chore(run-17): acceptance fixture — first P0a-lane action, hedge-free resume

[arcavenai]

Why

Run-17 is the first board where the P0a silent-data-loss lane moved: #342 drew a maintainer-authored fix (vsdd-factory PR #759, E-21 W1) with no measured-side diff in existence — the single most decision-relevant datapoint since the falsification protocol was registered, and it cuts against the strong form of our own fix-delivery-cost hypothesis. Committing this fixture makes run-17 the new shape canon for the regression gate, so the next refresh is compared against a body that carries the P0a-action rendering, the partial A4 discharge (#342 out, #365/#358 residue at 6/6 windows), and the run-17 classification index. Without it, the gate would keep measuring against run-16r3 and silently tolerate regressions in the newest sections.

What

• docs/fixtures/vsdd-factory-312-curated-run17.md — the posted run-17 body (watermark 753→758; 4 new issues incl. P0b #756; 10 merges + 4 closes; ceiling P0a; verdict 🟢 on-course v2).
• docs/fixtures/vsdd-factory-312-run17-classifications-rich.json — full axis records for #755–#758 (graded in the interrupted prep session, sync batch, zero hedges).

Verification

• §3 regression gate passed clean against both the before-snapshot and the run-16r3 canon: section presence + order, no heading loss, no coverage shrinkage (4 dropped numbers verified as GitHub closures — and retained anyway), no _unclassified_ rows, §7d render-integrity contract (blockquote/table spacing, attn rows visible, no tables in <details>).
• Posted body re-fetched and diffed: matches modulo CRLF normalization.
• Falsification meter reported per aae-orc#65: ramp prediction stays FAILED (evaluated run-16, permanent); ceiling movement one window later reported descriptively, not re-scored.