Final integration, page budget, and the regime-map promotion

[duckyquang]

The merge point for everyone's week, so it stays with me:

• Fold in whatever lands: Sparsh's 3D chains and R30 real-systems repair, Akash's wall-clock table, the R15 seed repeats.
• Promote fig_regime_map into the main text once the R28 section reaches final size (the port note documents the swap; it currently lives in the appendix purely for page budget).
• Re-cut to exactly 7 content pages under the official kit after all insertions; appendix absorbs overflow.
• Run the review board one final time on the built PDF (fresh agents, same five personas) ~day 6, fix what they find, freeze.
• Sign off ABSTRACT.md against the final body numbers before the full-paper deadline.

Everything here serializes on the paper source, hence one owner. Est 2 days spread across the week's tail.

[duckyquang]

Status toward this issue, since parts of it landed with the R28 close-out (PR #133, merged) and the review pass (PR #134).

Done

• Re-cut the body to exactly 7 content pages under the official kit; references start on page 8, appendix absorbs the overflow. A duplicate \bibliographystyle that was silently breaking bibtex is gone.
• Ran the review board on the built PDF (five personas) and fixed what they found — overclaims scoped to the evidence, one real logic error in the geometry section corrected. Details in Full-paper-week sprint: review fixes, R30/3D/oracle results, artifact hardening #134.
• ABSTRACT.md signed off against the final body numbers (both the full and the 150-word form versions synced).

Folded in

• Sparsh's third nonlinear family (Third nonlinear relation family for R10 transfer breadth #126) came in through the merge of main into the branch.

Still open (blocking the freeze)

• Nothing else has landed yet to fold: 3D chains (3D pruned chains: take the geometry repair family to the real DMDGP setting #123), R30 real-systems repair (R30: real-systems hardened variants - pick up the pilot and make it citable or kill it #124), and the wall-clock table (End-to-end wall-clock table + checkpoint archive with hashes #128) are still in flight. Once they merge I'll pull them in and re-run the board one final time before freezing.
• Regime-map promotion is deferred, not done. R28 has reached final size (a closed negative), but the 7-page budget has no room to promote fig_regime_map from the appendix into the main text — promoting it costs about a third of a page that would have to come out of caveat prose I'd rather keep. Flagging this as a call to make at freeze: either we find the space by cutting elsewhere, or the regime map stays in the appendix. Leaning toward leaving it in the appendix unless we open up room.

[duckyquang]

Pre-freeze pass done on quang/pre-freeze-fixes (PR to follow): a six-dimension audit with adversarial verification found and fixed 42 issues — headline ones: the n=6 'degrades to 0.250' caveat was a single-seed draw (3-seed rerun reads 0.983 ± 0.014 there; instability is actually n=4, 0.658 ± 0.484 — now stated as such), entrapment CI corrected to the committed JSON's ±0.109 (was ±0.086 in four places), R30's McNemar bound tightened to its true per-seed worst (p<1e-13), README still cited withdrawn v6 numbers, ABSTRACT.md was two re-cuts stale (both forms re-synced; short form exactly 150 words), and the regime map is promoted into the law section — content still ends on page 7, so the promotion this issue deferred now fits. R30 got its RESULTS section + PROVENANCE row. Verifier grew to 35 guarded numbers, all green; 431 tests pass. Left for the freeze: merge the PR, final repro_all.sh verify, re-run the supplementary packaging at the freeze commit (#129), checkpoint upload host decision (#128).