Write the gate-result memo — the kill/continue document

[myselfsiddharth]

Context

The founder has to make a kill-or-continue decision. The input cannot be an NDJSON file and a report directory — it has to be one document that states what was measured, under what caveats, and what the measurement means against the PRD section 9 gates.

This is the artifact the whole project has been building toward. docs/README-narrative.md names it as the outstanding item; every [PENDING TRACK-1] marker in docs/pitch/proof-points.md resolves here.

Depends on: the full matrix run issue.

What to write

docs/gate/gate-result-<YYYY-MM-DD>.md, standard frontmatter, doc_type: gate-result.

1. Verdict up front. PASS / EXTEND / FAIL against section 9, in the first line. A reader who stops after one sentence should have the answer.
2. The numbers, each with its measured value, numerator, denominator, and the artifact path it came from. Any quantity that came back no_data is listed as no_data — not omitted, because an omitted metric reads as an inconvenient one.
3. Which document controls. PRD section 9 gives a four-part gate (80% / 90% / ~50% / 70%); the pivot brief section 7 restates a single "~50%" kill line. docs/prd/pivot-brief-v0.3.md Open questions asks which governs when Track 1 reports. Answer it here, explicitly, and note that docs/INTEGRITY-AUDIT.md B-05 / D-03 flag the thresholds as proposed rather than accepted — so state whether you are treating them as binding, and on whose authority.
4. Caveats, unhedged. In particular:
   • Version-bump churn is a proxy for organic production churn (docs/gate/testbed.md), arguably harsher and different in kind. Pivot brief section 4 requires confirming against one live site once standing is unambiguous — say whether that happened.
   • Any version that could not run, and the reduced denominator that follows.
   • The load-bearing step count from the assertion audit. If most steps carry weak assertions, the headline number is softer than it looks and the memo must say so.
   • Run and step-execution counts against section 9's ≥42 / ≥400.
5. What this does and does not settle. It settles whether compiled trajectories survive controlled version churn on one OSS console. It does not settle whether they survive organic churn, whether the counterparty thesis is alive (ADR-0004 closed that), or whether anyone will buy this.
6. Handoff. Link the founder decision issue and state the 3-business-day clock from docs/research/vertical-search/DECISION.md if the verdict is PASS.

Constraints

• Every number cites an artifact path. If it is not in a committed metrics row or report file, it does not go in the memo. This is the document most likely to be quoted outside the repo, so it is the one where an unsourced figure does the most damage.
• Do not soften a FAIL. CONTRIBUTING rule 4. A FAIL memo is a successful outcome for a falsification exercise — Week-0 and Track-2 both produced FAILs and both are treated in-repo as assets, not embarrassments. Write this one the same way.
• Do not re-run the gate to get a better number. If the run was flawed, say the run was flawed and what would fix it.

How to test

npm run lint:docs

Then check every number mechanically — for each figure in the memo, open the cited artifact and confirm it matches. Have the check be reproducible: a short script under scripts/ that reads the report JSON and prints the memo's headline figures is better than trust.

Apply the reader test: hand the memo to someone who has not read the PRD and confirm they can state the verdict, the main caveat, and what happens next.

Before you open the PR

• Verdict in the first line
• Every number cites its artifact path and was checked against it
• no_data quantities listed, not omitted
• The PRD-vs-pivot threshold question answered explicitly
• Proxy caveat, reduced denominators, and load-bearing step count all stated
• No number that is not in a committed artifact
• docs/README.md gate table row added
• Founder decision issue linked
• Branch track1/b4-gate-memo