Re-measure the audit and proposal against the fixed code

[R204570]

Both documents were still written as if nothing had been fixed. The audit
described its findings in the present tense — "Live, right now: pydantic →
1.10" — and the proposal's acceptance table said "Today: 3/8". Neither is true
any more. Re-measured against 316c7f2 and updated.

AUDIT.md — now a two-pass document

Sections 1 to 5 are left exactly as written. They are the record of what was
true before anything changed, and rewriting them into the past tense would
destroy the evidence the fixes were built from. They are marked as-found.
Section 6 marks each planned item done or not. A new section 7 carries fresh
measurements:

	Before	Now
Resolution	3 correct / 3 wrong	7 correct / 0 wrong
The seven index technologies	155,442 chars	19,116,128 reachable (123×)
read_knowledge_base("pydantic")	24 blocks of 1.10	85 blocks of 2.11
is_forge("gist.github.com/…")	False	True
Tests	284	346 across both backends

§7.1 also records something worth noticing: fastapi and astro resolve
through the registry and are still correct. Domain-first is a preference,
not a rule — the identity checks are what make the registry path safe.

The thing the fix status must not blur

The code is fixed. The stored corpus is not. All seven affected
technologies were harvested before any of this, so they still hold their
indexes and still read complete: True.

That is stale data rather than a live defect, but it is called out at the top,
in §7.3, and in the summary table — because an audit that says "fixed" while
the store still shows the symptom would be exactly the kind of unearned
confidence this whole exercise was about. A re-harvest clears it and takes
the store from 8.4 M to roughly 27.5 M characters. It is the one action
outstanding.

§7.2 — three defects found by running, not reading

Recorded because two of them were regressions introduced by the fixes and
would otherwise have shipped:

• astro resolved to an astrology site. It owns astro.com, it is
  enormous, and it says "astro" constantly — every signal a name-plus-size
  check has, and none of the ones that matter.
• terraform.com, an unrelated company, outranked terraform.io on page
  size.
• F5 was half-fixed. Three version lookups ordered correctly; a fourth,
  PostgresStore._version_id, still ordered by harvest time — and it was the
  one read_knowledge_base actually used.

The first two came from the live fixture, the third from querying the real
database. Each looked correct in the source. That is a better argument for the
fixture than anything the document says about it in the abstract, and it is now
the closing point of the summary: the fixes are worth less than the thing that
will catch the next one.

PROPOSAL.md

• Acceptance criteria gain an achieved column — seven rows met, two openly
  not.
• Phases A, B, C and D3 marked done; D1 marked plainly not done. A long
  harvest still blocks past MCP client timeouts, and that is an architectural
  change rather than a fix.
• Two open questions answered by having built the thing. Chunk size is
  settled — heading level chosen by result, giving ai-sdk 2,605 pages and hono
  440. Conflict presentation turned out to be the wrong question: domain-first
  returns before consulting the registry, so it never has both answers in hand.
  The real conflict case was two live domains, settled by evidence.
• One new question opened in their place: what expected should compare
  against for a full dump. Making the sitemap count authoritative risks crying
  wolf; leaving it advisory misses a class of partial harvest. Undecided, and
  deliberately so.

Verified

• 346 passed, 22 skipped across both backends. No code changed — documents
  only.
• All four mermaid diagrams rendered in Chromium via mmdc, not parsed in Node,
  which reports false failures on flowcharts.
• Every number was measured today against the live store and live network.