How it works.
Tell Eri a goal or start from a catalog. Eri builds the program from 1,282 signals, including the driver's body. The program runs in the session and records the response. Eri proposes the next version, and you approve each change.
Run. Record. Revise.
How does a session run?
The runtime, recording and AI review are built. David trains with Eri on this loop today. Next, drivers preparing for 2027 run Eri-built programs alone.
Measure
Field-usedMeasure the person, the machine and the task together.
Car or simulator telemetry, driver inputs, timing, heart rate, HRV, breathing and EEG, aligned on one timeline with a quality check on each value.
Telemetry and timing were used at the Laguna Seca weekend in September 2026. A registry of 1,282 signals covers inputs, car, race state and physiology.
Write
LiveTell Eri a goal or start from a catalog.
A flow names the signal, condition, timing, cue and limits. Deterministic code runs it. Operators write their own. Performance engineers, coaches and pro drivers write catalogs to share or sell.
The compiled flow graph runs. Eri composes programs from 41 node types.
Run
LiveThe method runs in the session.
Sound, light, voice or on-screen cues, or a bounded change to a simulator, each with dwell, cooldown and suppression rules.
The compiled flow emits each command with its dwell, cooldown and suppression.
Record
LiveThe system records what was delivered and how the person responded.
Each cue, its timing and the next observation are stored together, including no response.
Each cue is stored with its timing and the observation that followed. A review decides whether the change helped.
Review
LiveEri and the operator review the results.
Eri compares the flows, drills and cues against deterministic measures of execution, operator input and vehicle output, and proposes a versioned revision.
One retained Eri review receipt. 209,825 tokens across three requests, 10,336 ms.
Revise
In developmentRun the approved version in the next session.
A cue that is not helping changes. A cue the operator has learned tapers off. The change can target the person, the setup, the interface, the workflow or the AI assistance.
Every version is kept. David trains with Eri on this loop, and the recorded session shows it end to end.
Stale signal, no cue. Probe laps hold the cue back to test whether the skill holds.
How does a sentence become a method?
Start from a shared catalog or create one with Eri. You approve the changes. The runtime executes the flows.
When she brakes past 85 percent for a second and a half, chime once, unless her heart rate is stale. Then stay quiet for three seconds.
A shared, machine-readable description of who is doing what, with which machine, toward which goal, and what the system may do about it.
- Signal
- Brake pressure
From the car or the simulator.
- Condition
- Above 85 percent
Normalized pedal travel.
- Timing
- 1500 ms dwell, 3000 ms cooldown
So the cue is not noise.
- Cue
- Single warning chime
Sound, light, voice, haptic or a simulator action.
- Taper
- Fade as the skill holds
Probe laps withhold the cue.
- Authority
- Approval required
Eri may propose. The operator approves.
- signal
- brake_pressure
- signal
- heart_rate
- when
- brake > 85%
dwell 1500 ms - unless
- heart_rate.stale
- cue
- chime once
- then
- quiet 3000 ms
- authority
- approval_required
Modeled example.
Know-how, written as a method.
Tell Eri a goal or build a flow yourself. Choose the signal, condition, timing, cue and limits. An expert's catalog gives you a head start.
- signal heart_rate
- when heart_rate > 170 bpm for 10000 ms
- cue say "breathe out"
- authority approval_required
Eri proposes. You decide.
− dwell 1500 ms
+ dwell 1200 ms
The chime arrived after the brake peak on most laps. Replayed on the last 10 laps.
Modeled example.
Eri reads the session and writes an exact change. Each proposal is replayed on recorded laps before you see it.
Nothing applies without approval unless the operator allows it. Then the loop runs again.
In the session, cues come from the approved flows. Eri proposes revisions after reviewing the record.
What Eri sends back.
Eri reads the whole session against the method. It answers in four kinds of result that the software can run, test and keep.
Each proposal is replayed on recorded laps. Every shift has to hold in the next one.
- Vehicle telemetry
- Heart rate and HRV
- EEG features
- Video and audio
- Competitors on track
- The operator's method
- Past sessions
eri.review(session, method)→ Hold | Shift | Proof | Revision
-
Hold In development
What already works. No later change may break it.
- skill
- brake release
- where
- Turn 1
- baseline
- frozen
-
Shift In development
One thing to change. Never more than three at once.
- cue
- "squeeze, then wait"
- landmark
- the 2 board
- reference
- best lap
-
Proof In development
The next session's verdict on the last shift, against the frozen baseline.
heldbuildingslippedunclear
- verdict
- held
- confidence
- medium
- signedOffBy
- operator
-
Revision Live
An exact edit to the method itself. It is compiled for review and replayed on recorded laps.
- op
- dwell 1500 → 1200 ms
- replayed
- 10 laps
- apply
- approval required
- CaptureEvery source on one clock.
- OntologyCorners, laps, cues and people as named things.
