[amaybaum]

Round: MATH-CON-1-S1B-1 (MATH-CON-1 Stage 1B, first stage: O-1, O-2 and O-3a–e only), a native V3 round under AGENTS.md §A.39, non-sealing. Planning Issue: #1169.

Drafting snapshot D: fd539e150759a6ea51ab279aeeac80fc06050ade (main at the cut; #1206 landed before it as 2cd49dc0). D's push run 42574043510 is green.

This pull request carries the control plane only. Its two single-parent commits add verification/programmes/oi-qm/reconstruction/round-mathcon1-s1b-1/preregistration.md: the independently accepted r13 draft, with the D-dependent fields filled and the owner-approved two-line Authority edit (#1169 c.6740495860). The V3 blocks are unchanged, and the governed-paths digest is 81897a90a068b0402e68393f82cffda8fd83f0bda5283316c38d8c58b28e5107. Six fields (PLAN items 3–8) stay PENDING until the last pre-F commit, and control P refuses each of them at F.

Commits: 2d2bf21df62a72befaec433bfed2914ad143b597 (parent D), which amends the unpushed e067870e5eee74e49b05cb59c55917be9d5809c1 under #1169 c.6740495860; and head aeab48e3662f3eb1302ff590f037e8d68319d037 (parent 2d2bf21d), which fills PENDING item 12 from the first comment (c.6740777773) and changes nothing else.

Independent exact-head review at aeab48e3: REVIEW-EXACTHEAD-aeab48e3.md, sha256 e0a373342cf604058ff8b5a234d0b2a3113c419784580b6f6d90077af542004c (model review), verdict ACCEPT WITH NITS, posted as c.6741128215. PR CI on aeab48e3 (run 42622568439, pull_request): 26 success, 1 skipped matrix placeholder.

Independent pre-push check: REVIEW-PREPUSH-2d2bf21d.md, sha256 7a7518599fbdcb094e15474deba19f01f67bfee52d8646230ed889ab1449166b, is a model review, not a human one. Its verdict is PASS (6/6), with one non-blocking wording note (N1) that needs no change. It confirms:

• the commit has one parent, D, and adds one file;
• the diff from r13 is only the D-dependent fills plus the two approved lines;
• the blocks are byte-identical, and precondition 7 exits 0 with the expected digest;
• SC-42 exits 0, matches its committed outputs, and fails with two hits on e067870e as a control;
• seven PENDING fields remained at 2d2bf21d (six at aeab48e3, after item 12);
• line 81's elision is faithful.

Authorization: #1169 comment 6740116927 (amaybaum, 2026-11-09): "Push the accepted r13 round to a branch, open a draft PR (ordinary PR CI runs, including the Lean builds), and run an independent exact-head review. No dispatch, F or merge." The two-line edit was approved in #1169 comment 6740495860.

Not authorized: any workflow_dispatch (including the F exact-head run), the designation of F or E, and the merge. Ordinary pull_request CI on this draft is not an F attestation (§A.39).

Push triggers checked (§A.43 item 4). The diff is the one preregistration file. verify.yml and hidden-sector-replay.yml push only on main. bg2-orientation-design.yml and bridge-compatibility-research.yml are path-filtered on paths this diff doesn't touch. The other workflows are dispatch-only. No governed launch is triggered.

Related: #1206 (the design record, merged).

---

Generated by Claude Code

[amaybaum]

Owner decisions this round relies on (quoted with author and time)

Each quote below is copied byte for byte from the GitHub record named with it. The account amaybaum posts the coordinating session's records; each record states that it quotes the owner's chat words.

1. Authorization of this pull request — #1169 c.6740116927 (amaybaum, 2026-11-09T12:31:12Z)

MATH-CON-1 draft PR — Push the accepted r13 round to a branch, open a draft PR (ordinary PR CI runs, including the Lean builds), and run an independent exact-head review. No dispatch, F or merge.

2. The halt-only decision on VOID handling at E — #1169 c.6740495860 (amaybaum, 2026-11-09T12:51:12Z)

The record states that the coordinating session asked the owner (amaybaum) this question on 2026-11-08 at 14:35:19Z, quoted verbatim there:

MATH-CON-1 round draft r10 now has an independent ACCEPT. One rule is still open. Your step-0 decision says: 'If the checker fails at E, halt, preserve the evidence, and request a separate decision. Do not silently repair or rerun.' At present the draft halts at E when a check fails or a countercontrol fails. But if the checker crashes and gives no verdict, the draft allows the evidence to be retaken by re-dispatch, with your approval. The reviewer recommends halt-only: a crash is deterministic, because the checker blob and inputs are pinned, so rerunning can't fix it, and fixing it would be a repair. How should a checker crash at E be handled?

Halt-only (Recommended): Any result at E other than a clean pass halts the round and preserves the evidence. No checker rerun path is pre-registered. Runs voided by the host (a cancelled run, a lost runner) can be retaken only by a separate decision you record after the evidence is preserved.

The owner answered "Halt-only (Recommended)" at 2026-11-08T15:04:42Z. The preregistration's Stop rule implements it.

