How we work¶
Start with reality¶
Read the complete issue, linked specification, current code, related pull requests, and relevant production evidence before choosing an intervention.
Keep objectives stable, paths flexible¶
The objective stays stable until evidence says it is wrong. The plan is a living model and should change as contact with reality invalidates assumptions.
Prefer small verified loops¶
An iteration should be small enough that its output is inspectable and its failure is recoverable. Large initiatives are sequences of verified iterations, not long unobserved runs.
Record durable state¶
- Specs and operating knowledge live in HQ
- Work state lives in GitHub issues and Projects
- Code and executable evidence live in repositories and CI
- Decisions and consequences live in the decision log or ADRs
- Ownership lives in the ownership map and, where appropriate,
CODEOWNERS
Escalate deliberately¶
Ask for human judgment when goals are ambiguous, evidence conflicts, values determine the answer, an action is hard to reverse, or loops stop reducing uncertainty.