MX·LAB

Methods

How it is measured.

Six analyses, one dataset. Each section states the question, how it is answered, what the source actually covers, and what the number does not mean.

Two rules apply to all six. A number is only produced where the source has it, and derived files are committed and checked — npm run derive --check fails if they no longer match the dataset.

6.1 — Lap chart

Question. Where was each rider on each lap?

Method. Read the position the timing system prints on every lap row for 2025–2026, and invert the LAP CHART sheet for 2020–2024: both series publish one per moto, a grid of positions by lap whose cells are plate numbers, so a plate's position on a lap is the row its number sits in. The file keeps the top twenty finishers of each moto: a full forty-line chart is unreadable, and the lapped riders are exactly the ones whose printed position needs care.

Coverage. Both series from 2020. Every 2020–2024 moto publishes a chart sheet; 2025–2026 print the position directly. Triple Crown rounds publish only three separate mains and no single chart, so they are the only rounds without per-lap positions. No position is estimated.

6.2 — Fade

Question. How much does a rider slow between the first and second half of a moto?

Method. Take laps 2 to the end, drop laps slower than 1.4× the rider's own median (a pit stop or a crash, not fading), then compare the median of the second half with the first. A moto needs at least six usable laps each side. The result is a percentage, so a 1:10 track and a 2:20 track are comparable.

Coverage. Every season and both series. The leaderboard needs ten measured motos in a season. Differences in the second decimal are stopwatch noise, not fitness.

6.3 — Lap 1

Question. Does the opening lap already decide the finishing order?

Method. Per moto, rank the riders who have a lap-1 time by that time and by finish, then take Spearman's rho. Report the median per season and the share of motos above 0.5. A moto needs eight riders with a lap-1 time to be measured at all.

Coverage. 2020–2026, with a gap: 2020 Motocross prints --.--- for lap 1 across the entire field, while 2020 Supercross does print it. This is a correlation within a moto, not a causal claim — fast starters are usually fast riders.

6.4 — Qualifying

Question. What is a fast qualifying lap worth once the gate drops?

Method. Within an event and class, join the qualifying sheet to the moto results by rider, then compute qualifying position minus mean finishing position. Positive means the rider gained places. Also report Spearman's rho between qualifying position and mean finish per event.

Coverage. All seasons and classes. The qualifying sheet lists the whole entry list, so riders who did not make the motos are excluded — otherwise "gained places" would be dominated by riders who never raced.

6.5 — Titles

Question. Who won, when did they clinch, and how did the points race move?

Method. Points are computed from finishing positions with the standard moto table — 25, 22, 20, 18, 16, then one fewer per place to 1 for 20th. Both motos score; a DNF scores 0 and still counts as a moto started. Ties break on wins, then best finish, then rider id. The clinch round is the first round after which the leader's gap exceeds the points still available at 50 per round.

Supercross. The 250 title is scored per coast, and an East/West showdown is recorded but not scored — it races two championships at once. Where a Triple Crown overall sheet prints the official award, that value is used instead of the table.

Caveat. A published standings total may differ where a series applied a tie-break or a penalty the sheets do not show.

6.6 — Ratings

Question. How good is each rider relative to the riders they actually raced?

Method. Elo per class over every moto: within a moto every finisher is compared with every other finisher, K = 32 spread across the field, and all deltas are computed from the pre-moto snapshot and applied together. A rider new to a class enters at that class's current median rather than a fixed 1500, so the scale does not drift.

Coverage. Both classes and both series; leaderboards need ten motos. Ratings are not comparable across classes.

Backtests

The models behind these pages are scored walk-forward on motos they have never seen, with the design written down before the results were read. See the backtests page for the errors, the bootstrap intervals and the by-season comparison against the field average.

What this site does not do