feat(webmd_doctor): add WebMD Doctor mirror (site 25, port 40024) - #95
Merged
Conversation
Offline Flask mirror of https://doctor.webmd.com/ ("WebMD Care"): deterministic synthetic directory of 224 doctors across 10 specialties and 8 cities around Newark, DE 19711, with hospitals, group practices, Choice Awards, real auth, saved providers, appointment requests and pending reviews. - Build-generated seed (.build-generated-seed): the Dockerfile regenerates instance_seed/webmd_doctor.db plus 224 Pillow initials avatars and 90 video poster frames from seed_data.py; no Hugging Face assets, .assets-revision untouched. - One seeded RNG (20260910), literal reference date 2026-09-10, hardcoded scrypt hashes for the four benchmark users, whole-function seed gates, seed_metadata version + expected-count validation; byte-identical after /reset and docker restart. - Deterministic longest-match query parser (specialty / condition / procedure / insurer / gender / virtual), conjunctive filters, Best Match / Distance / Average Rating / Number of Ratings ranking, numbered pagination. - Registered as index 24 in websyn_start.sh and control_server.py; Dockerfile EXPOSE raised to 40024, header comment bumped to 25 sites. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
Contributor-side task definitions only (web_name, id, ques, web, upstream_url): 9 lookups (0-7, 15), 4 multi-constraint (9-12), 2 compares (13-14), 5 stateful (8, 16-19). Every ques routes through a specialty + city, a hub page or an awards class before naming a doctor (name search is a non-goal, as upstream); credentials are embedded for login tasks; no answers, verifiers or rubrics. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
- seed: two extra West Chester cardiology slots appended after the grid (task 12 now has 8 doctors on the city page, 6 rated 4+ with exactly one male); deterministic language top-up for physicians covering three offices (task 1 target speaks 3 languages); EXPECTED_COUNTS refreshed (226 doctors, 202 in radius) - pagination: disabled prev/next rendered as spans instead of live links to an empty page - filter bar: empty and default params are dropped from auto-submitted URLs Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
Section-by-section pass against scraped_data/reference at 1440 px: - profile: booking widget overhangs the hero into the rail, tabs span the main column only, colleagues tile grid / hospital affiliation card in the stack, Basic rail below the tabs, NPI folded into the certification card, nearby + transparency panels full width, star picker, review controls, check-circle perspective icons, two-column bulleted top-20 lists - home: icon tile row for popular specialties, filled uppercase View Profile buttons, upstream heading sizes - filter pills in dark text with navy active state; typeahead no longer clipped by the search bar, prefix highlighting - hospitals / group practices hubs rebuilt as landing pages (state chips, name search, top-4 cards, search-all button, care-type chips); state lists drop the sliders button and gain a name filter - hospital / practice details: specialty select, map-on-top locations, ratings card, text flag lines - awards page art panels + larger headings; recipients grouped by state with a state filter; guidelines as a centred numbered card; sign-up terms line and mm/dd/yyyy date field; login/state/city copy per upstream - landing "Highest Rated" strip now the 25-mile ring so no task target is showcased there Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
README row counts match EXPECTED_COUNTS of the final seed (226 doctors, 348 locations, 91 posters). Freezer byte-identical across two runs and PYTHONHASHSEED=0/1; in-image instance == instance_seed before and after docker restart. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
Shared docs (README, AGENTS, CONTRIBUTING, CLAUDE, agent_demo/README) and the registry assertions in the walmart_careers and rotten_tomatoes tests now cover the appended webmd_doctor slot (index 24, port 40024). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GTvmqJcShyy3v3KfEPxMPe