3. The two-line SC-42 edit — same record, #1169 c.6740495860 (amaybaum, 2026-11-09T12:51:12Z)

Owner decision: the SC-42 fix (2026-11-09)

The round PR has not been pushed. The local round commit e067870e sits on D = fd539e15. Every local check passes except the SC-42 added-text scan, which CI runs in "Numerical probes / core B". The scan rejects the word "scratchpad" twice, both times in the accepted r13 preregistration's Authority section:

• line 73 describes the earlier one-word correction;
• line 81 quotes the owner's authorization of that correction.

The owner answered "Two-line edit (Recommended)", approving this edit:

• line 73 is reworded without the word;
• line 81 becomes "…the one-word […] → review correction", with the full words cited to MATH-CON-1: preregistration on PolC + B23 (WIP / UNVERIFIED) #1206 c.6722716262;
• an independent check confirms that only those two lines changed.

After that the authorized steps continue: push, open the draft PR, run CI, and an exact-head review.

The following remain unauthorized: the F exact-head dispatch, the F and E designations, and the merge.

4. Stage authority — #1169 c.6721662886 (amaybaum, 2026-11-08T12:46:00Z)

1. MATH-CON-1 Stage 1B: option (a) approved

Authorize the first stage only: O-1, O-2, and O-3a–e.

The objective is to establish whether the proposed construction works mathematically at this level, while explicitly recognizing that Hadamard is supplied rather than derived.

Conditions:

• Freeze F before execution.
• Require exact-head CI at F and E, with the prescribed receipt.
• Do not execute O-4 through O-10 without separate authorization.
• Do not claim that OI independently generates the quantum operation.
• A successful first stage establishes only a candidate for outcome (C). The complete outcome still depends on the remaining obligations, including valid instruments and repeatable composition.

The design is the MATH-CON-1 preregistration, accepted at 41a42cb9 (#1206).

2. MATH-CON-1 §4.5: all three choices accepted

Approve the three design decisions as recorded. In particular, I agree that characterizing Hadamard indirectly still counts as installing it. Uniform closure laws may remain unmarked, provided they do not explicitly identify a particular non-monomial matrix.

Important distinction: passing the calibration control must not be presented as evidence that such a closure law is independently physically justified. That remains a separate question.

Overall direction

The owner distinguishes three questions:

• Mathematical construction: can the extended framework reproduce the desired quantum operation?
• Independent physical justification: can the needed rule be motivated without inserting the desired quantum answer?
• OI derivation: does that rule follow from OI itself?

MATH-CON-1 is currently addressing the first question. Success there must not be promoted to success on the second or third.

5. Step 0: #1206 merge, sibling record directory, writer — #1169 c.6722109114 (amaybaum, 2026-11-08T13:30:26Z); the same record is #1206 c.6722109292 (2026-11-08T13:30:27Z)

The owner (amaybaum) decided this in chat with the coordinating session on 2026-11-08: "I approve the recommended path for all four decisions, with the existing scientific and execution safeguards preserved."

Decision	Authorization
Merge #1206	Approved. Preserve 41a42cb through a normal merge commit, unchanged.
Sibling directory	Approved. Use round-mathcon1-s1b-1/ and record the explicit deviation from §10.1.
Designated writer	Approved. The proposing Claude Code session is designated writer, limited to O-1, O-2 and O-3a–e.

The existing r2 / WIP header may remain unchanged. Record the accepted status and exact commit in the round's preregistration.

The Stage 1B round may proceed to PR preparation only after the revised round draft receives an independent ACCEPT and the design is integrated into main.

N6.

I do not approve running the actual exact-arithmetic checker on the real objects before F. … If the checker fails at E, halt, preserve the evidence, and request a separate decision. Do not silently repair or rerun the frozen round.

Execution boundaries.

The estimated 18–20 CPU-hours are acceptable as a planning budget, not as blanket execution authorization.

Each of the following still needs its own approval:

• the exact F designation and its controls;
• each CI dispatch at F, E, and optionally S1;
• the E designation and any re-dispatch;
• the final landing merge.

Most importantly, Stage 1B must not be described as deriving quantum mechanics from OI. The Hadamard operation is supplied by the proposed rule. A successful first stage establishes, at most, mathematical support for a candidate circular construction (C). It does not establish a final outcome or resolve the remaining instrument, composition, and noncircularity obligations.

No research execution, freeze, CI dispatch, or additional merge is approved beyond the explicit decisions above.

6. The DESIGN.md label correction — #1206 c.6722716262 (amaybaum, 2026-11-08T14:28:45Z)

Owner authorization (amaybaum, in the coordinating Claude Code session's chat, 2026-11-08; quoted verbatim):

Yes, you have my authorization to make the one-word scratchpad → review correction.

This was explicitly approved in my previous decision.

The authorization includes creating a new commit, obtaining independent confirmation of the editorial-only change, requiring green exact-head CI, and performing a normal merge after checking compatibility with current main.

Preserve the accepted 41a42cb9 commit in the ancestry. Verify green post-merge main CI.

No additional design edits are authorized.

Not authorized

The F exact-head workflow_dispatch, the designation of F or E, any other dispatch, and the merge.

