Physics decides. AI explains. A gate verifies.

The failure mode of AI coaching is a made-up number delivered confidently. TENTHS is built so that cannot reach you: the model that writes your debrief never computes anything — it narrates a verdict that deterministic code already produced, and a mechanical check proves it.

1 — The engine (in your browser)

Your RaceBox CSV or VBO file is parsed on your device. The engine segments laps (device markers, or start/finish-gate crossings on library tracks), excludes traffic laps from statistics, detects corners from path curvature, and computes per-corner speeds, brake points, spreads and the three session scores. All dynamics are GPS-derived — lateral load is v²·κ, lean is its arctangent — because a lean-mounted accelerometer does not measure earth-frame truth.

2 — The reference (19 US tracks)

For library tracks, your line is compared against a minimum-time racing line solved from track geometry: an exact minimum-curvature optimization with late-apex targets, refined for time, and a speed profile from a quasi-steady-state motorcycle model — friction ellipse, wheelie and stoppie limits, aerodynamic drag, grade. Assumptions (warm sport tires, dry track, μ 1.10/1.20) ship with every figure. These are simulation estimates and are always labeled as such — a ceiling to measure headroom against, not a lap record.

3 — The debrief (paid, and machine-verified)

When you buy a debrief, the verdict JSON — numbers only, never your GPS trace — goes to our API, which asks a frontier Claude model to write your debrief under strict rules: every number verbatim from the verdict, caveats respected, no safety-critical commands. Before you see it, a mechanical gate re-extracts every figure from the text and checks it exists in your data. A debrief that fails is regenerated; if it fails twice, you keep your purchase and we apologize instead of showing you an unverified debrief. The badge on your debrief ("47/47 figures machine-verified") is that gate's actual output.

Lineage

The structure is aviation's, not ours: flight-data monitoring plus a disciplined post-flight debrief is how pilots have improved for decades — measured against envelopes, scored, then turned into one or two focused items. Riding schools run the same loop with humans: session, debrief, goals, repeat. TENTHS automates that loop for the riders who don't have a data engineer in their pit.