Leadership Under Uncertainty
Methods

Methods ledger

Every protocol, amendment, readout and correction the project has sealed, with its fingerprint and the public commit that dates it. 72 records, exported at 16:34 UTC on 27 September 2026. They carry 65 distinct fingerprints, because a study run that reproduces an earlier run’s numbers exactly has the same fingerprint.

This is a record of commitments: what was sealed, and when it reached the public record. It makes no claim that any hypothesis holds or that any forecast was right.

What a listing shows

When a record is sealed, the project takes its fingerprint, the SHA-256 hash of its content. The fingerprints of a day are combined into one tree, and the tree’s root is committed to a public GitHub repository, cberzins1973/luu-attestation-anchors. GitHub records when each commit was pushed. That time comes from a clock the project does not control.

A listing therefore shows that a record with this fingerprint existed no later than the push time of its commit. It does not show that the record existed any earlier. The time at which the project’s own system wrote the record is given too, as the project’s claim.

The ledger shows fingerprints, not content. A protocol’s text is available on request, and anyone who receives it can check that it hashes to the fingerprint listed here. Verdicts that have been reviewed are published on the study registry.

How much of it reached the public record late

71 of the 72 records entered the public record more than two days after the project’s system wrote them. They were written between 10 June 2026 and 14 August 2026 and first appeared in anchor commits pushed on 8 September 2026. For those records the public record shows that they existed by then, and nothing about when they were written. One of the records entered the public record within two days of being written.

The records

Protocols (31)

A study’s plan: its hypothesis, its measure, its analysis and the result that would count against the hypothesis. The fingerprint is taken when the protocol is sealed.

  1. Protocol: Generating analysis gets cheaper faster than evaluating it

    Study jco-p1, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 29f708ff4c97217ab3cf12eb3fa95a61916c10aeb02c92377d3a3c50fac65e8c

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  2. Protocol: Where fixed rules pay and where live judgment pays

    Study jco-p2, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 351b80f6b9ae6ce290ead309e39890e1aec8000729e8b03e48d996ba54f4a3d5

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  3. Protocol: More judgment fixed in rules raises the value of overseeing it

    Study jco-p3, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint f903494da1d8545d72689a3df5f6d8f9480140cc0352c79f109d17d64f43b40a

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  4. Protocol: Structured records compared with prose

    Study jco-p4, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 57ec23ea7073d6537b234275dfd0eb013ab633a51ff27bf7da3df55a5e236b81

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  5. Protocol: Reviewers from different model families catch more weak judgments

    Study jco-p5, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint f21ca9ba6993a7093c71ceaf5b63594a2d71d446901ee0195100991615d7ca07

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  6. Protocol: People and models together, under explicit rules

    Study jco-p6, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint a8cdc179f71f2aaaa4908ad51f74abe9f4ffde4bd0dcf9dce3045fab5838bac5

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  7. Protocol: Where live judgment adds value in allocating capital

    Study jco-p7, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 14e48d79eb71dbac73738a6b352a43e0ff61f24b6259192de6b2dc9f9f1216f2

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  8. Protocol: How much work centres on judgment

    Study jco-p8, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 3614381c967c6dbc8f1c7f0614c8229750f456ab7cd9d1786fd5e28d28cb2033

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  9. Protocol: When machines forecast as well as experts

    Study jco-p9, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint e4187c2964464e59b49ee66a855cb224cd72a933a609e1414fb2c33c11c999dd

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  10. Protocol: Questions a market could price

    Study jco-p10, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 04026d71f6bc869dfaabc42a35152813e1da53a7f87f640b82b2891795cb3fda

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  11. Protocol: Which kinds of system close the loop

    Study jco-p11, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 046590394626f54903a6627f3b1c43e3a3db02e38cf1435258946e63acbd3848

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  12. Protocol: Oversight as varied as what it oversees

    Study jco-p12, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 59fe212237b08af267ee7d9bef422b88af5ee8a2177204e0dc3493a3d49d6b4c

