🚀 Participant briefing: ready in 2 minutes

A quick read – then the learning journey runs without friction.

1Connect your smartphone

At every station that uses the camera: scan the QR code (your normal camera app is enough, no installation).
The first time, the browser asks for camera permission → “Allow”. Without it, your measurement system delivers no data.
Keep the page on your phone open while you work at the station – it is your connection to the lab.

✅ Standard work:

❌ Known disturbances:

FIRST select the label (pip count), THEN take the photo – never the other way round. A wrong label is a wrongly recorded measurement: the model adopts it unchecked.
Misfire or wrong label? Press the × on the image in the gallery on the main screen – data hygiene is part of data work.
Diversity is sample quality: different dice, surfaces, angles, lighting. The model will only know the world you show it.

In the Inspection Lab and the Arena: commit to your own answer first, then reveal – otherwise you contaminate your own comparison.
The “true sum” is the reference standard for all three contenders – count honestly at the table.
In the Arena: press “Score the round” and “Next round” first, then send the next photo – running rounds do not accept new photos.

Dataset, model and progress live in the desktop's browser – nothing is stored server-side, no accounts. Therefore: do the entire journey on the same machine in the same browser (no incognito window).
The stations build on each other – the learning path works best in order.
Training metrics scatter (±3 percentage points are normal – random initialization). Treat them like measurements: consider repeatability before interpreting differences.
The 📖 deep dives and 🧮 math drawers at every station are optional – and the best source when you want the details.