenshots. | business_finance | Supply Chain & Logistics | full-spectrum | Goal:
Complete the recovered end-to-end Odoo supply-chain workflow and leave the final business state inside the task database.
Variant:
- odoo_compact
- Recovered compact legacy variant. The shared Odoo workflow is preserved, with this variant using its own DB and searchable run tag.
Recovered prompt:
Recovered comp... | [
"Launch Odoo through `software/launch_odoo.cmd` or the auto-opened browser session, and work exclusively inside the variant database (for example `odoo_compact`).",
"Follow the full workflow staged in `input/prompt.txt`: master data, tagged sales / purchases, landed cost, subcontracting, two-level manufacturing w... | [
"Odoo 19",
"PostgreSQL 17"
] | [
{
"name": "Agent-visible workflow prompt",
"format": "`.txt`",
"path": "input/prompt.txt",
"description": "Full recovered workflow contract for the active variant — master data codes, FX rates, warehouse routes, tagged purchases, subcontracting, shortage recovery, invoicing, expenses, and the mandat... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.8",
"subdomain_code": "supply_chain",
"subdomain_name": "Supply Chain & Logistics"
} | tasks/business_finance/odoo |
business_finance/pe_screening_memo_1 | PE Screening Memo 1 | Review a staged Zscaler diligence packet and write a structured private-equity screening memo with an explicit investment recommendation. | business_finance | Accounting & Finance | near-term | You are working on a Linux finance benchmark task.
Read these staged files first:
- `zscaler_fy2025/input/task_brief.md`
- `zscaler_fy2025/input/memo_template.md`
- `zscaler_fy2025/input/source_manifest.json`
Task-local software notes:
- `zscaler_fy2025/software/README.txt`
Your job is to write the requested private... | [
"Read the staged brief, memo template, and source manifest before starting.",
"Use only the staged Zscaler packet and do not use web search.",
"Write a Markdown screening memo that follows the visible template exactly enough to preserve the required section structure.",
"Make an explicit `Go`, `No-Go`, or `Ho... | [
"Python"
] | [
{
"name": "task_brief.md",
"format": "Markdown",
"path": "input/task_brief.md",
"description": "Primary agent-visible benchmark brief with the output contract and workflow constraints."
},
{
"name": "memo_template.md",
"format": "Markdown",
"path": "input/memo_template.md",
"desc... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/pe_screening_memo_1 |
business_finance/saas_onepager_brand_refresh_instance_1 | SaaS One-Pager Brand Refresh | Rebuild a flattened SaaS one-pager as an editable single-slide PowerPoint on Windows, apply the staged NorthstarOS rebrand changes, and export both a PPTX and a PNG. | business_finance | Graphic, Visual & Product Design | near-term | You are editing a SaaS one-pager on a Windows VM.
## Variant
`base`: NorthstarOS one-pager brand refresh
## Runtime Entry Point
- Open PowerPoint from: `base\software\OpenPowerPoint.bat`
## Visible Inputs
- Source one-pager PNG: `base\input\original_onepager.png`
- Edit brief: `base\input\edit_request.txt`
- Brand g... | [
"Read the staged source one-pager, edit brief, brand guide, KPI CSV, chart CSV, and asset bundle from `input/`.",
"Rebuild the page as a meaningfully editable single-slide PowerPoint on Windows.",
"Apply the requested NorthstarOS brand refresh while preserving the overall composition.",
"Save exactly `output/... | [
"Microsoft PowerPoint"
] | [
{
"name": "original_onepager.png",
"format": "PNG image",
"path": "input/original_onepager.png",
"description": "Flattened source one-pager to recreate and refresh."
},
{
"name": "edit_request.txt",
"format": "text",
"path": "input/edit_request.txt",
"description": "Natural-langu... | {
"domain_id": "8",
"domain_code": "visual_media",
"subdomain_id": "8.2",
"subdomain_code": "graphic_design",
"subdomain_name": "Graphic, Visual & Product Design"
} | tasks/business_finance/saas_onepager_brand_refresh_instance_1 |
business_finance/sec_10k_financial_parsing | SEC 10-K Financial Parsing | Parse a fixed 100-filing SEC 10-K corpus into normalized financial JSON outputs, preserve the raw text evidence used for extraction, answer three cross-filing analytical questions, and rerun a fixed validation subset deterministically. | business_finance | Accounting & Finance | last-exam | You are working on a Linux VM with a fixed corpus of SEC 10-K PDFs.
Visible files and tools:
- Filing corpus: `base/input/filings`
- Filing metadata: `base/input/filings/manifest.json`
- Output schema: `base/input/schema/extraction_schema.json`
- Normalization rules: `base/input/schema/normalization_rules.txt`
- Analy... | [
"Inspect the staged SEC 10-K corpus in `input/filings/` and use `input/filings/manifest.json` to understand company, ticker, and year coverage.",
"Produce one normalized extraction JSON per filing under `output/extractions/`, following the staged schema and normalization rules exactly.",
"Save one raw extractio... | [
"Python",
"uv",
"pdfplumber",
"pypdf",
"pydantic"
] | [
{
"name": "filings/",
"format": "Directory of PDFs",
"path": "input/filings/",
"description": "Fixed 100-filing SEC 10-K corpus."
},
{
"name": "manifest.json",
"format": "JSON",
"path": "input/filings/manifest.json",
"description": "Filing metadata covering ticker, company name, ... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/sec_10k_financial_parsing |
business_finance/sse_northbound_programmatic_trading_01 | SSE Northbound Programmatic Trading Research | Answer three closed-book Shanghai Stock Exchange reporting questions from staged Chinese source documents and return a structured JSON memo with Chinese citations and evidence. | business_finance | Compliance & Regulatory | near-term | You are acting as a regulatory consultant on Shanghai Stock Exchange Northbound programmatic-trading reporting.
## Closed-Book Rule
- Use only the staged materials under `base/input`.
- Do not use outside web search or external regulatory knowledge.
## Inputs
- Task brief: `base/input/task_brief.md`
- Question set:... | [
"Read the staged task brief, question set, output contract, document manifest, source documents, and extracted-text mirrors.",
"Answer all three client questions using only the staged Shanghai Stock Exchange materials.",
"Write exactly one JSON file to `output/research_answers.json`.",
"For each question, pro... | [
"Python"
] | [
{
"name": "task_brief.md",
"format": "Markdown",
"path": "input/task_brief.md",
"description": "Agent-visible task instructions and the three client questions."
},
{
"name": "question_set.json",
"format": "JSON",
"path": "input/question_set.json",
"description": "Structured list ... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.4",
"subdomain_code": "compliance",
"subdomain_name": "Compliance & Regulatory"
} | tasks/business_finance/sse_northbound_programmatic_trading_01 |
business_finance/taxform_4_1 | Taxform 4 1 | Prepare a U.S. tax return from staged wage documents and instructions, then save the required JSON outputs. | business_finance | Accounting & Finance | near-term | You are preparing a U.S. tax return inside a browser-based local task environment.
## Variant
`variant_1`: W-2 income with a single Form 1040 output.
