# Garage experiment

Status: human verification pending. Simulator tests and installation do not complete this experiment. Keep each original `.rideform` session and exported archive even when the result is inconclusive.

## Mount and calibration

Use a rigid sternum mount while seated on the stationary motorcycle. Record mount location, phone orientation, date, app/build and OS versions, and any mount movement. Select capture mode before calibration. Guided calibration collects a still neutral posture for2.5seconds, then deliberate left and forward gestures to establish rider axes. Return to neutral before recording. Verify left is positive lateral, right negative, forward positive forward angle, and neutral near zero. Recalibrate after changing mount, capture mode or returning from interruption. A standard mount assumes screen away/top up and still requires sign checks.

## Diagnostics and raw-stream comparison

Use Diagnostics to capture30seconds stationary, then slow and rapid left/right/forward/return drills. Follow the persisted phase prompts; manually mark movements as needed. Repeat three matched runs in fused-only and three in fused+raw, keeping posture, duration, movement order and UI activity comparable. Include raw-only diagnostic data if combined acquisition degrades fused capture.

Export every run. Compare achieved fused mean frequency, interval p99, gaps, receipt delivery age, writer loss and stationary noise across runs. Provisional combined-mode guardrails: at most5% fused mean decrease, at most20% p99 increase and zero unexplained writer loss. Report run variability; do not conclude equivalence from one favorable run. Delivery age and cue dispatch do not establish physical sensor or acoustic latency.

## Approximately50 transitions

Record at least two repeat blocks with neutral holds between movements. Suggested total:20 left departures/returns and20 right departures/returns, plus10 deliberately corrected movements across both sides. State whether the count means complete excursions or detected individual departure/return legs. Include slow/simple movements, faster movements and intentional overshoot/corrections. Use optional CENTER→LEFT→CENTER→RIGHT→CENTER speech cues as prompts; cues are not ground-truth movement times. GOOD/BAD are your annotations.

Keep an independent label sheet or timestamped video for a subset: direction, actual onset, peak, end of deceleration, stable hold and return. Synchronize comparison evidence explicitly and report uncertainty. A human observer or video identifies movement truth; app detections and speech dispatch cannot label themselves.

## Review and export checklist

- Stop and wait for saved status; note warnings and accepted/written counts.
- Inspect all timelines, markers, phase boundaries and gap flags. Count missed/extra/uncertain detections instead of hiding them.
- Compare ranges, durations, settling, overshoot/corrections and left/right distributions against labels; candidate jerk/spectral metrics have units and limitations, not a rider score.
- Change a filter setting, export the derived configuration/results, and retain the original master unchanged.
- Cancel one share, reopen a session, quit/relaunch, then reopen and export again.
- Save master ZIP and CSV through Files; test AirDrop too. Keep original filenames/UUIDs and all diagnostic runs.

## Evidence to return

Bring the master ZIP/CSV and derived result/configuration for each run, the mount/OS/build notes, your label sheet or video timestamps, and a short description of confusing UI or failed calibration/save/share. State which sign checks passed and whether the mount shifted. We will independently reload the files, reconcile every stream count, check rate/gap/noise comparisons and measure onset/phase error, misses/false positives and range/timing repeatability. An inconclusive smoothness result is valid; unexplained data loss requires investigation.

## Spoken calibration check

With the phone rigidly mounted and app open, set a clearly audible volume and leave spoken calibration enabled. Tap Calibrate once; follow the countdown and LEFT / CENTER / FORWARD / CENTER instructions. Breathe normally. Brief initial movement should age out of neutral capture. Check the spoken success or specific retry message. Enable automatic recording before calibration if reaching the screen afterward is difficult. Confirm left/right/forward/back signs with a helper before treating angles as validated. Handheld desk checks exercise interaction but do not validate torso mounting.