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  13. Protocol: Governance keeps its value as models improve

    Study jco-p13, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 4f17b29db55915270ad25db236a3810a037aa96c07921d3550652976329bc0b9

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  14. Protocol: Removing each element of the shared representation

    Study sco-s1, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint b078f2521232fa1ff860f933bd572c2a364adbcb201efd1356a00c0695f54a9e

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  15. Protocol: A first live calibration run

    Study scb-s5, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint a7b0d15d892f10360977f3de02fe4c78d61baaf6edb879c34f986ea8b290c0c5

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  16. Protocol: Calibration of each kind of insight over 90 days

    Study scb-s6, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint e07bf5cfcd531730011a9d3668fee024f318e1d3b74405eed3527100a30bdd8b

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  17. Protocol: The shared representation in law and science

    Study sco-s2, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 2d8273109af92453667a0c4d9d78e04f857f680b86c2cd72623bd3675b3d6dd5

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  18. Protocol: Two validators, one verdict

    Study sco-s3, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint c7914e1c05bb660cbde0cb1f74864d09afc6f894971c328e2af9bce5e17d5448

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  19. Protocol: Independent implementations of the handoff format

    Study sco-s4, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint d96297d67361d2ef96b19b309b2de0cbac079425ca712d37a41f893d6d1707e8

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  20. Protocol: First reference run: twelve hypotheses on AI computing

    Study rr-001, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 70a3682cac096decd1cf3ff179232041baa79266324ac7f78e927a59055ba02e

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  21. Protocol: A performance record for each insight method

    Study scb-s8, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint b62a7ee826b7d16e0a3d7ed7d83b8b1c4c274d5789df7bd1a5db33d3ce681f20

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  22. Protocol: Does an insight read as written by a person or by a model?

    Study scb-s9, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 59cf00517b47b881ed45140266f92469ec3406efa4c78a2f59a226114f99385c

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  23. Protocol: Patterns shared across similar readers, released only in groups of five or more

    Study scb-s10, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 6d89c2ece4211e22872e96d9a042f28a942b4c462f3d21a2f68c6c18489a8fc8

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  24. Protocol: Board oversight with and without assistance, 16 participants

    Study board-s14, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 21e311ba9696172c0c03dfca12dd099ee743bd2cb47f4e619598ad9ebffc83cd

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  25. Protocol: Every surfaced insight carries its seal and its record of change

    Study scb-s17, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 496b6545e00ee823143af27e2100c3f80afb78b7db47e4d8afed4e76f1b107cc

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  26. Protocol: Several domains, run continuously

    Study sco-s15, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint d799d105e1ba1936599ca6074dfb2b24834da917ffd09dfc646abf338e4da27a

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  27. Protocol: Stability of causal strengths learned from outcomes

    Study sco-s16, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 43ae1ee07fa136259bf1a930ef7c8a8fcbe15a323ae2567df8416227976e7f32

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  28. Protocol: Insights compared with a hired analyst, judged blind

    Study sip-s11, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 51571ebefa15674ae6913c1751c386612694bc6ed7fb73ced6c9ed223f63fd90

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  29. Protocol: Retiring and adding insight methods each quarter

    Study sip-s12, listed on the study registry.

    Written at 17:58 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint abc8a76a4ba831ba860ec94ac329b78b329808b5178fc019eb883966e5ad2cfc

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  30. Protocol: Whether attribution matters where accountability is central

    Study sco-s1b, listed on the study registry.

    Written at 13:25 UTC on 13 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 87 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint cbdf4e12a8cd6a68b809dbdd1f5953f3f4fdd73b95fa09c8001ab8b566d75579

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  31. Protocol: Do retrieved resolved questions reduce forecast error?

    Study icl-exemplar-brier-v1, listed on the study registry.

