feat(walmart_careers): add Walmart Careers mirror (site 20, port 40019)

[evanz37]

TL;DR

Adds a walmart_careers mirror of https://careers.walmart.com as site 20 on port 40019: a Flask + SQLAlchemy job board with deterministic search and filters, two job populations (hourly store roles / salaried corporate roles), local auth, saved roles, a two-step application flow, and 20 benchmark tasks. Paired assets PR: https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/54 (seed DB md5 b57631080969fe151e1717dbef3dd372, 35 harvested images). .assets-revision is temporarily pinned to refs/pr/54 so CI can build; I will repin to the merge sha as soon as the HF PR lands.

Motivation

The current catalog has no employer careers site. Careers sites are a distinct interaction shape — faceted search over a large uniform catalog, location-radius filtering, and account-scoped stateful actions (save / apply) — that agents meet constantly in real use. Walmart Careers is a good specimen: it mixes hourly store postings (store banner, shift windows, open-position counts) with salaried corporate postings (qualification options, worker type), so a single mirror exercises both list shapes and both detail layouts.

Design

What the real site is. Next.js + AEM front end; results are fetched through an LLM "job search assistant" endpoint whose ordering is non-deterministic; job detail pages are server-rendered with structured jobDetails; login is off-domain OIDC and apply is email + OTP. The mirror keeps the visible affordances (search pill, Filters, Add your location, Sort, tabs, cards, two detail layouts, save / apply) and replaces the non-deterministic backend with plain query parameters.

Data model (single Job table, population-aware):

model	rows	notes
Job	246	population = hourly 168 / salaried 78; brand Walmart 184 / Sam's Club 42 / Vizio 20; every exposed filter value has ≥20 postings; every one of the 7 shift values appears on ≥25
Store	51	6 offices + 45 field locations across 13 states/territories, with lat/lng for radius search; hub stores carry hub_name / hub_blurb / hub_image
Area / Category	7 / 33	hierarchical, mirrors upstream's areaMappings
User	4	benchmark users (alice.j@test.com … david.k@test.com, TestPass123!), werkzeug hashes hardcoded for byte-reproducible seeds
SavedJob / Application	13 / 4	pre-seeded account state (WMC-000001…WMC-000004)

Routes (23): /, /results (q, area, category, brand, shift, type, rate, loc, radius, sort, page; 10 per page), /jobs/<id>, /jobs/<id>/save|unsave (POST, login required), /jobs/<id>/apply → /apply/confirm → /apply/submitted, six /careers-areas/<slug> index pages (+ Military; students 302s to the filtered results), /resources/location, /resources/hiring-process, /resources/terms-and-conditions, /about-us, /login, /register, /logout (POST), /account, /account/edit, /candidate-home/saved-roles, /candidate-home/applications, /_health.

Auth & CSRF. Real password hashing and sessions (no auto-login shims); CSRFProtect on every POST, zero exemptions; every next parameter validated as a same-origin relative path.

Search. Scored token overlap over title + category + area + banner + city + state + brand (description excluded so answer strings never index); stem tier tightened so technician no longer matches Technology.

Answer-leak design. Result cards are faithful to upstream — hourly: title / banner #store / City, ST ZIP / shift • pay; salaried: title / City, ST / pay — and everything a task asks for (requisition ID paired with a second field, street address, open-position count, shift start window, worker type, qualification text) renders only on the detail page. The only visible count is the Open roles tab badge; no task answer is a count of list items.

Determinism. One seeded RNG, fixed reference date, no clock reads on any seed/bootstrap path, hardcoded password hashes, sorted iteration. The seed rebuilds to the same md5 under PYTHONHASHSEED=0 and =1.

Data policy. All postings, stores, users and qualification texts are synthetic; requisition IDs are synthetic and look like upstream's (R-24xxxxx, CP-<store>-1xxxx). Only site chrome (logo, spark, icons, fonts, lifestyle/hub photos) is harvested from the real site. No task is answerable from careers.walmart.com or from prior knowledge.

Registration

• websyn_start.sh and control_server.py: walmart_careers appended as index 19 (arrays verified identical); Dockerfile: EXPOSE 8101 40000-40019.
• Shared docs (README.md, AGENTS.md, CONTRIBUTING.md, CLAUDE.md, agent_demo/README.md): site count 20, port range 40019.
• History note: the branch merges upstream/main (phys_org and target landed as sites 18 and 19 while this was in progress); the site moved 40017 → 40019 in a9ab990, and the docs sweep is 4916213.
• Two small root changes: .dockerignore gains sites/*/scripts_dev/ (dev-only Playwright drivers never ship); root .gitignore gains /scraped_data/.

