earakely-scale/LocPhotosEnv-REA

loc-photos

loc-photos

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

Tables

• loc-photos_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• loc-photos_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/loc-photos.jsonl holds each run's record and native trajectory, and bundles/loc-photos/ the bundle itself.

Run it

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

Library of Congress JSON API (photos) on AgentEnv, from a REA capture

What this is. A replay environment of Library of Congress JSON API (photos) v1 (U.S. public domain (Library of Congress metadata); every item read is marked "No known restrictions on publication" or "on reproduction"), built from one recorded browser session against the public API at https://www.loc.gov on 2026-10-19. The responses are real public-domain records published by the Library of Congress, frozen as they were served then. REA projected the capture into an environment manifest (11 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 Library of Congress JSON API (photos). Tasks are graded deterministically, with no judge.

What's in it

• 11 tools (11 GET) over 29 recorded exchanges on https://www.loc.gov; no recorded group excluded.
• Path parameters: 1 numeric. 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 11 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/loc-photos-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_2395092c17034b218c3faa02d0d1a781661b12c9dcb93654e756a6de5074a1a6 (a capture_browser_scenario session REA recorded itself; capture file sha256 2c73066562964d681ad285268eb259c311971da57b33b2b80199d7f06977ca9f, 817,776 bytes), manifest view ev_63d6f1ccb521b43d066ba25cf8d51f2bc63d3c570ef44e3d2fd69d598caa1f03, confidence inferred, authority controlled-replay. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975, #4999 and #5182; 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_item_by_id	get_item_by_id	id=2002696569, fo=json, at=item	item.title	[Castle, Corfe Castle, England]
qa-02-get_photos	get_photos	q=lighthouse, fo=json, c=5, at=results,pagination	results[0].item.title	Lighthouse, Biloxi, Mississippi
qa-03-get_collections_photochrom_prints	get_collections_photochrom_prints	q=castle, fo=json, c=6, at=results	results[0].resources[0].caption	digital file from original
qa-04-get_collections_historic_american_buildi	get_collections_historic_american_buildings_landscapes_and_e	q=covered bridge, fo=json, c=3, at=results	results[0].resources[0].caption	Drawings from Survey HAER IN-33
qa-05-get_collections_stereograph_cards	get_collections_stereograph_cards	q=bridge, fo=json, c=3, at=results	results[0].title	East River Bridge
qa-06-get_maps	get_maps	q=grand canyon, fo=json, c=3, at=results	results[0].title	Grand Canyon National Park, Arizona
qa-07-get_collections_detroit_publishing_compa	get_collections_detroit_publishing_company	q=bridge, fo=json, c=6, at=results	results[0].item.title	High Bridge station, High Bridge, Ky.
qa-08-get_collections_panoramic_photographs	get_collections_panoramic_photographs	q=canyon, fo=json, c=5, at=results	results[0].item.title	Fish Creek Canyon
qa-09-get_item_in0214	get_item_in0214	fo=json, at=item	item.place[0].title	Indiana -- Ripley County -- Versailles
qa-10-get_item_by_id	get_item_by_id	id=2002696569, fo=json, at=item	item.place[0].title	England--Corfe Castle
qa-11-get_photos	get_photos	q=lighthouse, fo=json, c=5, at=results,pagination	results[0].item.summary	Photograph showing lighthouse and pier.
qa-12-get_collections_photochrom_prints	get_collections_photochrom_prints	q=castle, fo=json, c=6, at=results	results[0].item.rights_advisory	No known restrictions on reproduction.

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	11	92%	1.0	8,117
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	9,540

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

openai/gpt-oss-120b:groq missed qa-04-get_collections_historic_american_buildi. Miss by gpt-oss-120b:groq on qa-04-get_collections_historic_american_buildi: 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/LocPhotosEnv-REA --bundle loc-photos --task smoke          # Docker, no model
agent-env hf run earakely-scale/LocPhotosEnv-REA --bundle loc-photos --eval loc-photos-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 opened the public API's URLs in a browser REA launched, within the publisher's fair-access rules (28 navigations in one browser session (plus one server redirect), 900 ms wait between navigations (about one request per 1.3 s), no API key, no login; research before the capture used about 35 single requests at 1.2 s spacing).
2. REA's capture-browser-scenario visited each API URL (goto) and recorded every network transaction (Real public records read from the Library of Congress JSON API: 15 items by id (Photochrom prints of Corfe, Dover and Threave castles, the Rhone, Illecillewaet and Bossons glaciers, Glacier Point, a bridge in Glenariff; Detroit Publishing Company views of High Bridge over the Kentucky River, the Natural Bridge Hotel and Rascon Bridge; panoramic photographs of the Grand Canyon and Fish Creek Canyon; the Busching Covered Bridge survey record), every one marked no known restrictions on publication or reproduction, plus 9 searches or collection listings, 2 file-resource lists and 1 resource record); the environment-manifest view read that capture directly, with no HAR or scrubbing step.
3. 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	Library of Congress JSON API (photos) (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)	11 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 Library of Congress, 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://www.loc.gov, published by the Library of Congress (U.S. public domain (Library of Congress metadata); every item read is marked "No known restrictions on publication" or "on reproduction"); the dataset repeats them as recorded. REA is MIT software by morluto; AgentEnv Framework is Apache-2.0 by Scale AI.