Inner loop: one experiment¶
Hypothesis → run → evaluate → iterate. Keep model steps separate from the experiment’s wall-clock runtime.
Working notes¶
- Baseline gates and negative controls recur across independent projects.
- A surprising result triggers diagnosis rather than celebration.
- Artifact completion matters more than a process staying alive.
- Small pilots must be long enough for the suspected failure to occur.
- Diagnosis and recovery were stronger than unsupervised tuning.
Source: course evidence synthesis, §4. These are selected observations, not general performance guarantees.
Maturity, in Oct 2026
Mature where evaluation is cheap: the verification and control loop is well evidenced. Autonomous tuning is mixed, so maturity does not mean better science without supervision.
Before the course¶
- [TODO: select a compact worked example and its verification artifact.]
- [TODO: rehearse the live walkthrough and prepare a fallback.]