HF assets

• PR: https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/54 — one file, walmart_careers.tar.gz (11 MB): instance_seed/walmart_careers.db (md5 b57631080969fe151e1717dbef3dd372) + static/images/ (35 files). Zero AppleDouble entries, zero .gitkeep.
• .assets-revision: revision: refs/pr/54 for now (that ref is a strict superset of main); will be replaced by the merge sha in a follow-up commit.
• The two EverydaySansUI font files (176 KB) are tracked in git rather than HF so the mirror renders without the asset bundle; happy to move them if you prefer.

Tasks (20)

- Walmart Careers--0 — Search for Optician roles and open the posting at the Neighborhood Market in Wichita, KS. Report its requisition ID and the street address shown on the posting.

- Walmart Careers--1 — Find the Staff, Software Engineer posting located in Sunnyvale, CA and quote the exact text of "Option 2" under Minimum Qualifications.

- Walmart Careers--2 — Open the Freight Handler posting at eComm Whse Logistics #9054 in Porterville, CA. What shift start window does it list, and how many open positions does it show?

- Walmart Careers--3 — From the Healthcare career area page, go to its open roles and open the Pharmacy Technician posting in Bentonville, AR. What hashtag appears at the end of the "What you'll bring" section, and how many open positions does the posting list?

- Walmart Careers--4 — Using the filters, find Sam's Club Part time roles with a Weekend Overnight shift whose posted pay range tops out at $20.00/hr or less. Open the one in Texas and report its requisition ID and number of open positions.

- Walmart Careers--5 — Find Full time Technology roles in Hoboken, NJ whose salary range tops out above $200,000. Open the matching posting and report its requisition ID and the degree named in "Option 1" of its Minimum Qualifications.

- Walmart Careers--6 — Set your location to Cleveland, OH within 25 miles, filter to Full time roles on a Weekday Day shift, and open the Online Order Filling Team Supervisor posting. Report the street address and the number of open positions.

- Walmart Careers--7 — Filter to the Students career area, the Intern employment type and the Sam's Club brand, then open the merchandising internship in Bentonville, AR. Report the worker type chip shown on the posting and the street address listed for its location.

- Walmart Careers--8 — There are Auto Care Center Technician postings at two Mississippi stores. Open both and report which store number has more open positions and how many.

- Walmart Careers--9 — Two Freight Handler postings are located in Marcy, NY at different facilities. Which one has the earlier shift start time? Report its requisition ID and that start window.

- Walmart Careers--10 — Compare the Senior Manager, Delivery Search, Arrival & Matching (Last Mile Delivery) postings in Bentonville, AR and Hoboken, NJ. Which one requires more years of experience under "Option 2" of its Minimum Qualifications? Report that posting's requisition ID and the number of years.

- Walmart Careers--11 — Log in with the demo account (email: alice.j@test.com, password: TestPass123!), search for Yard Driver roles and save the Williamsburg, VA posting to your Saved roles.

- Walmart Careers--12 — Log in as bob.c@test.com (password: TestPass123!), open your Saved roles, and remove the saved role that is at a Neighborhood Market store.

- Walmart Careers--13 — Log in as carol.d@test.com (password: TestPass123!), apply to the Pharmacy Technician posting in Tacoma, WA using phone number 253-555-0142, and report the confirmation number shown after submitting.

- Walmart Careers--14 — Register a new account with an email and password of your choice, then save the eCom Warehouse Worker posting at eComm Whse Logistics #9046 in Marcy, NY to your Saved roles.

- Walmart Careers--15 — Log in as david.k@test.com (password: TestPass123!) and update your account so the city is Rogers and the state is AR, then open My applications and report the confirmation number of your existing application.

- Walmart Careers--16 — Set the location to Puerto Rico, then use the Shift filter for Weekday Day and the Rate filter for Hourly. Among the Cashier postings, report the requisition ID of the one with the most open positions and that number.

- Walmart Careers--17 — Log in as alice.j@test.com (password: TestPass123!). Exactly one of your saved roles is a Part time job; apply to it with your account email, then report that posting's shift start window and your confirmation number.

- Walmart Careers--18 — Open the Supply Chain and Transportation career area page, go to its Drivers category, and compare the Class A CDL Truck Driver postings at the Ottawa, KS and Williamsburg, VA facilities. For whichever of the two lists more open positions, report its requisition ID, its shift start window and that number of open positions.

- Walmart Careers--19 — Log in as bob.c@test.com (password: TestPass123!). From the Stores and Clubs career area page, open the Digital Pickup and Delivery category, filter it to Full time roles, and open the posting with the fewest open positions. Save that role to your Saved roles, then report its requisition ID and the street address shown on the posting.

