RoxBear

The first coach that reads your movement, station by station.

A HYROX training analyzer built on the watch IMU. Other apps show you a dashboard of charts. RoxBear tells you where the wheels came off — which station bled time, which rep set fell apart, where recovery lagged.

Apple Watch · 100 Hz IMU + HRPer-station analysisIndoor-native (no GPS)

The wedge

Nobody does per-station movement breakdown from a wrist IMU. Roxbase, HyroxDataLab, Edge — they program your training. COROS times your splits. None of them can tell you that your wall-ball cadence lengthened 30% after rep 40, or that your sled-pull form scattered while your pace looked fine.

The movement signal is the moat. It rides the same hardened watch pipeline shipping in PickleBear and Statterbox — so the capture is battle-tested, and the analysis is the new part nobody else has.

What it does — on the wrist today

Record

Watch captures 100 Hz IMU + HR through the workout. WR1 + a full forensic raw stream — nothing thrown away.

Mark

Tap LAP at each run↔station boundary. Manual segmentation now; auto-detect later.

Debrief

Phone shows a per-segment read: cadence/fade, rep consistency, HR recovery, run-split fade — plus a plain-English narrative.

THE READ — “Started strong. Wall Balls cadence lengthened 93% after rep 30 — that’s where the wheels came off. Recovery lagged into most runs. Run 5 was your slowest at 61% off Run 1 — the valley, and where to find time.”

The strategy — data first

Don’t over-engineer the analysis before there’s real signal. Collect a bunch of labeled raw data first, then mine it to learn what actually predicts a faster race.

Capture

Full-fidelity raw IMU + HR every session. The asset.

Label

Station tags + ground-truth reps + RPE — so the data can teach.

Learn

Per-athlete thresholds replace the seeded guesses. Then we coach.

Where it’s going

NOW

v0 — Training analyzer

Per-station cadence/fade, consistency, HR recovery, run-fade. On TestFlight, collecting data.

Station-specific analyzers

Burpee broad jump is the richest signal — jump power, airtime, landing load, power fade. Then wall ball, sled, lunge.

Learned coefficients

Replace seed thresholds with per-athlete models trained on the collected sessions.

Race-day live pacing

Real-time HR caps + pace targets, re-computed as the race evolves. The only app that optimizes during the race.

MOAT

PepBear integration

Recovery + peptide protocol data feeds training load. Bidirectional flow nobody else can replicate.

Honest status

Real

· v0 on TestFlight (phone + watch), branded.
· Validated consistency metric (clean r=0.999 vs 0.000).
· Hardened capture: WR1 v3 + forensic raw, watchdog, per-user RLS.
· GPS-independent — works indoors.

Unproven

· Zero real workouts recorded yet.
· Rep thresholds are seeds — need real-data tuning.
· Manual lap tapping; no auto-segmentation.
· No coaching layer / race-day pacing yet.