earakely-scale/TreasuryFiscalDataEnv-REA

treasury-fiscal-data

treasury-fiscal-data

An AgentEnv bundle with its recorded runs, published with agentenv-hf.

Tables

• treasury-fiscal-data_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• treasury-fiscal-data_episodes: one row per recorded run, with its status, model, reward and scores, the prompt and response, and the agent's transcript as chat messages when its trajectory format is one agentenv-hf reads.

raw/treasury-fiscal-data.jsonl holds each run's record and native trajectory, and bundles/treasury-fiscal-data/ the bundle itself.

Run it

pip install agentenv-framework huggingface_hub
hf download earakely-scale/TreasuryFiscalDataEnv-REA --repo-type dataset --local-dir agentenv-dataset
agent-env run agentenv-dataset/bundles/treasury-fiscal-data

Treasury Fiscal Data API on AgentEnv, from a REA capture

What this is. A replay environment of Treasury Fiscal Data API v1/v2 (U.S. public domain), built from one recorded browser session against the public API at https://api.fiscaldata.treasury.gov on 2026-10-19. The responses are real public-domain records published by the U.S. Department of the Treasury, Bureau of the Fiscal Service, frozen as they were served then. REA projected the capture into an environment manifest (16 tools), and rea environment-export wrote the bundle you can run from this dataset. Every tool call replays a retained record from the capture, cites the record and the Evidence it came from, and nothing executes Treasury Fiscal Data API. Tasks are graded deterministically, with no judge.

What's in it

• 16 tools (16 GET) over 52 recorded exchanges on https://api.fiscaldata.treasury.gov; no recorded group excluded.
• Path parameters: none. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the three opt-in rules (after-uid-literal, opaque-token, sibling-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 16 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/treasury-fiscal-data-retrieval.toml): each asks one factual question about a recorded record, with a distinct answer, answerable through a recorded exchange; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks replay through the MCP endpoint.
• 24 recorded episodes by openai/gpt-oss-120b:groq, zai-org/GLM-5.3-Flash:zai-org through the Hugging Face router, with their chat transcripts.
• Provenance: capture ev_522a428f32046d80249b50c984b3b677afe911d6009c0c3b380de67e789e6496 (scoped HAR sha256 3e41c0dfb7af43dcc87399c0976cc7bbc3ed6111e22f02e4f610cfee0f8b1413, 211,445 bytes), manifest view ev_c068c4ebf39cd85dc7535400efbe9bce560e5b02677b83e8c2fa18df8157f193, confidence derived, authority historical-reference. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975 and #4999; the bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What replay means, and does not

A tool call returns the recorded response whose path and query values best match the arguments, with replay.exact_match telling you whether it matched exactly or fell back to the closest record. Reads behave like the API did on 2026-10-19. The live API keeps changing, so every answer is about the recorded snapshot, not today, and a query the session never made falls back to the closest record. The API is read-only and so is the environment: it has no state and no write tools. The manifest's unknowns say what replay cannot know.

Tasks and grading

task	tool	arguments	field	answer
qa-01-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_auctions_query	filter=auction_date:eq:2026-10-22, fields=cusip,security_type,security_term,auction_date,issue_date,maturity_date,offering_amt,auction_format,high_yield,int_rate, page[size]=5	meta.labels.int_rate	Interest Rate
qa-02-get_services_fiscal_service_debt_tror	get_services_fiscal_service_debt_tror	filter=record_date:eq:2026-06-30,agency_name:eq:Department of Education, fields=agency_name,receivable_type_description,funding_type_description,beginning_fiscal_year_amt,new_receivables_amt,collections_total_amt,record_date, page[size]=5	data[0].receivable_type_description	Administrative Receivables
qa-03-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_avg_interest_rates	filter=record_date:eq:2026-09-30,src_line_nbr:lte:4, page[size]=5	data[0].security_desc	Treasury Bills
qa-04-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_debt_to_penny	filter=record_date:gte:2026-10-09,record_date:lte:2026-10-15, sort=-record_date, page[size]=5	meta.labels.debt_held_public_amt	Debt Held by the Public
qa-05-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_title_xii	filter=record_date:eq:2026-10-15, page[size]=5	meta.labels.state_nm	State Name
qa-06-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_interest_expense	filter=record_date:eq:2026-09-30,src_line_nbr:lte:3, page[size]=5	meta.labels.month_expense_amt	Current Month Expense Amount
qa-07-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_federal_maturity_r	filter=record_date:gte:2025-09-30, sort=-record_date, page[size]=5	meta.labels.one_year_or_less	1 Year or Less
qa-08-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_rates_of_exchange	filter=record_date:eq:2026-09-30,country:in:(Japan,Norway,Switzerland), fields=country_currency_desc,exchange_rate,effective_date,record_date, page[size]=5	data[0].country_currency_desc	Japan-Yen
qa-09-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_statement_net_cost	filter=record_date:eq:2025-09-30,src_line_nbr:lte:5, page[size]=5	meta.labels.agency_nm	Agency Name
qa-10-get_services_fiscal_service_accounting_o	get_services_fiscal_service_accounting_od_debt_outstanding	filter=record_date:gte:2022-09-30, page[size]=5	meta.labels.debt_outstanding_amt	Debt Outstanding Amount
qa-11-get_services_fiscal_service_accounting_d	get_services_fiscal_service_accounting_dts_operating_cash_ba	filter=record_date:eq:2026-10-15,src_line_nbr:lte:3, page[size]=5	meta.labels.sub_table_name	Sub Table Name
qa-12-get_services_fiscal_service_accounting_d	get_services_fiscal_service_accounting_dts_deposits_withdraw	filter=record_date:eq:2026-10-15,transaction_type:eq:Deposits,src_line_nbr:lte:4, page[size]=5	data[0].transaction_catg	Dept of Agriculture (USDA) - misc

