[GoGoKo699]

The common-probe route now has an actual native repair of its relative involution defect. Replacing the middle pulse makes both co-moving probe fibers equal and gives F_s=E_s Z_parity E_s† exactly. The base source has 4m T gates, four T bands, 4pm+20m−12 Cliffords, one arbitrary probe and zero clean inputs. This partial construction does not implement complete lookup.

The complete operator tests separate the remaining failures:

• At fixed native pulse angles, every static bank-diagonal dressing has direct-image error above 1/3 for p≥3. The proof compares exact transition-displacement mass, without extending the previous spectrum argument to the changed word.
• Every nonempty co-moving Hadamard mask has error at least √2 for p≥1. A separately paid cubic-phase/Hadamard interlayer collapses to the wrong Pauli image and has the same floor for p≥3.
• Three probe-asymmetric native repairs have exact nonreal relative entries at p=3. These are finite identity failures, not all-size exclusions. The classical degree observation counts completed coordinate rounds and does not rule out cross-round native synthesis.
• A genuinely four-path Bell/swap/relative-T source pays two-clean normalization, private-target controls and native depth at most 94, but its complete Gram error is exactly 1/2 at every canonical size.
• The bounded six-wire lower scout now excludes all 2^10 T/T† orientations of its fixed middle axes at one-band output endpoints, retaining the 1/4 complete-return error floor. Changed axes and additional non-Clifford endpoints remain outside this result.

Proofs are integrated into the existing source homes. Exact fixtures check native phases, actual inverses, relative fibers, complete coherent witnesses, normalization, transpose conventions and simultaneous ledgers. WORKSPACE.md remains 220 lines and supersedes the old unpaid-involution instruction: the next operation must supply complete lookup-dependent joint bank transitions. The tested single-dressing and swap/weight-phase mechanisms are paused.

R96, the exhaustive three-layer theorem's actual universal constant and tiny-error scope, approximate scalar zero reflection, paid recovery, clean removal and fusion are preserved. No general complete-frame frontier improvement, unrestricted native-depth obstruction or external mathematical review is claimed. Manuscript drafting is unchanged.

Starting main: 7fc9d32fd7bc70764ce09292897e248f315fa387 (PR #242).
Reviewed head: dccf25b66c8374911a7b127e96a2f465aa535b8b.
Reviewed tree: 69cf0de1acfd5cdfd89a2ba114eff3c5838b8d0b.

Focused exact tests, internal mathematical audits, navigation and typography checks passed. The review-tree comparison also exposed and corrected an existing carriage-return typo in the lower proof's maximum-over-Paulis formula. The final reviewed tree passed all ten local verification commands, including 1,716 tests in 1050.312 seconds on Python 3.12.14 / NumPy 2.3.5. Local gate SHA256: 99f9623c4e5f3b9479e47c6521c21fa86c81489f432f4a0a77c7ac12455902d2; validation log SHA256: a48ac4aabded0569163ff4cf96ccd166dc41ecdd360736b4b5e453861dda49df. Final-head CI is complete and successful.

The first verification run was intentionally stopped after rendered review clarified the chronological H-then-cubic-phase wording. The formulas and native tests were unchanged; earlier receipts are archived and will not be counted as final-tree validation.

Final-head focused presentation passed for all eight changed Markdown documents plus the presentation guide, all five diagrams, SVG MathJax and native MathML, at 1280 and 390 pixels. Thirty-six proof/navigation viewports were captured. The transition bound, cubic construction, asymmetric-trial table, Bell source and endpoint proof were manually inspected at desktop or narrow widths. Report SHA256: 3405bc7f96e96f41e7194b9a0c338de0c111e12ea976ba13c2ff226d81a871a0; visual receipt SHA256: fe83870d31edb3badbd1084e3cd2655112e6a7d387fa3b14a044d95abcb401f3.

Final-head CI: PR validation, push validation, and full presentation all completed successfully on dccf25b66c8374911a7b127e96a2f465aa535b8b. All five jobs report success. Four jobs expose 57 successful step records. The push Python 3.11 job's step-summary endpoint returned timeouts/empty metadata; its completed raw log independently confirms the exact head, all 1,716 tests passing in 1820.284 seconds, every later verification command, and final cleanup, with no error markers. Saved log SHA256: b126d0bf93b3c276d4344290b6176a1e1364661037ddb7e4c09ae9232b4f3acc. Missing step metadata is not represented as 16 observed successful step records.

The full presentation artifact is presentation-previews, ID 12566772353, 314,431,967 bytes, digest sha256:08d3f6583e9b6bb51e47277793ee622b7188c8ee6b9591ec37741873fca1ca94, expiring 2026-11-07T16:23:32Z. The PR test-merge commit 64b7ebb7106e1d56af1327b3a73b2d2c31fed2b9 has the same reviewed tree 69cf0de1acfd5cdfd89a2ba114eff3c5838b8d0b.

Merged as 4598492c870b81666f8b149a4c8ccd8efcd8d0b2. Its tree is exactly the reviewed 69cf0de1acfd5cdfd89a2ba114eff3c5838b8d0b, with parent 7fc9d32fd7bc70764ce09292897e248f315fa387. Local main was fetched and fast-forwarded to that commit and is clean. A separate internal CI audit independently confirmed the completed-run/job results and raw-log alternative described above. Combined review receipt SHA256: 83a98dc315088a8b91efa61221f20b90b833345eff741abaa2dc49cba2e7ad16. No post-merge CI result or external mathematical review is claimed.