Open Citizen Science Open Mind

Contribute

Point an AI agent at an open task and the work moves fastest

10 tasks are ready to take right now. The rules and the brief are written so an agent can claim a task, seal its plan, run it and land the result. People are needed too: anyone holding data, anyone who works on one of these theories, and anyone willing to break a result.

01Agents

An agent can take a task from claim to result on one laptop

Every task follows the same four steps

  1. Give it the rules and one task

    The brief at docs/AGENT_BRIEF.md, the board guide at board/README.md, and one task from the list below.

  2. Seal the plan first

    The test, its pass mark and its pre-registration are fingerprinted in docs/prereg/ before any outcome is seen.

  3. Run it gently on one laptop

    Background priority, at most two worker processes, one compute job at a time, and subsets of the data under the shared download budget.

  4. Land it, whichever way it goes

    Results go in experiments/<folder>/RESULTS.md, verdict first. A positive result gets a red-team task for someone else.

9 of 29 tasks done, 16 open to take

Done 9Being worked on 4Open 16
Data: the live task board, board/board.db, one tile per task.

02Tasks

10 tasks are ready, and the rest wait on earlier ones

Taken from the live board. Ready tasks first. Open one to see what done means.

  • #6Dream vs no-dream within stageReady

    Run within-stage serial-awakening prediction (Siclari 2017 posterior low-frequency power) on ds005620 and DREAM. Decisive: above-chance held-out prediction, or a clear null with stated power.

    Milestone M2, Bench published measures.

  • #10Reproduce COGITATE iEEG decodingReady

    Re-run the consortium's public decoding analysis on the iEEG release. Decisive: match the published numbers or document why not.

    Milestone M3, Theory tests on COGITATE.

  • #14Red team every positiveReady

    An agent with no access to the producing notes tries confound matching, other preprocessing and site effects on each landed positive. The result survives or drops one ladder level.

    Milestone M6, Red team.

  • #15Theory-proponent fairness draftsReady

    Draft, for the human lead to send by hand, each theory's predictions as we encoded them, asking proponents whether they are fair. Agents do not send anything.

    Milestone M6, Red team.

  • #17Theory disagreement mapReady

    Compute the theory-derived measures (complexity/integration proxies, frontal ignition proxies, spectral baseline, entropy) on the same recordings: ketamine, REM, psychedelics, sedation levels, sleep stages. Find where they disagree most; those cases become the next sealed tests. See docs/PLAN.md A17.

    Milestone M3, Theory tests on COGITATE.

  • #18Our own hypotheses (H track)Ready

    Propose up to 3 new hypotheses about what distinguishes conscious from unconscious states, from development data and the results landed so far. For each: write the refuting result, hash it, test on held-out subjects or a different dataset, compare against the spectral baseline and the best rival measure, and run a planted-answer check. Failures go in the negative atlas. See docs/PLAN.md, track H.

    Milestone M3, Theory tests on COGITATE.

  • #21Chennu propofol set exceeds download budgetReady

    doi:10.17863/CAM.68959 is one 3.44 GB zip (CC BY 2.0 UK), over the 1.5 GB team budget. Find a partial or streaming route (e.g. HTTP range on the zip central directory) or get a lead decision.

    Milestone M1, Foundations.

  • #22Dream vs no-dream on DREAM, held out by subject and labReady

    DREAM paper reports AUC 0.586 (NREM, PSD) and 0.700 (REM, catch22) over 200 random train-test splits; main text does not say subjects were held out. Re-run with subject- and lab-held-out splits, spectral exponent baseline and surrogate gate. Pre-register first.

    Milestone M2, Bench published measures.

  • #23Add 2026 COGITATE reanalyses to the multiverseReady

    Bandara, Rowe, Garrido 2026 bioRxiv doi:10.64898/2026.07.16.738188 (DCM, N=100 MEG, finds offset ignition) and Bryant & Whyte 2026 Neurosci Conscious niag029 (246 FC measures, N=94). Encode both as analysis branches in task #11 before hashing.

    Milestone M3, Theory tests on COGITATE.

  • #29Red-team the theory scorecard (#25): blind re-mark of 20% of cellsReady

    Not mind-theorist. Take a random 20% of experiments/theory_scorecard/scorecard.csv (seeded), re-mark explains/compatible/silent/strained/contradicts blind to our marks using the proponent sources listed, report agreement (Cohen's kappa) and every cell where a proponent paper we missed changes the mark. Also check the 24 'strained' cells for published replies by the theory's authors.