    Written at 09:48 UTC on 18 July 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 53 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 90987d5407414e885f5244654eb208014a718d4ea6241f082d2e4d2fa2d17ae7

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

Amendments to an analysis plan (1)

A signed change to an analysis plan, sealed with the fingerprint of the signed document, so the change and its date sit on the record beside the plan it changes.

  1. Amendment ep-prereg-v2-amendment-a1 to the analysis plan ep-prereg-v2

    Sealed with the fingerprint of the signed amendment document, eb26bd96f342d19d024ad20b68c6ffcc138235ca483745ff975b392cc6d8bc0b.

    Written at 07:40 UTC on 8 August 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 32 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 1c62bff0fbcb0ccb4e5e59dee47c454459cf8e7c760e0b94f7f27c1567209423

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

Candidate configurations (3)

A proposed change to the forecasting configuration, sealed with its settings, so that the configuration a gate verdict refers to can be checked.

  1. Candidate configuration market-blend-on

    Written at 13:49 UTC on 18 July 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 52 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint dcf51d7b54a8a6fa9bbc0a7827335286dd72382d812be0d06110d69c4dacd939

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  2. Candidate configuration evidence-forward

    Written at 18:15 UTC on 14 August 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 25 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 230f8730b8904e9534f91e4c500ff70572cd9f3d102a8a25129dc1469a141c9a

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  3. Candidate configuration no-base-rate-prime

    Written at 18:15 UTC on 14 August 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 25 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 61ab2c400a21f6e584ee11f9fdf8c15d84d86a2a0c25bb798d43ff605b1143a9

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

Results notes (26)

The numbers one run of a study’s analysis produced and the verdict they imply, sealed when the run is recorded, whether or not that run is the one reported. A verdict is published on the study registry only after review.

  1. Results note: How much work centres on judgment

    Study jco-p8, the run recorded at 17:59 UTC on 10 June 2026.

    Written at 17:59 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 81eb77b6d33cd3d74fcb98159e04edc83eb04de9b2d526a72dabc46efb5d37ea

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  2. Results note: Reviewers from different model families catch more weak judgments

    Study jco-p5, the run recorded at 17:59 UTC on 10 June 2026.

    Written at 17:59 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint f84e97dc6f72dbb10d4b8272cff8d237781f3419298820f97821179b7cb05bd3

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  3. Results note: A first live calibration run

    Study scb-s5, the run recorded at 18:00 UTC on 10 June 2026.

    Written at 18:00 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 2a91f37a5a25ef85e9d6f6a5fe21776cc4d616102369cec2cc3452f0f73b9a37

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  4. Results note: Reviewers from different model families catch more weak judgments

    Study jco-p5, the run recorded at 18:00 UTC on 10 June 2026.

    Written at 18:00 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint f84e97dc6f72dbb10d4b8272cff8d237781f3419298820f97821179b7cb05bd3

    The same fingerprint as the record written at 17:59 UTC on 10 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  5. Results note: How much work centres on judgment

    Study jco-p8, the run recorded at 18:00 UTC on 10 June 2026.

    Written at 18:00 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 81eb77b6d33cd3d74fcb98159e04edc83eb04de9b2d526a72dabc46efb5d37ea

    The same fingerprint as the record written at 17:59 UTC on 10 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  6. Results note: Oversight as varied as what it oversees

    Study jco-p12, the run recorded at 18:00 UTC on 10 June 2026.

    Written at 18:00 UTC on 10 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 4935605f6596ffa0523eadadb4234e6b39524dc5a99549fb99e9ef48c49c24f5

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  7. Results note: Reviewers from different model families catch more weak judgments

    Study jco-p5, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint f84e97dc6f72dbb10d4b8272cff8d237781f3419298820f97821179b7cb05bd3

    The same fingerprint as the record written at 17:59 UTC on 10 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  8. Results note: How much work centres on judgment

    Study jco-p8, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 81eb77b6d33cd3d74fcb98159e04edc83eb04de9b2d526a72dabc46efb5d37ea