## Runtime Entry Points
- Start the local HTTP server by running: `variant_1\software\start_http_server.bat`
- Read the taxpayer instructions at: `variant_1\input\instr... | [
"Launch the task-local HTTP server from `software/start_http_server.bat`, which serves the staged `input/` directory at `http://localhost:8080`.",
"Read `input/instruction.txt` and the staged source forms listed in `input/README.md`, combining free-text tax instructions with values shown on the pre-filled W-2 / 1... | [
"Python",
"Chrome"
] | [
{
"name": "README.md",
"format": "Markdown",
"path": "input/README.md",
"description": "Source-authored setup and page-order instructions"
},
{
"name": "instruction.txt",
"format": "UTF-8 text",
"path": "input/instruction.txt",
"description": "Taxpayer scenario, filing rules, and... | {
"domain_id": "6",
"domain_code": "business_finance",
"subdomain_id": "6.1",
"subdomain_code": "accounting",
"subdomain_name": "Accounting & Finance"
} | tasks/business_finance/taxform_4_1 |
computing_math/branch_bound_atsp | Branch-and-Bound ATSP Solver | Implement a Python branch-and-bound solver for deterministic asymmetric TSP instances and report certified results across three tiers. | computing_math | Mathematical & Operations Research | near-term | You are working on a Linux VM.
## Task Directory
`base`
## Visible Inputs
- Task prompt: `base/input/TASK_PROMPT.md`
- ATSP problem specification: `base/input/problem_spec.md`
- Runtime manifest: `base/input/runtime_env/pyproject.toml`
- Python entry point with task dependencies: `base/software/python`
## Your Task
... | [
"Regenerate the deterministic 3D city coordinates for the 12, 20, and 35 city tiers.",
"Implement branch-and-bound and assignment-relaxation based lower bounds without external optimization solvers.",
"Write a valid `results.json` containing tours, costs, bounds, gaps, node counts, and runtime metrics."
] | [
"Python",
"NumPy",
"SciPy"
] | [
{
"name": "problem_spec.md",
"format": "markdown",
"path": "input/problem_spec.md",
"description": "Authoritative ATSP problem statement and output schema."
},
{
"name": "TASK_PROMPT.md",
"format": "markdown",
"path": "input/TASK_PROMPT.md",
"description": "Normalized solve-time ... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.4",
"subdomain_code": "math_ops_research",
"subdomain_name": "Mathematical & Operations Research"
} | tasks/computing_math/branch_bound_atsp |
computing_math/cfr_game_theory_equilibrium | CFR Game Theory Equilibrium | Implement exact and regret-minimization solvers for three zero-sum games and write one combined JSON output. | computing_math | Mathematical & Operations Research | last-exam | You are working on a Linux VM.
## Your Task
Implement equilibrium solvers for three progressively harder two-player zero-sum games and save one combined `results.json`.
## Visible Input
- Authoritative problem specification: `base/input/problem_spec.md`
- Solve-time runtime manifest: `base/input/runtime_env/pyproject... | [
"Read the staged problem specification and implement all three tiers of the benchmark.",
"Use the staged runtime wrappers to install and run the NumPy-only solve-time environment if needed.",
"Return one combined `results.json` containing Tier 1 matrix-game, Tier 2 Kuhn CFR, and Tier 3 Leduc MCCFR outputs."
] | [
"Python",
"uv",
"NumPy"
] | [
{
"name": "Problem specification",
"format": "Markdown",
"path": "input/problem_spec.md",
"description": "Agent-visible task specification for all three game-solving tiers."
},
{
"name": "Runtime manifest",
"format": "pyproject.toml",
"path": "input/runtime_env/pyproject.toml",
"... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.4",
"subdomain_code": "math_ops_research",
"subdomain_name": "Mathematical & Operations Research"
} | tasks/computing_math/cfr_game_theory_equilibrium |
computing_math/clustered_cyclic_code_circuit_level_simulation | Clustered-Cyclic Code Circuit Simulation | Reproduce circuit-level logical failure-rate data for three clustered-cyclic CSS quantum error-correcting codes using the supplied construction notes and simulation grid. | computing_math | Quantum Computing | full-spectrum | You are working on a Linux VM.
## Variant
`base`: Three clustered-cyclic CSS codes with QUITS/Stim circuit-level memory simulation.
## Input Files
- Task notes: `base/input/cc_codes_quits_extraction.tex`
- Simulation grid: `base/input/simulation_grid.csv`
- Output requirements: `base/input/output_requirements.txt`
- ... | [
"Read `input/cc_codes_quits_extraction.tex`, `input/simulation_grid.csv`, and `input/output_requirements.txt`.",
"Install or inspect the open-source Python runtime from `input/runtime_env/pyproject.toml` or `input/requirements.txt`.",
"Construct and simulate the `[24,8,3]`, `[40,8,5]`, and `[54,18,3]` clustered... | [
"Python",
"Stim",
"ldpc",
"QUITS",
"NumPy",
"pandas"
] | [
{
"name": "cc_codes_quits_extraction.tex",
"format": "tex",
"path": "input/cc_codes_quits_extraction.tex",
"description": "Agent-facing construction and simulation notes from the submitter."
},
{
"name": "simulation_grid.csv",
"format": "csv",
"path": "input/simulation_grid.csv",
... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.7",
"subdomain_code": "quantum",
"subdomain_name": "Quantum Computing"
} | tasks/computing_math/clustered_cyclic_code_circuit_level_simulation |
computing_math/cost_optimization_1 | AWS Cost Optimization Analysis | Analyze a synthetic AWS billing dashboard screenshot plus usage and pricing tables to identify wasteful resources and produce a structured optimization report with projected monthly savings. | computing_math | Infrastructure Engineering & Cloud Operations | full-spectrum | You are an AWS cost optimization analyst working on Linux.
## Task Directory
`base`
## Visible Inputs
- Task brief: `base/input/task_description.txt`
- Dashboard image: `base/input/aws_billing_dashboard.png`
- Structured usage data: `base/input/resource_usage.csv`
- Pricing table: `base/input/aws_pricing_reference.cs... | [
"Read `input/task_description.txt` and inspect the dashboard image directly.",
"Use the CSV inputs plus the 3 dashboard-only findings to identify wasteful EC2, RDS, S3, NAT Gateway, Elastic IP, and CloudWatch spend.",
"Apply the explicit pricing and optimization rules from the task brief using `input/aws_pricin... | [
"Python",
"pandas",
"uv"
] | [
{
"name": "aws_billing_dashboard.png",
"format": "PNG",
"path": "input/aws_billing_dashboard.png",
"description": "Synthetic AWS billing dashboard screenshot containing both structured tables and 3 dashboard-only findings that are not present in the usage CSV."
},
{
"name": "resource_usage.c... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.6",
"subdomain_code": "infra_cloud",
"subdomain_name": "Infrastructure Engineering & Cloud Operations"
} | tasks/computing_math/cost_optimization_1 |
computing_math/cp_test_gen_1 | Competitive Programming Test Generator — Binary String Copying | Write gen.cpp, a C++ test case generator for the Binary String Copying problem that produces adversarial inputs to distinguish correct from incorrect solutions among 10 hidden submissions. | computing_math | Software Engineering | near-term | You are tasked with writing gen.cpp, a C++ program that prints one complete valid test case to stdout per invocation and accepts a single integer seed as argv[1] (e.g. ./gen 1, ./gen 2, ./gen 50) to vary output across invocations.
