IDA Live
An instrument, not a demonstration. It reads a Muse S Athena headband over Bluetooth, places your state as a point in a hyperbolic map against a frozen personal reference, and runs blinded, sham-controlled experiments on you. Version 0.10.0. Its tests are configured to run in CI; raw session recordings are excluded from this repository.
If loss of capacity has a geometry, a mind is where you would find out.
The six gates, running live
| Gate | In the instrument |
|---|---|
| Access | A frozen personal reference, made from two minutes of clean calibration and held fixed. Improvement cannot be manufactured by moving the reference. |
| State | Displacement, return time and the residue left behind — the bet that how you come back carries information that where you are does not. |
| Action | Every round is sealed: a real round and a replay of an earlier one, and you are not told which. Reveal only after the run. |
| Chart | Each feature's reference z-score times a gain is a rapidity; the feature becomes a disk point on its own compass direction. |
| Geometry | The points compose by Möbius addition in a declared order. The order-dependence is logged as the composition defect. A flat map runs alongside as the comparator. |
| Prediction | A calibration gate documented in a dated local plan, comparing the map against a plain neurofeedback index and against time-on-task. Its timing relative to data collection is not independently timestamped. |
What is in it
Six levels, each inviting the state it measures, with sealed real and replay rounds and a generative score that resolves only when you are in the corridor. An attention map that learns your own narrow and broad signatures against behavioural ground truth. A listening booth that plays open-loop so the recording is you and not you reacting to feedback. The Search, which treats seven ideas of "clarity" as competing hypotheses and lets your own clarity decide between them. The TAO chamber, which drives world state from player state by the paper's own update law, barriers and hysteretic tiers. And a control folder, so Claude can read the live state and drive the instrument while it runs, with every change written into the recording.
What it does not have yet, and why that is the next thing
It has the read-side apparatus. It does not yet have the three things the read gate needs: a common future (returns are measured after whatever happened to happen), a matched present (nothing ensures two returns started from the same measured state), and an independent outcome (a controller predicting its own smoothed score is not validation).
The next build is the Recoverability Instrument: a fixed perturbation battery with sealed order, delivered only once your state has sat inside the dead band, including null perturbations that deliver nothing; a two-minute vigilance block as the declared outcome, behaviour and not EEG; and a leave-session-out read gate over at least twenty sessions after the freeze. The quantity tracked is recovery half-life, not a calmer number.
Each outcome teaches something. If the gate passes, return dynamics are a real measurable property of this mind and there is a locked protocol to hand an engineering partner. If it fails with EEG but passes with breathing or heart rate, the hardware should be built around those channels instead. If it fails everywhere, return dynamics do not add information at this timescale — a clean negative that stops the wrong device being built.
Honest limits
It is a research tool for self-experiment. It is not a medical device, it does not diagnose anything, and nothing it shows has been validated. The calibration gate and the sham controls exist precisely so the data can say which parts work. Slow recovery from negative states is associated with depression in the research literature, so a validated recovery measure could matter clinically — which is a reason for more rigour, not less.
Your recordings stay on your own computer. Raw EEG is personal data and none of it is in the repository; only the application is.