…hanged) Switch to the walmart_careers asset shape: the Docker RUN block still regenerates instance_seed/webmd_doctor.db deterministically from seed_data.py, but the 226 initials avatars and 91 video posters under static/images/ now arrive through the pinned Hugging Face tarball (.requires-images) instead of being written at image-build time. - seed_data.py: image generation moves behind `--write-images` / build_images() for the local freezer only; PNGs are saved with fixed compression and no ancillary chunks (IHDR/IDAT/IEND only), identical under PYTHONHASHSEED=0 and 1 (317 files, list digest e1b14d92). - Dockerfile: assert the fetched avatars/posters are present, then build the DB only. In-image seed md5 unchanged (364c15c35fff64f0c25ef62ac0bb64fc). - Markers: .build-generated-seed reworded, .requires-images added so check_assets.sh / fetch_assets.sh treat the site like walmart_careers; .assetpaths globs already cover static/images/. scripts/ untouched. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qq2aHgZwwYWAS7cCepWKiu
…iscussions/68; temporary) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qq2aHgZwwYWAS7cCepWKiu
…ed-search notice, city-page sections, profile scale - seed: Medicare/Medicaid Insurance-card rows now follow the primary office flags (deterministic post-pass, no RNG; doctor_insurances 2274 -> 2208; in-image md5 8bf76b696f8e49ba59848a51bf47f24a; avatars/posters byte-identical, no HF change) - results: unresolved search text keeps the "ignore" semantics but the heading reads "All Providers" and a mirror-specific notice names the unmatched input - specialty city page: description blurb, FAQ and related-specialties grid below the list (FAQ removed from the landing page), Distance control + distance on cards - profile: type scale matched to upstream, sticky tabs + compact provider bar on scroll, "View Less" toggle, certification-card dividers, grey nearby/transparency panels - chrome: blue filter-pill labels, banner dismiss, larger stars, inline View Profile on regular cards, grey current page, footer badge cluster + privacy toggle, hub hero / Sort By / info-block styling, extra hospital & practice FAQ entries, recipients show only the active award chip, awards sub-nav chevrons, guidelines page without search row/footer, styled booking select, auth overlay dims the header Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QGf8U2ULBCevvaYfbBBuVt
…ve/hours consistency - results: a search term in which no token resolves to a vocabulary term now renders the upstream-style empty state (heading with the typed term, "0 Results", notice, filter bar kept, browse-by-specialty link); partially matched queries still ignore the unmatched tokens - seed (RNG-free post-passes, row counts unchanged): every primary office gets its own direct line (no office shares its practice's main number, all phones unique); a practice and every office located at it share one schedule (the longest weekly office schedule); Patients' Perspective votes per criterion are bounded by the ratings count and floored at the visible review marks - in-image md5 86cd0e7a8669151405f5c6b541fb91d0; avatars/posters byte-identical (no HF change); no task id shifted; T2 phone + Saturday hours and T15 practice Saturday hours change value, every other answer is unchanged Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QGf8U2ULBCevvaYfbBBuVt
…ecialty-consistent condition lists) - T15 ques names the real controls (Award Winning Hospitals > WebMD Patient's Choice); T4 ques uses the on-page heading "Certifications, License, & Education" (upstream label). - Booking reference is eight base-32 chars packed injectively from (row id, doctor id, office index, day, slot) with fixed constants - two bookings can never share a reference, and a first booking after a reset is no longer the same string for every doctor. - Most-Treated / Top-20 conditions and procedures draw only from the doctor's own specialty plus its SECONDARY_CHOICES pool (RELATED_SPECIALTIES removed); under-5-years bucket is 2-4 years; five training-institution names that doubled as hub hospitals replaced. - "1 Year Experience" pluralised on cards, profile and hub mini-cards. - <=720 px CSS: header nav, review controls, footer badges wrap; promo panel and location blocks stay inside the viewport. - Re-frozen seed (PYTHONHASHSEED 0 and 1 identical; images byte-identical to the shipped HF tarball); EXPECTED_COUNTS + README row counts updated; re-picked targets for tasks 4, 11 and 14 after the RNG reshuffle. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S8Mt6G2qwvqncZiKyVC5pR
