A forecasting site that does not publish its own archive is asking to be taken on trust. Here is everything the pipeline has produced: every estimate, every dated revision, every article kept on file, every pass of the resolver.
| Questions | 75 |
| Revisions | 170 |
| Articles | 340 |
| Generated on | 2026-10-02 11:00 UTC |
Every day, a SHA-256 fingerprint is computed over all the files above, then chained to the previous day’s. Quietly editing an old value would break every fingerprint after it. This is link no. 15 of the chain.
The fingerprints of each file are in chaine.json. To check that a downloaded file is really the one that was published:
sha256sum revisions.csv # compare to the fingerprints["revisions.csv"] field of the last link in chaine.json
And for precedence — the only thing that really matters — today’s fingerprint is deposited with two third parties we do not control: OpenTimestamps, which anchors it in the Bitcoin chain, and archive.org, which keeps dated copies of the pages.
ots verify ancres/chaine-YYYY-MM-DD.json.ots
A proof is deposited and kept every day in ancres/. What matters is that they accumulate: each one freezes the state of the site at its date, and none of them can be redone after the fact. Currently 17 proof(s), from 2026-09-18 to 2026-10-01.
This proves that a given figure existed on a given date, before the outcome was known. That is what makes a calibration score impossible to fake after the fact.
This does not prove the reasoning was sound, or that the sources were well chosen. Those judgements stay open to debate — and should.
This data is published to be reused, recomputed and criticised. Cite the source and the export date. Article titles listed belong to their publishers and appear only as references.