The problem statement is provided in Binary String Copying Problem Statement.pdf. The 10... | [
"Read the Binary String Copying problem statement PDF and submission IDs from the xlsx.",
"Analyze the problem structure to identify adversarial test case strategies (TLE, WA, boundary conditions).",
"Write gen.cpp that generates varied test cases across 50 seeds respecting all constraints.",
"Compile and ver... | [
"g++ (C++17)",
"bash"
] | [
{
"name": "Binary String Copying Problem Statement.pdf",
"format": "pdf",
"path": "input/Binary String Copying Problem Statement.pdf",
"description": "Full Codeforces problem statement with constraints."
},
{
"name": "Binary String Copying Inputs.xlsx",
"format": "xlsx",
"path": "inp... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.1",
"subdomain_code": "software_eng",
"subdomain_name": "Software Engineering"
} | tasks/computing_math/cp_test_gen_1 |
computing_math/data_pipeline_etl_instance_1 | Retail ETL Warehouse Cleanup | Build a cleaned SQLite star-schema warehouse plus two truthful JSON sidecars from messy retail CSV, JSON, and TSV source files on a Linux VM. | computing_math | Data & Analytics Engineering | full-spectrum | You are working on a Linux VM to build a cleaned SQLite warehouse from messy retail source files.
Visible task files:
- `base/input/task_prompt.md`
- `base/input/target_schema_spec.json`
- `base/input/output_contract.json`
- `base/input/raw_transactions/`
- `base/input/raw_customers/customers.json`
- `base/input/raw_p... | [
"Read the staged prompt, schema spec, output contract, and messy retail source files under `input/`.",
"Build a cleaned SQLite warehouse with deduplication, null handling, standardization, surrogate keys, and a full `dim_dates` table.",
"Write `output/warehouse.db`, `output/data_quality_report.json`, and `outpu... | [
"Python 3.10",
"uv",
"pandas",
"SQLite"
] | [
{
"name": "task_prompt.md",
"format": "Markdown",
"path": "input/task_prompt.md",
"description": "Agent-visible task brief for the ETL workflow."
},
{
"name": "target_schema_spec.json",
"format": "JSON",
"path": "input/target_schema_spec.json",
"description": "Semantic warehouse ... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.2",
"subdomain_code": "data_analytics",
"subdomain_name": "Data & Analytics Engineering"
} | tasks/computing_math/data_pipeline_etl_instance_1 |
computing_math/dit_pipeline_cfg_alignment_fid_256_001 | Repair Standalone DiT Pipeline CFG Alignment | Repair a buggy standalone DiT sampling pipeline so its classifier-free guidance behavior matches the intended reference implementation while preserving the one-file edit workflow. | computing_math | AI Engineering, Safety & CS Research | full-spectrum | You are repairing a standalone Diffusers-based DiT pipeline on a Linux VM.
Task directory:
- `base`
Visible files:
- Buggy starter pipeline: `base/input/pipeline_dit.py`
- Raw sampling harness context: `base/input/sample.py`
- Raw run script context: `base/input/run_sample.sh`
- Benchmark-owned prompt: `base/input/ta... | [
"Read the benchmark-owned task prompt and inspect the buggy standalone `pipeline_dit.py` plus the raw context files `sample.py` and `run_sample.sh`.",
"Repair the standalone DiT pipeline so the guidance behavior matches the intended reference behavior.",
"Keep the fix localized to the pipeline file rather than ... | [
"Python",
"Diffusers",
"PyTorch"
] | [
{
"name": "pipeline_dit.py",
"format": "Python",
"path": "input/pipeline_dit.py",
"description": "Buggy standalone starter pipeline."
},
{
"name": "sample.py",
"format": "Python",
"path": "input/sample.py",
"description": "Raw sampling harness context from the submission."
},
... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.3",
"subdomain_code": "ai_cs_research",
"subdomain_name": "AI Engineering, Safety & CS Research"
} | tasks/computing_math/dit_pipeline_cfg_alignment_fid_256_001 |
computing_math/ghidra_malware_config_extraction_01 | Ghidra Malware Config Extraction | Reverse engineer a packed Windows malware sample in Ghidra and recover the configuration fields requested by the visible JSON schema. | computing_math | Cybersecurity & Digital Forensics | near-term | You are a malware analyst working on a Windows VM.
## Your Task
Use Ghidra to reverse engineer the staged executable and recover the malware configuration requested by the visible schema.
## Visible Files
- Sample to analyze: `base\input\sample.exe`
- Output schema: `base\input\output_schema.json`
- Launch Ghidra fro... | [
"Launch the staged task-local Ghidra runtime from `software/launch_ghidra.bat`.",
"Import `input/sample.exe` into a local non-shared Ghidra project.",
"Recover the packer, C2, obfuscation, and architecture details requested by the visible schema.",
"Save exactly one schema-valid `output/malware_config.json`."... | [
"Ghidra 11.3",
"JDK 21"
] | [
{
"name": "Packed malware sample",
"format": "Windows PE",
"path": "input/sample.exe",
"description": "The executable the agent must reverse engineer in Ghidra."
},
{
"name": "Visible output schema",
"format": "JSON",
"path": "input/output_schema.json",
"description": "Defines th... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.5",
"subdomain_code": "cybersecurity",
"subdomain_name": "Cybersecurity & Digital Forensics"
} | tasks/computing_math/ghidra_malware_config_extraction_01 |
computing_math/go_game_reconstruction_1 | Go Game Reconstruction 1 | Reconstruct a professional 19x19 Go game inside Sabaki from a final board image, five known opening moves, and game metadata, then export the result as SGF. | computing_math | Sports | full-spectrum | You are reconstructing a professional 19x19 Go game in Sabaki on Ubuntu.
## Your Task
Use the final board image plus the known metadata to reconstruct the game move by move inside Sabaki, then export the reconstructed game as SGF.
## Visible Inputs
- Final board image: `base/input/input-board-position.png`
- Sabaki A... | [
"Inspect the staged final-board image under `input/`.",
"Launch the staged Sabaki AppImage on the Ubuntu VM and use the GUI rather than bypassing the task with a browser or Go engine.",
"Reconstruct the move sequence using the visible board image, five fixed opening moves, and the provided metadata.",
"Export... | [
"Sabaki v0.52.2",
"Python",
"sgfmill"
] | [
{
"name": "Final board image",
"format": "PNG image",
"path": "input/input-board-position.png",
"description": "The final 19x19 board state shown to the agent."
},
{
"name": "Sabaki runtime",
"format": "Linux AppImage",
"path": "software/sabaki-v0.52.2-linux-x64.AppImage",
"descr... | {
"domain_id": "14",
"domain_code": "other",
"subdomain_id": "14.1",
"subdomain_code": "sports",
"subdomain_name": "Sports"
} | tasks/computing_math/go_game_reconstruction_1 |
computing_math/ising_post_measurement_1 | Post-Measurement States for a Quantum Ising Chain | Compute the post-measurement output bundle for several 1D quantum Ising chain variants, including the critical ground state, measurement probabilities, one-site reduced density matrices, and optional one-body correlators. | computing_math | Quantum Computing | near-term | You are working on a Linux VM.
