Review: complete BoardGameGeek environment and grading contract (#35)

[jackjin1997]

Summary

Reviewer-owned continuation of #35, preserving the original BoardGameGeek contribution and attribution while integrating it after current main as the 31st mirror (index 30, port 40030).

The review fixes environment, task-contract, verifier, responsive-layout, and broad visual-fidelity issues. All 21 accepted tasks retain deterministic grading. The branch is synchronized with current main, fully verified, and ready for maintainer review.

Changes

• preserve the contributor implementation and public attribution to @hqhq1025
• correct direct Flask startup, rating aggregation, registration errors, and task-required collection/GeekList state
• replace the inherited retro shell and restore the full homepage, browse, and game-detail hierarchy
• register BoardGameGeek after the existing 30 sites and align every task URL with port 40030
• keep one deterministic verifier and one non-answer-leaking rubric per task
• repair four task-contract defects: Task 8 answer leakage, hidden filter/sort requirements in Tasks 11/18, and click-only recognition for stateful form submission
• merge current main and resolve the registry, port-range, documentation, and asset-pin conflicts without changing BoardGameGeek behavior

Task quantity and quality

• accepted tasks: 21; verifiers: 21; mapped executions: 21
• portfolio: 16 read/compare tasks, 5 persistent-write tasks, 7 authenticated tasks, and 10 tasks with at least five recorded UI actions
• 21 is already above the public 15–20 recommendation; no weak task was added merely to increase the count
• all rows contain the required task/rubric/verifier contract and no answer key

Verification completed

• exact-head 31-site full-environment smoke: PASS at 9432776b35c2c8a2c6ff1ed1146e34d591af818b against main aa95bd80fb760a0921949477b6f79f5c673f4884
• complete image build: PASS, 4.97 GB; the runner began with 86 GiB free
• BoardGameGeek environment/verifier test methods: 11/11 passed
• all-site checks before and after reset-all: 31/31 HTTP 200 and 31/31 healthy
• BoardGameGeek dirty-state reset: byte-identical PASS; reset-all: 31/31 ready in 7 seconds, with the target DB still seed-identical afterward
• public smoke evidence artifact, GitHub artifact digest sha256:e9b8387955064e855936cc1368829837d52f384b4c16838554945434b7b9ee7a
• verifier suite: 91/91 passed, covering positives, legal alternatives, incorrect/incomplete answers, answer/state without navigation, prompt-permitted paths, and keyboard form submission
• mapped deterministic evaluation: 21/21 passed; Task 8 uses its affected homepage-origin rerun and the other 20 reuse unchanged application/asset executions
• independent Claude Code blind review: manifest valid (228/228 references), 21 PASS / 0 FAIL, using Claude Fable 5.1
• desktop and 390 px captures of home, browse, and Gloomhaven detail: HTTP 200, no broken images, and no page-level horizontal overflow; the restored homepage renders 101 local images
• owner visual comparison accepted after the lower-page restoration: “这个还挺像的”
• reviewed asset archive: 11,345 safe members; archive SHA-256 970c6618e39a37920b228abd976d4756db126239ab4b4fa48b37b03b81cad6d1; seed SHA-256 5a00124fa1e523678c1010ea74a2c1d8ed5febd18f71cf54632bec83a07612b1

Compared with the prior reviewed head, the BoardGameGeek application, templates, static assets, asset inventory, task semantics/rubrics, and all 21 deterministic verifier implementations are unchanged. The only BoardGameGeek delta is the mechanical task URL move from port 40024 to 40030 plus its matching environment-test expectation. The exact-head smoke above revalidates that affected integration surface; the prior mapped execution and blind-review evidence therefore remains applicable.

Guided/source-aware runs prove task feasibility and scoring behavior; they are not independent exploration-performance measurements.

Current integration

• review head: 9432776b35c2c8a2c6ff1ed1146e34d591af818b
• review tree: 05f1edc584457265f11d74e76c09eb8ddc8ab01d
• current-main parent: aa95bd80fb760a0921949477b6f79f5c673f4884
• GitHub reports this PR as CLEAN / MERGEABLE
• launcher, control server, Docker exposure, docs, skills, and all task URLs agree on 31 sites and port range 40000–40030

Asset dependency and maintainer handoff

The matching asset remains open in HF PR #25 by the original contributor. Its immutable revision 6737719a7eeb8d9b5fba77072d455e0d9afd2a93 is the revision tested above, and HF currently reports no file conflicts.

The GitHub branch is ready for review, but the contribution is not directly mergeable end-to-end until maintainers:

1. merge HF PR Add Compass real-estate mirror (port 40015) #25;
2. update .assets-revision here to the resulting HF main commit and confirm the BoardGameGeek archive hash remains unchanged; and
3. merge this GitHub PR after final review.

Non-blocking P2 polish remains: the generic designer index shows a zero game-count label for designers, and user search uses uppercase-first SQLite binary collation. Neither affects the accepted tasks or their results.

Reviewer agents will not merge either PR.

Original PR: #35

[QianhuiWu]

Thanks @jackjin1997 for the BoardGameGeek review and grading work, building on @hqhq1025’s #35. The combined implementation and follow-up fixes were delivered in #127. At the maintainer’s request, #128 now incorporates both original commit histories into main, preserving this PR through merge 464d1ef. GitHub confirms both #35 and #88 are merged. No application code or asset pins changed from the tested #127 result; the only file change records the future integration preference in AGENTS.md. Full build, all 37 homepages/health checks, and BoardGameGeek byte-identical reset passed.