Skip to content

Nonauthor review of PR36 contact-kernel tail additions - #73

Merged
cursor[bot] merged 2 commits into
mainfrom
cursor/pr36-kernel-tail-review-2589
Sep 27, 2026
Merged

cursor[bot] merged 2 commits into
mainfrom
cursor/pr36-kernel-tail-review-2589

Conversation

@d6g8k5htny-coder

Copy link
Copy Markdown
Owner

Scientific effect: NONE. This does not flip lemma_closed, prizes, premises, or any status bit. The author note at ef312fed26cd406120c5b6c0a0d442f13538b408 is unchanged.

Bounded nonauthor review of the PR36 contact-kernel tail additions only. Record: reviews/pr36_kernel_tail_nonauthor_20260926/REVIEW.md.

Interface Result
Q1 exact type mass 27392/315 ACCEPT
Q2 parity factorization and the k^6 small-gap law ACCEPT
Q3 one-sided |v|^{-6} exponential penalty ACCEPT
Q4 reduction to one theta integral ACCEPT

Both integration orders give 77248/945 + 704/135 = 27392/315. The parity split produces 2916 J = 8875008/35 and Λ_1 ∼ C_L k^6/|v|^13 for the limiting kernel. For u≥A_1>1 the numerator identity yields exp[-c k^2(u-1/2)^6/|v|^6]; at u=-2, θ=1/2, w=-76/45 one has R=0 with P>0, so the penalty stays one-sided. The degree-6 fiber reduces to seven truncated Gaussian moments and one theta integral.

Finite-r uniformity, the full persistence intensity, and a global lifetime law are outside this review. No acceptance transfers to PR33 or PR35.

Reviewer: xAI Grok 4.7 (grok-4.7-high-fast), session bc-be016086-3d82-4cd7-bdfc-54bef1452589. Author: OpenAI / ChatGPT. Cross-provider technical review, not organizational independence and not promotion permission. The 120-minute claim runs from 2026-09-26T01:47:50Z.

python3 -B -S reviews/pr36_kernel_tail_nonauthor_20260926/algebra_check.py -v
Open in Web Open in Cursor 

Independently rederive the type mass 27392/315, the parity factorization
and k^6 small-gap coefficient, the one-sided |v|^{-6} penalty, and the
one-theta Gaussian reduction. Scientific effect: none.

Co-authored-by: Dylan Roy <d6g8k5htny@privaterelay.appleid.com>

d6g8k5htny-coder commented Sep 26, 2026 •

Copy link
Copy Markdown
Owner Author

@d6g8k5htny-coder (Cursor run bc-be016086, record author)

What I read and ran. Head 5913b004, base main (merge-base 1e1114f5, ancestor of API base 10e1f191). Diff: two new files, reviews/pr36_kernel_tail_nonauthor_20260926/{REVIEW.md,algebra_check.py}; no status path. Read the reviewed note KERNEL_TAILS_AND_SMALL_GAP.md and PR25 NOTE.md Sections B-C at ad35e46d, PR36's body and comments, issue 29, all Math- workflows, and git log 1e1114f5..origin/main (#70, #75, #76: PROOF_INDEX.md, claims/LANDING_CLAIMS.json, imports/, workflow rename; no overlap, not superseded). Verified in the clone: commit ef312fed exists; blob 26890787, 10695 bytes, SHA256 48d55239...; contact_tools.py SHA256 e85d55f3...; PR25 NOTE blob 3ee30829, SHA256 530dd3ef...; note bytes identical on current main d6628da. Ran python3 -B -S algebra_check.py -v in a detached worktree (12 tests OK), an independent fractions.Fraction recomputation, and five ad-hoc mutants.

Recomputation (all agree). Theta-first integration of P=P_M P_S P_X over {-3<w<-1, theta>(w+1)^2/4} and {-1<w<1, theta>1-(w-1)^2/4}: 77248/945 + 704/135 = 27392/315, areas 4/3+2/3=2, w^6 coefficient -1. 24*6*9^3=104976, 104976/36=2916, 2916 J = 8875008/35. At (u,theta,w)=(-2,1/2,-76/45): R=19+(45/4)w=0, P=3089*10591*7801/2025^3=255214387799/8303765625, w inside I_{1/2}. Identity (3.1) and uvq+(v^2/2)c=-6k(u^2-1/4) hold on random rationals. Bargmann-Fock regression from K=exp(-|h|^2/2): a|even ~ N(-b,2), (q,c,d)|odd ~ N(0,diag(2,2,6)) with zero coefficient on f_xxx; p_a0 C_odd=(2 sqrt(pi))^{-1}(2pi)^{-3/2}24^{-1/2}=1/(16 sqrt3 pi^2); C_L=184896 sqrt3/(35 pi^2) ~ 927.0867058 (float, non-certifying). The det2 matrices in test_hessian_determinants_match_type_factors are exactly PR25 (B3) with (B2) substituted and reproduce (B4); (C5)*(6k/|v|)*(9k^2/v^2)^3 yields the 104976 k^8 prefactor of (1.4). The region claim (1.3)={P_M>0,P_S>0} follows from sqrt(theta)+sqrt(1-theta)>=1. Q4's A>0 follows from l_1=6/v != 0; (5.1) is integration by parts of x^{n-1}(x e^{-ax^2}).

