Review: Adopt-a-Pet mirror + task verifiers (site by @sarendis56, verifiers by reviewer)

[JeremyJC67]

Summary

Review vehicle for #94 (adopt_a_pet, site by @sarendis56). Carries the contributor's commit unchanged (8e11e7d7, authorship preserved) plus reviewer commits that add the grading contract and the two mechanical fixes the review found. The full evidence-backed review is posted on #94; this PR is what should merge once the contributor repacks the HF tarball.

What the reviewer commits add

• sites/adopt_a_pet/verify/ — verify_0.py … verify_19.py, shared verify_lib.py (package validation, loopback origin + port check, navigation gates with filter/page params, negation-aware answer matchers, SQLite snapshot contract with exact row deltas, werkzeug-scrypt password check with stdlib only), ground_truth.py (frozen catalog; a drifted seed fails every verifier closed), README.md (per-task contract table).
• sites/adopt_a_pet/tasks.jsonl — every row now carries verifier_path + an English judge_rubric (rules and fact-checkpoints, no answers; no answer key). Three questions were re-anchored because they were not gradable as written: T1 ("find the kitten" matched two kittens → "the kitten whose listing is in Scottsdale itself"), T2 (the correct answer flipped between Phoenix, AZ = 7 dogs / Batman $165 and Phoenix = 2 dogs / Horus $210 → the query is now quoted and every results page must be checked), T19 ("young dogs near Arizona" returned exactly one dog, so the "lowest fee among those good with children" comparison was vacuous → all Arizona dogs, good with children, lowest fee: 8 profiles on 2 pages).
• sites/adopt_a_pet/verify/tests/ — 272-test harness (synthetic snapshots + agent.py-shaped trajectories, every verifier run as a subprocess; dockerignored via the existing sites/*/verify/tests/ rule).
• sites/adopt_a_pet/requirements.txt — was Flask only; import app failed on flask_sqlalchemy outside the image. Pinned to the image's versions.
• Docs/tests sweep to 25 sites / 40000-40024 (README, AGENTS, CONTRIBUTING, CLAUDE, agent_demo/README, sites/walmart_careers/README.md, sites/walmart_careers/tests/test_integration.py, sites/rotten_tomatoes/tests/test_environment_quality.py) — the same sweep feat(webmd_doctor): add WebMD Doctor mirror (site 25, port 40024) #95 did; the doc-range test now derives the range from control_server.SITES.

Validation

Check	Result
Byte-identical seed (standalone, 4 boots/resets)	376b1ca1… every time
20 tasks driven through the UI with Playwright, run dirs in the agent.py shape	20/20 PASS
no-op / shortcut / wrong-answer variants (×20) + state-mismatch (×6)	66/66 FAIL, unexpected: none
agent_demo/eval_judge.py --verifier True	PASS on the genuine T0 run, FAIL (visited_dog_search_near_phoenix) on its shortcut
uv run python -m unittest discover -s sites/adopt_a_pet/verify/tests	272 tests OK
Registry / docs tests (walmart_careers, rotten_tomatoes) on this branch	green
Docker (bundle extracted by hand because its layout is rejected)	build ok (4.69 GB); 25/25 sites 200; /health 25/25; POST /reset/adopt_a_pet byte-identical to the build-generated seed (993d6113…) before/after a dirty request and after docker restart; /reset-all 25/25 in 1.46 s; no-op verifier matrix in the container 20/20 FAIL; registry tests green; site suite 272 passed

Still required from the contributor (see the review on #94)

1. HF tarball layout — members must sit under adopt_a_pet/ (currently static/images/…, instance_seed/… at the root plus 25 ._* AppleDouble members); scripts/validate_asset_archive.py rejects it, so fetch_assets.sh and a fresh-clone Docker build fail. Repack, push to HF Add GOV.UK mirror + task verifiers (site by @lamawmouk, verifiers by reviewer) #67, then pin .assets-revision to the merged SHA.
2. Photos vs. listings — 12 of the 20 pets show a photo of the wrong species, breed or colour (e.g. Archie the Australian Shepherd is a grey cat; Olive the Siamese and Teddy the Poodle share one dog-plus-tabby photo; the six article-*.avif files double as the /blog banners). Re-harvest per-pet photos or rewrite those rows.