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	8	67%	1.0	6,401
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	7,789

Every episode is in the episodes config with its transcript; rewards are the deterministic verifier scores.

openai/gpt-oss-120b:groq missed qa-05-get_services_fiscal_service_accounting_o, qa-06-get_services_fiscal_service_accounting_o, qa-09-get_services_fiscal_service_accounting_o, qa-10-get_services_fiscal_service_accounting_o. Misses by gpt-oss-120b:groq on qa-05-get_services_fiscal_service_accounting_o, gpt-oss-120b:groq on qa-06-get_services_fiscal_service_accounting_o, gpt-oss-120b:groq on qa-09-get_services_fiscal_service_accounting_o, gpt-oss-120b:groq on qa-10-get_services_fiscal_service_accounting_o: the model's answer did not contain the expected value.

Run it yourself

uv tool install agentenv-framework --with 'agentenv-hf @ git+https://github.com/earakely-scale/agentenv-hf-plugin@v0.3.0'
agent-env hf run earakely-scale/TreasuryFiscalDataEnv-REA --bundle treasury-fiscal-data --task smoke          # Docker, no model
agent-env hf run earakely-scale/TreasuryFiscalDataEnv-REA --bundle treasury-fiscal-data --eval treasury-fiscal-data-retrieval --model openai/gpt-oss-120b:groq

The retrieval eval needs a model endpoint: set [model] base_url and api_key in .agentenv/config.toml (the Hugging Face router works with any hf_ token). The bundle's agents/replay-qa is a small OpenAI-compatible tool-calling agent; any A2A agent can replace it.

How it was made

1. No instance was run: a throwaway VM called the public API directly, within the publisher's fair-access rules (a declared User-Agent, one request at a time and at most 2 requests per second (52 capture requests, no key)).
2. Playwright opened the API origin and ran a scripted sequence of read requests (exchange rates for ten currencies, average interest rates on Treasury securities, Treasury-owned gold by facility, interest expense, Title XII advances for two states, agency net cost, Daily Treasury Statement cash and agency deposits/withdrawals, five sets of securities auctions, revenue collections, Monthly Treasury Statement outlays for three departments, savings bond sales, maturity rates, debt to the penny and receivables for two departments); the HAR was scoped to the API, with headers and cookies removed.
3. rea inspect-web-network-capture retained the exchanges as Evidence; rea inspect-analysis-view projected the environment-manifest view; rea environment-export wrote this bundle; gen_tasks.py derived the retrieval tasks from the manifest and the recorded bodies; agent-env run recorded the episodes; agent-env hf publish wrote this dataset. Scripts: https://github.com/earakely-scale/rea-environments.

Compared with PortSimEnv on AgentEnv

	PortSimEnv-AgentEnv	Treasury Fiscal Data API (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	A replay of a real application's API, captured from one session
Observation	Tool results computed by the simulator as the week unfolds	Recorded responses, cited to Evidence
Action space	Planning tools (check_plan, confirm_berths, advance, message_agent)	16 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	None: every call is independent; the API is read-only
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic, recorded-value match; 0 or 1 per task
What it measures	Planning under uncertainty and negotiation	Tool discovery, parameter inference and faithful retrieval in a real API
Data provenance	Real 2024 port calls and weather; simulated workloads	Real public-domain records from the U.S. Department of the Treasury, Bureau of the Fiscal Service, as served on 2026-10-19
How it was built	Hand-written plugin, months of work	Generated from a capture in one scripted pass
Where it is strong	Long-horizon decisions with hard rewards	Breadth: any app you can capture becomes an environment

The two are complementary: PortSimEnv shows what a hand-built simulation can grade; this dataset shows how far a capture alone gets you, and the manifest's unknowns list exactly what a stateful version would still have to specify.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses of the public API at https://api.fiscaldata.treasury.gov, published by the U.S. Department of the Treasury, Bureau of the Fiscal Service (U.S. public domain); the dataset repeats them as recorded. REA is MIT software by morluto; AgentEnv Framework is Apache-2.0 by Scale AI.