## Variant
`n10_critical_u01_correlators`: N=10, u=0.1, critical ancilla equal to the computed ground state, with one-body correlators
## Input Files
- Task specification: `n10_critical_u01_correlators/input/task_specification.md`
- Numerical parameters: `n10_critical_u01_correlators/in... | [
"Read the staged task specification and variant parameters from `input/task_specification.md` and `input/config.json`.",
"Use the staged `ancilla_state.npy` only for the paramagnetic-ancilla variants; otherwise follow the critical-ancilla rule from the visible task spec.",
"Optionally install task-specific Pyth... | [
"Python",
"NumPy",
"SciPy",
"uv"
] | [
{
"name": "config.json",
"format": "JSON",
"path": "input/config.json",
"description": "Variant-specific numerical parameters such as system size, coupling strength, ancilla mode, and site index."
},
{
"name": "task_specification.md",
"format": "Markdown",
"path": "input/task_specifi... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.7",
"subdomain_code": "quantum",
"subdomain_name": "Quantum Computing"
} | tasks/computing_math/ising_post_measurement_1 |
computing_math/k3_abelian_extensions | Finite Abelian Extension Classification | Solve a staged finite abelian group extension classification problem in GAP and produce the exact canonical JSON answer. The benchmark contains 6 variants with different abelian groups H and search ranges for m. | computing_math | Mathematical & Operations Research | near-term | You are working on a Linux VM.
## Your Task
Solve the staged finite abelian group extension classification problem for variant `h_4_4_4_m_1_8` ((Z/4)^3, m=1..8).
## Visible Files
- Variant config: `h_4_4_4_m_1_8/input/config.json`
- Mathematical task specification: `h_4_4_4_m_1_8/input/task_specification.md`
- GAP la... | [
"Read the staged `input/config.json` and `input/task_specification.md` for the active variant.",
"Use GAP to enumerate every finite abelian group `G` that fits into the required short exact sequences across the full `m` search range.",
"Determine the `product_type` classification for every valid extension group... | [
"GAP",
"Python"
] | [
{
"name": "config.json",
"format": "JSON",
"path": "input/config.json",
"description": "Variant parameters defining `H_invariant_factors`, `H_order`, and the inclusive `m_search_range`."
},
{
"name": "task_specification.md",
"format": "Markdown",
"path": "input/task_specification.md"... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.4",
"subdomain_code": "math_ops_research",
"subdomain_name": "Mathematical & Operations Research"
} | tasks/computing_math/k3_abelian_extensions |
computing_math/k8s_migration_1 | Docker Compose To Kubernetes Migration | Migrate a 3-tier Docker Compose web app (React, Flask, PostgreSQL) to a production-grade Helm chart on local Minikube, plus matching Terraform and a CI/CD pipeline; deliverables are scored on a weighted Helm/Terraform/CI/CD/verification rubric. | computing_math | Infrastructure Engineering & Cloud Operations | near-term | You are migrating a Docker Compose 3-tier web application (React frontend, Flask backend, PostgreSQL) to a production-grade Kubernetes deployment on a local Minikube cluster.
## Working Directory
Task root on this VM:
- `base`
Agent-visible inputs:
- Compose source + app code: `base/input/app/`
- Architecture spec: ... | [
"Read `input/requirements.md` and `input/README.md` first; these list the architectural targets and which resources, replicas, probes, and policies must end up in the Helm chart.",
"Use the task-local wrappers under `software/` instead of raw PATH commands: `software/docker`, `software/minikube`, `software/kubect... | [
"Docker",
"Minikube",
"kubectl",
"Helm",
"Terraform",
"Trivy",
"Python"
] | [
{
"name": "Architecture spec",
"format": "markdown",
"path": "input/requirements.md",
"description": "Detailed Kubernetes spec: replicas, resource limits, HPA, NetworkPolicy, probes, CI/CD stages."
},
{
"name": "Task README",
"format": "markdown",
"path": "input/README.md",
"desc... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.6",
"subdomain_code": "infra_cloud",
"subdomain_name": "Infrastructure Engineering & Cloud Operations"
} | tasks/computing_math/k8s_migration_1 |
computing_math/k8s_payment_api_root_cause_analysis | Kubernetes Payment API Root Cause Analysis | Analyze static Kubernetes incident artifacts for a flapping payment-api deployment and produce a grounded JSON root-cause report. | computing_math | Data & Analytics Engineering | near-term | You are the on-call SRE for a Kubernetes production incident.
## Task Directory
`base`
## Visible Inputs
- Task prompt: `base/input/task_prompt.md`
- Cluster state capture: `base/input/cluster_state.txt`
- Current Deployment and HPA YAML: `base/input/deployment.yaml`
- Crashing pod log: `base/input/failing_pod.log`
... | [
"Read the task prompt and three static incident artifacts.",
"Identify Kubernetes root causes and cite literal evidence from the input files.",
"List affected resources and prioritized remediation steps.",
"Write output/root_cause_analysis.json."
] | [
"Python"
] | [
{
"name": "task_prompt.md",
"format": "md",
"path": "input/task_prompt.md",
"description": "Sanitized task instructions."
},
{
"name": "cluster_state.txt",
"format": "txt",
"path": "input/cluster_state.txt",
"description": "Captured kubectl pod, event, top, deployment, and HPA ou... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.2",
"subdomain_code": "data_analytics",
"subdomain_name": "Data & Analytics Engineering"
} | tasks/computing_math/k8s_payment_api_root_cause_analysis |
computing_math/mp_checkpoint_consolidation_v2 | Checkpoint Consolidation V2 | Infer the non-standard TP/PP/EP checkpoint layout from staged framework code and shard files, then reconstruct a single HuggingFace-compatible `model.safetensors` artifact. | computing_math | AI Engineering, Safety & CS Research | near-term | You are working on a Linux checkpoint-consolidation task.
## Task Directory
`base`
## Visible Inputs
- Checkpoint shards: `base/input/checkpoints`
- Framework source: `base/input/framework`
- Reference model code: `base/input/reference_model/model.py`
- Reference model config: `base/input/reference_model/config.json`... | [
"Inspect the staged framework code and shard filenames to infer the checkpoint layout.",
"Reconstruct one single-device checkpoint compatible with the staged reference model.",
"Save exactly one file named `model.safetensors` under the active variant's `output/` directory.",
"Produce a checkpoint that matches... | [
"Python",
"PyTorch",
"safetensors",
"uv"
] | [
{
"name": "Checkpoint shards",
"format": "PyTorch .pt files",
"path": "input/checkpoints/",
"description": "The staged TP/PP/EP shard files for the active variant."