Notes for maintainers

• Port 40024 collides with feat(webmd_doctor): add WebMD Doctor mirror (site 25, port 40024) #95 (webmd_doctor, also index 24); whichever merges second needs re-indexing (websyn_start.sh, control_server.py, Dockerfile EXPOSE, all 20 web fields, docs/tests).
• Verifiers are deterministic-first and decide with --no_llm True; llm_* helpers exist only for parity with merriam_webster.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AkNTAZMziDodR6uqa6QBtP

[sarendis56]

Contributor follow-up for the remaining pet-photo and archive findings:

Focused photo mapping / credits PR, rebased onto your latest 1166ad6 review tip, targets this review branch directly. It includes desktop/mobile screenshots and preserves your tasks, verifiers, and every non-image catalog field.

HF PR #67 is updated at 8316809f5eed94078c9b188e5c315f67df751be1: 20 distinct pet photos at usable resolution, corrected species/breed/coat pairings, article banners kept separate, and matching seed image paths. The archive uses supported roots without Mac metadata; the downloaded upload passes the archive validator. The code PR provides complete photo provenance and a public credits page.

Verification: 272 existing verifier tests passed; 80 visible-browser checks passed across 1440, 768, 390, and 320 px; the rebased container build passed with 27/27 sites ready; reset returned ready and a direct byte comparison matched the seed database.

Please land the small mapping fix with the asset update, then pin the HF merge revision in .assets-revision after HF merge.

[JeremyJC67]

Both halves check out. Thank you for turning these around so quickly.

Assets. Downloaded refs/pr/67 at 8316809f on the review host and ran the repo's own validator rather than trusting the summary:

bytes: 2422612   md5: 70273804f142687affaab540b9dab424
members: 34      AppleDouble/DS_Store: 0
archive roots: adopt_a_pet/    second level: adopt_a_pet/instance_seed  adopt_a_pet/static
validate_asset_archive.py -> validated 34 managed members for adopt_a_pet

That closes the archive-layout blocker.

Code PR (JeremyJC67#1). Reviewed against the current review tip. It merges cleanly, touches no tasks.jsonl and no verify/ file, and the app.py change is confined to the image column — the reused article-*.avif and rehome*.avif banners are replaced by per-pet files, which is exactly the finding. The footer gains a Photo credits link. I will land it once the container run currently in flight finishes, so the verification covers the final tree rather than an intermediate one.

One small thing I will add on top when I merge, rather than ask you for another round: this branch's .dockerignore does not exclude review-reports/, so the two remediation screenshots would be COPYd into the published image. I added that exclusion on the 4shared branch for the same reason and will do the same here.

After that the only open item is release mechanics: a maintainer merges HF #67, then I pin .assets-revision to the merge sha.

[JeremyJC67]

Rebased onto the 26-site main — this branch is now site 27 / port 40026

main merged FedEx (40024) and WebMD Doctor (40025) overnight, which put every open site
branch into conflict. This branch has been brought up to date by merging main rather than
rebasing, so every contributor and reviewer commit is preserved verbatim, and re-slotted to
index 26 / port 40026: registry, Dockerfile EXPOSE, the shared docs and every
site-local port reference follow. The slot is provisional as always — say the word at merge
time and I will re-slot in whatever order you land things.

The whole environment was then rebuilt and re-verified at the new slot:

check	result
asset fetch	fetch_assets.sh adopt_a_pet at refs/pr/67 — validated 34 managed members through the normal path, no hand extraction
HTTP sweep	27/27 return 200
GET /health	ok, 27/27 alive and ready
POST /reset/adopt_a_pet → md5	2d56013ab6ea2b115e13ad4a59d530bd, identical to the seed
POST /reset-all	ok, 27/27 ready, 1.51 s
verifier no-op matrix	20/20 FAIL
registry + docs tests	4 passed, 3,343 subtests

@sarendis56 — your photo remediation is merged. JeremyJC67/WebHarbor#1 landed as decb0d6;
the branch head is 6c56371. Before merging I confirmed it touches no tasks.jsonl and no
verify/ file, that the app.py change is confined to the image column, and that all 20 image
paths it references resolve against HF #67 (20 referenced, 28 present in the bundle, none
missing). I added one commit on top: .dockerignore now excludes review-reports/, so your two
remediation screenshots do not ship inside the published image.

That the fetch above succeeded through the ordinary path is itself the proof that the archive
layout fix is real — the previous bundle had to be extracted by hand. The audit assertions also
hold in the image: GET /logout 405, POST /favorite/<slug> without a token 400,
POST /alerts with radius=abc 400 rather than a 500, a branded 404, the hero placeholder no
longer naming the location tasks 0 and 2 dictate, and five read-only page loads leaving the DB
byte-identical.

Still blocked only on HF #67 being merged, after which I repin .assets-revision.