docs(2026-08-27): morning operating record, CEO brief, and the corrections the evidence forces

[petterlindstrom79]

The two artifacts DAILY-RUN.md requires for 27 August, plus the GOALS.md and
LESSONS.md updates today's evidence forces.

Yesterday produced no run — no handoff, no brief, no commit on main — so
this covers two days.

What the measurement changed

Three corrections to GOALS.md, all through lib/metrics on the canonical
external population:

1. "The card customer is buying the compliance wedge" does not survive.
   They have now bought on three days, the third (08-26) one no session was
   watching. €7.00 of the €10.09 they have spent — 69% — is competitor-compare
   at €1.00 a call. The eleven-capability compliance burst the 08-25 entry built
   its conclusion on was 12 calls for €1.09 in total, and the very next day
   they went back to the expensive capability. Under GOALS' own "follow the
   money" rule this is a competitive-intelligence buyer who sampled compliance
   once.

2. "Seven of their ten euros are unspent, so waiting costs weeks" is false.
   Wallet holds €3.91 — under four more competitor-compare calls.

3. They are assembling N-way comparisons by hand. All three 08-26 calls
   were different domain pairs covering three companies, inside 49 seconds —
   every pairwise combination of a three-company set. €3.00 for something we
   could sell once, and six page renders for three distinct domains. Strongest
   product lead on the board.

Two things closed without escalation

The cache unverified: is closed. The 08-25 handoff could not confirm the
24h result cache in production because STRALE_API_KEY is dead. Production
traffic settled it without a key — same input, 37 seconds apart:

08-25 08:43:28Z  cache_hit=false
08-25 08:44:05Z  cache_hit=true, cache_age_hours=0

Self-second-sourcing: cache_hit exists only in the post-f992fd5 build and
every earlier row has it null, so the rows also evidence which build served
them. Also recorded: the cache cannot help the card customer, who never
repeats a pair.

OpenRegister is not a founder decision. All 36 external
german-company-data calls landed on one day (08-24) and every one came from
e9e672ef719ee934 — the dominant wallet, not a new customer. €1.80 against a
€59/month plan needing ~295 calls/month to break even. Investigated rather than
escalated.

LESSONS.md — F5 incident 8

The orphaned-handoff false positive fixed in #407. Filed as incident 7's
direct consequence: incident 7 left the primary checkout on a stale branch,
and incident 8's predicate reads that branch's index, so the first produced the
second. It inverts the family's usual direction — a check that failed when it
should have passed rather than passing when it should have failed — and the
entry says why it is filed under F5 anyway rather than split across two
families.

Also in the record

• 3 PRs merged, 0 open at end; deployed commit verified equal to main's tip
  at both checks.
• 3 remote branches deleted, each with its commit id recorded first so every
  deletion is restorable; verified against git ls-remote, not the ref cache.
• fix: declare paid preflight output contract #402 is the repository's first outside contribution ever — what was
  checked before merging, and why running the held fork CI was a technical
  rather than a trust decision.
• A job_schedule shape that reads exactly like a silent bulk-operation
  failure and is documented, deliberate behaviour — written down so the next
  session does not spend a morning on it as this one did.
• Twelve branches carry unmerged content against zero open PRs, with owners and
  a deadline. The Italian ownership capability is deliberately not being
  finished, and the record says so rather than letting it appear a fifth time.

check-ceo-brief.ts over all four briefs: 0 errors.

🤖 Generated with Claude Code