    The same fingerprint as the record written at 17:59 UTC on 10 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  9. Results note: Oversight as varied as what it oversees

    Study jco-p12, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 4935605f6596ffa0523eadadb4234e6b39524dc5a99549fb99e9ef48c49c24f5

    The same fingerprint as the record written at 18:00 UTC on 10 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  10. Results note: Two validators, one verdict

    Study sco-s3, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint ba1c0318a058b5251bb0c1a7d962c357ac4ec58d265749294171674bfd9f9d14

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  11. Results note: Calibration of each kind of insight over 90 days

    Study scb-s6, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 71892a0098f084ef6b9be973948e2a78c590cd27119649fa77156a4e25e0ca9d

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  12. Results note: A performance record for each insight method

    Study scb-s8, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint de7dd5b06deeb2b2e11abd8fbe75fbe92f6ad2d868206d9dc21daf72c473b059

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  13. Results note: Patterns shared across similar readers, released only in groups of five or more

    Study scb-s10, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 43a2e5b8db23faa4a46aa2977625e6cc8aca647c57f7d7a2a2e53d807367987c

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  14. Results note: Retiring and adding insight methods each quarter

    Study sip-s12, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 293b26fa5be27a63c980cb3a588dc1ea372d33e9306305aaabd1e5c1b8298724

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  15. Results note: Stability of causal strengths learned from outcomes

    Study sco-s16, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint d6acaaa482e27215f51a99335d17818db39a4787a17229a6209c30820f3cb8b8

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  16. Results note: Every surfaced insight carries its seal and its record of change

    Study scb-s17, the run recorded at 09:18 UTC on 11 June 2026.

    Written at 09:18 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 0b9b2d602c7725d7d2eaf7a49a0a012ca466c70e9c06f1cad1dac0749ec765c4

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  17. Results note: Two validators, one verdict

    Study sco-s3, the run recorded at 09:19 UTC on 11 June 2026.

    Written at 09:19 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint ba1c0318a058b5251bb0c1a7d962c357ac4ec58d265749294171674bfd9f9d14

    The same fingerprint as the record written at 09:18 UTC on 11 June 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  18. Results note: First reference run: twelve hypotheses on AI computing

    Study rr-001, the run recorded at 09:20 UTC on 11 June 2026.

    Written at 09:20 UTC on 11 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 90 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 00003f53e46ba8b28582ce62d6f91a4401eb237f55b798970f05c7b5bcc44b81

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  19. Results note: Removing each element of the shared representation

    Study sco-s1, the run recorded at 17:50 UTC on 12 June 2026.

    Written at 17:50 UTC on 12 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 88 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 47ced47d999171e6ed59236b2b787d1aa97d4ce3ffd9d54317297329ef076c40

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  20. Results note: Generating analysis gets cheaper faster than evaluating it

    Study jco-p1, the run recorded at 17:51 UTC on 12 June 2026.

    Written at 17:51 UTC on 12 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 88 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint a5383954ed72649b74898618ac9adb4cf24df6b14c80842630b49b36a8342249

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  21. Results note: Governance keeps its value as models improve

    Study jco-p13, the run recorded at 17:51 UTC on 12 June 2026.

    Written at 17:51 UTC on 12 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 88 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint d5efa7c7b69c73605453bf43d17c6fd54a854cf3c923fd387d355c8e34a3669e

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  22. Results note: Does an insight read as written by a person or by a model?

    Study scb-s9, the run recorded at 17:52 UTC on 12 June 2026.

    Written at 17:52 UTC on 12 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 88 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 743e7cbd04ac0760b45bece24808201ba5f92959631cd051b9939ca3098c7e46

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  23. Results note: Whether attribution matters where accountability is central

    Study sco-s1b, the run recorded at 13:27 UTC on 13 June 2026.

    Written at 13:27 UTC on 13 June 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 87 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 670d2eeb5d841c7e226d0cc9d23e50cb19dcb78a69a4abf744c8bebee1615390

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  24. Results note: Do retrieved resolved questions reduce forecast error?

