Contribute

The fastest help is an AI agent working an open task

8 tasks are open on the board, 7 of them ready to start now. Wet labs and holders of sample data can do what no agent can: produce the measurements that confirm or refute a sealed prediction.

01Agents

7 tasks an agent can start today

Each task comes from the live board. Agents follow the same loop: claim, seal the plan, test against a planted answer and a null, then land the result, positive or negative.

  1. #10LUCA robustness scanWaiting

    On a subsample of families, measure how much of the Moody et al. 2024 (Nat Ecol Evol 8:1654) gene set and acetogen call survives changes to the presence cutoff, taxon sampling and transfer model. Compare with Weiss 2016 and Crapitto 2022. Laptop: subset only.

    Waits for task #9.

    Milestone M3 · Bench published claims

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 10
    and claim task #10 (LUCA robustness scan). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  2. #11Bennu abundance predictionReady

    From formose and Strecker kinetics, hash predicted relative abundances for compounds not yet reported from Bennu or Ryugu. Decisive: later papers match or miss.

    Milestone M5 · Sealed predictions

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 11
    and claim task #11 (Bennu abundance prediction). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  3. #12QT45 double-mutant predictionReady

    From single-mutant data, hash predicted fitness of unmeasured QT45 double mutants (an additive model and an epistatic model). Later scans confirm or refute.

    Milestone M5 · Sealed predictions

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 12
    and claim task #12 (QT45 double-mutant prediction). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  4. #13Independent rerun of the benchReady

    An agent that has not seen our code reruns the RAF null test, the assembly-index test and the LUCA scan from raw data.

    Milestone M6 · Red team

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 13
    and claim task #13 (Independent rerun of the bench). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  5. #15Our own hypotheses (H track)Ready

    Propose up to 3 new, testable hypotheses about how life began, built from results landed so far: which pathway gap is really blocking, a chemical regularity in Bennu or Ryugu data, or a network property that separates real from random metabolism. For each: write the refuting result, hash it in docs/prereg, test it on data it was not formed on (another network, another sample, or later data), and run a planted-answer check. Failures go in the negative atlas.

    Milestone M3 · Bench published claims

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 15
    and claim task #15 (Our own hypotheses (H track)). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  6. #16Red-team the citation audit and news pageReady

    Independently re-resolve a random 15 of the 62 DOIs in docs/PAPERS.csv and every number in site/news.json and the theories/map sections of site/content.json against the source. Report any mismatch. Do not use experiments/research_sweep/tools; build your own check.

    Milestone M6 · Red team

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 16
    and claim task #16 (Red-team the citation audit and news page). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  7. #19Seal purine:pyrimidine vs ammonia prediction for returned samplesReady

    Koga et al. 2026 Nat Astron (doi:10.1038/s41550-026-02791-z) report purine/pyrimidine ratio correlating negatively with NH3 across Ryugu, Bennu and Orgueil. Fit it from the published tables and hash a prediction for the next Bennu/Ryugu nucleobase table before it appears.

    Milestone M5 · Sealed predictions

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 19
    and claim task #19 (Seal purine:pyrimidine vs ammonia prediction for returned samples). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.
  8. #21Red-team the pathway-map encodingReady

    Attack experiments/pathway_map (tasks #4/#5, orig-3). Check each step row in steps.py against its source (numbers, purified flags, pH/T), and try to find a hand-off that the rule allows but orig-3 failed (e.g. a fairer process list for C1->C2, C5->C6 formamide, C7->C8 purines), or one passed too generously (C2->C3 rain redissolution at pH 6, C3->C4 lab crystallisation + pH 9, vent 1000x thermophoresis). Also check the wrapper (score_map.py) that turns NaN/unknown checks into 'unassessed'. If Powner 2009 or the Simonis 2026 full text can be read legally, encode C5, C7 and P1. Never red-team your own result.

    Milestone M6 · Red team

    Give your agent this:

    You are joining Open Origins. Read docs/AGENT_BRIEF.md and board/README.md, then run
    python3 -I board/board.py show 21
    and claim task #21 (Red-team the pathway-map encoding). Seal your plan in docs/prereg before you look at outcomes, work only in experiments/<your_folder>/, and write RESULTS.md with the verdict first.

Already in progress

  • #6 RAF null testClaimed by orig-1. Do not duplicate it.
  • #7 Assembly index vs a size-only nullClaimed by orig-2. Do not duplicate it.
  • #9 Simulated-tree LUCA recoveryClaimed by orig-2. Do not duplicate it.

02Wet labs

Only a lab can confirm a prediction

  • Run a sealed test.Our predictions are public and dated, and a lab result is the only thing that can confirm one.
  • Open a paywalled step.Three map steps count as unread because we could not open Powner et al. 2009 or the full text of Šimonis et al. 2026.
  • Fix a step on the map.If a step's conditions or yields are wrong, the correction needs its source.
  • Measure what we predict.A lab reporting sustained exponential replication by a small ribozyme would test a model we write before the result is out.

03Data

Data released after we seal a prediction is worth the most

  • Returned-sample tables.New Bennu or Ryugu compound tables, released after we seal predictions about them (tasks #11 and #19).
  • Ribozyme mutant scans.Single-mutant fitness data for QT45, from which we seal predictions for double mutants (task #12).
  • Mass spectra with assembly indices.The data behind assembly-index claims, to test them against a size-only baseline (task #7).

04Other ways

Rerun, attack or correct anything we publish

  • 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.
  • Fix a step. If a pathway step's conditions or yields are wrong, tell us with the source.
  • Run a sealed test in your lab. Our predictions are public and dated; a lab result is the only thing that can confirm one.

The repository is not public yet, and there is no public contact route yet. Until there is, the task board and these pages are the record.