…aining timelines, per-city streets) - SECONDARY_CONDITIONS / SECONDARY_PROCEDURES: curated per-specialty pools (9-10 conditions, 6-8 procedures incl. the primaries); Top-20 lists draw only from the primaries + the doctor's own specialty pool (conditions 40 -> 82, procedures 30 -> 62) - TRAINING + training_plan(): residency length per specialty after the MD, fellowship (Cardio/GI 3 y required, optional 1-3 y elsewhere), board = end of training + 0/1, years of experience = 2026 - board year; subspecialty board only after a fellowship - SECONDARY_STREETS per city; every street name belongs to exactly one city (Deer Park Digestive Health moves from E Main St to Library Ave) - _bound_review_stars(): review means stay within 1.0 of the profile average (RNG-free) - doctor stream unchanged (names, slugs, NPIs, cities, ratings, years, offices, phones); EXPECTED_COUNTS + README rows updated; images unchanged (list digest e1b14d92) - tasks: T4 -> Dr. Charles Villanueva, T11 reports NPI + residency institution, T13 pair -> Greenwood vs Dr. Emerson Huang, T14 compares the two Psychiatrists Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YKXfhRp7xMmrgixw5Ccohe
…next, no __pycache__ in the image) - logout(): session.clear() before logout_user() - safe_next(): reject leading/trailing whitespace (raw and decoded), strip fragments explicitly - Dockerfile: rm -rf __pycache__ after the build-time seed run Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YKXfhRp7xMmrgixw5Ccohe
20 verifiers (verify_0..19.py) plus verify_lib.py and ground_truth.py under sites/webmd_doctor/verify/, in the walmart_careers contract: argparse CLI, zero LLM calls, fail-closed snapshot discovery (run_dir or docker cp), pinned 28-table schema hash + seed marker + counts, catalog tables row-identical, read-only tasks require the four runtime tables unchanged, stateful tasks enforce exact row deltas (saved provider removed/added, appointment row matched by user/doctor/office/slot with the answer's reference taken from the new row, review row by rating/text/status, registration by the e-mail typed on /signup). Route gates follow the task text (results query params incl. the Newark rule and combined filters, specialty menu path, hospital and awards pages, second review page); comparison tasks gate on both profiles. ground_truth.py re-derives every target from the initial snapshot and fails closed on disagreement. verify/tests: 372 subprocess tests (genuine, shortcut, wrong answer, state mismatch, over-action, read-only write, packaging and contract negatives) plus 25 library tests; no docker, no API key. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017yDUT66ukMXuASicT7yd6H
tasks.jsonl now carries the seven reviewer keys. The five contributor keys are byte-identical to the submitted rows; verifier_path points at sites/webmd_doctor/verify/verify_N.py; judge_rubric opens with the shared scoring preamble (step list authoritative, unverified is not contradicted, the verifier owns values and the database) followed by rule-only checkpoints. A dev-side validator asserts the key shape, ids 0-19, verifier files, byte identity of the contributor keys, and that no rubric contains a ground-truth token or a doctor, hospital, practice or city name absent from its own ques. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017yDUT66ukMXuASicT7yd6H
LLM-free matrix (116 scripted Playwright runs graded through eval_judge.py --verifier True) showed the five stateful verifiers ran row-identity / exactly_one_* checks before <table>_exact_delta, so a run that wrote nothing failed under the wrong name. Exact delta now comes first; the redundant exactly_one_* count checks are gone; unit tests updated (397 green). README gains the validation summary (counts only). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WM3oHCY7Vi612v5c6JSfay
Rules-only rewrite from 40 dual-graded agent runs (gpt-5.4-nano and gpt-5.4-mini, judge on nano): the shared preamble now states the screenshot window, that the final answer text is the evidence, and that an uncontradicted checkpoint set means success; "must be visible" checkpoints (8-11, 16-19) and "as shown in the section" facts (0, 4, 5, 9) are phrased against the step list. validate_tasks: contributor keys byte-identical, no ground-truth token. README records the real-run counts (verifier 1/20 nano, 7/20 mini; 0 verifier defects; 0 judge false PASSes; agreement 19/20 and 15/20). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UB1bWfYQV98foy5iCKxP1H
