substrate attestation · live · v0
Every commit to this repository writes a substrate measurement to a per-SHA JSON record. The cache-tier latency, the ballistic-gate speed, and the full 12×12 verifier walk land in the stream below — measured unprivileged on the host that produced each commit. Read this page as the receipt the FIM patent claims silicon will issue at line rate; today it is software-prosthesis on consumer laptops, and that is already enough to settle the discriminating questions an underwriter or examiner needs answered.
Mean nanoseconds per dependent load at each tier, with σ. Sharp separation between tiers is the discriminating signal — when a commit shifts a workload across a tier boundary, the change is visible at this scale, no privileged counter access required.
| Tier | Mean ns/access | σ ns | n samples |
|---|---|---|---|
| L1 | 0.00 | 0.00 | 0 |
| L2 | 0.00 | 0.00 | 0 |
| SLC | 0.00 | 0.00 | 0 |
| DRAM | 0.00 | 0.00 | 0 |
Per-host, per-tier z-score against that host's own rolling mean. A zone is a commit whose substrate signal moved outside the stationary band — the chip noticed the change. v0 surfaces single- host zones; cross-host correlation arrives with the v1 pilot.
No zones above the 3σ threshold yet — need more samples to establish per-host baselines.
Expected loss = frequency × severity. The frequency half is measured, twice, from two different corpora, read from src/data/actuary-ledger.json and data/pmu/sealed-calibration.json at build time — never typed onto this page.
The two numbers disagree and we publish both. The sealed rate is pre-registered and small; the live rate is larger, post-hoc, and recomputed on every commit. Neither supersedes the other — we have not chosen a favourite. What separates a real scope departure from an artefact of a badly declared scope — the false-alarm rate an underwriter actually prices against — is unmeasured. The full accounting, including what we refuse to claim and why, is in docs/underwriting/false-alarm-rate.md. This page states the frequency; it does not claim an underwriter has accepted it.
Newest first. Each row is a per-commit reading the post-commit hook wrote next to that commit's diff.
No measurements in the stream yet.
Stream generated 2026-10-15 00:08:39 UTC · v0 storage = file-based ·source
Sufficient for: where it landed, re-runnable from the commit: same bytes, same hash. NOT sufficient for: whether the work was good, which is undecidable (Rice, 1953); nothing here claims it.
