The standardised motor-intelligence test battery for coaches and teachers.
Phone-native. Measurements of 8 key motor capabilities. Same protocol, same scoring · comparable across classes, schools, regions, countries.
PE teacher · sports coach
School · DoE · academy · parents
Comparable cohorts at scale
- Standardisation · same protocol, same scoring, same camera anywhere
- Coverage · 8 motor capabilities tracked in <10 min / athlete
- Comparability · class · school · regional · national benchmarks
- Group throughput · multi-athlete capture, no proctor bottleneck
WHO youth-movement guidelines · ISB testing protocols · FITNESSGRAM · extended for AI-camera scoring at school scale. Designed with Dr G.F. Joubert · Human Movement, CPUT.
5 paying schools · POPIA pack complete · bib OCR ≥ 90% · ~50K labelled reps in corpus.
| Capability | Test | Output |
|---|---|---|
| Locomotor | 10 m Sprint | seconds |
| Standing Long Jump | cm | |
| Coordination | Jumping Jack | reps · rhythm · errors |
| Strength | Push-Ups (modified, younger grades) | reps |
| Endurance | Repeated Sprint | distance · fatigue index |
| Balance | Single-Leg Balance (eyes closed) | seconds held |
| Reaction | Ball Drop Reaction | milliseconds |
| Agility | Zig-Zag Agility | seconds |
| Laterality | Cross-Crawl March | reps · symmetry |
One vision pipeline powers all 9 tests. New tests slot in as scoring modules · they don't fork the system. Every captured rep aggregates into a single corpus that improves every other test.
Every test runs on the same engine · different scoring module, same capture, same review flow.
Each tile below is a real capture from a school pilot. Pose · tracking · classifier · scoring all live.
9 tests, 1 pipeline. Pose + tracking + classifier are shared. Adding a new test is a new scoring module and a config · not a forked product. Every captured rep aggregates into a single corpus that makes every other test more accurate.
From measurement to coaching.
Schools fund the wedge · corpus funds the coach.
PE teacher · youth-development trainer
School orchestrates · parents + school pay subscription
Schools relationship from V1 → parent acquisition funnel
V1 is a measurement tool (commoditises over time). V2 is the coaching companion · sticky, high-LTV · trained on the dataset V1 generates at school-test scale.
V1 generates the three things V2 needs to coach: classifier labels · progression curves · error patterns. New in V2: movement template engine + on-device inference. Carried over: data-gen stack refined on V1 + movements dataset compiled through Phase 1 collaborations.
V1's moat (protocol library + POPIA + child-population data) + V2's (classifier labels + progression curves + error patterns) unlocks Tier 1B (PTs / gyms) and 1C (insurance). V2 is the bridge into Stages 3 and 4.
A number. A rank. Just the score.
"Strong jump. Arm swing softened on rep 3 · try a sharper drive."
Per-rep biomechanics · progression curves · population-aware cues · auto-narrated reports.
3 movements with form-correction live · trainer-rated cue quality > 80% · cohort improvement ≥ 8% on re-test · movement engine in prod.
Phone records the rep
Engine extracts the measurable
Trainer fixes any AI mistake
Correction trains the next session
The trusted health partner. Insurers engaged from day one.
Personal trainer · gym owner
Gym chains pay subs · insurers buy outcomes
~10M reps/gym/yr · every trainer override = free label
Gyms are the surface. Insurers buy the outcomes. Verified-movement points + member benefit ride on top of every gym deployment.
Primary entry, zero travel. Virgin Active SA ~100 sites @ R1.5K$80£65/site/mo = R1.8M$95K£78K ARR from one chain.
Mature trainer culture · sector consolidating (Pure Gym, David Lloyd). Cofounder network unlocks chains. GDPR over POPIA = ~4 weeks.
Largest TAM, most fragmented. Enter after 3+ chain proof points. NIL-funded colleges + chains in parallel.
- Consumer-direct competitors miss the trainer · the channel that owns the client
- They train on adults; ours generalises across populations from day one
- They add movements as engineering; we add as configuration
- No youth or insurer-grade data anywhere else; we have both
50+ paying gyms · >$500KR9M£400K ARR · cross-pop pose validated (<5% drop) · 50+ movement library · health partner engaged.
$0.5–1.5M ARRR9–27M ARR£400K–1.2M ARR
Verified movement quality becomes the new health currency.
Insurer member · gym client
Life & health insurer · wellness programme
Insurers want verified outcomes · we are the proof layer
Discovery's Vitality is the global model · points for gym check-ins, steps, healthy buys. Today those signals are presence, not quality. We add the missing layer: verified movement quality · every rep, every test, scored. That's a new tier of points, claim evidence, and underwriting input · direct API into the insurer's reward engine.
Nobody bridges school + gym + insurer-grade verification. Every gym session feeds the points engine. Every points payout reinforces the gym. The same engine, two business models, one data flywheel.
- Verified-quality points API · every scored rep can mint reward points
- Member benefit surface · premium reductions, in-app rewards, partner perks
- Risk-segmented coaching · same engine, member-tier-aware cues
- Programme attribution · verified outcomes at 90 / 180 days
- Compliance pack · POPIA · GDPR · per-jurisdiction data residency
Discovery · Momentum Multiply · Bonitas
Vitality UK · Bupa · AXA Health
John Hancock Vitality · Oscar · employer wellness
AIA Vitality · Manulife · global re-insurer pilots
$0.3–0.8M ARRR5–14M ARR£240K–640K ARR + per-member-month rev-share unlocks here.
A phone-native AI vision stack that turns video into structured human-performance measurables.
Phone capture
60fps · landscape-locked · pre-flight scene QA · bib OCR
On-device pose
Quantised pose model · on-device runtime · privacy-preserving
Tracking & segmentation
Multi-athlete tracking · phase & rep windows · scene-stable IDs
Movement classifier
Per-movement scoring module · fault & quality labels
Measurables
Reps · ROM · velocity · symmetry · protocol scores · audit log
- Protocol-scored results · FitnessGram-equivalent scores on every phone, no proctor
- Per-rep biomechanics · joint angles · ROM · tempo · symmetry · fatigue curve
- Classifier labels · exercise · phase · fault tags (knee-valgus, butt-wink, depth, tempo)
- Demographics-linked records · age · sex · school · sport · longitudinal student ID
- Auditable trace · raw video → pose → score, end-to-end reviewable
- Paying customer from day one · schools + DoE contracts fund the build
- Class-scale labelled capture · 30 students × 9 tests in < 60 min, per teacher
- POPIA-cleared · pack portable to GDPR & FERPA in weeks, not quarters
- The workflow is the label loop · trainer override in-app = a labelled rep · fastest path to a real-world annotated dataset
The same stack, now feeding a vision-language coaching layer · real-time feedback, Q&A, audience-aware reports.
+ corpus
+ reports
- Real-time form feedback · "knees out, rep 3" mid-session
- Population-aware cues · same fault, different cue for a 12-yr-old · 45-yr-old · returning trainee
- Natural-language Q&A · "why was rep 3 bad?" · "what should I work on?"
- Auto-generated reports · parent narrative · trainer brief · clinician note · insurer summary
- Multi-language output · UK/US English · Afrikaans · isiZulu · ES · FR · DE · NL · PT
- Self-hosted, corpus-tuned VLM. Our model, our weights, our infrastructure.
- Library is a framework, not engineering. Adding a new exercise or insight is configuration. Adapts fast to any vertical.
- Data compounds. Every session sharpens the classifier and re-tunes the VLM.
- Structured signals in, not raw pixels. Pose + classifier outputs feed the prompt. Auditable end-to-end.