Contributor
Author
Track B — grading contract for webmd_doctorFollowing #86's pattern, this adds the reviewer-side contract to the same PR: Reading the pass rates. Verifier PASS is 1/20 on gpt-5.4-nano and 7/20 on gpt-5.4-mini. Every task is solvable: the scripted Playwright walk from the ques text is 20/20 in the matrix, and every real-run FAIL adjudicates to the agent — a name search that returns 0 results (as on upstream) followed by a give-up, The number that matters: 0 judge false PASSes before and after polish. All 7 verifier↔judge divergences were judge false FAILs (evidence-window or self-contradictory nano-judge rationales); a gpt-5.4-mini judge agrees with the verifier on 5 of the 6 residual ones. Observation for maintainers, not actioned: the 0-result name search is the single largest source of agent give-ups. It is faithful to upstream and the empty state already points to specialty search; if you would prefer a more explicit hint there, it is a template-only change and I can push it separately so it does not muddy these runs. Total spend $1.17 (cap $2.50). Details below. Full reportVerdict on the site (A): unchanged from the PR description's Verification table: 25/25 alive, Contract (B): Validation (C):
Agreement tablePass — agent nano
Notes for maintainers: |
Seed and data integrity: - All 226 doctor NPIs are now checksum-valid (CMS Luhn rule over 80840+9 digits) and verified unassigned against the NPPES registry (August 2026 monthly + weeklies through 2026-09-06 + deactivated report); the verified list is embedded in seed_data.py and re-checked on every build. RNG stream preserved, so slugs, images and education rows are unchanged. - Alice holds six saved providers (seven rows total) so the saved-provider removal task has real breadth; _seed_is_complete hardened to all immutable counts + foreign-key check. - Booking confirmation references use range-checked deterministic packing. App hardening: - SECRET_KEY from env or per-process random; session cookie hardening; login timing equalization via a dummy scrypt hash; bounded integer/argument parsing; safe_next rejects control characters; IntegrityError and duplicate guards on save/booking/review; booking enforces each office's new-patient policy. - Specialty hub pages count primary-OR-secondary specialty (matching results search); hub header totals reflect the filtered population; /_health added with legacy /health alias; inject_globals degrades safely so error pages render even when the database fails. Verifier hardening (447 tests): - Clause-level negation matching, first-value parameter binding, signed-in checks bound to the login email+password, minimum screenshot 320x240, booking reference recomputation, per-task ordering/role/comparison checks, and an EXPECTED_FACTS cross-check in ground truth covering all 20 tasks. UI, accessibility, responsive: - Sitewide mirror notice and synthetic-data disclaimers; heading-order fixes; skip link; focus-visible outlines everywhere; fieldset/legend radio groups; role=alert error lists; aria-controls/expanded on disclosures; contrast fixes for muted/red tokens, booking steps and map placeholder; 320px overflow fixes (scrollable header nav row, single-column mini grids, wrapping footer badges); no-JS progressive enhancement (menus/popovers render as static lists, noscript submit fallbacks). Asset governance and repo integration: - generated_asset_inventory.json (317 PNGs with sha256) + site-local check_generated_assets.py, gated in the Dockerfile and check_assets.sh. - fetch_assets.sh all-sites mode now enumerates local sites and tolerates extra archives on HF (drugs_com, fedex). - .assets-revision pinned to merged HF main sha ad6f424f72cada9e6f5c09a580 93d0ceeab9c52b (HF assets PR merged); walmart pin assertion updated. - Root README site count 23 -> 25; site README documents the NPI policy, known deviations and benchmark accounts. Tests: new sites/webmd_doctor/tests (seed quality + determinism + Luhn audit, task breadth, rendered-page sweep, generated assets); verify suite 447 passed; cross-site battery green (compass 94, RT 56, walmart 51, ikea 119, walmart verify 246); 20/20 Playwright E2E tasks pass against the packaged image.
