Drop in a recorder file and read what happened in those 16 seconds.

RoadRec applies the same fixed physics rules the recorder uses, shows the evidence behind every verdict, and tells you when it cannot tell.

Choose an event file, or drop it hereCSV with columns t, ax, ay, az, gx, gy, gz and optional speed (for example evt_0001_CRASH.csv)

Recorder over USB

Plug the XIAO in with a USB cable and press Connect. Needs Chrome or Edge on a computer.

No file loaded yet. Pick a sample above to see a full report.

Orientation through the event

The device outline tilts by the measured change from its pre-event orientation. Direction of tilt is not shown, only how far it turned. Drag the slider through time.

Four events, same time scale. Every trace is lined up on its trigger. Three are real phone recordings; the hard brake is synthetic (generated by this page) and only shows what the rule looks for. None come from the recorder hardware.

What this page is for. The honest version: what the rules are, how well they did on data recorded after the code was frozen, and where they fail.

The rules, in the order they are applied

Label	Physics test
DROP	Free-fall of 120 ms or more and the vehicle was not moving. A falling device is weightless; a ridden vehicle is not.
CRASH	Impact of 4 g or more, vehicle was moving, then lying down or stopped.
FALL	Orientation changed 45° or more and stays there, peak under 4 g.
HANDLING	Orientation changed but the device kept moving, or hard braking with no confirmed speed. Carried, not ridden. No alert.
POTHOLE	Spike of 100 ms or less, 55% or more of the energy along gravity, upright, ride continues.
NEAR_MISS	Hard deceleration with a swerve (35°/s yaw) or 0.7 g or more, upright, ride continues, speed 2 m/s or more.
HARD_BRAKE	Sustained horizontal deceleration of 0.35 g for 400 ms, no swerve.
MINOR_IMPACT / BUMP	High impact but still moving upright / below every signature.
INCOMPLETE	Not enough data after the event to judge the outcome. An honest “cannot tell”.

An event starts when any trigger fires: impact over 2 g, free-fall under 0.5 g for 80 ms, horizontal push over 0.30 g for 300 ms, or rotation over 200°/s.

Accuracy on the frozen test set (20 phone recordings)

Event type	Files	Correct	Accuracy	95% interval
Drop	7	7	100%	65 to 100%
Tip-over	8	0	0%	0 to 32%
No alert	5	5	100%	57 to 100%
All	20	12	60%	39 to 78%

The code was frozen on 4 October 2026, 14:05 IST and fingerprinted before this test set was recorded. Tuning-set numbers are not quoted because thresholds were adjusted while looking at them.

Known weaknesses

Tip-overs are the main failure. On the test set 7 of 8 were labelled CRASH and 1 DROP. A phone toppling onto a hard surface hits well above the 4 g crash line, and indoors there is no GPS speed. This is our reading of the misses, to be tested on a bicycle.
Hard brake, near miss, pothole and real crash have no recordings yet. Those rules are written but unproven.
Speed-based rules have never run on real GPS data, and the false-alert rate during ordinary riding is not measured.
The sample files here come from a phone, not the recorder. The MPU6050 will differ and clips at 16 g, so crash peaks are lower bounds.

Has this file been changed?

The page computes the file's SHA-256 fingerprint here in your browser. Paste the fingerprint the device wrote for this event; if one character differs, the file was edited.

Load a file first.

Fingerprint written by the device Or load the .sha256 file that sits next to the event on the SD card

With the updated firmware, every event on the SD card has a matching .sha256 file. A mismatch means the CSV differs from what the recorder saved. It cannot prove who changed it, and anyone who can edit the card can edit both files.