Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
140 changes: 140 additions & 0 deletions reviews/lb_rate_thm023_landing_20260926/REVIEW.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,140 @@
# LB-RATE / KIMI-THM-023 landing review

Review date: 2026-09-26.
Reviewer: xAI / Grok 4.6 (source-exposed to the 2026-08-04 addendum packet, the 2026-09-26 Drive landing, GP-LB-REC-001, and AO48-AUD-064). Packet author: Kimi / AO48 campaign. Technical pass on one shared operator account. **Zero organizational-independence credit** — the ground is source exposure and single-operator authorship, not provider identity.

Amended 2026-09-27 at queue reconciliation, answering the review in [PR #161 comment 5850569939](https://github.com/d6g8k5htny-coder/main/pull/161#issuecomment-5850569939): F2 intensity figure, §1 C031 identities, two §3 constants, and the grading of the 0.0334 far term. The HOLD disposition is unchanged.

**Disposition: HOLD — CONDITIONAL ASSEMBLY DRAFT. Do not accept `c = 0.9144` as a proved uniform theorem constant.**

This is not D0–D7 3D lifetime work and does not inherit SIDE24 coefficient status.

---

## 1. Exact objects

Normalized 2026-08-04 addendum carriers (hashes match GP-LB-REC-001):

| Object | Bytes | Whole-file SHA-256 | Body SHA-256 |
|---|---:|---|---|
| KIMI-THM-023.md | 14,936 | `61e14810c55f688561bacd5a62bfa3e0872e0672fa387161c60a0455be65f92a` | `fedcc4b67a792efd5d43496b9a6f26c3412a0e863aac038f3e2aa37654ded3a7` |
| KIMI-AUD-023.md | 5,448 | `df6565acf9231b3ae41e0f5f27f55384d786119013bebc06b79bd87a8e70e874` | `23377bd4257832490cf9e5e863dc7afba7ea1fc4f719b2abfe1e02081cc1fe3d` |
| KIMI-AUD-024.md | 3,369 | `309aefb8bc5c6dab656cf2d6ddae0a570e4ba593a3e327a2008f785c97bdc47c` | `0633402b95a586331848bf12bceed0db519f3425f7e2f8b93ceb7110e514d25f` |
| MANIFEST.sha256 | 1,199 | `73e20364804275cadb84f5452dbec95c3a9f7c171fae79a17670e731274a9a48` | — |

Identity split (do not collapse):

- THM-023 cites AUD-023 as `61f2b702…`. That is AUD-023 **v1.0** body. The landed file above is the normalized carrier of the same campaign. AUD-023 v1.1 body is `a7decb6f4b8b3a896d7e0aadddcd698853e57a9a253ef787173bb3c702136a32`.
- C031 is two documents, not one file with two hashes: `C031_LBRATE_Integration.md` (13,690 bytes, blob `ec2087e6`) is `e7998ef0d17d951bc89978f9fe32e510019059dd0650c8f0e0ae33273f40f32e` (the THM/AUD-023 cite), and `C031_Freeze.md` (1,739 bytes, blob `f61db50c`) is `e165821b…`, which is what LB-1 line 193 cites. Both are on the hardening mirror at `7caac254`. The Drive copy of the Integration file was not re-hashed this session.

Drive landing of these exact bytes: 2026-09-26T19:30Z, owner Dylan Roy, Drive root. Operator card `README925.txt` is a landing label, not a status upgrade.

---

## 2. Exact estimand

Field: exact normalized periodized Bargmann–Fock on the flat 2-torus $T^2_{24}$, covariance the periodization of $K(u)=e^{-|u|^2/2}$, level $b=6/5$.

Estimand: under the typed maximum–saddle pair-Palm law at separation $r$,

$$
1-q(r,\,6/5).
$$

Claim printed by the addendum: there exist $r_0>0$ and $c>0$ such that

$$
1-q(r,\,6/5)\;\ge\;c\cdot r^3\qquad\text{for all }0<r\le r_0,
$$

with candidate intercept

$$
c \;=\; (1-0.0334)\cdot 0.946 \;=\; 0.9666\times 0.946 \;=\; 0.9144036.
$$

Domain firewall: this object is **not** the 3-dimensional Side-24 first-moment density $\nu_{(3,24)}(\ell)=c_{(3,24)}\ell^{-1/3}(1+o(1))$. Gap convention in the packet: $\ell=r^3/6$.

---

## 3. Arithmetic that holds

- $1-0.0334=0.9666$.
- $0.9666\times 0.946=0.9144036$. Printed $0.9144$ is a 4-decimal truncation, not a round-up.
- WP component sum at $r=0.025$: $1.95\times 10^{-6}+1.37\times 10^{-6}=3.32\times 10^{-6}=0.21248\,r^3$, consistent with printed $0.213\,r^3$ within rounding.
- R2 printed line with denominator $(9-6.25)$ evaluates to $5.18\cdot(\ell/2)$, not the printed $2.1\cdot(\ell/2)$.
- Corrected reading $(9-2.25)$: $0.76\times(18.75/6.75)=19/9=2.\overline{1}$ on $(\ell/2)$; with $\ell=r^3/6$ this is $(19/9)\cdot(r^3/12)=\tfrac{19}{108}\,r^3=0.175925\ldots\,r^3$.
- Finite-rung measured $\Lambda$-window: $C^*(0.025)=0.946$.
- C031 derived-on-grid limit (Drive C031 file): $C^*_\infty=0.9091\pm 0.038$ (quad) $\pm 0.009$ (shell). Then $0.9666\times 0.9091=0.87873606$, not $0.9144$. The printed intercept multiplies a theorem-grade reduction (DER-009) whose far constant $0.0334=\bar p(0.05)$ is *measured* (THM-023 line 122, $\nu_T$ instrument, C025; GP-LB-REC-001 line 77 lists both $0.0334$ and $0.946$ as measured inputs) by a finite-rung measured $\Lambda$ constant.