},
{
"name": "Framework source",
"format": "Python",
"path": "input/framework/",
"description": "The visible source... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.3",
"subdomain_code": "ai_cs_research",
"subdomain_name": "AI Engineering, Safety & CS Research"
} | tasks/computing_math/mp_checkpoint_consolidation_v2 |
computing_math/newyear_keygen2 | Newyear Keygen2 | Reverse engineer the bundled program, recover the key-generation logic, and write the correct answer for the staged UID. | computing_math | Cybersecurity & Digital Forensics | last-exam | Provided UID:
- `20252025`
Goal:
Reverse the bundled Windows crackme and compute the correct password for the provided UID.
Inputs:
- Executable to analyze: base\input\crackme.exe
Output:
- Save EXACTLY one UTF-8 text file to: base\output\key.txt
- The file must contain exactly one line in the form `flag{...}`
- Do ... | [
"Reverse the bundled crackme.",
"Use the provided UID `20252025`.",
"Compute the password for the most recent UTC half-hour slot.",
"Write exactly one line to `output/key.txt`."
] | [
"Python"
] | [
{
"name": "crackme.exe",
"format": "Windows PE",
"path": "input/crackme.exe",
"description": "agent-visible crackme"
}
] | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.5",
"subdomain_code": "cybersecurity",
"subdomain_name": "Cybersecurity & Digital Forensics"
} | tasks/computing_math/newyear_keygen2 |
computing_math/os_log_permission_guard_v1 | OS Log Permission Guard | Safely update sandboxed Linux log-file permissions from a filesystem snapshot while preserving protected system-owner, active-writer, and non-log files. | computing_math | Infrastructure Engineering & Cloud Operations | near-term | You are working on a Linux VM.
## Task Directory
`base`
## Visible Inputs
- Filesystem snapshot: `base/input/fs_snapshot.tar.gz`
- Ownership metadata: `base/input/ownership.csv`
- Initial permission metadata: `base/input/permissions.csv`
- Active-writer exclusion list: `base/input/active_writers.json`
- Workspace set... | [
"Read the staged task instructions and metadata.",
"Create a writable sandbox workspace under `output/sandbox_fs`.",
"Change eligible `.log` files to mode `444` while leaving protected files unchanged.",
"Write `output/final_state.json` with final metadata for every listed file."
] | [
"Bash",
"GNU coreutils",
"tar",
"Python"
] | [
{
"name": "fs_snapshot.tar.gz",
"format": "tar.gz",
"path": "input/fs_snapshot.tar.gz",
"description": "Sandbox filesystem snapshot containing `/var/logs` files."
},
{
"name": "ownership.csv",
"format": "csv",
"path": "input/ownership.csv",
"description": "Owner/group metadata so... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.6",
"subdomain_code": "infra_cloud",
"subdomain_name": "Infrastructure Engineering & Cloud Operations"
} | tasks/computing_math/os_log_permission_guard_v1 |
computing_math/paper_reproduction_instance_1 | LCA-on-the-Line Table 2 Reproduction | Reproduce the 40-cell correlation table (Table 2) from the ICML 2024 paper "LCA-on-the-Line: Benchmarking Out-of-Distribution Generalization with Class Taxonomies" (arXiv:2407.16067). The provided codebase ships precomputed per-model metric dictionaries sufficient to regenerate all 40 R² / Pearson correlations without ... | computing_math | AI Engineering, Safety & CS Research | near-term | You are reproducing Table 2 of the ICML 2024 paper "LCA-on-the-Line: Benchmarking Out-of-Distribution Generalization with Class Taxonomies" (arXiv:2407.16067).
## Inputs (under `input/`)
- `paper.pdf` - local copy of the paper; do not re-fetch from arXiv.
- `codebase.zip` - snapshot of the official repo `github.com/E... | [
"Read `paper.pdf` and identify Table 2 as the correlation table (R² and Pearson between ID LCA/Top1 and OOD Top1/Top5 across five ImageNet-OOD datasets and 75 pretrained models).",
"Unpack `input/codebase.zip` and use the precomputed `ICML_dict_result_metric_dict` + `ICML_dict_agreement_dict_*.npy` artifacts to r... | [
"Python",
"PyTorch",
"NumPy",
"pandas",
"SciPy",
"scikit-learn",
"statsmodels"
] | [
{
"name": "Paper PDF",
"format": "pdf",
"path": "input/paper.pdf",
"description": "ICML 2024 paper \"LCA-on-the-Line: Benchmarking Out-of-Distribution Generalization with Class Taxonomies\" (arXiv:2407.16067v1). Local copy."
},
{
"name": "Codebase snapshot",
"format": "zip",
"path": ... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.3",
"subdomain_code": "ai_cs_research",
"subdomain_name": "AI Engineering, Safety & CS Research"
} | tasks/computing_math/paper_reproduction_instance_1 |
computing_math/particle_filter_nonlinear_tracking | Particle Filter Nonlinear Tracking | Implement a three-tier particle-filter benchmark in Python, covering a linear-Gaussian validation case, 2D range-bearing tracking, and a coordinated-turn smoother case with Student-t process noise. | computing_math | Mathematical & Operations Research | last-exam | You are working on an Ubuntu coding task about particle filtering for nonlinear tracking.
Task directory:
- `base`
Visible input files:
- Problem specification: `base/input/problem_spec.md`
- Python runtime manifest: `base/input/runtime_env/pyproject.toml`
- Python lockfile: `base/input/runtime_env/uv.lock`
- Canonic... | [
"Read the staged problem specification in full before implementing the solver.",
"Use the staged Python wrapper if a task-local NumPy/SciPy runtime is needed.",
"Implement the solver in `output/pf_solver.py`.",
"Produce `tier1_results.npz`, `tier2_results.npz`, `tier3_results.npz`, and `results.json` under th... | [
"Python",
"NumPy",
"SciPy",
"uv"
] | [
{
"name": "problem_spec.md",
"format": "Markdown",
"path": "input/problem_spec.md",
"description": "Full public task specification for the three-tier particle-filter benchmark."
},
{
"name": "runtime_env/pyproject.toml",
"format": "TOML",
"path": "input/runtime_env/pyproject.toml",
... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.4",
"subdomain_code": "math_ops_research",
"subdomain_name": "Mathematical & Operations Research"
} | tasks/computing_math/particle_filter_nonlinear_tracking |
computing_math/pcap_enterprise_triage_01 | Enterprise PCAP Triage In Wireshark | Analyze a staged enterprise packet capture in Wireshark, identify the compromised internal host, reconstruct the infection chain, and deliver a structured triage report. | computing_math | Cybersecurity & Digital Forensics | last-exam | You are a cybersecurity analyst working on a Windows VM.
## Your Task
Use Wireshark to analyze the staged enterprise packet capture and produce a structured triage report.
## Input Files
- PCAP: `base\input\capture_enhanced.pcap`
- Output schema: `base\input\output_schema.json`
- Software notes: `base\software\README... | [
"Open the staged packet capture in Wireshark on the Windows VM.",
"Identify the compromised internal workstation from the network evidence.",
"Reconstruct the infection timeline in chronological order with the required timestamps and endpoints.",
"Recover the initial vector, malicious delivery URL, and C2 ser... | [
"Wireshark"
] | [
{
"name": "Enterprise packet capture",
"format": "PCAP",
"path": "input/capture_enhanced.pcap",
"description": "The staged network capture the agent must inspect in Wireshark."
