[ShixiangWang]

Closes the governance half of #74 (repair-order step 5: ROADMAP sync + the P8-vs-deferred ruling).

Why

The parity campaign's phases are named P0-P7 only in vignettes/articles/research-09-sigminer-v3-parity-plan.qmd section 5. The roadmap tracks M/G milestones. A reader who lands on the roadmap cannot locate a phase in the version plan without re-reading the plan — #74's first charter line: "The P-series parity campaign (P0–P7), including all current-branch work, lives outside the ROADMAP M/G vocabulary."

#74's second charter line: "P8 treatment-forensics collides with the deferred list ("therapy forensics" is a hard v1.x non-goal) yet ms_treatment_report() shipped (PR #56) with no PI adjudication and no ROADMAP revision."

What

vignettes/articles/roadmap.qmd gains one additive section after the dependency graph (line 270 onward; nothing above it moves, so every repo-wide roadmap.qmd:NNN pin still resolves):

• a P0-P7 index table: phase -> landing repo -> this roadmap's milestone -> acceptance gate. Rows are transcribed from the parity plan section 5 (landing repo, gate) and section 4 (the milestone each phase actually advances). P7 is marked "campaign, not a PR", operator-gated.
• the plan's own sequencing sentence.
• the P8-vs-deferred adjudication: the shipped ms_treatment_report() is the Pillar-3 case-study MVP surface (a report over already-extracted exposures with committed synthetic fixtures); the v1.x non-goal continues to cover the full therapy-forensics program. The deferral line stands as written and is read through the new paragraph. Moving that boundary either way is a PI call, and the collision is now recorded rather than silently absorbed.

vignettes/articles/performance.qmd re-pins the roadmap digest in the integrity seal (5147c7da... -> 51ce5888...) and names the superseded digest in the census-exempt waiver with its reason, per the pinned-byte source-audit law.

No code, no workflow, no gate changes. Documentation only.

Constraint checks (all run in the lane, all green)

• Rscript tools/docs-sync.R -> rc 0. Half one (L-H API-surface drift): every ms_* token named in the new section (ms_select_k, ms_qc_report, ms_update_refdb, ms_treatment_report) is already a row of the CAPABILITY-MATRIX L-H table. Half two (checkbox truth): the new heading contains none of M6s, M7\b, G-gates, so the section is out of in_scope and its unchecked boxes are not policed — the 4 in-scope checked boxes from before are unchanged. Half three (pkgdown index): 84 exports, 96 entries, OK.
• Rscript tools/cjk-census.R -> rc 0; 577 files, 0 normative CJK lines outside quote/code context. The roadmap is classed normative, so the new text is English-only.
• testthat::test_file("tests/testthat/test-docs-sync.R") -> [ FAIL 0 | WARN 0 | SKIP 0 | PASS 15 ]. The red half asserts expect_length(grep("ms_hrd_report", lines), 1L); the new text does not spell that token, so the census stays at one line.
• python3 tools/ci-selfcheck.py -> PASS (all CI wiring assertions hold).
• Seal round-trip: shasum -a 256 vignettes/articles/roadmap.qmd equals the digest now pinned in the performance.qmd fenced block (51ce5888...), and the superseded 5147c7da... is still named exactly once in the waiver.

Test plan

• CI: docs-sync + cjk-census guard shards stay green on this PR
• CI: the three R-CMD-check-fanout legs stay green (docs-only change)
• post-merge: attest the main push run, then close [audit] 2026-11-04 full-codebase audit — consolidated fix backlog (1×P0, 7×P1) #74