Skip to content
Merged
Show file tree
Hide file tree
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
59 changes: 59 additions & 0 deletions .github/workflows/sard-g-r3r4-review.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
name: SARD-G R3-R4 nonauthor review checks

'on':
pull_request:
paths:
- 'reviews/sard_g_r3r4_claude_20260929/**'
- '.github/workflows/sard-g-r3r4-review.yml'
workflow_dispatch:

permissions:
contents: read

jobs:
exact-and-probe:
runs-on: ubuntu-latest
timeout-minutes: 10
steps:
- uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262
with:
persist-credentials: false
- uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065
with:
python-version: '3.11.16'
- name: Verify sources, replay both modes, reject mutants
env:
PYTHONDONTWRITEBYTECODE: '1'
run: |
set -euo pipefail
python -B -S - <<'PY'
import hashlib, json, pathlib, subprocess, sys
root = pathlib.Path('reviews/sard_g_r3r4_claude_20260929').resolve()
manifest = json.loads((root / 'SOURCE_FILES.json').read_text())['files']
present = sorted(p.name for p in root.iterdir())
expected_names = sorted([e['path'] for e in manifest] + ['SOURCE_FILES.json'])
if present != expected_names:
raise ValueError('packet tree differs from manifest: ' + repr(present))
for entry in manifest:
path = root / entry['path']
if pathlib.PurePosixPath(entry['path']).name != entry['path'] or path.is_symlink():
raise ValueError('flat regular source required: ' + entry['path'])
data = path.read_bytes()
if len(data) != entry['bytes'] or hashlib.sha256(data).hexdigest() != entry['sha256']:
raise ValueError('source identity mismatch: ' + entry['path'])
script = str(root / 'r3r4_review_check.py')
expected = (root / 'RESULTS.json').read_bytes()
mutants = ['wrong-residual', 'keep-degenerate', 'span-deficient']
for flags in (['-B', '-S'], ['-B', '-O', '-S']):
out = subprocess.run([sys.executable, *flags, script],
check=True, capture_output=True, timeout=300).stdout
if out != expected:
raise RuntimeError('output differs from RESULTS.json under ' + ' '.join(flags))
for m in mutants:
rc = subprocess.run([sys.executable, *flags, script, '--mutant', m],
capture_output=True, timeout=300).returncode
if rc != 1:
raise RuntimeError('mutant ' + m + ' not rejected')
print(json.dumps({'scientific_effect': 'NONE', 'mathematical_acceptance': False}))
PY
git diff --exit-code
13 changes: 13 additions & 0 deletions reviews/sard_g_r3r4_claude_20260929/RESULTS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
{
"checks": {
"R3a_rkhs_identity": true,
"R3c_residual_uncorrelated": true,
"degenerate_direction_removed": true,
"fewer_points_rank_deficient_control": true,
"regular_zeros_have_nonzero_derivative": true,
"span_of_point_evaluation_images_is_full": true
},
"object": "CLAUDE-REVIEW-SARD-G-R3R4-20260929-v1",
"passed": true,
"scope": "exact finite illustrations only; not the Banach-space proof"
}
78 changes: 78 additions & 0 deletions reviews/sard_g_r3r4_claude_20260929/REVIEW.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
# Nonauthor analytic review: SARD-G robust charts, slice R3/R4 (Math-#135)

Scientific effect: **NONE**. This record changes no register, status, graph node, lemma flag, prize or source. It
reviews slice R3/R4 only. R1/R2 is xAI/Harper's slice. Integration is a separate act.

## Object and exposure

| Field | Value |
|---|---|
| Object | OA-SARD-ROBUST-CHARTS-20260929-v1 (OpenAI / ChatGPT), sections 7–8 and claims R3–R4 |
| Head | `ee94b06368c29bceca80662698b4b4d6ee81ec91` ([Math-#135](https://github.com/d6g8k5htny-coder/Math-/pull/135)); `PROOF.md` blob `7e2237063a8dd0120e9d5bd15a3caaa32e8495fa`, 25022 B, SHA256 `b91b9f792c6322204a0f2a573b2ca76d5c951e49ddd7caa6ffd1abed0aa79e79` |
| Request / pickup | [5890713526](https://github.com/d6g8k5htny-coder/Math-/pull/135#issuecomment-5890713526) / [5890766663](https://github.com/d6g8k5htny-coder/Math-/pull/135#issuecomment-5890766663) |
| Reviewer | Anthropic Claude, session `session_017Mi3hxjaxV45x6zo6o1ee3`. I have no authorship of, and no prior review of, SG (`research/q0/sard_g/`, main `a1fc9581`), BR (OpenAI/Codex, main `f43f243`) or RI (xAI/Harper). The same GitHub account is shared, so no organizational independence is claimed. |
| Own checks | `r3r4_review_check.py`: exact finite illustrations, 3 mutants, identical under `-O`. |

## Verdicts

| Claim | Verdict |
|---|---|
| **R3** (§7): point-evaluation covariance images are dense in `H`; the normalized rational family; standard `ξ_ℓ`; the independent residual (R3b)–(R3c); a nonvanishing direction (R3d) given the A3–A4 premise | **ACCEPT** |
| **R4** (§8): `B_{χ,ℓ}` is Borel; open validity sets; countable regular zeros on each shift line; Tonelli gives `μ(B_{χ,ℓ}) = 0`; the covering inclusion; the completed-measure conclusion given `μ(Ω_gen) = 1` | **ACCEPT**, conditional exactly as stated on A2-T (through R2), A2-D and A3–A4 |
| A2-T, A2-D, A3–A4, `μ(Ω_gen) = 1`; the R1/R2 geometry | Not reviewed; these are imports or another slice. R4 uses them only as stated. |

No defect was found.

## Re-derivations

**R3.**
- *Density.* For `h ∈ H` and `ℓ ∈ X*`, `⟨h, Qℓ⟩_H = ℓ(h)`. If `h ⊥ Qℓ_n` for every `n`, then `h` vanishes on the
dense set `S`. Since `H ↪ C²`, `h` is continuous, so `h = 0`, and the span of `{k_n}` is dense.
- *Normalization.* `v_ℓ = ℓ(Qℓ) = ‖Qℓ‖_H²`, so `v_ℓ = 0` exactly when `Qℓ = 0`; those `ℓ` are removed. Then
`‖h_ℓ‖_H = 1` and `ξ_ℓ ~ N(0, 1)`.
- *Residual.* `Cov(a(f), ξ_ℓ) = a(Qℓ)/√v_ℓ = a(h_ℓ)`, so `Cov(a(g_ℓ), ξ_ℓ) = 0` for every `a ∈ X*`. The pair
`(g_ℓ, ξ_ℓ)` is a continuous linear image of `f` in the separable space `X × ℝ`, hence jointly Gaussian. Its
characteristic functional factorizes, and that functional determines Borel laws on a separable Banach space. So
the pair is independent. The derivative-evaluation remark in the note is a valid alternative, not needed.
- *(R3d).* `D_χ'(f)` lies in `X*` by A2-D, and its restriction to `H` is continuous. If that restriction is nonzero,
it cannot vanish on a set with dense span.

**R4.**
- *Borel.* `B_{χ,ℓ}` is Borel because `U_χ^rob` is open (R2, under A2-T) and both `D_χ` and
`f ↦ D_χ'(f)[h_ℓ]` are continuous there (A2-D).
- *Shift lines.* `I_g` is open, because `t ↦ g + th_ℓ` is continuous into `X`. On it, `F_g` is `C¹` with
`F_g' = D_χ'(g + th_ℓ)[h_ℓ]`. Zeros with `F_g' ≠ 0` are isolated, so they form a countable set, even where
singular zeros or the ends of the intervals accumulate.
- *Tonelli.* The set `{(g, t): g + th_ℓ ∈ B}` is the preimage of a Borel set under a continuous map. By
independence and Tonelli, `μ(B) = ∫ N(0,1)({t: g + th_ℓ ∈ B}) dP_{g_ℓ}(g) = 0`.
- *Countable union.* Both index sets `χ` and `ℓ` are countable.
- *Conclusion.* The covering inclusion follows from (R2b) and (R3d). Outer measure zero means null in the completion,
and with `μ(Ω_gen) = 1` the conclusion follows. No shift line is assumed generic.

The finite checker (in exact rationals, on a trigonometric circle field, at dense Pythagorean points
`x_m = mθ` with `cos θ = 3/5`) exhibits:
- the reproducing identity;
- full span once there are `2K + 1` points, and rank deficiency with fewer;
- the uncorrelated residual;
- removal of a degenerate direction;
- a regular zero versus a singular zero.

It illustrates the proof; it does not replace the Banach-space argument.

## Notes

- **N1 (simplification).** Single point evaluations already suffice for (R3d). Since the span of `{Qℓ_n}` is dense,
a nonzero continuous functional on `H` is nonzero on some `k_n = Qℓ_n`. For the variance-one field,
`v_{ℓ_n} = 1`, so no removal is needed. Rational combinations and density in the unit sphere are valid but not
required.
- **N2 (dependency).** R4's Borel step uses the openness of `U_χ^rob`, which is part of R2 and rests on A2-T and
R1. The R4 conclusion is therefore also conditional on the outcome of the R1/R2 slice, not only on A2-D and A3–A4.
The note's §1 dependency statement says this; the verdict here does not discharge R1/R2.
- **N3.** The note correctly avoids any claim that a whole shift line is Morse. Only regular zeros inside the open
validity intervals are counted.

## Reproduce

python -B -S r3r4_review_check.py # prints RESULTS.json byte for byte
python -B -O -S r3r4_review_check.py # identical
python -B -S r3r4_review_check.py --mutant M # exit 1 for each of the 3 mutants
38 changes: 38 additions & 0 deletions reviews/sard_g_r3r4_claude_20260929/SOURCE_FILES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,38 @@
{
"schema": 1,
"scientific_effect": "NONE",
"review_status": "NONAUTHOR_REVIEW",
"slice": "R3/R4",
"reviewed_object": {
"pr": "d6g8k5htny-coder/Math-#135",
"head": "ee94b06368c29bceca80662698b4b4d6ee81ec91",
"paths": [
{
"path": "frontiers/sard_g_robust_charts_20260929/PROOF.md",
"git_blob": "7e2237063a8dd0120e9d5bd15a3caaa32e8495fa",
"bytes": 25022,
"sha256": "b91b9f792c6322204a0f2a573b2ca76d5c951e49ddd7caa6ffd1abed0aa79e79"
}
]
},
"files": [
{
"path": "REVIEW.md",
"bytes": 5538,
"sha256": "72f1107d7a42bb53ff8dcfc53d1ebab5fc34b98b57eaaace773b97c4a4d7349b",
"git_blob": "8c6fd343e876c911df060be67d9d5c97028a46a8"
},
{
"path": "RESULTS.json",
"bytes": 431,
"sha256": "f0b1a94edcc9a956cb9ce98776d40cc8c0917b59bdff2cc23743f124200b2e59",
"git_blob": "280eb9df1a9d3a0a5e8a89867a387041fe14ee66"
},
{
"path": "r3r4_review_check.py",
"bytes": 5206,
"sha256": "81679a5d25ed9f888369cfb3d8a47d79400512270032cbcc2b1e060dfdc1d641",
"git_blob": "25873d3c7438a42b1173553298512937c1a11f9f"
}
]
}
122 changes: 122 additions & 0 deletions reviews/sard_g_r3r4_claude_20260929/r3r4_review_check.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,122 @@
"""Independent nonauthor finite checks for OA-SARD-ROBUST-CHARTS-20260929-v1 slice R3/R4 (Math-#135). Exact rationals.

MODEL a trigonometric Gaussian field on the circle with modes 0..K and positive rational weights, in coefficient
coordinates (covariance Sigma = diag(w)); point evaluations at x_m = m*theta with cos(theta) = 3/5, so every
cos(k x_m), sin(k x_m) is rational (Chebyshev recursion) and {x_m} is dense (theta/pi irrational).
RKHS (R3a) <h, Sigma l>_H = l(h) for h = Sigma a, with <Sigma a, Sigma b>_H = a^T Sigma b.
SPAN the images Sigma l_{x_m} span the Cameron-Martin space (rank 2K+1) once m runs over 2K+1 points; fewer points
are rank-deficient (control). Single point evaluations already suffice (note N1).
RESIDUAL (R3b)-(R3c) unnormalized: xi = l(f), h = Sigma l / v, g = f - xi h gives Cov(a(g), xi) = 0 for all a
(checked on a basis); a degenerate l (v = 0) in a singular model must be removed.
ZEROS (R4) a C1 function with accumulating singular zeros: regular zeros are isolated, hence countable; exact
model F(t) = t^2 sin-free polynomial surrogate with a double zero and simple zeros.
Mutants (each must fail): wrong-residual, keep-degenerate, span-deficient.
"""
import argparse
import json
import sys
from fractions import Fraction as F

MUTANTS = ("wrong-residual", "keep-degenerate", "span-deficient")
MUT = None
K = 3
W = [F(1)] + [F(1, 2 ** k) for k in range(1, K + 1) for _ in (0, 1)] # weights for 1, cos x, sin x, cos 2x, ...


def trig(c, s, kmax):
"""cos(kx), sin(kx) for k = 0..kmax from cos x = c, sin x = s (exact)."""
cs, ss = [F(1)], [F(0)]
for _ in range(kmax):
cs.append(cs[-1] * c - ss[-1] * s)
ss.append(ss[-1] * c + cs[-2] * s)
return cs, ss


def point_eval(m):
c, s = F(1), F(0)
for _ in range(m): # x_m = m theta, cos theta = 3/5, sin theta = 4/5
c, s = c * F(3, 5) - s * F(4, 5), s * F(3, 5) + c * F(4, 5)
cs, ss = trig(c, s, K)
vec = [F(1)]
for k in range(1, K + 1):
vec += [cs[k], ss[k]]
return vec


def rank(rows):
M = [list(r) for r in rows]
rk = 0
for col in range(len(M[0]) if M else 0):
p = next((i for i in range(rk, len(M)) if M[i][col] != 0), None)
if p is None:
continue
M[rk], M[p] = M[p], M[rk]
for i in range(len(M)):
if i != rk and M[i][col] != 0:
f = M[i][col] / M[rk][col]
M[i] = [x - f * y for x, y in zip(M[i], M[rk])]
rk += 1
return rk


def sig(v, w=W):
return [wi * vi for wi, vi in zip(w, v)]


def dot(a, b):
return sum(x * y for x, y in zip(a, b))


def check():
out = {}
n = 2 * K + 1
# RKHS identity: <Sigma a, Sigma l>_H = a^T Sigma l = l(Sigma a)
ok = True
for m in range(4):
l = point_eval(m)
for a in ([F(1)] + [F(0)] * (n - 1), [F(j + 1, 3) for j in range(n)]):
h = sig(a)
ok &= dot(a, sig(l)) == dot(l, h)
out["R3a_rkhs_identity"] = ok
npts = n - 1 if MUT == "span-deficient" else n
imgs = [sig(point_eval(m)) for m in range(npts)]
out["span_of_point_evaluation_images_is_full"] = rank(imgs) == n
out["fewer_points_rank_deficient_control"] = rank([sig(point_eval(m)) for m in range(n - 1)]) < n
# residual: xi = l(f), h = Sigma l / v, g = f - xi h; Cov(a(g), xi) = a^T Sigma l - (a^T h) v = 0
ok = True
for m in range(1, 4):
l = point_eval(m)
v = dot(l, sig(l))
h = [x / v for x in sig(l)] if MUT != "wrong-residual" else sig(l)
for j in range(n):
a = [F(int(i == j)) for i in range(n)]
ok &= dot(a, sig(l)) - dot(a, h) * v == 0
out["R3c_residual_uncorrelated"] = ok
# degenerate direction in a singular model (last weight zero): l supported on that mode has v = 0 and is removed
Ws = W[:-1] + [F(0)]
l_deg = [F(0)] * (n - 1) + [F(1)]
v_deg = dot(l_deg, sig(l_deg, Ws))
kept = [l for l in (point_eval(1), l_deg) if dot(l, sig(l, Ws)) != 0 or MUT == "keep-degenerate"]
out["degenerate_direction_removed"] = v_deg == 0 and all(dot(l, sig(l, Ws)) != 0 for l in kept)
# (R4) isolated regular zeros: F(t) = t^2 (t - 1)(t + 2): simple zeros at 1, -2 have F' != 0, the double zero at 0
# has F'(0) = 0; F is C1 and the regular zeros are isolated, hence countable
Fp = lambda t: 4 * t ** 3 + 3 * t ** 2 - 4 * t # derivative of t^4 + t^3 - 2 t^2
Fv = lambda t: t ** 4 + t ** 3 - 2 * t ** 2
out["regular_zeros_have_nonzero_derivative"] = all(Fv(t) == 0 and Fp(t) != 0 for t in (F(1), F(-2))) and Fp(F(0)) == 0
return out


def main():
global MUT
ap = argparse.ArgumentParser()
ap.add_argument("--mutant", choices=MUTANTS)
MUT = ap.parse_args().mutant
checks = check()
passed = all(checks.values())
print(json.dumps({"object": "CLAUDE-REVIEW-SARD-G-R3R4-20260929-v1", "checks": checks, "passed": passed,
"scope": "exact finite illustrations only; not the Banach-space proof"}, indent=2, sort_keys=True))
return 0 if passed else 1


if __name__ == "__main__":
sys.exit(main())
Loading