feat/sec-seam-1-session-bound-writes #5 / 8
aaronrene · 40 days ago · Jul 27, 2026 · Diff

SEC-SEAM-1a round 3: ratify D1-D4, close BLOCKER N1, record round-3 blocked

Operator ratification obtained and quoted verbatim in freeze 12.1 before any rule was edited: D1=A (stamp type:'session' at all mint sites; absent = legacy_session, propose-OK but self-apply-ineligible), D2=A (media out of scope), D3=start empty (S10 ships dormant), D4=A (classification reuses the apply path's own predicate).

The paste-ready prompt this round was launched with said "fix N1 per D4 option A", but that text was written by the round-2 session itself. Treating it as ratification would have been an authoring session clearing its own escalation - the defect that reverted SEC-KN-4a 12.1. The session stopped and asked instead.

BLOCKER N1 (security) closed per D4=A. S3 rewritten: isSeamSurfaceProposal now calls normalizeCanisterProposalForTaskPrecheck and the sibling per-surface predicates, so a proposal that can trigger a seam apply is seam by construction. New frozen rule S3.0 forbids any hand-written seam field list in built code - two consecutive rounds wrote a parallel key list and both were wrong in the same way. Evasion and fix reproduced by executing the predicates, not only by reading them.

All 18 round-2 findings addressed: new ground truth G27-G31, 14-step refusal precedence S6.1, corrected guard table (N5), four mint sites incl. both refresh paths (N9), four mcp_access sites (N11), S10.4 qualification (N14), N15 resolved as the audit row being stale, N18 fixed with a corrected line number.

Round-3 independent review = blocked, 11 findings (11.2). It confirmed 15/18 round-2 fixes and that N1 is closed by construction, but disproved three claims this round asserted: V1 (security) a machine-credential mint path does exist and is the premise D3 was ratified on; V2 a fifth learner-session mint site; V3 "no live behavior change" disproved by execution. Loop halted per the skill's security/blocked hard stops. New operator decision D5 recorded (are Flow / Flow-capture seam surfaces?).

Mechanical ok review --freeze = pass, 0 findings; the auto-written review_stamp was removed by hand, since the semantic verdict is blocked.

SEC-SEAM-1b not started. T1-T5 unexecuted. No merge.

sha256:f9737f4e7c965b7a476b6eeada15b943dba8ff9cfb90f7e4a8dd7dfbba8f9421 sha
+23 ~40 −9 symbols
sha256:efa85795edb284a0ade0db129aec66f1456f8c7690db4864d2cddd0eeb682787 snapshot
+23
symbols added
~40
symbols modified
−9
symbols removed
0
dead code introduced
Semantic Changes 72 symbols
~ docs/OVERSEER-HANDOVER.md .md 3 symbols added, 2 symbols removed, 7 symbols modified
NEXT SESSION — SEC-SEAM-1a fix round 3 (PRIMARY) section NEXT SESSION — SEC-SEAM-1a fix round 3 (PRIMARY) L20–86
code[text] variable variable code[text] L46–61
+ NEXT SESSION — SEC-SEAM-1a fix round 4 (PRIMARY) section NEXT SESSION — SEC-SEAM-1a fix round 4 (PRIMARY) L20–141
+ Prior session — SEC-SEAM-1a round 3: D1–D4 ratified, N1 closed, loop blocked at round 3 section Prior session — SEC-SEAM-1a round 3: D1–D4 ratified, N1 closed, loop blocked at round 3 L96–117
+ code[text] variable variable code[text] L75–95
~ table
~ docs/ROADMAP.md .md 3 symbols modified
~ docs/SEC-SEAM-1-SESSION-BOUND-IDENTITY-FREEZE.md .md 20 symbols added, 7 symbols removed, 30 symbols modified
~ table
1 Round-2 findings — full list, none yet fixed section 11.1 Round-2 findings — full list, none yet fixed L646–674
S3 — The seam surface intent set is frozen and closed section S3 — The seam surface intent set is frozen and closed L307–358
1 — Classification is a union, and it is fail-closed class section S3.1 — Classification is a union, and it is fail-closed L323–337
2 — Obligations attach to admission, not as a skippable overlay section S3.2 — Obligations attach to admission, not as a skippable overlay L337–358
code[text] variable variable code[text] L311–318
table@L394 section table@L394 L394–401
table@L408 section table@L408 L408–417
+ 1 Round-2 findings — full list, all addressed in round 3 section 11.1 Round-2 findings — full list, all addressed in round 3 L852–907
+ table@L863 section table@L863 L863–883
+ 2 Round-3 findings — the work list for round 4 section 11.2 Round-3 findings — the work list for round 4 L907–940
+ table section table L917–930
+ table section table L1054–1060
+ D3 — reopened by round-3 V1 (the ratified premise is false) section D3 — reopened by round-3 V1 (the ratified premise is false) L1034–1048
+ D5 — scope — are Flow and Flow-capture seam surfaces? (raised by round-3 V8) section D5 — scope — are Flow and Flow-capture seam surfaces? (raised by round-3 V8) L1014–1034
+ table section table L1025–1029
+ 4 — What S10 does and does not deliver (N14, ratified D3) section S10.4 — What S10 does and does not deliver (N14, ratified D3) L635–655
+ S3 — Seam classification reuses the apply path's own predicate class section S3 — Seam classification reuses the apply path's own predicate L363–462
+ 0 — The frozen anti-drift rule section S3.0 — The frozen anti-drift rule L368–386
+ 1 — The predicate section S3.1 — The predicate L386–432
+ table@L391 section table@L391 L391–398
+ table@L405 section table@L405 L405–413
+ 2 — Why this closes the evasion round 1 left open section S3.2 — Why this closes the evasion round 1 left open L432–451
+ 3 — Self-hosted input shape (N10) section S3.3 — Self-hosted input shape (N10) L451–462
+ 1 — Refusal precedence is frozen (N6) section S6.1 — Refusal precedence is frozen (N6) L541–576
+ table section table L548–564
+ table@L498 section table@L498 L498–504
+ table@L525 section table@L525 L525–534
~ table
~ table
~ table
~ table
~ table
~ table
~ table
~ table

0 comments

No comments yet. Be the first to start the discussion.

To add a comment, use the Muse CLI: muse hub commit comment sha256:f9737f4e7c965b7a476b6eeada15b943dba8ff9cfb90f7e4a8dd7dfbba8f9421 --body "your comment"