Add NBA mirror

[lyc61c]

Summary

Adds a new WebHarbor mirror site for NBA.com:

• Real site mirrored: NBA.com
• Upstream URL: https://www.nba.com/
• Site slug: nba
• Local port: 40015 in the container, typically exposed as http://localhost:41015/ during testing

The mirror implements an NBA.com-style sports information portal with homepage stories, games/schedule, playoffs, draft, fantasy, stats, standings, teams, players, news/articles, search, tickets, store/cart/checkout, and account flows. Runtime behavior is deterministic and uses SQLite plus local static assets only.

Seeded data

Major seeded model counts:

Model	Rows
Users	4
Teams	30
Players	28
Games	10
Articles	22
Products	16
Cart items	8
Orders	4
Order items	4
Favorites	12
Ticket requests	0

Functional coverage

Implemented routes and flows include:

• NBA.com-style homepage with top game strip, hero story, story cards, headlines, trending/news sections, recaps, and right rail
• Games and schedule pages with official-style game cards and playoff schedule layout
• Playoffs bracket/news page
• Draft page with combine, lottery, mock draft, key dates, draft info, and classic moments sections
• Fantasy page with featured stories, watch cards, Yahoo High Score sections, and dense player-stat table
• Stats and standings pages with official-style leader panels, tables, conference standings, and bracket context
• Teams by division, team profiles, rosters, schedules, and team stats
• Players table, filters, player profiles, and stat summaries
• News/category browsing and article detail pages
• Search grouped by teams, players, games, articles, stats, and store items
• Tickets request flow, local NBA Store, bag, checkout, account, orders, and favorites

Benchmark tasks

sites/nba/tasks.jsonl contains 20 WebVoyager-schema benchmark tasks.

Task coverage includes:

• Player filtering and profile lookup
• Games and season leader comparison
• Schedule lookup
• Standings-to-team-profile reasoning
• Player stats and team stats comparison
• Team division browsing
• Draft page/article/key-date tasks
• Fantasy stat table comparison
• Search-to-article navigation
• Watch/highlights disambiguation
• Tickets comparison
• Store filtering/product-detail reasoning
• Authenticated cart/checkout/account update flows

The tasks were revised to reduce benchmark leaks:

• No repeated seeded credential text in tasks
• No alice.j@test.com / TestPass123! task leakage
• No stale TBD arena task from an older schedule layout
• Fewer login-dependent tasks; only tasks that genuinely require account/cart/profile state ask the user to sign in

Paired HuggingFace dataset PR

Paired asset-side PR:

<PASTE_HUGGINGFACE_DATASET_PR_LINK_HERE>

The HF asset bundle is nba.tar.gz and contains:

• nba/instance_seed/nba.db
• nba/static/images/
• nba/static/external_cache/

After the HF PR merges, this code PR should bump .assets-revision to the merged dataset revision.

Visual fidelity

The NBA mirror is designed to closely follow NBA.com rather than a generic Bootstrap demo:

• Dark NBA-style top navigation with major nav entries: Games, Schedule, Watch, Playoffs, News, Draft, Stats, Standings, Teams, Players, NBA Play, Fantasy, BAL 2026, League Pass, Store, Tickets, Sign In
• Homepage portal layout with top score strip, hero story module, story grids, headlines rail, trending sections, game recaps, Around the NBA content, social/affiliate-style right rail sections
• Schedule, Draft, Fantasy, Playoffs, Stats, and Standings pages modeled after NBA.com page structures
• Teams grouped by NBA divisions with local profile/stats/schedule/roster links
• Dense player/stat/standings tables for benchmark comparison tasks
• Local NBA.com-style imagery served from static/images/; no runtime network fetches

Screenshot evidence to attach to this PR:

• NBA.com homepage vs local http://localhost:41015/
• NBA.com Schedule page vs local /schedule
• NBA.com Draft page vs local /draft
• NBA.com Fantasy page vs local /fantasy
• NBA.com Playoffs page vs local /playoffs
• NBA.com Stats page vs local /stats
• NBA.com Standings page vs local /standings

Test plan

• python3 -m py_compile sites/nba/app.py
• ./scripts/build.sh webharbor:dev
• Started test container:

docker run -d --rm --name wh-test \
  -p 8201:8101 -p 41000-41015:40000-40015 webharbor:dev

• Control plane reported nba alive on port 40015
• Key pages rendered successfully during local checks:
  • /
  • /games
  • /schedule
  • /playoffs
  • /draft
  • /fantasy
  • /stats
  • /standings
  • /teams
  • /players
  • /news
  • /search
  • /tickets
  • /shop
  • /login
• POST /reset/nba returned "ready": true
• Byte-identical reset verified