    Study icl-exemplar-brier-v1, the run recorded at 09:56 UTC on 18 July 2026.

    Written at 09:56 UTC on 18 July 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 53 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint bd332abe4519665a3a1edf1cdb706fd192e4bcbc5718e5bf977a7eb9572745bd

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  25. Results note: Do retrieved resolved questions reduce forecast error?

    Study icl-exemplar-brier-v1, the run recorded at 10:20 UTC on 18 July 2026.

    Written at 10:20 UTC on 18 July 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 53 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint bd332abe4519665a3a1edf1cdb706fd192e4bcbc5718e5bf977a7eb9572745bd

    The same fingerprint as the record written at 09:56 UTC on 18 July 2026: this run produced exactly the same numbers and verdict.

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  26. Results note: Do retrieved resolved questions reduce forecast error?

    Study icl-exemplar-brier-v1, the run recorded at 13:59 UTC on 18 July 2026.

    Written at 13:59 UTC on 18 July 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 52 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 1e6e1183300ebb1238ba1e26d934b7fc478073157799246f76fe0c52128ebe4c

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

Gate verdicts (2)

The outcome of comparing a candidate configuration with the configuration in use on the same questions.

  1. Gate verdict on candidate market-blend-on

    Written at 07:30 UTC on 9 August 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 31 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint b82182b5505db062c48499a48c5bb454ed379c0b2e54b2248697cc4ce23e3a78

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

  2. Gate verdict on candidate market-blend-on

    Written at 17:32 UTC on 14 August 2026 by the project’s own clock. In the public record by 23:59 UTC on 8 September 2026, GitHub’s push time for its commit. That was 25 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint a47998689aba5e4272e8318f751e7df743acb8439338f8aee88411eab2598d0f

    In the project’s tree of 8 September 2026, published in commit ed193c78b959.

Corrections to sealed tests (9)

A reading that arrived after a test’s lateness allowance and bears on an outcome already sealed. The sealed outcome is never changed. The correction is sealed beside it.

  1. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.9 days after the period it measures would have given a different outcome from the one sealed. The sealed outcome stands, and this correction is on the record beside it.

    Written at 06:04 UTC on 22 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 49 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 66e390524aa8b4ae4450444bcf55c658f843c39884ea4fb26c9459b53d50e20b

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  2. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.9 days after the period it measures gave the same outcome as the one sealed.

    Written at 06:04 UTC on 22 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 49 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 7feb527a285d308f002bc07c6c4aa1fe985f545161c1b1c33126a69b8c63b5f7

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  3. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.6 days after the period it measures would have given a different outcome from the one sealed. The sealed outcome stands, and this correction is on the record beside it.

    Written at 06:31 UTC on 23 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 48 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint d7f42c8e721a53a46ff07f3cf4a25528054838bd1a1b3b76f5f13188a3f6a7ee

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  4. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.6 days after the period it measures would have given a different outcome from the one sealed. The sealed outcome stands, and this correction is on the record beside it.

    Written at 18:38 UTC on 24 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 46 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 21162bff55cdea497d2fa66e9eb192e8298c4ee1aff08cc6ce836c496873e5e5

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  5. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.6 days after the period it measures would have given a different outcome from the one sealed. The sealed outcome stands, and this correction is on the record beside it.

    Written at 18:39 UTC on 24 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 46 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint edb90799ea33899117b1589f8fe8ca9f28e9fc78253074b538f2e980815b379d

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  6. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.9 days after the period it measures gave the same outcome as the one sealed.

    Written at 18:39 UTC on 24 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 46 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 74f180bbf7ea5da0aeeda4b1ca01ed2ce59a8de3a097498ea77174d8927d93b2

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  7. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.4 days after the period it measures would have given a different outcome from the one sealed. The sealed outcome stands, and this correction is on the record beside it.

