WHAT THIS RECORD ISEvery quantitative claim in the certificate log that touches a measured constant, recomputed and set against its COD value in one table. Six EXACT (≤ 1e-8), four PPM-class, and five graded honestly with their attributions — the two corpus-known ppm deficits (m_p/m_e vs 6π⁵ at 18.8 ppm; α = π²e²×10⁻⁴ at 637 ppm) are recorded as the corpus records them: real, linked, and part of the friction programme, not glossed. No claim in the log grades worse than its certificate already states.

FORWARD RUN — 2026-08-31 (second pass)The fixed equations driven FORWARD to observables not in the first table. Ten land EXACT or exact-class against COD data — including all four CMB observables from the Planck carrier and the full α-ladder. One audit catch worth the record: the first forward pass used the frequency-lane root (x = 2.821439) for the wavelength-lane constant — the two Wien peaks are LANE-ADDRESSED (same carrier, two lanes, two peaks), and the audit itself demonstrated the lane rule by tripping on it and correcting to x = 4.965114232, landing b at 6.4e-11. The Debye rows carry their attributed model channels; the beta-dial row is a registered prediction awaiting machine data.

SECOND FORWARD SWEEP — 2026-08-31 (completeness pass; the first was a subset)Everything remaining that runs forward without external survey/machine data. Wins: Thomson and Rydberg at 1e-11, the 2π identity exact, Koide at 9.2 ppm, para-Ps and the 21-cm line at exact-class. Two of my own errors were caught BY the COD comparison and are kept on the page: the 21-cm factor-2 (Ry vs α²m_ec²/h) and a mislabeled ppm figure in the c-split row. One honest NOT-A-MATCH recorded (naive H₀ reading). NOT RUNNABLE and therefore not run: the FUSION-BULK design identities (no COD counterpart — they are design equations), the staked falsifiers awaiting survey/machine data (quasar stretch, coil-phase transfer, ITER dial row), and AMERICONIUM’s stability lane (experiment).