Reset evidence

POST /reset/nba  ->  {"pid":524,"ready":true,"site":"nba"}

acb697f83c108e21b8da4c069cfef373  /opt/WebSyn/nba/instance/nba.db
acb697f83c108e21b8da4c069cfef373  /opt/WebSyn/nba/instance_seed/nba.db

Notes for reviewers

• The mirror does not implement real video streaming, League Pass playback, ticket purchasing, payment processing, or external authentication.
• Tickets, store, checkout, favorites, and account flows are local deterministic benchmark workflows.
• HTTP handlers use SQLAlchemy/SQLite and local static files.
• The runtime does not depend on live NBA.com data or runtime JSON files.
• sites/nba/image/ is a local scratch/reference directory and should not be included in the final GitHub PR.

[MufanQiu]

Review — nba (PR #30)

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 nba.tar.gz is not on main (CI fail-closed)
• BLOCKER: Port-slot 40015 collides with merriam_webster on current main (merge conflict / clobber)
• MINOR: NBA--18 'pre-order total' is non-deterministic due to non-empty seeded cart
• MINOR: NBA--13 answer leaks in the search-result dek (no detail page needed)
• MINOR: NBA--16 listing card exposes name+team+color+price (no product page needed)
• MINOR: NBA--5 has a tie for second-highest BPG, making 'the top two' ambiguous

Mechanical checks: FAIL

• Byte-identical reset: OK. Idempotent seed gated on Team.query.count()>0 and seed_benchmark_users() on alice's email; fresh boot reproduces the seed md5.

Visual fidelity: PASS

High-fidelity clone. Homepage (/tmp/nba_home.png) reproduces the utility nav, full main nav (Games/Schedule/Watch/Playoffs/News/Draft/Stats/Standings/Teams/Players/Store/Tickets/Sign In), hero story with a real player photo, Stories carousel, Trending Now, 2026 Postseason, Game Recaps, footer.

Functional depth: PASS

All flows driven with Playwright. Login (alice.j@test.com/TestPass123!) → /account; wrong password → "Invalid email or password." Register creates + auto-logs-in; duplicate email → "already registered." Search is scored/additive (multi-word does not zero out).

Task quality: PASS (with nits)

20 tasks; difficulty bar met (≥5 tasks need ≥5 actions; several multi-step/comparison tasks; no task solvable by clicking result #1, since listings hide PPG/APG and standings hide arena/coach).

Required fixes before approval

1. (BLOCKER) Resolve the 40015 collision with merriam_webster. Rebase on current main and register nba at index 16 → port 40016 in all three files in sync: websyn_start.sh SITES, control_server.py SITES, and Dockerfile EXPOSE (raise to 40000-40016).
2. (BLOCKER) Bump all 20 task URLs in lockstep. After the rebase, change every "web":"http://localhost:40015/" in sites/nba/tasks.jsonl to the new nba port (:40016/) so tasks target nba, not merriam_webster.
3. (MAJOR) Pin .assets-revision to a merged dataset SHA that provably contains nba.tar.gz (not the floating main, not a PR ref).
4. (MAJOR) Fix NBA--18 determinism. Either empty the benchmark users' seeded carts in seed_benchmark_users() (so the total is the intended $259.78) or rephrase the task to instruct clearing the bag first / report only the added jersey line.
5. (MAJOR) Fix the NBA--13 leak. Shorten the "Bucks focus on half-court spacing" dek so it no longer enumerates both balanced elements; keep "Antetokounmpo rim pressure / Lillard range" only in the article body.

[jackjin1997]

I’m claiming PR #30 for reviewer takeover. I’ll preserve @lyc61c’s original contribution, carry it on a separate reviewer-owned branch, audit and fix the environment/tasks/verifiers, and submit a linked Draft Review: PR with reproducible evidence. I will not merge either PR.

[jackjin1997]

Reviewer takeover continues in Draft PR #101: https://github.com/aiming-lab/WebHarbor/pull/101\n\nPaired reviewed asset candidate: https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/71. The original mirror contribution and authorship are preserved; the Draft remains pending independent blind review and final visual reconciliation.

[lyc61c]

Thanks for the reviewer takeover and follow-up work in #101.

Since #101 has now been merged and explicitly supersedes this PR while preserving the original NBA mirror contribution and attribution, I’m closing #30 to avoid keeping a duplicate PR open.

Thanks @jackjin1997 and the maintainers for reviewing, fixing, and integrating the NBA mirror.