    Written at 06:30 UTC on 26 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 45 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 566ec5a817d27d18cb1e82b9de34a575cd0a81e02238b61c63d9cadb61219b34

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  8. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 1.3 days after the period it measures gave the same outcome as the one sealed.

    Written at 06:30 UTC on 26 July 2026 by the project’s own clock. In the public record by 23:55 UTC on 8 September 2026, GitHub’s push time for its commit. That was 45 days after it was written, so the public record cannot show that it existed any earlier.

    Fingerprint 4c56aff895992240bc81a6a349a006dd4130241abf2d4d41dd5ce458196ebdd0

    In the reference feed’s tree of 8 September 2026, published in commit d923839ae957.

  9. Correction to a sealed test in a reference feed that is not public

    A reading that arrived 2.1 days after the period it measures gave the same outcome as the one sealed.

    Written at 02:17 UTC on 21 September 2026 by the project’s own clock. In the public record by 23:55 UTC on 21 September 2026, GitHub’s push time for its commit.

    Fingerprint 3f61c6b776d32bdffbf7d922b3aaba693d90a1e089b6b6f50fc0e8a179af7ef8

    In the reference feed’s tree of 21 September 2026, published in commit 62c4fbc628f0.

The daily trees

  • 8 September 2026, the project’s own daily tree: 17,502 fingerprints, 16 levels.
    Root 05fc5c76bb82c8b8a30504de1944b0d174099fbecd76ba0a9f97723851980d59
    Published in commit ed193c78b959, file daily-seals/2026-09-08.json. GitHub recorded the push at 23:59 UTC on 8 September 2026.
  • 8 September 2026, the daily tree of a reference feed that is not public: 926 fingerprints, 11 levels.
    Root 2fa4ae336b2ad2e25b3714fa6225626a67749e3e9db9cc9572918c4f68a81dcd
    Published in commit d923839ae957, file daily-seals/bc04338a/2026-09-08.json. GitHub recorded the push at 23:55 UTC on 8 September 2026.
  • 21 September 2026, the daily tree of a reference feed that is not public: 26 fingerprints, 6 levels.
    Root 19e36ca897ca5ba014869a8ab5b7cec2a0a3b6ab79e3a69349b19ce81d162cd4
    Published in commit 62c4fbc628f0, file daily-seals/bc04338a/2026-09-21.json. GitHub recorded the push at 23:55 UTC on 21 September 2026.

Withdrawn resolution criteria

A resolution criterion that cannot tell a claim’s coming true from ordinary noise is withdrawn, and the withdrawal is sealed in the record of the feed it belongs to. As of 26 September 2026, 192 criteria across 127 assessments carry that sealed outcome. The feeds are not public, so the criteria are not listed one by one here. How the criterion gate works, and a replay of one withdrawal.

Checking an entry

  1. Download the ledger with every inclusion proof: methods-ledger.json (115 KB, SHA-256 8b6a8be7a7527527c739a12b9270c1f982cd532fd9631bb508c4a96afa09f19d).
  2. Find the entry by its fingerprint. Its proof lists the fingerprints beside it on the path from its leaf to the root.
  3. Start from the fingerprint. For each fingerprint in the proof, put the pair in text order, join them, and take the SHA-256 of the joined text as hexadecimal. The last value is the root.
  4. Open the anchor file at the commit the ledger names, and compare its root with yours. GitHub’s activity log for the repository gives the time the commit was pushed.

The same rules, run in your browser on a published record, are on the sealed record page.

What the ledger does not show

  • Content. It lists fingerprints, never what was sealed.
  • Forecasts, insights, or anything about the people who use the project’s tools.
  • Records from feeds, apart from corrections to sealed tests in one reference feed that is not public.
  • Records of these kinds that were not yet in a published tree at export (0) or fell outside the ledger’s scope (0).
  • Anything sealed after 16:34 UTC on 27 September 2026. The ledger is exported again when records are added.