build-pipeline: freeze evaluator fingerprint @9f024d0; add terminal sentinel to build.sh

[jjohare]

Baseline — evaluator fingerprint frozen @9f024d0

Measured in the 2026-09-01 dream night (pure-measurement, ACCEPT) via the two checked-in script evaluators:

• build (output tree): exactly 6 entries — CNAME (20 B), css/, img/, index.html (62,523 B), js/, video/.
• build-check: pages: 1 bytes: 59768236 → BUILD-OK (≈ 56.99 MiB).
• links: internal-refs-checked: 9 missing: 0 → LINK-INTEGRITY-OK.
• Derived (A-class arithmetic): non-index.html bytes = 59,705,713 (99.90 %); index.html = 0.1046 % of payload.

Proposal (next build-pipeline night)

Append a success sentinel (e.g. SITE-BUILT / BUILD-COMPLETE) to build.sh stdout so a bare tail proves completion — raw build.sh output currently ends with the video/ line and carries no sentinel of its own; completion-proof lives only in the check scripts. This extends the 2026-08-31 sentinel lesson to the build step itself.

Evaluate the change by byte-identity: build-check must still report bytes: 59768236 and links 9/0 after the patch — a terminal sentinel must not perturb the output tree.

HANDOFFs (sidecar-equipped run)

• npm run test:perf: the 59.77 MB single-page payload (99.9 % non-HTML, presumed media — C-class) will dominate Lighthouse weights; measure before any content decision.
• npm run test:a11y + Rich Results on the 62.5 KB index.html; meta/JSON-LD state still unobserved (2026-08-31 carry-over stands).

Biggest uncertainty

Whether 59,768,236 B is a clean-build figure — stale-output/orphan risk is unverified (single run; nested img//video/ contents unitemized).

---

Filed from the 2026-09-01 dream-cycle report (annexe had no GitHub capability; this is the ready-to-file text). Commit 9f024d0618f5af32fd3a7a809e4b3a93f4b6ae39.