Design notes for the reviewer / verifier author:

• Every non-stateful task has a unique detail-page URL as its navigation gate; the answer is always a detail-only field (requisition ID is never the sole answer because card hrefs expose it).
• Stateful tasks leave DB traces: saved_jobs (11, 14, 19), unsave (12), applications (13, 17), new users row (14), users update (15).
• Tasks 13 and 17 both yield WMC-000005 from a fresh reset; verifiers should match the applications row (user, job, phone), not the string.
• Task 15 ends on /candidate-home/applications, reading david's pre-seeded WMC-000004.
• Hard tasks: 10, 16, 17, 18, 19 (multi-page comparison and/or three-hop navigation with a stateful action).

Verification

check	result
python3 -m py_compile sites/walmart_careers/*.py	clean
./scripts/build.sh → run on -p 8201:8101 -p 41000-41019:40000-40019	all 20 ports 200; /health 20/20 alive
POST /reset/walmart_careers	{"pid":213,"ready":true,"site":"walmart_careers"}
md5 instance vs instance_seed after reset	b57631080969fe151e1717dbef3dd372 == b57631080969fe151e1717dbef3dd372
same after docker restart	identical
seed rebuilt twice (PYTHONHASHSEED=0) and once with =1	identical md5
POST /reset-all	1.4–1.7 s, 20/20 ready
Playwright walkthrough, all 20 tasks, control-plane reset before each	20/20 solvable through the UI only
leak audit: home / natural results page / area page / locations, rendered HTML incl. hidden nodes and comments	0 leaks (task 8: store number visible on card, count detail-only)
distractor invariants (build-time): every non-stateful task	≥6 results, ≤50% full matches, target never rank 1
cross-field consistency (FK check, category↔area, store coordinates within state, pay_frequency↔population, Intern⊂Students)	0 violations
smoke (24 flow assertions) / robustness (26)	24/24, 26/26
horizontal overflow at 768 / 1024 / 1440 px on home, results, both detail layouts	none
tarball dry-run	1 DB + 35 images, 0 ._*

Deviations & non-goals

• Maps are deterministic server-rendered SVGs (cluster map on results, pin card on detail), not Google Maps.
• No LLM search assistant / Chat panel, no "Go back" link, no count badges on the Future roles / Content tabs (they render explicit empty states), Filters is a multi-column panel rather than a single scrolling column.
• Auth is local email + password; apply is a two-step contact → confirm form (upstream: OIDC + OTP).
• Upstream serves two salaried detail templates; the mirror implements the masthead variant captured in the reference screenshot.
• (USA) title prefixes are stripped; testimonial copy is mirror copy; no video blocks or carousels.
• Location search additionally accepts a whole state/territory (mirror-only affordance; task 16 also works through a city + 60-mile radius).
• Dallas hub photo is a real Global Tech office interior from the same CMS (upstream has no Dallas image; its dc-metro-area.png is a Hoboken duplicate).
• Job count (246) sits above the 50–200 guideline so that every exposed filter value has ≥20 postings and every task query returns ≥6 results; categories (33) keep a floor of 4 postings each.

Not actioned, with reasoning

• .claude/skills/* port/site-count mentions left as-is (maintainer-owned; the phys_org PR skipped them too).
• AGENTS.md "~3 GB image" statement (already inaccurate on main; unrelated to this site).
• Dockerfile-side seed regeneration (as IKEA does) would let the image build without the HF bundle — a good follow-up, out of scope here.
• CONTRIBUTING nit: the hf upload example lists arguments as <file> <repo> <path>, but the CLI signature is hf upload <repo_id> <local_path> <path_in_repo>.

Review history

Before opening this PR the branch went through four independent read-only audit passes (CONTRIBUTING compliance, guide-page hardening, a reviewer simulation using the review-env skill, and a regression audit for the failure modes caught in #9 / #53 / #71 / #77 / #83). Findings that were fixed: an answer table that had crept into a tracked file, six task targets at rank 1, three under-populated filter values, an over-broad stem matcher, real upstream requisition IDs and boilerplate qualification text on three postings, a 768 px header overflow, and a mislabeled hub photo. The build was driven with Claude Code; every finding was verified against the code before it was actioned.

Screenshots

page	upstream (reference)	mirror
Home
Results
Job detail
Career area (Stores and Clubs)
Sign in

[evanz37]

Port slot note: this site takes index 19 / port 40019 on top of main@90afddb (19 sites). If another site merges first I'll rebase and migrate — port references are confined to the registration files, tasks.jsonl, the site's CLAUDE.md and the shared docs.

[evanz37]

Pushed f7f5bf1: Benefits cards, Life at Walmart, footer icons and plain-text requisition ID / hashtag now match the reference layouts on job detail and shared sections; mirror screenshots in the description refreshed from this commit.