},
{
"name": "Visible output schema",
"format": "JSON",
"path": "input/output_schema.json",
"descriptio... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.5",
"subdomain_code": "cybersecurity",
"subdomain_name": "Cybersecurity & Digital Forensics"
} | tasks/computing_math/pcap_enterprise_triage_01 |
computing_math/ranking_node_feature_parity_recovery_instance_1 | Ranking Node Feature Parity Recovery | Recover a ranking-service node by rebuilding the feature manifest, deleting only safe rollback debris, and preserving required serving shards. | computing_math | Software Engineering | full-spectrum | You are the on-call ML platform engineer for a Linux ranking-serving node.
The evaluator will prepare a writable task workspace at:
- `/workspace`
Start by reading:
- `/workspace/instruction.md`
- `/workspace/config.json`
- `/workspace/logs/service.log`
You must repair the node by implementing:
- `/workspace/safe_re... | [] | [
"Python 3",
"pytest"
] | [] | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.1",
"subdomain_code": "software_eng",
"subdomain_name": "Software Engineering"
} | tasks/computing_math/ranking_node_feature_parity_recovery_instance_1 |
computing_math/recsys_cold_start_instance_1 | Hybrid Recommender Cold-Start Benchmark | Build a warm-plus-cold hybrid recommender on Linux, using temporal splitting, metadata/embedding cold-start ranking, and a combined output contract. | computing_math | Data & Analytics Engineering | near-term | You are building a hybrid recommender system on a Linux VM.
Visible inputs:
- `base/input/interactions.csv`
- `base/input/item_metadata.csv`
- `base/input/item_embeddings.json`
- `base/input/user_profiles.csv`
- `base/input/runtime_env/pyproject.toml`
- `base/input/runtime_env/uv.lock`
- `base/software/setup_runtime_e... | [
"Build a warm-item recommender that uses both `rating` and `watch_percentage`.",
"Build a cold-start model from `item_metadata.csv` and `item_embeddings.json` without using cold-item interactions.",
"Use the canonical per-user temporal split: `70%` train, `10%` validation, `20%` test.",
"Fuse warm and cold sc... | [
"Python 3.10",
"uv",
"numpy",
"pandas",
"scikit-learn",
"scipy"
] | [
{
"name": "interactions.csv",
"format": "csv",
"path": "input/interactions.csv",
"description": "Warm-item user interaction history with rating, watch percentage, and timestamps."
},
{
"name": "item_metadata.csv",
"format": "csv",
"path": "input/item_metadata.csv",
"description":... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.2",
"subdomain_code": "data_analytics",
"subdomain_name": "Data & Analytics Engineering"
} | tasks/computing_math/recsys_cold_start_instance_1 |
computing_math/synthetic_causal_structure_inference | Synthetic Causal Structure Inference | Inspect 40 synthetic tabular datasets, infer each dataset's causal scenario and graph structure, and submit one combined JSON benchmark prediction file. | computing_math | Data & Analytics Engineering | near-term | You are working on a Linux VM to solve a synthetic causal-structure benchmark.
## Input Files
- Task brief: `base/input/task_brief.md`
- Benchmark prompt: `base/input/AGENT_PROMPT_FOR_SYNTHETIC_CAUSAL_TASK.md`
- Sanitized dataset manifest: `base/input/manifest.json`
- Dataset directories: `base/input/public/dataset_##... | [
"Read the staged task brief, benchmark prompt, and sanitized manifest.",
"Inspect each dataset directory under `input/public/dataset_###/`.",
"Infer the scenario family, identification strategy, identifiability, variable roles, directed edges, and latent confounders for every dataset.",
"Write one final `subm... | [
"Python",
"pandas",
"numpy",
"statsmodels",
"scikit-learn",
"networkx"
] | [
{
"name": "Task brief",
"format": "`.md`",
"path": "input/task_brief.md",
"description": "High-level wrapper with output path and runtime notes"
},
{
"name": "Benchmark prompt",
"format": "`.md`",
"path": "input/AGENT_PROMPT_FOR_SYNTHETIC_CAUSAL_TASK.md",
"description": "Detailed... | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.2",
"subdomain_code": "data_analytics",
"subdomain_name": "Data & Analytics Engineering"
} | tasks/computing_math/synthetic_causal_structure_inference |
computing_math/tris_crackme | Tris Crackme | Analyze the bundled registration crackme and make the program appear successfully registered. | computing_math | Cybersecurity & Digital Forensics | full-spectrum | Goal:
Solve the registration crackme and make the bundled Windows program appear registered.
Target program:
- Executable: base\input\tris.exe
Requirements:
1. Launch and analyze the program if needed.
2. Make the registration succeed. You can use any method to achieve this, such as reverse engineering, patching, or ... | [
"Reverse the bundled registration crackme.",
"Make the program appear registered, typically by writing the correct registry values.",
"Leave the registered state persistent across restarts."
] | [
"Python"
] | [
{
"name": "tris.exe",
"format": "Windows PE",
"path": "input/tris.exe",
"description": "agent-visible target"
}
] | {
"domain_id": "9",
"domain_code": "computing_math",
"subdomain_id": "9.5",
"subdomain_code": "cybersecurity",
"subdomain_name": "Cybersecurity & Digital Forensics"
} | tasks/computing_math/tris_crackme |
education_info/homework_grading_numerical_pdes_instance_02 | Homework Grading Numerical PDEs 02 | Grade five synthetic graduate numerical-PDE homework submissions from released materials, then emit scores, error tags, per-student feedback, a class summary, and a grader manifest. | education_info | Educational Technology | near-term | You are grading synthetic graduate numerical-PDE homework submissions on Linux.
Task directory:
- `base`
Visible inputs:
- Instructions: `base/input/TASK_PROMPT.md`
- Released materials: `base/input/released/`
- Starter scaffold: `base/input/starter_project/`
Your task:
1. Read the grading protocol, rubric, solution... | [
"Read the released grading materials and student submissions.",
"Produce rubric-aligned grades for all five students.",
"Assign the appropriate error tags.",
"Write all five required deliverables under `output/`."
] | [
"Python"
] | [
{
"name": "Released materials",
"format": "Markdown, PDF, JSON",
"path": "input/released/",
"description": "Problem set, rubric, solution key, protocol, and student submissions."
},
{
"name": "Starter scaffold",
"format": "Python + JSON",
"path": "input/starter_project/",
"descri... | {
"domain_id": "11",
"domain_code": "education_info",
"subdomain_id": "11.1",
"subdomain_code": "edtech",
"subdomain_name": "Educational Technology"
} | tasks/education_info/homework_grading_numerical_pdes_instance_02 |
education_info/marc_remediation_folio_overlay | MARC Remediation FOLIO Overlay | Complete a Python starter project that remediates legacy MARC batches, applies cataloging policy rules, and prepares deterministic FOLIO Data Import overlay artifacts. | education_info | Library & Information Science | full-spectrum | You are working on a Linux VM to complete a MARC remediation and FOLIO overlay starter project.
