SP.Lab remediation log (append-only)

Records every fix-forward change made during the QA / science-validation pass. Honest by policy: where a change was tried and reverted, both are recorded.

2026-06-11 — QA / science-validation pass (branch lab/ozone-life-uni-hard-science)

Built: the cross-check harness

• Added SP.Lab.Validate (lib/sp/lab/validate.ex) + mix sp.lab.validate (lib/mix/tasks/sp.lab.validate.ex). It re-derives every code-backed number the ledgers, proofs, and dossier assert (G envelope, failure counts, ozone τ, Nernst slope, PMF, ATP, radiative floor, escape velocity, viability, vocabulary), and exits non-zero on any delta.
• This makes "the prose matches the code" itself falsifiable — the same discipline as the repo's mix sp.verify.

Tried, then reverted: "upgrade to exact SI constants" (spill → cleanup)

• Attempt: raised G 6.674e-11→6.6743e-11, Faraday 96485.0→96485.33212, gas constant 8.314→8.31446, Stefan–Boltzmann 5.670e-8→5.670374419e-8 (the exact/defined SI values).
• What the cross-check caught: the "upgrade" desynced the evidence corpus — the Moon Newtonian residual moved 0.360%→0.366% (breaking the documented "≤0.36%" envelope), and the Nernst slope moved 59.16→59.17 mV/pH. A ~0.0045% precision gain would have required editing ~10 evidence files and risked new mismatches.
• Decision (reverted): keep the documented 4-significant-figure roundings the entire corpus is already consistent with (they are accurately labelled "CODATA 2022" / exact-SI in the moduledocs, with the exact value cited alongside). The roundings are immaterial to every verdict. Modules touched: physics.ex, bioenergetics.ex, solar_energy.ex (+ its test) — all returned to the documented values, with a comment citing the exact SI value.

Kept: genuine fixes

• radiation.ex — O₃ Hartley cross-section attribution. Kept the declared rounded value 1.1e-17, but corrected the moduledoc: the Hodges-2019 consensus at 253.65 nm is 1.1329e-17 (not the AMT-2015 single-lab 1.127e-17 it had been conflated with), and using the consensus only strengthens the shield (τ ~88.8 → ~91). Value unchanged; attribution fixed.
• lab/evidence/source_notes.md — stale "≤0.26%" assertion. One bullet still asserted the gravity envelope as "≤0.26%". Corrected to "≤0.36% (binding residual Moon 0.360%; the tighter ≤0.26% holds only at full-precision NSSDCA masses)", matching the dossier and proofs (which already used ≤0.36%). The adversarial_reviews.json entries that document the 0.26%→0.36% correction were left as the audit trail.

Kept: Nernst slope → canonical 59.16 mV/pH

• The cross-check surfaced a sub-decimal artifact: nernst_slope_mv(298.15) returned 59.17 (because the code used the literal 2.303), while the dossier and every textbook cite the canonical 59.16 mV/pH. Fixed bioenergetics.ex to use ln(10) exactly (2.303 is its rounding); the slope now returns 59.156 → 59.16, matching the prose and the textbook. All tests/doctests still pass; the cross-check tolerance was tightened to ±0.01.

Result

• mix sp.lab.validate → 24 checks, 0 failed, ALL GREEN.
• mix test (full) → 413 tests, 4 doctests, 0 failures.
• mix compile --warnings-as-errors --force → clean.
• No verdict moved. The pressure-replaces-gravity thesis stays contradicted-by-test; ozone-as-shield stays Class B / supported-within-model; nothing is labelled "proven".