feat: add twelve environmental observation request workflows

[yangziao56]

Summary

Completes the requested environmental-observation category in this same PR: two reviewed pilots plus ten additions, 12 public-data request workflows across NOAA, USGS, NDBC, NASA POWER, Open-Meteo and EPA. This addresses the request to merge the rest of the tasks together. Released V1/V2 and framework code are unchanged.

The evaluated outcome is initiation of the correctly configured data request, not final file saving or downstream analysis. Exact station/series, dates, units, product and format constraints remain. The interceptor credits the first qualifying request; it does not penalize unrelated earlier requests. No site API-key values or private user data are included.

Validation completed

• 808/808 offline checks: 12 schema, 49 must-accept URL, 697 must-reject URL, 50 supplied-data checks, across all 12 tasks. Nine tasks have data identity/shape checks; NDBC/Open-Meteo/EPA response data is not covered by that part.
• 12/12 official-base-runtime website UI paths verified: each produced a matching interception record. Codex independently rechecked all captured schemas and URLs against the final predicates, plus saved interception hashes for the ten additions.
• Runtime pin: 9dd9d442c8ec720ba4d6d4c95f2bbc0ff2a2d0df; its server source is unchanged at reviewed upstream 187cd252bc60af8ac3a2c98a87c9316e49a5ac75.
• Independent Claude Opus 5.5 review of the earlier draft identified instruction/evaluator and validation-coverage issues; the final corrections were verified by Codex. Older runs with subsequently clarified instructions are explicitly distinguished from identical full-task-hash runs in the README.
• JSON schema and whitespace checks pass. CHANGELOG has an Unreleased entry.

Important corrections and failure history

• Earthquake magnitude now accepts equivalent 3, 3.0, and 3.00 without accepting other, missing or duplicate values. The actual UI run typed 3.0, which the site serialized as 3.
• NASA must leave optional custom site elevation unset. A direct paired official-API check showed site-elevation=0.000 adds a third corrected-pressure series, so the matcher remains strict. NASA attempts 1–2 timed out at 600s; attempt 3 was manually stopped after requests included that unwanted extra field and did not intercept. Attempt 4 passed without it. Only NASA's timeout changed to 30 minutes; the other 11 remain 10 minutes.
• Open-Meteo's first attempt timed out during operator input; the second passed within its unchanged 10-minute limit. All unsuccessful attempts are retained.
• The official USGS continuous run produced the requested EDT date window; this is an observed container result, not a universal browser-timezone claim. The temperature UI and data confirm the non-discontinued multiparameter-sonde Mean series.
• The already-addressed limit/f review thread is not duplicated: omitting the former truncates the default page; omitting the latter changes CSV to JSON. The existing constraints remain.

Final two addition runs (2026-09-24 UTC):

Task	Runtime finished	Interception artifact SHA-256
USGS earthquake	23:32:38, eval_matched	8d5451299531b16456e0397e1197354118a4b91ca32c3c269bab0eb0caa49b3b
NASA POWER	23:42:14, eval_matched	ace84e08c5f799918282d9a86dc03ecea76d8875c3d89540d9448dacee0ead49

Raw request/action/video artifacts are retained locally, not uploaded in this task-only diff.

Explicit boundaries and assistance

These were Codex-operated browser checks in the official base image, not Ziao's personal human run or timing study. Human completion in under one minute is not measured. The full runner CLI, Stage 2 model judge, saved-file success and paid frontier-model performance were not evaluated. Request-initiation evidence is not a claim of full benchmark acceptance.

Implementation and validation used Codex for Ziao Yang, with the earlier independent Claude Code Opus 5.5 review disclosed above. Only public workflows and data sources were used. Category acceptance and any individual authorship invitation for a concrete follow-up paper remain unconfirmed.

[Perry2004]

The two tasks LGTM. Please also add the rest of the tasks to this PR as well so that they can be merged together.

[yangziao56]

Added the remaining ten workflows to this same PR in 2319e25, so the category now contains all 12 tasks requested in your review. Thank you for asking for the complete category together.

Final local checks pass 808/808 (schema, accepted/rejected request variants and supplied-data checks). All 12 have matching request records from actual website UI paths in the pinned official base runtime. I independently rechecked the recorded schemas/URLs against the final predicates; the README and updated PR description distinguish instruction-only changes from identical-task-hash runs and retain failed attempts.

The final checks also caught NASA's optional advanced elevation field adding an unwanted pressure series; the instructions now leave it unset, and the matcher was not loosened. NASA alone now has a 30-minute watchdog; other tasks remain at 10 minutes. Earthquake magnitude handling accepts equivalent 3/3.0/3.00 values, and the real UI run passed.

These are disclosed Codex-operated UI checks, not Ziao's personal human timing study, full runner/Stage 2 results, saved-file verification or paid model trials. The new-head CI is still pending. The limit/f explanation remains in the existing review thread; no duplicate run request is being made.

[Perry2004]

LGTM🫡
Thanks for the contribution.