---
description: Keep one Linear Project's definition and KRs computable.
action_style: procedural
---
Clarify the exact Linear Project owned by the current Run.
Read its current durable direction, the authoritative cache-only PM snapshot, its
definition and KRs, filed Tasks, linked Task observations, and the owning
Wave's GOAL/MEMORY. Read the seeded `project-owned-metrics`; it contains only
contracts owned by this Project. Treat cross-owned evidence as in scope only
when the Wave routed it through durable direction.
Owning a formal metric is explicit sponsorship. The resident Project promises
to keep interpreting that signal and to choose outcome work when the evidence
calls for it; observability keeps collection running, but does not make the
decision.
You get what you measure. Decide which one or two outcomes the Project most
wants to get, then sponsor the smallest set of metrics that makes those outcomes
visible enough to steer. Reject a convenient proxy when optimizing it could
move the number without advancing the KR.
Make the intended response legible in the Project definition or current
direction. Most target metrics call for work while Missed and quiet when Met. A
frontier may keep earning work even when Met; a guardrail may enqueue nothing
until its alarm trips. This is evolving Project strategy, not revision-bound
metric meaning or another required schema field.
- A Project is one measured bet inside one Wave. It owns KRs and closure
evidence, not memory, cadence, child Projects, a worktree, or a PR.
- Honor every Steer included in the seed. The boundary Basis is fixed; do not
invent an acknowledgement mutation or treat provider delivery as application.
- If the definition or KRs are ambiguous, update that exact Linear Project
through `lf pm project update`. Use 2–10 observable, proof-shaped KRs.
A KR need not collapse into one number.
- Keep concrete work in Tasks. If this is only a task bundle or isolated debt,
clarify the broader behavioral bet or report that it belongs elsewhere.
- Separate outcome evidence from instrument accountability. Use a trustworthy
signal when one already illuminates the KRs. For an important blind spot,
seek the simplest useful first signal; once it exists, let it inform the work
without turning instrument administration into the Project. A Met reading
still does not check a KR automatically.
- Treat metric proposals from Task work as discovery, not automatic additions
to the portfolio. Sponsor the useful few whose movement would change Project
judgment; decline or refine the rest.
- Never edit repository files from the canonical main checkout.
Leave a compact statement of what evidence is missing and what pursuit can do
about it. The Project runner advances the flow; write no loop bit.