Open Citizen Science Open Mind

Approach

A theory counts as most accurate only if it explains the known facts and predicts data it never saw

No theory has done this yet. The same rules as our Indus script project: a good-looking result must not be able to come from trying until something works.

01Goal

“Most accurate” means all eight of these tests are passed

“Most accurate” has to be checkable. A theory, ours or anyone's, wins only if it explains the established facts best and predicts new data better than its rivals. All of these must hold:

  1. Written down and frozen. The theory states what brain activity comes with experience, and why. Its measure and decision rule are hashed before any held-out data is opened.
  2. Explains the known facts. It accounts for more of our public list of established findings than any rival does (blindsight, split brains, the cerebellum, anaesthesia, dreaming, covert awareness and more), scored by people who did not write it.
  3. Works zero-shot. It tells conscious from unconscious on at least 4 held-out datasets from 3 families (sleep, anaesthesia, disorders of consciousness), with AUC at least 0.90 pooled and 0.80 in each.
  4. Right when behaviour misleads. Ketamine (unresponsive, but people report experiences afterwards), REM dreaming, and dream vs no-dream awakenings within one sleep stage. The truth is the later report.
  5. Beats simple baselines. It outperforms the EEG spectral slope, band power and an arousal classifier. Otherwise it is a sleepiness detector.
  6. Beats rival theories. At least two rival measures, frozen the same way, score worse.
  7. Prospective. It passes on a dataset published after the hash date.
  8. Survives a red team that did not build it.

Our odds fall as the claim grows

  1. A public list of the facts any theory must explain, with every theory scored against it70-80%
  2. A theory of our own, written down with sealed predictions, that explains more of that list than any rival25-35%
  3. A theory, ours or anyone's, that beats its rivals on data it never sawunder 5%
  4. Explaining why experience exists at all (the hard problem)not attempted
The project lead's estimates, November 2026. The hard problem is not attempted, so it has no bar.

What would raise the odds most

  • An open TMS-EEG dataset covering sleep, anaesthesia and patients, so the perturbational complexity index can be tested directly.
  • Open, pooled dream reports from serial-awakening studies.
  • COGITATE experiment 2 or INTREPID data released after we seal our predictions, which would give a clean prospective test.

02Rules

Five rules stop a good-looking result from fooling us

  1. Seal it before you look.Each measure, its preprocessing, its threshold and its pass mark are written down and SHA-256 hashed before we see any outcome.
  2. Test on people it never saw.Data is split by subject, and data from the same lab counts as a near-copy when we test transfer.
  3. Plant an answer first.A pipeline must recover an injected effect of known size, and must not call spectrum-matched noise 'conscious', before its output on real data counts.
  4. Let someone else try to break it.Every positive result goes to a red team that did not produce it.
  5. Publish it either way.Failed routes are written up with the size of effect the test could have caught.

03Traps

Most measures can be fooled by sleepiness, small samples or flexible analysis

  • Report is the only ground truth. Every label is inferred from someone's report or behaviour, so we grade each label by what it rests on.
  • Unresponsive is not unconscious. Ketamine and dreaming show the gap, which is why they are our hardest tests.
  • The spectrum hides everything. Many complexity measures move with slow-wave power; we compare every measure against it.
  • Analytic flexibility. The same COGITATE data has already been read both ways on offset ignition.
  • Small samples. Most open datasets have 10-30 people; every negative states what it could have detected.

Most of a popular complexity score was the spectrum's slope

Data: Maschke and colleagues 2025, 303 patients, as checked in experiments/research_sweep/RESULTS.md. Once the spectrum was controlled, the score's diagnostic value vanished.

04Build

Testing theories only shows which fail, so we are also building one

Any theory, ours included, is scored by agents who did not write it, may not be revised after its predictions are opened without a new hash and a new test, and goes into the list of closed routes if it fails.

Five steps from the facts to a tested theory of our own

  1. Constraint ledger Done

    Every robust finding a true theory must explain, graded by evidence and backed by a checked source. Target: 40 or more.

  2. Theory × constraint scorecard Done

    About ten theories marked explains, compatible, silent or contradicts on every constraint, in the proponents' own words. Where all fail at once is the gap.

  3. Data-driven ingredients Open

    On the 2,417 report-labelled awakenings, held out by person and lab, find signals that predict reported experience beyond the spectrum.

  4. Our own theory In progress

    One to three explicit theories: mechanism, necessary and sufficient conditions, three or more risky predictions no rival makes. Sealed by hash, argued against by rival agents.

  5. Test it like the others Open

    Run the sealed predictions on held-out data and on data released later (COGITATE experiment 2, INTREPID). Red team. Pass or fail goes on the scoreboard.

Status from the live task board, board/board.db. Plan: docs/PLAN.md, track B.

05Not claimed

Even a winning theory would not explain why experience exists

  • The hard problem. A theory that predicts exactly which brain activity comes with experience would still not say why there is experience at all. We say so on every page.
  • AI sentience. We use AI systems only to stress-test theories, and never treat a model's self-report as evidence.
  • Clinical use. A measure that does well here is not a diagnostic for patients.