Everything it does, without the varnish

Two ways to read the same session.

The driver asks where. The engineer asks why. One analysis sits behind both, so the two answers cannot contradict each other.

Driver mode

Circuit map coloured by how much time changes hands, a corner-by-corner diagnosis, speed trace, deltas by phase, two laps compared with playback, degradation, and a sixty-second session verdict.

8 views

Engineer mode

Multi-channel WebGL overlay with up to seven laps, a MoTeC-style lap table with the ideal lap stitched sector by sector, tyres, a strategy simulator, and a tabbed workbench.

5 views

The workbench

Eighteen panel types across six tabs: setup advisor, dual track map, cockpit, driver trace, channel and section reports, suspension, G-G diagram, 3D elevation, histograms, scatter, FFT, maths channels, video and export.

18 panels

Open what you already have.

A file is routed by its CONTENT, not by the button you clicked: anyone can write a filename and it is not a trustworthy source.

Every import returns a decode score

0 to 100, with itemised deductions and a high / medium / low verdict. Instead of a silent success you know whether to trust what was just read, before you make a decision with it.

0-100

Laps are never discarded

They are classified: race, out, in, pit, warm-up or anomalous. Parity with the vendor's own lap counter is a hard invariant, verified on five real sessions of 5, 7, 34, 13 and 15 laps. You see the out-lap and you know it is an out-lap.

5 sessions

An unknown format raises a named error

No attempt is made to guess at a binary layout that has changed. The error points at the vendor's CSV export, which carries their own numbers. Degrading to the vendor's ground truth beats inventing on top of yours.

It knows which circuit you are on.

Without sending your coordinates anywhere: the catalogue ships inside the application.

Circuit library

46 layouts, 38 verified within 0.5 % of their official length and 24 with real surveyed asphalt widths. Four ship deliberately WITHOUT coordinates, because inventing a position would let the matcher name a venue the car cannot be shown to have visited.

46

Or you draw your own

Four routes in: your own GPS, the catalogue, a .gpx or .kml file, or freehand. Corner detection runs in the browser at 60 fps, ported from the same Python detector and pinned against it by a test.

The projection is metric and measured

Against pyproj the residual is 8.8 cm at Barcelona, 15.9 at Monza, 37.0 at Spa and 1.04 m at the Nordschleife, against 1-3 m of receiver error. A dependency is not added to win centimetres below the noise floor.

8.8 cm

It closes the loop between one run and the next.

This is the part no other tool on the comparison does: not just what to work on, but whether what it told you last time worked.

Every objective names its own check

A metric, a corner and a numeric target. The target is your own best execution of that corner, so it is achievable by construction: you have already done it at least once.

The verdict can be “we cannot tell”

Improved, worse, unchanged or inconclusive - and the last is first-class and the most common. If the conditions were not recorded it says so; if the car was a different one it says so. A verdict nobody can check is flattery, not a measurement.

The history is yours and it is local

Saved sessions carrying their conditions on the row, a setup-change log per session, comparison across dates, and an audited pilot merge for when two names are the same person.

See it on a run of your own.

Send us a file and we will send back the analysis done on your data, with the corner-by-corner diagnosis and the numbers behind it.

Send us a file