---

## 4. Findings

### F1. Gap: printed R2 denominator

The printed factor $(9-6.25)$ is a ledger defect. The assembly constant $2.1$ stands only under the corrected reading $(9-2.25)$. Falsifier: a C030 source showing collar radius $2.5$ on both zones was intended.

Kind: gap in the printed derivation, not a counterexample to an $O(r^3)$ class.

### F2. Gap + decertification: Lemma WP quantitative upper bound

Two independent defects.

**Intensity (LB-1 §9, as corrected by AO48-AUD-064).** LB-1 §9 (lines 204–210) prints a rigidity-zone integral $I_{cs}(0.025)\approx 1.30399\times 10^{-5}$ with a hot spot near $(-0.05,-0.575)$ / $(-0.04,-0.58)$, about $1.313\times 10^{-4}$ per unit area, against a printed rigidity figure $\sim 3\times 10^{-15}$. AO48-AUD-064 (lines 138–142 and §7.5, lines 207–222) **decertifies** those two figures as bounds: the $-y$ argmax is a phantom (true intensity there $\lesssim 10^{-21}$). The figure that survives is the derived-on-grid $I_{\text{true}}=4.7602\times 10^{-6}$ at $(0,\,0.5875)$ (KIMI-DER-025 / AUD-064 §6), which is about $1.43$ times the printed *total* WP budget $0.213\,r^3=3.328125\times 10^{-6}$ at $r=0.025$, not $3.9$ times. The gap stands either way; the record carries the surviving number.

**Cauchy–Schwarz (AO48-AUD-064, DEF-WP-CS-01; not re-derived this session).** The implemented envelope used a factor $\min(\mathrm{Cantelli},P_W)$ where the displayed CS bound on the indicator requires $\sqrt{\min(\mathrm{Cantelli},P_W)}$. The printed computation is therefore not a proved upper bound. Quantitative WP ub **DECERTIFIED**. Remainder order is not proved $O(r^3)$.

AUD-023 item 4 “VERIFIED at ledger arithmetic” is true of the component sum only. C029-(1) “DISCHARGED at derived-grade-with-measured-constants” is **SUPERSEDED**.

### F3. Scope mismatch: `0.9144` is not a proved uniform constant

Even ignoring F2 and the unquantified outer remainder, the displayed finite-$r$ coefficient is $(0.9666-0.3889\,r^3)\times 0.946$. It equals $\approx 0.91440$ at $r=0.025$ and $\approx 0.91436$ at $r=0.05$, both below $0.9144$. Reaching $0.9144$ requires $r<0.02139\ldots$. LB-2 is certified only at the frozen rungs $0.025$ and $0.05$. Measured $c_\Lambda=0.946$ is not a rigorous $r\to 0$ limit; C031’s own on-grid limit is $\approx 0.909$.

Kind: scope / hypothesis mismatch with the consumer banner “proved floor $0.9144$”. The far term $0.0334$ is itself a measured constant inside a theorem-grade reduction, which is a further reason for HOLD.

---

## 5. Inherited statuses (do not upgrade)

From GP-LB-REC-001 (2026-08-04), which intaked these exact normalized bytes:

- KIMI-THM-023: **HOLD — CONDITIONAL ASSEMBLY DRAFT**
- KIMI-AUD-023: **AMEND REQUIRED**
- KIMI-AUD-024: **HOLD — CONDITIONAL IMPLICATION ONLY**
- LB-1: localized finite-rung computational evidence only
- LB-2: deterministic finite-rung conditional-mean evidence only
- LB-3: R0 conditionally closed; $\gamma$-LOC(ii-c) open
- No Boolean change, no promotion

From AO48-AUD-064 (2026-08-05): quantitative WP ub DECERTIFIED; C030 qualitative kill STANDS; THM rebuild gated.

LB-1 above-$b$ channel SURVIVES as a different functional (confirmation gate: it does not use the defective WP product).

---

## 6. What this review did NOT check

- Byte-level re-execution of the LB-1/2/3 certificates.
- Bodies of KIMI-DER-009 / 009b / 010, C030 CountingLemmas, `c031_verify.json`.
- WP certificate source bytes for DEF-WP-CS-01 (taken from AUD-064).
- C031 quadrature that produces $C^*_\infty=0.9091$ (taken from the Drive C031 file / prior read).
- GP-DER-118-v1.10 (AUD-024 premise). Status inherited: HOLD / NOT PROMOTED.
- Any 3D Side-24 coefficient identity or lifetime-remainder theorem.
- Organizationally independent review of GP-LB-REC-001 or AUD-064 themselves.

---

## 7. Forbidden statements

- “proved constant $c=0.9144$”
- status transfer from 3D Side-24 coefficients
- promotion of GP-DER-118-v1.10
- treating measured exit counts such as $1750/1750$ as a proved zero probability
- using this lower-bound assembly to close the two-sided rate or issue #116 (Q0-C101 qualitative *upper* cubic)

---

## 8. Public-surface fit

Closest existing public issue: [#116](https://github.com/d6g8k5htny-coder/main/issues/116) (Q0-C101 qualitative *upper* rate $1-q\le C_{Q0} r^3$). This landing is the sibling lower bound and does not close that upper-rate audit.
Loading