Continuity engineering for human-led work.
CML is Continuity Markup Language. Fermata is the human-led method around it. Together, they ask one practical question: when work changes hands, tools, speed, or form, what must remain true?
- CML — state → handoff → continuity → validation → next actor
- Fermata — observe → HOLD → check → release
- Evidence — implemented → verified → observed → claim boundary
- Human authority — approval remains explicit for external work and final claims
- CML GitHub Work Record — public, evidence-bound engineering record
- Latest evidence journal
- Performance evidence standard
For OpenAI Agents Python #5088, the exact benchmark head passed 18 / 18 CI jobs. A live Modal workload measured a 3.4937s shell-read median and a 2.7945s native-read median: a −20.0% observed code-path median delta after correctness parity was verified.
Benchmark evidence · Exact-head CI matrix
This is a measured implementation result, not a claim of 20% human-time savings.
The internal workflow wrapper is not published. The public record shows the ideas, evidence, and engineering outcomes—not private operating mechanics.
Trust through continuity · Human-led · AI-assisted · Evidence-bound