- Rules engineThe method runs every 50 ms.
- EriReads the session. Returns exact results, not paragraphs.
- AuthorityThe operator sets how far a decision may go.
A better model makes each decision sharper. The kinds of result, the method and the approval stay the same.
Modeled example.
Eri adapts each driver's program.
In development
Eri proposes each change from that driver's own sessions. The driver approves it, and every version is kept. Five drivers who start from one catalog each end with their own program.
Eri adapts each catalog to the driver who runs it. In development
- Cue that lands
- Chime, before the brake point
- Laps to hold a change
- 6
- Says less when
- Heart rate above 175
- Next
- Hold the cue back on lap 4
Measured in time to skill.
Each method records the laps, minutes and sessions it took to reach the operator's standard. That is the number the loop is built to shrink.
- Laps
- 38
- Minutes
- 52
- Sessions
- 3
Modeled example.
Standard. 4 of the last 5 probe laps.
- Pass
- Pass
- Fail
- Pass
- Pass
How far may Eri go?
Six levels, from watch-only to bounded automatic. Automatic apply is off by default.
A patch applies only with trusted and fresh source quality, remaining recursion budget and a passed cooldown. Model output cannot write arbitrary graph paths, policy fields or the recursion budget.
This is a built boundary, not proof that a revision improved an outcome.
What runs today?
| Status | Meaning | Carried by |
|---|---|---|
| Live | Built and runs in the product today. | Loop. Write, Run, Record, Review Signals. Telemetry, Heart rate Authority. Observe only, Approval required Products. DRIVERFORM, DRIVERFORM for iPhone, DRIVERFORM for Watch, Eri, Machine Nerve |
| Field-used | Used at real race weekends. | Loop. Measure Products. DRIVERFORM Desktop |
| In development | Built in part. An integration gap remains. | Loop. Revise Authority. Recommend only, Auto-apply local, Auto-apply feedback, Auto-apply adaptation Products. FlightDeck, STINTFORM |
| Research | Experimental feature or estimate. Not a diagnosis or a validated measure. | Signals. EEG features |
| Proposed | Designed. Not built yet. |
Not tied to one logger.
Source-agnostic by design. Simulator streams and real-car recordings feed a common performance workflow, during the session and after it.
Live sources run methods in the session. Imported files feed review and trends. Each value carries its source, format, version and quality.
Physiological signals are context for performance review. They are not diagnoses, emotion detection or fitness-for-duty measures.
Machine Nerve is not a medical device and does not provide clinical diagnosis, treatment, or medical readiness certification.
| Source | Path | Boundary |
|---|---|---|
| iRacing | Live relay and file import Live telemetry, CSV and IBT Built paths | Live delivery and recorded-file analysis have different clocks and rates. |
| Microsoft Flight Simulator | Live adapter and replay Enabled SimConnect profiles and replay bundles Built paths | Flight simulator profiles run the same methods as racing. |
| MoTeC | File import i2 CSV exports Built CSV path | i2 CSV exports import with their channels and units. |
| Cosworth | File import Qualified PDS recordings Qualified variants | Direct decoding covers the qualified Porsche and Cosworth layouts. |
| VBOX | File import CSV and native VBO Built import paths | Channels, timebase and companion video alignment depend on the recording. |
| AiM | File import Recognized Race Studio CSV exports Built CSV boundary | The CSV adapter keeps the raw source data. |
| RaceChrono | File import CSV exports Built import path | Column mappings follow the exported recording. |
| Garage61 | File import CSV exports Built import path | Recorded sessions import for review and trends. |
| Bosch | Compatibility request Representative export required Not verified | Send a representative export and we qualify the format and channels. |
Which signals does it record?
The machine, the track, the rivals, the body, the eyes, the voice and the video. Recorded together, so Eri reads one moment whole.
Body
- ECG Live
Polar H10 · 130 Hz
- PPG, multi-site Live
Polar, Muse, WHOOP · Up to 176 Hz
- EEG Research
Muse headband · 256 Hz
- Motion Live
Strap and headband · Up to 416 Hz
- Skin temperature, SpO2, respiration Research
WHOOP strap
Eyes and head
- Gaze, blinks and head pose In development
iPhone face tracking
Track and field
- Rival positions and gaps Live
Sim relay · 25 to 32 Hz
- Live timing and pit records Field-used
Timing feeds · Per lap
- Weather Live
With telemetry
Machine
- Live sim telemetry Live
iRacing, Microsoft Flight Simulator · 60 Hz
- Recorded car data In development
MoTeC CSV, Cosworth PDS, VBOX, AiM CSV, RaceChrono
Video
- Camera video In development
VBOX video and cameras
Voice
- Speech to text Live
Whisper, Qwen3-ASR · 16 kHz
- Voice physiology Research
Audio physio pipeline
Start a pilot.
Start from a catalog or create one with Eri. Run your session and review Eri's proposals yourself.