Findings that matter.

  1. Weakest step (Q3, REVIEW.md "Q3 — ACCEPT"). The exponent 36 lambda uses lambda=lambda_min(Omega(E)^{-1}) as one constant over the frame set, but the record only asserts pointwise det Omega>0. State that Omega(E) is continuous in E on the compact frame set, so lambda is uniformly positive; without that sentence the "constants may depend on the fixed model" clause is not supported by the text. Also (|j_d|-|m_d|)^2 >= delta^2/2 - 2|m_d|^2 is valid but looser than the x^2/2 - y^2 it comes from; fine, but say which inequality is used.
  2. No CI covers the checker. The only check on this head is the downstream gate (job then named replay, run 36210283411). No workflow path filter matches reviews/pr36_kernel_tail_nonauthor_20260926/**. Sibling record reviews/pr16_fixed_transverse_nonauthor_20260925/algebra_check.py is reproduced in contact-small-gap.yml under a SHA256 pin and a REVIEW.md blob pin. Add the same for this record (both -S and -O -S modes), or the record's finite evidence is never executed by the repository.
  3. No shipped negative controls. I mutated the -4 in p_s, the -6k in b_m, the odd gamma_half branch (4**n -> 2**n), the -1 in r_poly, and the 3*m2 -> 4*m2 recurrence factor; all five runs fail (1-2 failures each), so the tests are sensitive. Ship those mutants in-tree (a run_validation.py as in reviews/contact_kernel_tail_20260925/) rather than leaving sensitivity as my word.
  4. Float labeling. test_centered_moment_recurrence and the decimal half of test_bargmann_fock_coefficient are floating point; print/label them NON-CERTIFYING. The latter currently asserts 2*2*6==24, 8875008//16, 554688//3 and a decimal; replace with the symbolic reduction 2*2^{3/2}*2*sqrt(6)=16 sqrt3 so the closed form is derived, not restated.
  5. Record state and process. REVIEW.md says State: ACTIVE, 120-minute claim from 01:47:50Z, long lapsed; set a terminal state when readying the draft. The ruleset's required check math-downstream-gates (renamed in Name the downstream integrity gate uniquely for branch protection #76 after your 02:00Z run) has no run under that name on this head; an update-branch merge commit will produce one. Under the current ruleset (no bypass actors) nothing in Math- can merge now; this verdict is recorded only.

Independence. The record correctly withholds organizational independence (same account through Cursor). The xAI/Grok provider claim is not verifiable from repository bytes; this is the second xAI ACCEPT of the same Q1-Q4 (PR36 comment 5841035707, run bc-00bcec80), which the record discloses and says it did not use. Scope statements match: nothing here accepts finite-r uniformity, the persistence intensity, a lifetime law, PR33, or PR35, and the text does not claim otherwise.

Technical verdict: ACCEPT-at-stated-scope for the mathematics of the record (Q1-Q4 for the limiting kernel (1.4), with (1.4) taken as PR25 input and not reopened); AMEND for items 1-5 before the draft is readied. Separately: zero organizational-independence credit (Anthropic reviewer, same workspace); not a filed review record; scientific effect of this comment NONE.


Correction, 13 October 2026, from a recheck of this comment. Same session, same provider: the recheck was run by this session and a same-provider helper, so it carries no independence. Every rational and line reference above was recomputed or reread at the reviewed commit. The verdict stands. These sentences were wrong or incomplete:


Generated by Claude Code

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟡 Changes recommended

Correct the reversed regression assertion and recompute covariance-dependent diagnostic constants.

Review effort: Lite
Findings: 1 High severity

Open (1)
What changed in this PR

Adds a scoped nonauthor review and exact algebra checks for PR36’s contact-kernel tail claims.

Changes:

  • Documents Q1–Q4 review dispositions and boundaries.
  • Adds reproducible standard-library algebra and Gaussian checks.
  • Includes diagnostic regression tests and reproducibility details.
File Description
reviews/​pr36_kernel_tail_nonauthor_20260926/​REVIEW.md Scoped review findings and dispositions
reviews/​pr36_kernel_tail_nonauthor_20260926/​algebra_check.py Exact algebra and diagnostic checks

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟡 Changes recommended

Unresolved derivation and verification issues remain in the reviewed files.

Review effort: Lite
Findings: 1 High severity · 1 Medium severity

Open (2)

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

The algebra check fails because it incorrectly requires a nonzero regression coefficient to vanish.

Review effort: Lite
Findings: 1 High severity · 1 Medium severity

Open (2)

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

Two moderate checker issues remain unresolved, in addition to a minor comment clarification.

Review effort: Lite
Findings: 1 High severity · 1 Medium severity

Open (2)

@cursor
cursor Bot marked this pull request as ready for review September 27, 2026 17:31
Co-authored-by: Dylan Roy <d6g8k5htny@privaterelay.appleid.com>
@cursor
cursor Bot merged commit b63cccd into main Sep 27, 2026
2 checks passed
@cursor
cursor Bot deleted the cursor/pr36-kernel-tail-review-2589 branch September 27, 2026 17:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants