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

Large diffs are not rendered by default.

Large diffs are not rendered by default.

210 changes: 210 additions & 0 deletions frontiers/planar_soft_layer_chain_20261003/C124/CHECKER_COMMENT.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,210 @@
## C124 nonauthor review: independently designed executable controls

Dylan Roy — delegated AI review. Actual reviewer/control author: OpenAI/Codex c99_custody_audit; root publishes the frozen bytes. Full analytic review: [5974327730](https://github.com/d6g8k5htny-coder/main/issues/229#issuecomment-5974327730). These controls were frozen before execution and the C124 author checker/output remained unopened. Prior source exposure and organizational independence 0 are disclosed in that review. The finite controls do not prove the analytic source theorems.

Extraction: each file is the exact text within its five-backtick fence, including its final LF. independent_controls.py: 8,759B SHA256 1d77a571133c7803704ee5e0c0dfd5f13bb0d33249748b09291205646d2c2cdc; DESIGN.md: 3,213B SHA256 457b7a19db5ae1a9ee7dc24e08f4b277d4493d36787c9bb7d84747a3e49c0bdb; baseline stdout: 255B SHA256 988e0f97d1a4cda0cd92c01800991d5cbd89b84e6eaa6349e477ee75b7faf25e.

Python3.11.16, both `-B -S` and `-O -B -S`: baseline601checks/0failures. Each positional mutation GAMMA_POWER, UNWEIGHTED_BAND, DIAGONAL_RESIDUAL, UNCENTERED_RESIDUAL, UNSAFE_EPSILON, FACTOR_TAIL, NORMALIZER_POWER, H_SCHEDULE exited1 in both modes; UNKNOWN exited2. Twenty invocations and forty transcripts are bound by the immutable review evidence seal.

### independent_controls.py
`````python
#!/usr/bin/env python3
"""C124 nonauthor finite falsification controls; no author checker imported."""
from fractions import Fraction as F
from math import factorial
import json
import sys

MODES = ('GAMMA_POWER', 'UNWEIGHTED_BAND', 'DIAGONAL_RESIDUAL',
'UNCENTERED_RESIDUAL', 'UNSAFE_EPSILON', 'FACTOR_TAIL',
'NORMALIZER_POWER', 'H_SCHEDULE')
if len(sys.argv) > 2 or (len(sys.argv) == 2 and sys.argv[1] not in MODES):
print('UNKNOWN_CONTROL_MODE')
sys.exit(2)
mode = sys.argv[1] if len(sys.argv) == 2 else 'BASELINE'
counts = {}
failures = []


def check(group, truth, label):
counts[group] = counts.get(group, 0) + 1
if not truth:
failures.append(group + ': ' + label)


# The chart Jacobian and pole cancellations are checked on signed gamma,
# including arbitrarily small rational values; no numeric tolerance is used.
for gamma in (F(-7), F(-1, 1000), F(-1), F(1, 1000), F(2, 3), F(5)):
for b in (F(-2), F(0), F(1, 7)):
for c3 in (F(-1, 5), F(0), F(3)):
d = gamma**2 - 12*b
j = 8*gamma**3 - 144*b*gamma + 576*c3
c, rr = d/gamma**2, j/gamma**3
psi = abs(c) + F(7, 3)
lam = gamma**2 * psi / 24
wl = (24*lam-d)*(24*lam+d)/16
power = 4 if mode == 'GAMMA_POWER' else 6
check('chart', wl*gamma**2/24 == gamma**power*(psi**2-c**2)/384,
'w d-lambda = gamma^6 polynomial d-psi')
check('chart', gamma**6*abs(c)**3 == abs(d)**3, 'absolute c pole')
check('chart', gamma**6*rr**2 == j**2, 'signed R pole')

# Correlated atomic measures, not products of marginal laws. The finite model
# uses exact band constants extracted from the actual weighted atomic bands.
for depth in (2, 3, 8, 20, 50):
a = F(1, 2**depth)
m = 0
while 2**m*a < F(1, 4):
m += 1
check('dyadic', F(1, 4) <= 2**m*a < F(1, 2), 'terminal band')
shells = [F(2)*a*F(1, 2**j) for j in range(m)]
floors = [F(1, 4**j) for j in range(m)]
check('dyadic', sum(shells) <= 4*a, 'width sum')
check('dyadic', sum(floors) <= F(4, 3), 'finite-r floor sum')
# A boundary-focused, strongly correlated finite population.
atoms = [(a/2, F(1), F(1, 7)),
(a, F(1), F(1, 13)),
(2*a, F(2), F(1, 17)),
(3*a, F(4), F(1, 19)),
(F(1, 3), F(2**depth), F(1, 23)),
(-F(3), F(4*2**depth), F(1, 29)),
(F(10), F(1), F(1, 31))]
band = lambda delta, p: sum(mass*n**p for u,n,mass in atoms if abs(u)<=delta)
floor = F(1, 100)
h3 = F(8)
deltas = [2**j*a for j in range(m+1)]
moment2 = sum(mass*n*n for u,n,mass in atoms)
cc = max([band(delta,p)/(delta+floor) for delta in deltas for p in (0,2)]
+ [moment2/h3])
actual = sum(mass for u,n,mass in atoms if abs(u)<=a*n)
bound = cc*(5*a + F(7,3)*floor + 16*h3*a*a)
check('dyadic', actual <= bound, 'correlated decomposition bound')
part = band(a,0)
for j in range(m):
selected = [(u,n,mass) for u,n,mass in atoms
if 2**j*a < abs(u) <= 2**(j+1)*a and abs(u)<=a*n]
check('dyadic', all(n>2**j for u,n,mass in selected), 'shell norm implication')
part += sum(mass for u,n,mass in selected)
outer = [(u,n,mass) for u,n,mass in atoms if abs(u)>2**m*a and abs(u)<=a*n]
check('dyadic', all(n>1/(4*a) for u,n,mass in outer), 'outer norm implication')
# Inner band can include points not in the moving band, so it is a cover.
part += sum(mass for u,n,mass in outer)
check('dyadic', actual <= part, 'all moving-band atoms covered')

# The unweighted small-ball condition cannot be multiplied by an independent
# moment. U uniform and N=ceil(log2(1/U)) gives a factor m on the stated event.
for m in (2,4,8,16,32):
em = F(1,2**m*m)
lower_mass = F(1,2**m)
inferred = em if mode == 'UNWEIGHTED_BAND' else m*em
check('correlation', lower_mass <= inferred, 'unweighted O(e) shortcut fails')

# Orthogonal projection onto the complement of (3/5,4/5). Its negative
# off-diagonal entries increase variance along (1,-1); they are not zero.
a0,a1 = F(3,5),F(4,5)
cov = ((1-a0*a0,-a0*a1),(-a0*a1,1-a1*a1))
check('projection', cov[0][0]+cov[1][1] == 1, 'rank-one trace')
check('projection', cov[0][0]*cov[1][1]-cov[0][1]**2 == 0, 'PSD rank-one')
actual_var = cov[0][0] + cov[1][1] - 2*cov[0][1]
used_var = cov[0][0]+cov[1][1] if mode == 'DIAGONAL_RESIDUAL' else actual_var
check('projection', used_var == F(49,25), 'correlation retained in linear-form variance')
for value in (F(-4),F(0),F(7)):
mean, coeff, obs, obsmean = F(100),F(3,7),F(19),F(11)
field = mean+coeff*(obs-obsmean)+value
residual = field-coeff*(obs-obsmean)
if mode != 'UNCENTERED_RESIDUAL':
residual -= mean
check('projection', residual == value, 'centered endpoint residual')

# Explicit Gaussian even moments and the exponential-series coefficient
# bound implied by Minkowski; no independence of Fourier residuals is used.
moment = 1
for n in range(1,81):
moment *= 2*n-1
check('exponential', moment <= (2*n)**n, 'Gaussian moment upper bound')
divisor = 2 if mode == 'UNSAFE_EPSILON' else 12
coefficient = F((2*n)**n, divisor**n*factorial(n))
check('exponential', coefficient <= F(1,2**n), 'exponential series majorant')
check('exponential', sum(F(1,2**n) for n in range(81)) < 2, 'geometric series partial sum')

# Exact independent endpoint-scalar integration. r^2 J^4 times the cubic
# endpoint integral is r^5 J^10, and r^-3/Z_r has Z_r proportional to r^2.
for radius in (F(1,2),F(1,7),F(1,100)):
for jr in (F(1),F(2),F(17,3)):
for depth_const in (F(1),F(4),F(7,2)):
for cross_const in (F(1),F(5,2)):
upper = depth_const*radius*jr**2
integral = upper**3/3 + cross_const*radius*jr**2*upper**2/2
numerator = radius**2*jr**4*integral
expected = radius**5*jr**10*(depth_const**3/3+cross_const*depth_const**2/2)
check('scalar_tail', numerator == expected, 'r^5 J^10 numerator')
zpower = 4 if mode == 'NORMALIZER_POWER' else 2
scaled = numerator/radius**zpower/radius**3
check('scalar_tail', scaled == jr**10*(depth_const**3/3+cross_const*depth_const**2/2),
'one full r^2 normalizer and r^-3 scaling')
weights = ((F(1),F(99,100)),(F(10),F(1,100)))
weighted_tail = sum(prob*j**10 for j,prob in weights if j>5)
tailprob = sum(prob for j,prob in weights if j>5)
moment10 = sum(prob*j**10 for j,prob in weights)
used_tail = moment10*tailprob if mode == 'FACTOR_TAIL' else weighted_tail
check('scalar_tail', weighted_tail <= used_tail, 'same residual tail indicator retained')

# Independently encode each term as powers (r,H,w), then substitute w.
rows = {
'radius':(0,0,-8),'radius_error':(1,3,-4),
'elder':(1,4,4),'elder_error':(F(3,2),5,2),
'endpoint':(1,3,2),'endpoint_error':(F(3,2),4,1),
'band':(1,0,4),'leak':(1,4,0),'outer_band':(2,3,8),
'hessian1':(2,5,4),'hessian2':(2,7,2)}
expected = [(F(2,3),F(8,3)),(F(4,3),F(13,3)),
(F(2,3),F(8,3)),(F(4,3),F(13,3)),(F(5,6),F(7,3)),
(F(17,12),F(11,3)),(F(2,3),F(-4,3)),(F(1),F(4)),
(F(4,3),F(1,3)),(F(5,3),F(11,3)),(F(11,6),F(19,3))]
wh = F(-1,4) if mode == 'H_SCHEDULE' else F(-1,3)
for (label,(pr,ph,pw)),want in zip(rows.items(),expected):
got = (F(pr)-F(pw,12),F(ph)+pw*wh)
check('ledger', got == want, label+' substituted powers')
check('ledger', got[0]>F(2,3) or (got[0]==F(2,3) and got[1]<=F(8,3)),
label+' target domination')
check('ledger', F(pr)-F(pw,12)>=F(2,3),label+' fixed-layer endpoint')
for label,(pr,ph,pw) in {'rH':(1,1,0),'embedding':(1,0,1),
'e':(1,0,4),'H2e':(1,2,4),'rw2':(1,0,2)}.items():
check('ledger', F(pr)-F(pw,12)>0,label+' cutoff decays')
check('ledger', F(3)+F(2,3)==F(11,3),'physical fixed-layer power')

# Variation norm on atomic signed measures is sum of absolute masses.
for nu0 in ((F(1),F(2)),(F(1,5),F(4,5)),(F(2),F(0))):
for changes in ((F(1,10),F(-1,20)),(F(0),F(0)),(F(-1,10),F(1,10))):
nu=tuple(a+b for a,b in zip(nu0,changes))
if min(nu)<0:continue
m,m0=sum(nu),sum(nu0)
err=sum(abs(a-b) for a,b in zip(nu,nu0))
normalized=sum(abs(a/m-b/m0) for a,b in zip(nu,nu0))
check('normalization', normalized<=err/m+abs(m-m0)/m,'probability normalization')
if m>=m0/2:
check('normalization', normalized<=4*err/m0,'positive mass floor')

print(json.dumps({'mode':mode,'checks':sum(counts.values()),'groups':counts,
'failures':failures,'scope':'finite controls only; no analytic proof'},sort_keys=True))
sys.exit(1 if failures else 0)
`````

### DESIGN.md
`````markdown
# C124 independent review design and exposure record

Reviewer: OpenAI/Codex /root/c99_custody_audit, delegated child of root01a0bbb5 under existing C124 claim33e06044-3ffa-4d61-81d8-937ad54e3fb4. No C124 proof, control or design contribution. Prior exposure includes C99/C103 source/author-side project work, C101 author-control custody, C102/C104 analytic reviews, C107 authorship and subsequent engineering/analytic review/custody roles. The immediately preceding source-only frontier audit identified the published beta<2/3 boundary and need for quantitative tails, but did not inspect or contribute to this new JB/DB/ES derivation. No fresh-context, blind-source, provider-independent or human-review claim. Organizational independence0. Dylan's owner-reported ongoing review is acknowledged under5973603003; this review itself is AI work, not his human verdict.

Target PROOF.md20668B SHA90148657397dcce31e8039afa9015c022f74b2ed0d41e75f2f2339074a8a520f; source manifest SHAa0371bd1985358820250e0d75936d6ec5112f165824df7a9f6e29e25e3f46f6d. I read the entire proof and exact C82/C101/A4/P interfaces before this design. I have not opened C124 controls.py, CONTROLS.md, author outputs, or author development notes. Existing sources' own equations are inputs, not independently rediscovered theorems.

Analytic obligations: reconstruct JB from weighted-before-conditioning Q10 and fixed-width LB, verify gamma-six cancellation and actual finite-r error; reconstruct DB as a partition including m=0 and the outer band, with fixed moment order2; identify centered endpoint residual as joint Gaussian projection after U_r,A_M and verify summable Fourier/Lp and exponential-series constants without coefficient independence; retain the residual indicator in Q30's scalar integral and one full normalization; verify failure-marked model tail separately; map every ledger/cutoff to the imported geometry and distinguish fixed11/3 from global2/3 log8/3 and conditional normalization.

Independent finite controls will use Fraction, not author code: signed/small-gamma chart cancellation; finite atomic correlated band decompositions and geometric sums; explicit correlated Gaussian projection and noncentered-regression falsifiers; exact Gaussian even moments/exponential-series coefficient majorant; endpoint-scalar integral powers and a two-atom residual-weight correlation countermodel; independently composed rational r/H exponent ledger and cutoff powers; signed finite-measure normalization. Negative modes will delete a gamma power, discard weighted correlation, assume diagonal residual covariance, omit centering, enlarge epsilon beyond the stated safe coefficient, omit the residual tail indicator, drop one normalizer power, or change the H schedule. Unknown modes must be rejected. These controls can refute algebra/finite implications but do not prove LB, Gaussian series convergence, topology or infinite-dimensional analytic estimates.

Freeze this design and independent_controls.py before execution or any author-checker exposure. Run only these new bounded controls in normal and optimized Python modes. Do not rerun parent/master suites. Preserve any actual failures rather than edit the candidate.
`````

### normal-baseline.stdout
`````text
{"checks": 601, "failures": [], "groups": {"chart": 162, "correlation": 5, "dyadic": 103, "exponential": 161, "ledger": 39, "normalization": 16, "projection": 6, "scalar_tail": 109}, "mode": "BASELINE", "scope": "finite controls only; no analytic proof"}
`````
Loading
Loading