Add Best Buy WebHarbor mirror

[Lxr-max]

Summary

Add Best Buy WebHarbor mirror.

Status

Closed as superseded by Review PR #135 (Best Buy mirror takeover).

Original implementation lived on branch add-bestbuy-site.

[MufanQiu]

Review — bestbuy (PR #48)

Verdict: REQUEST CHANGES.

We deployed the environment locally and ran the full review checklist; every finding below was independently reproduced, and fidelity was checked against the real upstream site.

Issues (summary)

• BLOCKER: .assets-revision pins 'main' but bestbuy.tar.gz only exists on the HF PR ref -> CI fail-closed
• MAJOR: 73/156 product images broken: unescaped '&' in generated SVGs (XML invalid)
• MINOR: Task BestBuy--11 is ill-posed: no deal title mentions a 'big screen event'
• MINOR: Answer-leak + single-result: Task 0 (price) readable on results card; product searches return 1 result
• MINOR: Port-slot collision with released merriam_webster (both claim slot 15 / :40015)
• MINOR: Task 3 amenity leaked on stores list page

Mechanical checks: FAIL

Most mechanics are clean: byte-identical idempotent reset (instance md5 == seed md5 == 71c41f6f..., function-level idempotency gate at seed_data.py:1515), no cross-site imports (from sites./import sites = 0 hits), no AppleDouble files, no scraped_data reads in handlers, .gitignore matches .assetpaths, and py_compile passes on app.py + seed_data.py.

Visual fidelity: FAIL

Brand fidelity is strong: yellow/blue palette (#0046be/#fff200), category nav, hero, deals row, featured grid, store cards, footer.

Functional depth: PASS (with nits)

Every flow was driven live. Login (alice.j@test.com → /account "Jordan"; bad password rejected), register (success + mismatched-password and duplicate-email validation), rewards (alice 1840), wishlist, cart add + reload-persistence, full checkout end-to-end as david producing deterministic order BBY-240061, compare page with alice's 3 seeded items, order lookup/detail.

Task quality: FAIL

19 tasks over a deep, balanced catalog (156 products = 13×12 categories, 25 brands, 15 stores, 24 support articles, 60 orders, 339 reviews, 312 protection plans), so per-category browse distractors are real.

Required fixes before approval

1. (BLOCKER) Repin .assets-revision from revision: main to the merged HuggingFace dataset SHA that actually contains bestbuy.tar.gz (after HF PR 31 merges).
2. (MAJOR) Fix _write_svg (seed_data.py:258-292) to XML-escape title/subtitle/badge (xml.sax.saxutils.escape/html.escape) before interpolation; regenerate all SVGs and reseed.
3. (MAJOR) Fix the Task 0 price leak and the single-result pattern: hide price on result cards for price-answer tasks and/or phrase product tasks with looser queries that return ≥6 results with near-misses, so the answer requires opening the PDP and the agent must disambiguate.
4. (MAJOR) Resolve the :40015 slot collision with the released merriam_webster: append the later-merged site at index 16 (:40016) and bump EXPOSE 40000-40016 in websyn_start.sh, control_server.py, and Dockerfile.
5. (MINOR) Repair Task 11's wording/location mismatch: rename deal 14's title/badge to mention "Big Screen Event," or surface product.deal_badge on the /deals page, or reword the task to "find the product tagged Big Screen Event" without constraining it to the deals page.

[jackjin1997]

I'm taking over the review and remediation of this Best Buy contribution. I'll build a separate Review: PR on top of @Lxr-max's original commit, preserve the original authorship, and link the reviewer PR here.

The review will cover current-main integration (port slot/registry and the .assets-revision pin against the paired HF asset PR #31), the findings already raised by @MufanQiu, browser/UI fidelity against bestbuy.com, task solvability and quality, deterministic verifiers backed by recorded UI executions, positive/negative and valid-alternative grading tests, and an independent review of the frozen runs. Maintainers retain final approval and merge responsibility. Please avoid duplicating this review while it is in progress.

[jackjin1997]

Reviewer takeover is now tracked in Draft Review PR #135: #135

It preserves @Lxr-max’s Best Buy contribution and links the reviewer-owned HF asset candidate. Blind review is still pending; please do not merge either artifact yet.

[Lxr-max]

Closing as superseded by Review PR #135 (Best Buy mirror takeover).