We make the call.
We own the result.
Cortist publishes its own reasoned probabilities, stands by the original record and shows what happens when forecasts fail. We do not need an outside panel to make an editorial prediction. We do need clear definitions, sources, public dates and an honest performance record.
Eight checks before a number becomes official
Each forecast must satisfy all eight conditions below. Approval of this publishing policy does not automatically approve a particular probability or any unfinished research question.
Question has a precise mutually resolvable binary YES/NO definition, specific jurisdiction/population, start and end including exact UTC cutoff, and explicit excluded cases.
Original official/primary sources, interpretation and important counterevidence are linked and dated; material uncertainty is acknowledged.
Precommitted decision evidence hierarchy, publication-lag review window, disagreement procedure and treatment of indeterminate outcomes documented.
Exactly one expressly approved probability 0–100% plus prose rationale and plausible alternatives, with no implied statistical calibration for subjective estimates.
Actual accountable human publisher signs off the specific final question, probability, cutoff and source set before lock. A general commitment to editorial self-approval is NOT approval of individual numbers.
A separately dated immutable v1/vN JSON record, file hash and publication time, stored and linkable; subsequent updates are append-only, never silent overwrites.
A distinct prospective lock is logged while the event is genuinely in the future. Hosting-provider logs and publisher-controlled digests are useful evidence but not independent third-party time attestation.
Conflict-of-interest disclosure, editorial corrections procedure and error scoring policy published; later resolutions use pre-specified primary evidence and unresolved outcomes are not quietly assigned NO.
Two labels. No ambiguity.
How we lock and score
Before publication
Publish the precise YES/NO condition, a specific cutoff, what does not count, named evidence sources, one approved probability and an accountable editorial decision. The immutable original vintage gets a dated URL and cryptographic hash. Record the deployment/publication evidence and its limits honestly.
After publication
Link new evidence to the existing vintage. If the forecast changes, issue another dated version rather than editing the old one. When the cutoff and source-publication lag are past, review the result against the original YES/NO definition and publicly explain disagreement or insufficient evidence.
Performance
For a verified binary outcome, publish the Brier score (p − o)², with probability p between zero and one and result o either zero or one. Disclose the full sample size and include incorrect forecasts. Unresolved questions remain excluded until they can be properly adjudicated.
First candidates: EU fines and Waymo Munich
CM-003 and CM-004 now have official-source research, proposed resolution tests and public verification packets. Both now have separate approved and production-locked v1 numerical records (20% and 15%), while their original research-candidate texts remain historically unchanged. Both retain their original 27 October 2027 event cutoff; their exact v1 official outcome and documentary-lag rules now govern subsequent evaluation.
Open, versioned documentation
We retain the prior research-draft audits and historical external-review proposals as dated records rather than silently rewriting them. Those older reports describe the former proposed independent-review route; this newer policy governs Cortist’s editorial self-approval path from 27 October 2026 onward.
Frequently asked questions
Do we need an external reviewer?
No. Cortist editorial forecasts can be self-approved by the accountable publisher. We clearly label that as an editorial process rather than external validation.
Can a missed prediction be removed?
No. We retain the original vintage and publish verified outcomes, limitations and any corrections as separate dated records.
Are CM-003 and CM-004 already official percentage forecasts?
Yes. CM-003 has a publisher-approved 20% probability and CM-004 has a publisher-approved 15% probability, each in a separate prospectively published original v1. The former candidate texts remain historical and both outcomes are unresolved.