## Task Directory
`base`
## Visible Inputs
- Full prompt: `base/input/TASK_PROMPT.md`
- Harness summary: `base/input/task_spec.md`
- Editable starter project: `base/input/starter_project`
- Visible public case: `base/input... | [
"Copy `input/starter_project/` to `output/submission/` before editing.",
"Preserve the CLI contract: `python scripts/remediate_catalog.py --case-dir <case_dir> --output-dir <output_dir>`.",
"Apply the visible cataloging policy uniformly to MARCXML/MRK records, FOLIO matches, authority mappings, and 949 holdings... | [
"Python"
] | [
{
"name": "TASK_PROMPT.md",
"format": "Markdown",
"path": "input/TASK_PROMPT.md",
"description": "Full agent-facing task prompt."
},
{
"name": "task_spec.md",
"format": "Markdown",
"path": "input/task_spec.md",
"description": "Harness-facing summary of the expected starter-copy w... | {
"domain_id": "11",
"domain_code": "education_info",
"subdomain_id": "11.2",
"subdomain_code": "library_info",
"subdomain_name": "Library & Information Science"
} | tasks/education_info/marc_remediation_folio_overlay |
education_info/moodle_gradebook_closeout_reconciliation | Moodle Gradebook Closeout Reconciliation | Repair a broken offline Moodle-style course backup and rebuild the registrar plus OneRoster closeout exports from the repaired bundle. | education_info | Educational Technology | full-spectrum | You are working on Linux on an offline Moodle gradebook closeout task.
## Task Root
- `base`
## Visible Inputs
- Task prompt: `base/input/TASK_PROMPT.md`
- Bundle README: `base/input/README.md`
- Shared helper library: `base/input/bundle_lib.py`
- Starter project: `base/input/starter_project`
- Starter backup: `base/... | [
"Read `input/TASK_PROMPT.md` and the starter materials under `input/starter_project/`.",
"Repair the offline Moodle-style backup `input/starter_project/course_backup.mbz` by editing only the benchmark-designated editable backup files.",
"Rebuild the registrar export and the OneRoster package with the staged hel... | [
"Python",
"pandas",
"uv"
] | [
{
"name": "TASK_PROMPT.md",
"format": "Markdown",
"path": "input/TASK_PROMPT.md",
"description": "Agent-visible benchmark contract for the offline Moodle gradebook closeout workflow."
},
{
"name": "README.md",
"format": "Markdown",
"path": "input/README.md",
"description": "Top-l... | {
"domain_id": "11",
"domain_code": "education_info",
"subdomain_id": "11.1",
"subdomain_code": "edtech",
"subdomain_name": "Educational Technology"
} | tasks/education_info/moodle_gradebook_closeout_reconciliation |
education_info/yi_manuscript_translation_1 | Yi Manuscript Translation 1 | Inspect a classical Yi manuscript image, identify which glyph a phonetic gloss refers to, localize that glyph, and write a cited translation report from bundled source texts. | education_info | Translation & Localization | last-exam | You are analyzing a classical Yi manuscript image on a Windows VM.
## Task Folder
`base`
## Visible Input Files
- Question: `base\input\question.txt`
- Manuscript image: `base\input\inscription.png`
- Bundled source texts: `base\input\reference_materials\`
## Software
- Image launcher: `base\software\launch_viewer.c... | [
"Read `input/question.txt` and inspect the staged `input/inscription.png` image via `software/launch_viewer.cmd`.",
"Determine which glyph the phonetic gloss refers to by using only the bundled `input/reference_materials/` files.",
"Write `output/bounding_box.json` with integer pixel coordinates for the target ... | [
"Windows Image Viewer",
"Text Editor"
] | [
{
"name": "Task brief",
"format": "`.txt`",
"path": "input/question.txt",
"description": "Explains the manuscript-inspection and reporting task."
},
{
"name": "Manuscript image",
"format": "`.png`",
"path": "input/inscription.png",
"description": "Classical Yi manuscript image co... | {
"domain_id": "11",
"domain_code": "education_info",
"subdomain_id": "11.3",
"subdomain_code": "translation",
"subdomain_name": "Translation & Localization"
} | tasks/education_info/yi_manuscript_translation_1 |
engineering/2d_drawings_to_3d_bridge_model | Bridge The Gap — Bridge + Site 3D Model | Extend an existing 3D site model (terrain + surrounding buildings) by adding a bridge designed in the supplied posters. Output must contain BOTH the unchanged site AND the new bridge geometry. The model is exported as model.obj plus model.3dm; an image-only multimodal LLM judge scores 12 rendered views (4 elevations + ... | engineering | Civil, Architectural & Geospatial Engineering | null | You are a BIM modeler using Rhino 8 to add a bridge design to an existing 3D site model.
This is NOT a build-from-scratch task. The site (terrain + surrounding buildings) already exists — you must place the bridge into it and preserve the rest of the site unchanged.
## Inputs
All input materials are under:
E:\agent... | [
"Read input/README.md and input/output_contract.json to understand the brief, the required filenames, and the coordinate convention.",
"Open Rhino 8 via the staged software/Rhino 8.lnk shortcut.",
"Open input/site_model/3D Model_Site.3dm in Rhino to inherit the world coordinate frame and the existing terrain + ... | [
"Rhino 8"
] | [
{
"name": "README.md",
"format": "Markdown",
"path": "input/README.md",
"description": "Natural-language task brief that mirrors the prompt and lists the required bridge components."
},
{
"name": "output_contract.json",
"format": "JSON",
"path": "input/output_contract.json",
"des... | {
"domain_id": "1",
"domain_code": "engineering",
"subdomain_id": "1.3",
"subdomain_code": "civil_geo",
"subdomain_name": "Civil, Architectural & Geospatial Engineering"
} | tasks/engineering/2d_drawings_to_3d_bridge_model |
engineering/2d_drawings_to_3d_building_model | Betonwerk Katzenberger 3D Model | Build a 3D architectural model of the Betonwerk Katzenberger building (Bavarian concrete plant — workshop Hall + residential Tower) in Rhino 8 by reading six PDF drawings, a 3D snapshot, and a footprint-only positioning anchor. Export as model.obj plus model.3dm; an image-only multimodal LLM judge scores 14 rendered vi... | engineering | Civil, Architectural & Geospatial Engineering | last-exam | You are a BIM modeler using Rhino 8 to build the Betonwerk Katzenberger building in 3D, given the architectural drawings of the project.
The building is a Bavarian concrete-plant complex with two main parts:
- a wide, low workshop Hall housing three prefabricated workshop modules
- a tall narrow residential Tower ... | [
"Read input/README.md and input/output_contract.json to understand the brief, the required filenames, and the coordinate convention.",
"Open Rhino 8 via the staged software/Rhino 8.lnk shortcut.",
"Open the 6 PDF drawings in input/architectural_drawings/ and the base_model.3dm to study form and lock the world c... | [
"Rhino 8"
] | [
{
"name": "README.md",
"format": "Markdown",
"path": "input/README.md",
"description": "Natural-language task brief that mirrors the prompt and lists the required components."
