[frankbria]

Brings the landing page and the README into line after #1531.

1. The TDD contrast on the site

#1531 put this on the README's first screen, under the demo. The site had the same headline, install line and demo, but not this:

TDD protects your code. This protects your tests. Your agent can edit a test as easily as it can edit the code, and when a test is inconvenient that is often what it edits. A requirement is a test it cannot quietly drop: releasing one takes a waiver with a name, a reason and an expiry date.

Placed after the figure, inside <header>, mirroring the README. .tdd has no stylesheet rule, so it renders at body size and colour rather than the muted .lede — deliberate, since it is the punchline.

2. One verified basis for the figures, on both surfaces

The README and the site stated different figures for the same thing, and one of the site's had gone stale.

	README before	Site before	Measured	Both now
Tests	5,000+	7,500+	6,035 passed	6,000+
Coverage	80% floor	85%+, 80% floor	86.13%, floor 80	85%+, 80% floor

The site's 7,500+ was correct when written (7,891 at d26a5c8, 14 Oct) and is no longer true. The README's figures were true but undersold.

Source: the Backend Unit Tests job on main @ 507c6e0c, 2026-10-28 — 6035 passed, 38 skipped, 2 deselected and Required test coverage of 80.0% reached. Total coverage: 86.13%.

Rounded down, following the site's existing convention, and both places now carry an HTML comment with the measurement and a pointer to the other file, so the next person can re-check rather than trust.

Verification

• uv run pytest tests/ci/ -q → 81 passed
• <blockquote id="limitations"> untouched; still pinned to the README's beta note
• No limitations list added to the site (decided: not yet)

🤖 Generated with Claude Code

https://claude.ai/code/session_019cEymaSpmZ1arjngxx1VNM

[claude]

Review: LGTM. This is a docs-only change with no code or security impact.

• The site and README now state the same figures (6,000+ tests, 85%+ coverage, 80% floor). Both are rounded down from the measured 6,035 passed and 86.13%. Both files carry a comment with the source and a pointer to the other file.
• The .tdd paragraph mirrors the README copy. The PR says .tdd has no stylesheet rule on purpose, so it renders at body size.
• The README's HTML comment sits directly above the paragraph and doesn't render. It does not touch the #limitations blockquote.
• The count dropped from 7,891 to 6,035 since 14 Oct. That is plausibly the [P2.79] Remove the web UI (decision recorded) #1477 web/server removal, and the PR explains the new basis, so the lower figure is fine.
• I didn't run the tests. The PR reports tests/ci/ at 81 passed.

Optional: nothing checks that the README and site figures agree, so a small tests/ci guard could stop them drifting again. That isn't needed for this PR.

🤖 Generated with Claude Code

[claude]

Re-review: no changes since my earlier LGTM, and I still have nothing blocking. This is a docs-only diff. The README and site figures now match (6,000+ tests, 85%+ coverage, 80% floor), and each file has a comment saying where the numbers came from. The earlier optional suggestion stands: a small tests/ci check that the two files agree would stop them drifting again. I didn't run the tests.

🤖 Generated with Claude Code

[github-actions]

Claude finished @frankbria's task in 1m 24s —— View job

---

Third-party review — ✅ no defects found

• Read the diff (README.md, site/index.html — docs-only)
• Verify figure parity across both surfaces (6,000+ tests, 80% floor; no stale 7,500+/5,000+/85%+ anywhere)
• Verify the [P1.30] Fix CI gates that pass without verifying anything #948 guard still holds (coverage-80%25 badge present; all coverage claims ≤ fail_under = 80)
• Verify TDD copy mirrors the README verbatim and the HTML placement is valid
• Post summary comment

Docs-only change. Figures are consistent on both surfaces and rounded down from a plausible measurement (5,290 static test definitions + parametrization ≈ the cited 6,035 collected; I could not re-query CI from this environment). The #948 coverage-claim guard remains satisfied. Summary posted as a separate review comment.
· Branch: site-tdd-line

[claude]

Re-review: the diff has changed since my earlier LGTMs, and I have nothing blocking. It is still docs-only.

• The site no longer says 85%+. The stat is now 80% / "test-coverage floor, enforced in CI", and both HTML comments say coverage is stated as the floor and never the measured value. That fits the tests/test_lifecycle_gates_948.py rule (README coverage claims must match fail_under). It also means the PR description is stale: its table and "Both now: 85%+, 80% floor" no longer match the diff. Please update it.
• Minor: in site/index.html the two comment lines about tests/test_lifecycle_gates_948.py lost their indentation. It doesn't render, but it is inconsistent with the lines around it.
• Drift guard: my earlier optional suggestion stands. Nothing checks that the README and site test counts agree, and the [P1.30] Fix CI gates that pass without verifying anything #948 test only covers coverage in the README. A small tests/ci check would cover both. This is not needed for this PR.

I didn't run the tests.

🤖 Generated with Claude Code

[github-actions]

✅ Third Party Review: no defects found.

Docs-only diff (README.md, site/index.html). Verified against the checked-out tree:

• Figures agree on both surfaces (6,000+ tests, 80% coverage floor) and no stale figures (7,500+ / 5,000+ / 85%+) remain anywhere in either file.
• Rounded-down claim is plausible: 5,290 static def test_ definitions plus parametrization is consistent with the cited 6,035 collected; fail_under = 80 confirmed in .coveragerc.
• The [P1.30] Fix CI gates that pass without verifying anything #948 guard (tests/test_lifecycle_gates_948.py) still holds: the coverage-80%25 badge is present and every coverage claim the regex extracts (80, 80) is ≤ the enforced floor.
• The TDD paragraph on the site mirrors the README wording verbatim; placement inside <header> after </figure> is valid HTML, and no .tdd stylesheet rule exists, matching the PR's stated intent.
• CI figures (6,035 passed / 86.13%) could not be independently re-verified (no CI API access from the review environment), but the claim is rounded down and internally consistent.

Model: k3[1m]