Add GOV.UK mirror site (port 40015)

[lamawmouk]

TL;DR

Adds a Flask mirror of gov.uk as the 16th WebHarbor site (port 40015), with topic browse, guidance article detail, department directory, announcements, and search. Uses the official MIT-licensed govuk-frontend v6.1.0 for canonical Design System DOM.

Companion HuggingFace PR: https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/22

What's in this PR

sites/gov_uk/:

File	Purpose
app.py	5 SQLAlchemy models, 9 routes
seed_data.py	Idempotent seed: 16 topics, 44 subtopics, 15 departments, 73 articles, 20 announcements
templates/*.html	base + 9 page templates using canonical govuk-frontend DOM
static/{css,js,fonts,icons}/	Official govuk-frontend v6.1.0 bundle (MIT)
tasks.jsonl	Stub; WebVoyager tasks in follow-up

Registration (sync per AGENTS.md): gov_uk added to websyn_start.sh and control_server.py, Dockerfile EXPOSE bumped to 40000-40015.

Verification

All checks in AGENTS.md § Pre-PR checks pass: image builds clean, 16/16 sites alive, every gov_uk route returns 200, POST /reset/gov_uk byte-identical pre/post (md5 f6931b6c…), and identical after docker restart.

Notes

• Content is synthesized (no upstream copy); OGL v3.0 would permit direct copy but synth keeps the seed at 128 KB and deterministic.
• govuk-frontend.min.css only patched with one sed to rewrite url(/assets/...) → relative paths so they resolve through Flask's /static/.
• .assets-revision still points at main; will bump to the HF merge SHA after that PR is reviewed.

[lamawmouk]

@Raibows would you be able to review this when you have a chance? Thanks! 🙏

[YuanDaoze]

Review: gov_uk

I ran the full pipeline in an isolated git worktree using alternate ports 42000-42015.

Mechanical / Visual / Functional all PASS, but Task Quality is N/A because tasks.jsonl is empty. See notes below.

✅ Mechanical checks: PASS

• 16/16 sites return HTTP 200, including gov_uk on port 42015

• Control-plane /health reports all 16 sites alive

• Byte-identical reset PASS

  • md5 == f6931b6c99df43d3ce1ffbef4566d2ef
  • Verified at startup, after /reset/gov_uk, and after /reset-all

• /reset-all across 16 sites completes in parallel in ~1.3s, under the 10s requirement

• Reasonable seed scale:

  • 16 topics
  • 44 subtopics
  • 15 departments
  • 73 articles
  • 20 announcements

✅ Visual fidelity: PASS

This site uses the official MIT-licensed govuk-frontend v6.1.0, and the homepage, detail pages, and search page closely match the real [gov.uk](https://www.gov.uk/).

Checked pages include:

• Home page

  • Crown logo
  • “Welcome to GOV.UK” hero
  • Popular-on-GOV.UK links
  • 16-topic 4×4 grid
  • Departments / Announcements two-column footer
  • Open Government Licence v3.0 footer

• Detail page: State Pension forecast

  • Breadcrumb: Home > Working > State Pension
  • DWP department label
  • Related content sidebar
  • Published / Last updated metadata

• Search page

  • Tri-category grouping:

    • Guidance and services
    • Announcements
    • Departments

  • Each result has a short summary, department, and updated date

The “Mirror” banner correctly identifies this as a benchmark mirror.

Note: having 0 <img> elements on the home page is not a bug. The real gov.uk is also an almost text-only portal with just an SVG crown, so this is a deliberate and faithful departure from e-commerce-style mirrors.

✅ Functional depth: PASS

• Search quality is good:

  • passport → 12 results across 3 categories
  • tax return → 6 results
  • drive → 6 results
  • Search uses token-overlap scoring rather than strict AND matching, which works well here

• 8/8 sampled detail pages have populated content with more than 200 characters

• 4/4 sampled navigation links return non-404

• No authentication system is intentional and correct

  • The real gov.uk also has no on-site user accounts
  • It delegates account-related flows to Government Gateway
  • This matches the upstream design

❌ Task quality: N/A

sites/gov_uk/tasks.jsonl currently contains only:

[]

The PR description acknowledges that “WebVoyager tasks” will come in a follow-up, so Phase 2 is not done yet.

This is the most important step in the review-env skill, specifically Step 5. Without tasks, I cannot audit:

• answer leaks
• distractor density
• task difficulty
• whether the site is benchmark-ready

Recommendation:

• Either add 15-20 tasks in this PR before merging, per the contribution guide
• Or, with maintainer agreement, split the tasks into a follow-up PR

However, if tasks are split into a follow-up, that follow-up should land before the author-list cutoff. Otherwise, this contribution should probably count as infrastructure scaffolding rather than a complete benchmark-ready site.

⚠️ Required fixes

1. .assets-revision should point at the HF PR ref, not main

Currently .assets-revision points to revision: main, but gov_uk.tar.gz is still in HF PR #22 and has not been merged yet.

Because of this, when a reviewer runs:

./scripts/fetch_assets.sh

only the existing 15 sites are pulled, and gov_uk is missing.

I had to manually fetch from refs/pr/22 in order to build and review the site.

Suggested fix, following the pattern used by other paired PRs:

repo: <your-hf-fork>/WebHarbor
revision: <commit-sha-containing-gov_uk-assets>

Then, once HF PR #22 is merged, bump it back to:

repo: ChilleD/WebHarbor
revision: <merge-sha>

2. The tarball contains macOS AppleDouble files

Extracting gov_uk.tar.gz writes 7 ._* files into sites/gov_uk/, for example:

• ._instance_seed
• ._gov_uk.db
• ._.gitkeep

It also emits warnings such as:

Ignoring unknown extended header keyword 'LIBARCHIVE.xattr.com.apple.provenance'

Suggested fix:

Re-pack on macOS with:

COPYFILE_DISABLE=1 tar --no-xattrs -czf gov_uk.tar.gz ...

or re-pack the tarball from Linux.

Summary

The infrastructure is solid:

• visual fidelity is strong
• functional depth is good
• byte-identical reset passes
• data modeling looks reasonable
• using govuk-frontend was a great call

The main blocker is the empty tasks.jsonl, which prevents this site from being benchmark-ready.

Once the tasks are added and the two asset issues are fixed, I would be happy to approve.

Looking forward to the follow-up.