… overflow and saved-link contrast; add answer-leak sweep Review-round 2 (checklist re-audit): - Booking widget no longer hardcodes New Patient as checked: task 17 requires the runner to choose the patient type, so every form field the tasks ask for now arrives empty (README documents the policy). - Enhanced-profile layout forced horizontal overflow below 1181px (absolute 384px sidebar plus 424px tabs/hero offsets; 791px document width measured at a 390px viewport): an appended max-width:1180px block makes the sidebar a static full-width card and drops the offsets. - Saved-providers card links used #7d92ff (2.85:1 on white, below the 3:1 large-text floor); switched to var(--blue) (5.31:1). - New tests/test_answer_leaks.py: derives all 20 ground-truth answers from the frozen seed and asserts unique answer facts (NPIs, phones, institution names, office addresses, websites, review dates, years) appear on none of the ~100 pre-discovery chrome surfaces, are absent from templates/CSS/JS, and that no booking-reference pattern appears on chrome. Shared-vocabulary classes (condition browse tiles, criterion callout chips, small integers, hour strings) are covered by entity-bound tests: the T5 target's cards never render her condition names or tier wording, the T7 target's own card chip never reads the answer criterion, and the card template renders no conditions at all. Mutation-tested: injecting a ground-truth school name and a TODO marker into index.html fails both guards; corrupting one seeded NPI fails the Luhn audit; an extra image file fails the asset checker. - test_seed_quality and verify/tests/_support surface subprocess stderr in assertion messages on build failure. - walmart shared-documentation test derives the expected 40000-400NN port range from the control_server.py SITES registry via ast instead of hardcoding the site count.
…inventory test The T5 entity-bound test asserted that none of the target's most-treated or Top-20 condition names appear on her pre-profile cards. That is stricter than the leak model: enhanced cards quote patient reviews, and a quote may mention a non-answer condition in prose (the target's snippet mentions Irritable Bowel Syndrome while the task answers are the More-Than-Most condition and the first Top-20 condition). The prose mention is non-discriminative because the tier data that identifies the answer lives only on the profile, and the verifier binds scoring to a required profile visit. The old assertion also depended on regex extraction from the rendered profile, which made it sensitive to whitespace changes across Jinja2 versions (3.1.4 vs 3.1.6 gave different extracted name sets from byte-identical seeds). The test now derives the two answer facts from ground truth, pins them fail-closed, and asserts they and the tier wording never appear on the target's own cards. Also restores test_tarball_contains_only_declared_generated_images, which was lost to a truncated file write before the previous commit. It asserts the published tarball at the pinned revision contains exactly the 317 declared generated images (no seed database, no ground-truth data, no undeclared members in either direction). Both tests are mutation-verified: injecting the answer fact into the card template and adding a fake inventory entry each fail the respective test, and reverting passes.
The enhanced-profile bio lead-in ('Meet Dr. X:') and .meet accent rendered
#d9541e on white, which measures 4.01:1 - below the 4.5:1 WCAG AA threshold
for normal-size text. The replacement #c24a19 keeps the same hue (17 deg)
and measures 4.90:1 on #fff and 4.70:1 on #fafafa. Found by the pixel- and
computed-style-based contrast sweep over all routes at 1440/768/390/320.
This was referenced Sep 11, 2026
…WebMD Doctor branch Registry: append `webmd_doctor` after `fedex`, so FedEx keeps index 24 / port 40024 and WebMD Doctor moves to index 25 / port 40025 (26 sites, 40000-40025). Conflict resolutions: - `.assets-revision`: keep `ad6f424f` (current HF dataset main, which carries `webmd_doctor.tar.gz` and the byte-identical `fedex.tar.gz` blob) instead of main's FedEx-only pin `68dcbf2c`. - `scripts/fetch_assets.sh`: take main's registry-scoped fetch implementation. - `Dockerfile`: keep both site blocks (FedEx asset gate + WebMD generated-asset gate and source-built seeds) and widen `EXPOSE` to 40000-40025. - `README.md`, `AGENTS.md`, `CLAUDE.md`, `CONTRIBUTING.md`, `agent_demo/README.md`, `.claude/skills/*`: 26 mirrors, port range 40000-40025, alt ports 41000-41025. - `sites/walmart_careers/tests/test_integration.py`, `sites/rotten_tomatoes/tests/test_environment_quality.py`: take main's registry-derived assertions and extend them for `webmd_doctor` (40025) and its generated-asset gate. Follow-on work required by the port move: - `sites/webmd_doctor/tasks.jsonl`, `README.md`, `verify/verify_lib.py` comment, `tests/test_task_breadth.py` (now derives the port from the registry), `verify/tests/test_verify_lib.py`: port 40025. - new `sites/webmd_doctor/tests/conftest.py`: rebuild the build-generated seed when it is absent and install the runtime copy the app binds, so the suite runs after `scripts/fetch_assets.sh` alone (the Walmart mirror's convention). In that state the suite previously failed 7 tests. Verified after the merge: webmd site tests 19, webmd verify 447, walmart 51, fedex 91 (+1341 subtests), rotten_tomatoes 56 (+4129 subtests), compass 94.