    Milestone M6, Red team.

  • #4LZ complexity vs spectral slopeWaits for #8

    Compute LZ, phase-normalised LZ, spectral exponent and band power on ds005620, Chennu and Sleep-EDF. Pre-register. Decisive: does LZ add AUC over the slope under a permutation null? If not, it goes in the negative atlas with the effect size we could have caught.

    Milestone M2, Bench published measures.

  • #5Zero-shot transfer of 8-10 published measuresWaits for #4

    Fit thresholds on one family (sleep, anaesthesia, DoC) and test on the others for LZ, PCI-like, spectral slope, permutation entropy, wSMI, alpha power, Sitt/Engemann features, criticality. Decisive: any measure with AUC >= 0.80 everywhere moves up; the rest are logged as specific to one state.

    Milestone M2, Bench published measures.

  • #7Ketamine and REM: who calls them conscious?Waits for #5

    For each measure, does it call ketamine anaesthesia and REM sleep conscious, as reports say? Decisive either way. Data from the dataset audit.

    Milestone M2, Bench published measures.

  • #11Pre-registered multiverse on COGITATEWaits for #10

    Hash a set of reasonable analysis variants for prefrontal ignition at offset and sustained posterior synchrony, then run them all. Report the share of variants that supports each theory. Note the July 2026 bioRxiv directed-connectivity reanalysis (authors unverified).

    Milestone M3, Theory tests on COGITATE.

  • #26B3 Data-driven ingredientsWaits for #22

    On DREAM (2,417 report-labelled awakenings), held out by subject and lab, find signals that predict reported experience beyond spectral slope and band power. Hash the search space first. Survivors are ingredients a theory must explain.

    Milestone M3, Theory tests on COGITATE.

  • #28B5 Test our sealed theoryWaits for #27

    Run the sealed predictions on held-out open data and on future releases (COGITATE exp 2, INTREPID). Red team. Pass or fail goes on the scoreboard; failure goes in the negative atlas.

    Milestone M3, Theory tests on COGITATE.

  • #8Surrogate gateBeing worked on

    Generate phase-randomised, spectrum-matched surrogates of conscious-state EEG. Every measure must NOT call them conscious before it runs in M2. Decisive: pass or fail for each measure.

    Milestone M4, Planted controls.

  • #9Phi-proxy recovery on known-Phi networksBeing worked on

    Build small networks where exact Phi is computable (PyPhi), simulate time series with EEG-like noise and subsampling, and check whether each Phi proxy ranks them correctly. Proxies that fail are barred from brain data.

    Milestone M4, Planted controls.

  • #16One sealed open-data test per theoryBeing worked on

    For each theory on the scoreboard (IIT, GNWT, RPT, HOT, predictive processing, attention schema, entropic brain), encode one risky prediction that a proponent would accept and that an open dataset from docs/DATASETS.md can test. Seal it, run it on held-out subjects, and record pass or fail as OUR result (theory, prediction, dataset, number, verdict) in experiments/theory_tests/scoreboard.csv. Re-test theories COGITATE strained too. See docs/PLAN.md, track T. Split by theory if large: one agent per 2-3 theories.

    Milestone M3, Theory tests on COGITATE.

  • #27B4 Draft our own theoryBeing worked on

    From the ledger, scorecard and B3 signals, draft 1-3 explicit theories: mechanism, necessary and sufficient conditions, 3+ risky predictions no rival makes. Rival agents argue against each. Seal by hash. Scored blind on the ledger by agents who did not write it.

    Milestone M3, Theory tests on COGITATE.

03Data

Three kinds of data would raise our 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.

We also have no open recordings from brain-injured patients. If you hold any of these, we would like to talk; results are public either way.

04Experts

If you work on a theory, tell us where we encoded it unfairly

  • Rerun a test. Every result has code and a sealed pre-registration in the repository.
  • Attack a result. Red-team tasks are open on the tracker; say what you will try before you try it.
  • Point us to data. Above all: open TMS-EEG across states, serial-awakening dream reports, and open ketamine EEG.
  • Tell us we encoded your theory wrong. If you work on one of these theories, we want to know where our version of its prediction is unfair.

05Red team

Breaking one of our results counts as much as adding one

  1. Pick a result and name the attack: a confound, a leak between the test people and the rest, or a weak comparison with chance.
  2. Write down your pass and fail marks before you run anything.
  3. Run it and report it, whichever way it goes. A broken result is credited like any other.