Programmes
Four lines of work.
Each has its own output, its own rhythm, and its own failure condition. No line moves to the next period without publishing.
A
Measurement
Task horizon, compositional generalisation and agent reliability on public models. Independent reproduction of the METR and ARC protocols.
- Output
- Raw data, method and code
- Rhythm
- Every six months
B
Translation and archive
Turkish editions of safety frameworks, foundational papers and policy documents, with version history.
- Output
- Archive with version diffs
- Rhythm
- Continuous
C
Forecast registry
Dated, probabilistic forecasts with written resolution criteria, scored every year with the Brier rule.
- Output
- Registry and score report
- Rhythm
- Annual
D
Document audit
Monitoring the labs' commitment documents; detecting version changes and weakenings.
- Output
- Change bulletin
- Rhythm
- Quarterly
Three-year roadmap
2026
- Archive opens: 766 studies, full-text sources
- Forecast registry opens
- Advisory panel formed, two dissenting members included
2027
- First independent run of the METR protocol
- Three safety frameworks in Turkish, with diffs
- First audit bulletin and first Brier report
2028
- ARC protocol with declared compute budgets
- Campaign for generation-time reporting
- First published advice to a public body
2029
- Three-year self-assessment
- Charter review; clause 10.2 applied if required
Methods, traps and the failure conditions of each line: charter, chapter III.