},
{
"name": "output_contract.json",
"format": "JSON",
"path": "input/output_contract.json",
"descriptio... | {
"domain_id": "1",
"domain_code": "engineering",
"subdomain_id": "1.3",
"subdomain_code": "civil_geo",
"subdomain_name": "Civil, Architectural & Geospatial Engineering"
} | tasks/engineering/2d_drawings_to_3d_building_model |
engineering/Analog_Active | Analog Active | Design and simulate the required analog circuit in LTspice from the staged specification and deliver the completed schematic output. | engineering | Electronics Engineering | full-spectrum | You are given an LTspice schematic with the LTM4648 µModule DC/DC buck regulator IC already placed on the canvas. Your task is to complete the circuit design by adding all required external components, wiring them to the correct IC pins, and running a transient simulation to verify the output.
... | [
"Open the staged LTspice starter schematic for the active variant.",
"Read the design specification and LTM4648 datasheet to choose the required external component values and wiring.",
"Complete the buck-converter schematic, add a `1 ms` startup transient analysis, save the updated `circuit.asc`, and run simula... | [
"Python"
] | [
{
"name": "design_spec.txt",
"format": "text",
"path": "input/design_spec.txt",
"description": "Active-variant targets for VIN, VOUT, load current, ripple, and startup behavior"
},
{
"name": "ltm4648_datasheet.pdf",
"format": "PDF",
"path": "input/ltm4648_datasheet.pdf",
"descrip... | {
"domain_id": "1",
"domain_code": "engineering",
"subdomain_id": "1.2",
"subdomain_code": "electronics",
"subdomain_name": "Electronics Engineering"
} | tasks/engineering/Analog_Active |
engineering/abb_irb6700_asset_to_urdf_instance_1 | ABB IRB6700 Asset To URDF | Reconstruct a complete ABB IRB6700 robot URDF from staged mesh assets and structural metadata. The agent must preserve the required links, joints, limits, mimic behavior, and auxiliary frames while producing exactly one valid `submission.urdf` file. | engineering | Robotics & Autonomous Systems | near-term | You are reconstructing a robot URDF on a Linux VM.
## Variant
`base`: ABB IRB6700 URDF reconstruction
## Input Files
- Mesh assets: `base/input/meshes`
- Structural metadata: `base/input/metadata`
- Task brief: `base/input/task_brief.md`
## What You Must Do
1. Read `base/input/task_brief.md`.
2. Use the mesh assets ... | [
"1. Read the staged task brief and inspect the ABB IRB6700 meshes plus metadata.",
"2. Reconstruct a valid URDF with the required links, joints, limits, mimic behavior, and auxiliary frames.",
"3. Save exactly one final file named `submission.urdf` under `output/`."
] | [
"Python"
] | [
{
"name": "meshes/",
"format": "stl directory",
"path": "input/meshes/",
"description": "Visual and collision STL meshes for the ABB IRB6700 links."
},
{
"name": "link_manifest.json",
"format": "json",
"path": "input/metadata/link_manifest.json",
"description": "Lists the require... | {
"domain_id": "1",
"domain_code": "engineering",
"subdomain_id": "1.11",
"subdomain_code": "robotics",
"subdomain_name": "Robotics & Autonomous Systems"
} | tasks/engineering/abb_irb6700_asset_to_urdf_instance_1 |
engineering/aerospace_low_thrust_trajectory | Low-Thrust LEO To GEO Trajectory | Compute a three-tier orbital transfer from inclined low Earth orbit to geostationary orbit using analytic, numerical low-thrust, and optimal-control methods. | engineering | Aerospace & Mechanical Engineering | last-exam | You are working on a Linux VM.
Task root:
- `base`
Visible files:
- Problem specification: `base/input/problem_spec.md`
- Task prompt: `base/input/task_prompt.md`
- Output contract: `base/input/output_contract.json`
- Python runtime manifest: `base/input/runtime_env/pyproject.toml`
- Task-scoped Python entry point: `... | [
"1. Compute the Tier 1 Hohmann transfer quantities.",
"2. Integrate the Tier 2 continuous tangential low-thrust spiral and save the required trajectory array.",
"3. Solve the Tier 3 fixed-time low-thrust transfer with inclination change and save the required trajectory and control arrays.",
"4. Write `output/... | [
"Python",
"uv",
"NumPy",
"SciPy"
] | [
{
"name": "problem_spec.md",
"format": "md",
"path": "input/problem_spec.md",
"description": "Full low-thrust LEO-to-GEO problem statement with formulas, state definitions, and output requirements."
},
{
"name": "task_prompt.md",
"format": "md",
"path": "input/task_prompt.md",
"d... | {
"domain_id": "1",
"domain_code": "engineering",
"subdomain_id": "1.1",
"subdomain_code": "aerospace_mech",
"subdomain_name": "Aerospace & Mechanical Engineering"
} | tasks/engineering/aerospace_low_thrust_trajectory |
End of preview. Expand in Data Studio
Agents Last Exam — Task Card Metadata (v1.0)
A metadata-only release (v1.0) of 153 tasks from the Agents Last Exam (ALE) benchmark for evaluating computer-use agents on long-horizon professional work.
The Agents Last Exam dataset family
ALE is published as three companion HuggingFace datasets:
| Dataset | Contents | Access |
|---|---|---|
| Task Card Metadata | One row per task: titles, prompts, taxonomy, input-file descriptors | Open |
| Task Input Data | The input/ files each task hands the agent at run start |
Open |
| Reference (Ground-Truth) Data | The reference/ outputs used to score runs |
⚠️ Gated (manual approval) |
You are viewing: Task Card Metadata.
- Source repository: https://github.com/rdi-berkeley/agents-last-exam
- Official site: https://agents-last-exam.org
This dataset contains task cards only (one row per task). No input data, software binaries, reference outputs, scoring fixtures, or VM images are included here. Use the HuggingFace Dataset Viewer above to browse, filter, and search the table; fetch the input files from the companion data repo and the ground-truth from the gated reference repo.
What's Included
| Column | Type | Notes |
|---|---|---|
task_id |
string | Stable identifier, <category>/<task_name>. |
title |
string | Human-readable task title. |
summary |
string | One-paragraph summary. |
category |
string | Top-level domain (e.g. business_finance, health_medicine). |
subdomain |
string | Subdomain name (e.g. Quantitative Finance & Trading). |
task_split |
string | One of near-term, last-exam, full-spectrum. |
task_prompt |
string | Full task instruction the agent receives. |
agent_must_do |
array | Required action checklist. |
software |
array | Software the task expects the VM to provide. |
input_files |
array | Descriptors for staged input files: name, format, path, description. No file contents. |
taxonomy |
struct | Full taxonomy: domain_id, domain_code, subdomain_id, subdomain_code, subdomain_name. |
source_repo_path |
string | Path inside the source repository for this task. |
What's Excluded
- Input data, reference outputs, software binaries, VM images.
- Private evaluation logic (
evaluation), reference file lists (referenceFiles), VM provisioning details (vm). - Internal audit trails (the classifier model run, submission IDs, internal file paths).
- Downloads last month
- 32