Contributor
|
Thanks for your contribution! @evanz37 @alidezhihui |
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Mirrors https://doctor.webmd.com/ (the WebMD Care physician directory only; PR #62 mirrors www.webmd.com and is unrelated). Registered at index 24 / port 40024, bringing the registry to 25 sites. Ships 20 benchmark tasks in
sites/webmd_doctor/tasks.jsonl. Every provider, practice, hospital, address, phone, NPI, review and account is synthetic; only the specialty / condition / insurer / city names are real vocabulary, and only site chrome (layout, icons, CSS) was harvested.What's included
app.py, SQLAlchemy models, CSRFProtect, Flask-Login) plus Jinja templates for home, results with filters and sorting, physician profiles (Basic and Enhanced layouts), hospital and practice hubs, specialty index / state / city pages, Choice Awards recipients, login / signup, saved providers, appointment requests and reviews..build-generated-seedmarker and a Dockerfile RUN block in the same shape as walmart_careers / compass (PYTHONHASHSEED=0 python seed_data.py,rm -rf instance __pycache__); one seeded RNG, no wall-clock reads, hardcoded benchmark password hashes.alice.j/bob.c/carol.d/david.k@test.com, passwordTestPass123!), plus 4 saved providers, 1 appointment request and 1 pending user review pre-seeded.web_name, id, ques, web, upstream_url); no verifier, rubric or answer lives in any tracked file.websyn_start.sh,control_server.py,DockerfileEXPOSE 8101 40000-40024) plus a separate docs/tests bump commit (docs: bump site count to 25 (ports 40000-40024)).static/images/only (317 generated PNGs: 226 initials avatars + 91 video poster frames, md5-list digeste1b14d92); no DB in the tarball..assets-revisionis pinned torefs/pr/68until it merges.Verification
GET //healthPOST /reset/webmd_doctor{"pid":293,"ready":true,"site":"webmd_doctor"}instance/==instance_seed/12b4c269cae06ccfd41e3d8116565e3abefore reset, after reset, afterdocker restartPYTHONHASHSEED=0and=1; two--write-imagesruns give identical PNG md5 lists equal to the shipped digeste1b14d92git clone --depth 1+ per-sitefetch_assets.sh+build.sh)scripts_dev/, not shipped)Notes for the reviewer
appointment_requestsrow (carol.d, Dr. Sarah Keller, "Riverfront Heart & Vascular - Wellness Center", New Patient, 2026-09-14 10:30 AM), not on the printed confirmation string. T6: the oldest review is on the profile's second review page, so?rpage=2is a valid navigation gate../scripts/fetch_assets.shin all-sites mode currently fails on every revision ("expected 25 site archives, found 26") because the dataset already holdsdrugs_com.tar.gz, which is not in GitHub main. Per-site fetch works:for s in sites/*/; do ./scripts/fetch_assets.sh $(basename $s); done.sites/walmart_careers/testsimmutable-pin assertion is red while.assets-revisionpoints atrefs/pr/68; it passes again once the pin is bumped to the HF merge sha (same history as feat(walmart_careers): add Walmart Careers mirror (site 23, port 40022) #86).sites/rotten_tomatoes/testsregistry assertion relaxed from the literal array tail to an ordered prefix (ted osu rotten_tomatoes compass walmart_careers); the literal-tail form breaks for any appended site.Tasks (20)
Screenshots
Reference captures from doctor.webmd.com (left) vs the mirror on port 40024 (right), both at 1440 px.