---

Generated by Claude Code

[amaybaum]

Exact-head review of aeab48e3 (independent model reviewer, not human): ACCEPT WITH NITS

Object. aeab48e3662f3eb1302ff590f037e8d68319d037, the PR head and the remote branch head. Chain: D = fd539e150759a6ea51ab279aeeac80fc06050ade → 2d2bf21d → aeab48e3. Review record: REVIEW-EXACTHEAD-aeab48e3.md, sha256 e0a373342cf604058ff8b5a234d0b2a3113c419784580b6f6d90077af542004c.

There is no blocking finding.

#	check	result
1	Commit structure	PASS. The chain is linear and every commit has one parent. 2d2bf21d adds only …/round-mathcon1-s1b-1/preregistration.md, and aeab48e3 modifies only that file. D is main's tip and is on its first-parent chain.
2	Final commit	PASS. One line changes: line 59, PENDING → #1215 comment 6740777773 (amaybaum, 2026-11-09T13:05:43Z). It matches the first comment's REST id, user and created_at exactly.
3	Content	PASS. Both V3 blocks are byte-identical to r13. Precondition 7's exact command (bytes identical to r13) on D aeab48e3 exits 0 with digest 81897a90a068b0402e68393f82cffda8fd83f0bda5283316c38d8c58b28e5107. As a countercontrol, D D exits 1 (t1:record-directory). The diff from r13 consists only of D, the drift sentence, the splice anchors, the XS lines, the two approved Authority lines (73, 81) and item 12. I re-measured every D-dependent value at D and each one matches: 2cd49dc0 is D's first parent; the only path changed in 2cd49dc0..D under lean/tools/.github is verify.yml, which gains #1208's probes_arp1marc; the three blobs; import OIBridge.PC1Controls at :286; 122 census families, the last being SUBSTATE-1 / GradientLinkFlatness; verify.yml :200/:713/:750/:752/:781/:806/:834; all 20 XS lines. scratchpad occurs nowhere in the file.
4	First comment	PASS, with nit N2. Quotes 1, 2, 3, 5 and 6 are each one contiguous verbatim substring of their source. That includes the halt-only question, the option text, the asked time 14:35:19Z and the answer time 15:04:42Z, all matching c.6740495860. Every author and time matches REST.
5	PR body	PASS, with nit N1. It states that no dispatch, no F/E designation and no merge is authorized. It contains no closes/fixes/resolves keyword. D's push run 42574043510 is push/D/success.
6	CI	PASS. The check-runs for aeab48e3 number 27 (total_count 27): 26 success and 1 skipped (the A42 exclusion (${{ matrix.part }}) placeholder). All are in one suite of run 42622568439, which has event: pull_request, pull_requests: [1215] and attempt 1. ?head_sha= for both round commits and ?branch= return only pull_request runs. No workflow_dispatch exists, so none of these is an F attestation.
7	Remaining PENDING	PASS. Six unbackticked fields remain: lines 131, 194, 1437, 432, 518 and 597, which are PLAN items 3–8. Items 1, 2, 9 and 12 are filled. Items 10 and 11 are in their pre-F state: status "drafting, not a candidate F", the drafting note present, no pins yet. That is expected before steps 4–6.
8	§A.42 item 8	PASS. All 12 full ids resolve. The 7 commits are reachable from main and the 5 blobs are in D's history. The cited comments either carry quoted words that the record itself quotes (6721662886, 6722716262, 6722109292, and 6740777773, which relays 6740495860) or are provenance (6721645433, run 42489463543, issue and PR numbers). Nothing rests on support off main.

Findings

• N1 (non-blocking; correct the description, §A.43 item 2). The PR body is stale at aeab48e3:

  • "Its one commit": there are two commits;
  • "Commit: 2d2bf21d…": it should name the head aeab48e3 and its item-12 fill;
  • "Seven fields stay PENDING": six remain.

  This needs a description edit only, with no commit.

• N2 (non-blocking). The first comment says "copied byte for byte". Its quote 4 joins two verbatim segments of c.6721662886 and omits that comment's §3 (CONS-0) with no elision mark. Nothing quoted is altered. The preregistration relies on quote 2, which is exact. If this is to be recorded, use a follow-up comment rather than an edit to c.6740777773, since that comment is now cited by id and time.

• Observation (no action). Line 59 keeps "to be quoted in the round PR's first comment" beside the filled citation, which is the form PLAN item 12 prescribes. I concur with the pre-push reviewer's N1 on line 73.

I made no push, dispatch, approval or merge. The only git operation that was not read-only was a fetch of origin main, which remains fd539e15.

---

Generated by Claude Code

[amaybaum]

A note on the first comment (c.6740777773).

The first comment says its quotes are "copied byte for byte". Quote 4, from #1169 c.6721662886, joins two verbatim segments of that comment. It omits the comment's §3 (CONS-0) between them without an elision mark. No quoted words are altered.

The preregistration relies only on quote 2, which is exact. The first comment is not edited, because the preregistration cites it by id and time.

The exact-head review raised this as N2 (c.6741128215).

---

Generated by Claude Code