[rlfleming93]

What

• Ride HUD: cadence dial (0–130 rpm, 80–100 band, zone-colored) and crank torque in Nm (crankTorqueNm) on both the dashboard and the TV card. In ERG, faster pedaling now shows as lower torque.
• Devices: device labels (Rename); kg/lb weight; profile form sets the rider's trainer/HRM through the existing registry assignment; session builder auto-fills both and shows live status + driver messages; Devices page shows a Garmin pairing hint.
• Device status replay: the hub records each driver's last status and /ws replays it on connect (status was event-only, so late pages showed "unknown").
• Disconnects: engine auto-pauses a rider when the trainer drops or goes stale (pauseReason: 'trainer') and auto-resumes on the next fresh sample; no 0 W dropout rows, clock frozen, no ERG guard trip. HR older than 5 s is dropped instead of repeated. Clear pause banners on both views.
• Power math: standard NP (full 30 s windows only); summary workKj, non-zero avgCadenceRpm; FIT session gets normalizedPower, IF, TSS, totalWork, avgCadence.
• OPENCYCLE_PORT for a side-by-side sim server.

Evidence

• pnpm check clean; vitest 631/631.
• Sim server (:4100, temp data dir): created rider with 115 lb → 52.2 kg + devices; builder auto-filled both, chips "Connected"; HUD screenshots game on/off at 1080p; manual pause banner; 47 s ride decoded with @garmin/fitsdk: NP 76, IF 0.507, TSS 0.3, totalWork 3438 J, timer 47 s (12 s pause excluded, no rows during pause).
• Real KICKR CORE ride (read-only DB): avg 30.9 W, NP 53.4 W, 2.284 kJ.
• NP fixture change 230 → 227 verified by an independent Python computation.

Note: branch also carries local main commit 4ccee4c (site move) that was never pushed.

[rlfleming93]

Review gate: PASS

• Head: 09724a8a39c62f7e702cda0eed96e34c4ea64832
• Reviewer: opus / claude-opus-5-5 (claude)
• Rounds used: 2/2

ID	Severity	Status	Location	Finding	Proof
—	—	—	—	None	—

MINOR findings (0)

None.