๐Ÿ“ฆ Where your data lives

local-first darts โ€” every byte accounted for

The trainer is built local-first: your board plays perfectly with zero internet, forever. The network exists for exactly one reason โ€” the shared detection model that every board helps train and every board gets back. This page shows everything that is stored, where, and why. The row counts below are live.

On your board (the Raspberry Pi)

๐ŸŽฏ Every throw โ€” stats.db

SQLite on the Pi ยท the local-first source of truth ยท works offline forever

๐Ÿ“ธ Frames in flight โ€” RAM staging

ramdisk (/dev/shm) ยท ephemeral ยท never wears the SD card

๐Ÿ† Boards, players, identity โ€” small JSON stores

on the board, not in the browser

๐Ÿง  The model & calibration

the seeing part

On central (dart-room.com)

One small service holds the shared brain: the training data the boards contribute, the model registry they update from, and the cross-board leaderboards. The darts themselves are tiny โ€” nearly all of the storage is the camera frames.

TableRowsWhat it holds
game_throwโ€ฆevery uploaded throw: label, model proposal, correction flag, game context, capture timestamp
capture / capture_imageโ€ฆ / โ€ฆthe capture sets โ€” 3 camera frames per throw, lossless WebP, content-addressed (SHA-256), stored in the 13 GB blob volume
highscore + highscore_tagโ€ฆ + โ€ฆleaderboard rows with extensible tags (board:, group:, variant:โ€ฆ) โ€” new scopes need no schema change
elo_ratingโ€ฆthe ranked-match ladder (K=32, saved players only)
player / player_identityโ€ฆ / โ€ฆdisplay names ยท opt-in 12-word key bindings (only the public key is stored โ€” never the words)
rig / rig_tokenโ€ฆ / โ€ฆenrolled boards and their per-board access tokens (each board holds only its own key; any one can be revoked alone)
rig_health / rig_eventโ€ฆ / โ€ฆheartbeats + the shipped decision log โ€” a friend's board can be debugged without SSH
rig_calibrationโ€ฆcalibration snapshots โ€” per-throw provenance for training
modelโ€ฆthe published model registry: every trained model with its provenance and metrics โ€” boards pull updates from here
feedbackโ€ฆin-app bug reports, arriving with screenshot + rig context as evidence
eval_run / eval_resultโ€ฆ / โ€ฆreserved: detector-vs-truth evaluation slices

๐Ÿ” Access

The flywheel โ€” how a dart becomes a better model

you throwโ†’ 3 cams captureโ†’ model proposes a tip positionโ†’ you accept โ€” or tap to correct
โ†ณ scored instantly, stored in stats.db and frames + label queue in the outboxโ†’ central capture store
โ†ณ training pulls the throws โ€” your corrections are the gold (a tap marks exactly where the tip really was)
โ†ณ new model trained & field-testedโ†’ published to the registryโ†’ every board updates over the air

The loop's honest rule: models are judged by field acceptance โ€” how often real players accept proposals untouched at a real board โ€” not by lab metrics alone. A model that aces the test set and annoys players does not ship.

What never leaves your board

install ยท game modes ยท statistics
dart-room.com โ€” open dart detection, local-first