earakely-scale/OpenFdaEnv-REA

openfda

openfda

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

Tables

• openfda_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• openfda_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/openfda.jsonl holds each run's record and native trajectory, and bundles/openfda/ the bundle itself.

Run it

pip install agentenv-framework huggingface_hub
hf download earakely-scale/OpenFdaEnv-REA --repo-type dataset --local-dir agentenv-dataset
agent-env run agentenv-dataset/bundles/openfda

openFDA on AgentEnv, from a REA capture

What this is. An environment of openFDA v1 (U.S. public domain), built from one recorded browser session against the public API at https://api.fda.gov on 2026-10-19. The responses are real public-domain records published by the U.S. Food and Drug Administration, frozen as they were served then. REA projected the capture into an environment manifest (9 tools) and an environment state, the entities the capture's reads showed, and rea environment-export wrote the bundle you can run from this dataset. Reads the state can render are answered from it, the API offers no writes, and data/reset restores it. Nothing executes openFDA. Tasks are graded deterministically, with no judge.

v0.2.0 rebuilds this environment from the same capture with a REA build that serves a state, confirmed writes and list queries; the tag v0.1.0 keeps the earlier replay-only version.

What's in it

• 9 tools (9 GET) over 53 recorded exchanges on https://api.fda.gov; no recorded group excluded. 9 are offered.
• Path parameters: none. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the four opt-in rules (after-uid-literal, opaque-token, sibling-variance, prefix-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 9 tools.
• State: 7 entities of 1 types. 1 read tools answer from it and 8 replay their recorded responses. 1 list tool also answers filters, search, sorting or paging from it. No write tool is served.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/openfda-retrieval.toml): each asks one factual question about a recorded record, with a distinct answer that the app does not ship by default; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks it through the MCP endpoint.
• 24 retrieval episodes, plus the smoke run, 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_831755436c6fab21a1196cc47650cc102eb756bc8286b8c8528cda928dcfaf90 (scoped HAR sha256 2d19831b1c27073e37639eb593dceb429dc8190cdd612c767e07fc0262671568, 347,854 bytes), manifest view ev_1738db5785f791947ee11c57afe5335a5b28db3ee60299b01f8764339048e2f8, confidence derived, authority historical-reference. The REA build is rea-agents 6.7.0 from main 335c042d with open pull requests merged: morluto/rea#5600 at fe8ab584 (which carries #5351 and #5352), #5664 at 03b66df9, #5787 at edc09dba, #5800 at 6db41d29 and #5699 at ac3ca8a3. The bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What the state means, and does not

Reads the state can render, item, list and document views of the entities the capture's responses showed, answer from it. List reads filtered, searched, sorted or paged answer from it too wherever the capture's recorded calls explain what each query parameter does, including values no call recorded; a parameter the recordings do not explain, or a flag's unrecorded value, replays. Other reads replay the recorded response whose path and query values best match the arguments, with replay.exact_match saying whether it matched exactly; a replayed read does not change after a write. The API is read-only, so the environment offers no write tool; data/add can still load a state for a task. The data plane at /agentenv takes data/reset (restore the extracted state), data/add and data/get (return it), which is how write tasks are graded.

Not modelled: side effects a write has beyond its own entity (counters, copies in other collections, related entities), preconditions the application enforces before a write, list query parameters the recorded calls do not explain, paging members such as page counts and next links when a call asks for another page size or page, and anything the capture never showed. The manifest's unknowns list what a capture cannot establish.

Tasks and grading

Retrieval

task	tool	arguments	field	answer
qa-01-get_device_enforcement	get_device_enforcement	search=classification:"Class I", sort=report_date:desc, limit=2	results[0].product_quantity	771,127 total
qa-02-get_drug_shortages	get_drug_shortages	search=status:"Current", limit=2	results[0].generic_name	Lidocaine Hydrochloride Injection
qa-03-get_food_enforcement	get_food_enforcement	search=state:"CO" AND report_date:[20250101 TO 20261231], limit=2	results[0].reason_for_recall	Package integrity compromised.
qa-04-get_drug_enforcement	get_drug_enforcement	search=classification:"Class I" AND status:"Ongoing", limit=3	results[0].address_1	6555 State Highway 161
qa-05-get_device_classification	get_device_classification	search=device_name:"stethoscope", limit=2	results[0].device_name	Stethoscope, Esophageal
qa-06-get_other_nsde	get_other_nsde	search=proprietary_name:"Advil", limit=2	results[0].application_number_or_citation	NDA018989
qa-07-get_drug_drugsfda	get_drug_drugsfda	search=sponsor_name:"PFIZER", limit=1	results[0].application_number	NDA019617
qa-08-get_device_recall	get_device_recall	search=root_cause_description:"Software design", sort=event_date_initiated:desc, limit=1	results[0].openfda.medical_specialty_description	General, Plastic Surgery
qa-09-get_drug_ndc	get_drug_ndc	search=brand_name:"Tylenol", limit=2	results[0].labeler_name	Kenvue Brands LLC
qa-10-get_device_enforcement	get_device_enforcement	search=classification:"Class I", sort=report_date:desc, limit=2	results[0].recalling_firm	Medline Industries, LP
qa-11-get_drug_shortages	get_drug_shortages	search=status:"Current", limit=2	results[0].company_name	Hospira, Inc., a Pfizer Company
qa-12-get_food_enforcement	get_food_enforcement	search=state:"CO" AND report_date:[20250101 TO 20261231], limit=2	results[0].code_info	BEST BY 14 AUG 2026 and BEST BY 15 AUG 2026

Episodes

model	retrieval episodes	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	6,617
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	7,583

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

Every recorded episode passed.

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/OpenFdaEnv-REA --bundle openfda --task smoke          # Docker, no model
agent-env hf run earakely-scale/OpenFdaEnv-REA --bundle openfda --eval openfda-retrieval --model openai/gpt-oss-120b:groq

The evals need 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, no API key, one request at a time and at most 2 requests per second).
2. Playwright opened the API origin and ran a scripted sequence of read requests (drug recalls, food recalls in four inland states, device recalls and Class I device enforcement reports, NDC listings for Kenvue and ibuprofen tablets, two drug applications, drug shortages, NDC SPL data elements for Advil and stethoscope device classes, each re-read by its public key); 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 and the state it serves; 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	openFDA (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	An application's API rebuilt from one captured session
Observation	Tool results computed by the simulator as the week unfolds	Answers rendered from the extracted state, or recorded responses, cited to Evidence
Action space	Planning tools (check_plan, confirm_berths, advance, message_agent)	9 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	A state extracted from the capture (7 entities); read-only; data/reset restores it
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic: recorded-value match for retrieval; 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. Food and Drug Administration, 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 and the state's limits above say what it still leaves out.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses of the public API at https://api.fda.gov, published by the U.S. Food and Drug Administration (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.