Skip to content

Add CarMax mirror + task verifiers (site by @Violet24K, verifiers by reviewer) - #57

Merged
QianhuiWu merged 7 commits into
aiming-lab:mainfrom
DEM1TASSE:carmax-verifiers
Sep 17, 2026
Merged

QianhuiWu merged 7 commits into
aiming-lab:mainfrom
DEM1TASSE:carmax-verifiers

Conversation

@DEM1TASSE

Copy link
Copy Markdown
Contributor

Brings the CarMax mirror + its verifiers into main as one PR, per the maintainer's
request to merge via the main repo rather than the cross-fork PR #24.

Authorship is preserved — both contributors stay credited. This branch is Zihao Li's
CarMax site commits (from PR #24) with the reviewer's verifier commit on top; each commit
keeps its original author (site = @Violet24K, verifiers = the reviewer). Please merge with a
regular merge (not squash) so both signatures are kept.

Contents:

  • the full sites/carmax/ site (Flask app, templates, seed, 20 tasks) — @Violet24K
  • sites/carmax/verify/verify_<0..19>.py + verify_lib.py and verifier_path/judge_rubric
    in tasks.jsonl — reviewer deliverable (deterministic verifier per task: navigation +
    DB after-state + answer; validated against agent_demo/agent.py)

Known conflict (please resolve on merge): main already merged Merriam-Webster at
index 15 / port 40015, and CarMax was authored at the same slot. On merge, register CarMax as
the next site — index 16 / port 40016 — in websyn_start.sh, control_server.py,
Dockerfile (EXPOSE …-40016), and bump the web port in sites/carmax/tasks.jsonl
(40015 → 40016). The verifiers are port-agnostic (they match URL paths, not the host port).

Supersedes the cross-fork PR #24 and the verifier-only PR #56.

Violet24K and others added 7 commits May 14, 2026 22:28
…com. - 13 SQLAlchemy models (User / Store / Vehicle / SavedVehicle / Comparison + ComparisonItem / Reservation / TestDrive / Appraisal / FinancePreQual / Order / Review / Article) - 59 routes covering search / browse / detail / research / compare / saved / sell-my-car / pre-qual / reserve / test-drive / checkout / account / articles / FAQ / MaxCare / stores / auth - Token-overlap scored search with multi-field weighting - 141 deterministically-seeded vehicles across 31 templates - 12 real CarMax store locations - 5 benchmark users with pre-populated saved/reservation/test-drive/ appraisal/order data - 20 WebVoyager tasks in tasks.jsonl (6 Easy / 9 Medium / 5 Hard, including 2 disambiguation tasks) - Idempotent seed at function level; byte-identical reset verified
Per-task deterministic verifiers (verify_<N>.py) + shared verify_lib.py under
sites/carmax/verify/, plus verifier_path + judge_rubric on each task in tasks.jsonl.
Each verifier checks navigation (anti knowledge-shortcut), DB after-state for stateful
tasks, and the answer against frozen ground truth. Validated against the react agent.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jackjin1997 added a commit to jackjin1997/WebHarbor that referenced this pull request Sep 12, 2026
Refresh review-reports/PR-33-IMDB.md to the fixed candidate (code 1c8a1eb,
HF PR aiming-lab#57 revision f9ddfd25): drop the superseded HF pin and Owner-pending
text, record the formal 20-task blind review and reconciliation, the
fixed-candidate engineering and guided impact regression, Owner visual
acceptance, known limitations and the maintainer handoff order.

Add public summaries and current-candidate screenshots under
review-reports/assets/pr33-imdb/ and mark the historical probe summaries
as superseded. Docs-only: no application, task, verifier, seed or asset
bytes change.
Raibows added a commit to jackjin1997/WebHarbor that referenced this pull request Sep 17, 2026
…ex 30 / 40030 after aiming-lab#116 took index 29

Conflict resolution against current main (aiming-lab#116 merged: nvidia index 28, UC Berkeley
index 29, 30 sites):

- websyn_start.sh, control_server.py: keep main's 30-site order and append imdb as
  index 30 (container port 40030).
- Dockerfile: keep main's build steps unchanged (the PR adds no build step for imdb,
  whose seed ships in the HF archive) and set the canonical header/EXPOSE lines:
  "31 Flask mirror sites" / "EXPOSE 8101 40000-40030". The merged file differs from
  main's by exactly those two lines.
- sites/imdb/tasks.jsonl: 20 web URLs 40024 -> 40030. No task semantics changed
  (ques/judge_rubric/verifier_path untouched); sites/imdb/app.py keeps its
  PORT-env default of 5000, so no other file in the site referenced 40024.
- nvidia (40028) and berkeley (40029) task URLs are unchanged from main.
- .assets-revision: kept as the PR's value (revision:
  f9ddfd2596229f2610418d57fc88c3051e1056bb, the head commit of HF dataset PR aiming-lab#57).
  This pin is NOT usable after this merge: refs/pr/57 carries 26 archives and has no
  archive for fedex, webmd_doctor, healthline, kaggle, nvidia or berkeley, and
  c32018ca (the pin main uses) has no imdb.tar.gz. The pin will be replaced with the
  merged HF dataset commit once HF PR aiming-lab#57 is merged; this commit does not change it.
- Documentation (README.md, AGENTS.md, CLAUDE.md, CONTRIBUTING.md) took main's side
  here; the 31-site / 40000-40030 statements follow in the next commit.
QianhuiWu added a commit that referenced this pull request Sep 17, 2026
Integrate CarMax PR #57 with reviewed fixes and harder tasks
@QianhuiWu
QianhuiWu merged commit 47c9bb5 into aiming-lab:main Sep 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants