Cycling analytics, reimagined

Predict
performance.
Not potential.

Divergence applies machine learning to your physiological data — surfacing insights that tell you not just how you rode, but how you will ride.

Connects to Garmin, Wahoo & Strava

Heat Adaptation — last 90 days

ML model active

live

FTP (estimated)

418w

Heat Adaptation

72/100

Recovery score

39/100

Form index

109

Model detecting upward trend — updated daily from connected devices.

Platform

Built for riders who need
answers, not more charts.

Every feature is oriented around a single question: what should you do next to get faster?

01

Predictive Power Modelling

A model trained on your ride history forecasts where your functional threshold is heading — and what's driving the change.

02

Physiology Fingerprinting

Your heart rate response, power curve shape, and fatigue signature are unique. Divergence builds a model of you, not a population average.

03

Race & Event Simulation

Drop in a course file. Get a lap-by-lap pacing model, projected energy expenditure, and likely fade points — before you pin on a number.

04

Recovery Intelligence

Multi-signal recovery scoring that blends training stress, sleep quality, and HRV into a single, actionable daily readiness score.

05

Adaptive Training Zones

Zones recalculated from your actual recent performance — not anchored to a single test you did six months ago.

06

Segment Benchmarking

Understand exactly where you gain and lose time on climbs relative to your own baseline, with power-to-gradient breakdowns.

ML engine

A model that learns
you.

Divergence's inference pipeline ingests raw sensor streams and constructs a personalized physiological model — updated continuously, not on a quarterly test schedule.

  • Personalized, not generic

    The model weighs your own history more heavily than population-level priors. Accuracy improves as your ride log grows.

  • Uncertainty-aware

    Every prediction includes a confidence bound. Divergence tells you when to trust the model and when to trust your legs.

  • Explainable outputs

    Feature attribution shows which signals drove each prediction. No black boxes, no inexplicable numbers.

  • Continuous learning

    The model updates after every activity — not on a quarterly test cycle. Your current form is always reflected.

Inference pipeline

01Raw sensor ingestion94%
02Feature extraction88%
03Physiological encoding96%
04Personalization layer91%
05Uncertainty-aware output85%

Transformer-based architecture. SHAP attribution on every prediction.

Integrations

Meets you where
your data already lives.

Divergence connects to the devices and platforms you already use. No manual uploads, no data gaps.

Garmin Connect

Device & platform

Wahoo ELEMNT

Device

Strava

Activity platform

Polar Flow

Device & platform

TrainingPeaks

Training platform

Oura Ring

Recovery & sleep

Whoop

Recovery & sleep

FIT file import

Universal format

More integrations in development.

Early access

Your next breakthrough
starts with the data you already have.

Divergence is currently in private development. Request access and we'll be in touch when a spot opens.

No spam. No marketing cadence. Just a note when we're ready for you.