KAIROSVerify the frozen protocol
Don't take our word for it. The check is one command.
The protocol for the 30-day study was written, hashed with SHA-256 and frozen on 23 April 2026 — before the study launched at 09:00 CET the next morning, and before any data existed.
The manifest is published, and so is its hash. If the manifest had been edited after the fact — an input added, an endpoint moved, a hypothesis softened — the hash would not match.
SHA-256 of protocol_frozen_2026-04-23.json:
0972a2c650a562909e53832845ec226ab897f6094db14645c4a0d5ed000d709a
Check it yourself:
curl -s https://kairos-experiment.com/protocol/protocol_frozen_2026-04-23.json \ | sha256sum
On macOS use shasum -a 256. On Windows, certutil -hashfile protocol_frozen_2026-04-23.json SHA256.
The manifest covers the hypothesis, the two conditions, the endpoint, the judging rubric and the analysis plan — plus the five conductor files, which are the frozen inputs themselves and the day-zero crontab:
- input_30giorni.json — the ninety inputs, in order, for both conditions
- input_giampy.json — the human interaction inputs
- input_neutri.json — the neutral inputs
- input_sorpresa.json — the surprise inputs
- crontab.day0.txt — the schedule as it stood on day zero
Independent timestamps exist: the manifest and the protocol page were archived by the Internet Archive's Wayback Machine on 24 April 2026, the day of launch. A third party's clock, not ours.
It proves: that the manifest you are reading today is byte-for-byte the one whose hash was published on 23 April 2026 and archived by a third party the next day. The inputs could not have been chosen after seeing the results. The endpoint could not have been moved.
It does not prove that the study was well designed, that the analysis was right, or that the conclusion held. It did not: the study's central claim was retracted on 14 July 2026, by its author, after he ran the controls it had promised. A frozen protocol is not a guarantee of truth — it is a guarantee that the goalposts stayed where they were put. Ours stayed, and the ball still missed.
The five later pre-registrations — the ones covering the validity checks — are hashed locally and are not notarised by a third party. That is a weaker claim than this page, and we say so on Methods & data rather than let this page's Wayback timestamps launder it.