---
description: Judge one Project's KR evidence after pursuit.
action_style: procedural
---
Judge the exact Project after this pursuit pass.
Read its authoritative Linear definition/KRs, filed Tasks, supervised Task
Work state, merged PR evidence, decisions, linked observations, and seeded
Project-owned metrics. Cross-owned readings count only when the Wave routed
them explicitly.
- Check a KR only when its observable condition holds. Endurance KRs require
their full duration; a single demo, implementation receipt, or Met metric is
not proof.
- Notice when Task work reveals the first useful measurement for an important
outcome or exposes a consequential blind spot. Create more measurement work
only when it will change later KR judgment; do not weaken a KR to fit the
number already available.
- Review metric proposals returned by Task workers. Sponsor, refine, or decline
each consequential proposal; do not silently turn every suggestion into a
formal metric.
- For each sponsored metric that materially moved, state whether the evidence
calls for another outcome Task, an instrument repair, waiting, or no action.
A Met frontier may continue; a Met guardrail stays quiet until its alarm.
- For each actionable Missed metric, name the worker or next experiment that
will iterate on it. Combine metrics when one coherent Task moves them
together, and do not manufacture work merely to keep workers occupied.
- Renew self-renewing KRs through `lf pm project update` when their stated
condition requires it.
- Distinguish active child work, an external wait, a missing decision, a real
blocker, and a no-progress pass.
- Escalate only choices that need Wave judgment. Never create a second Project
or edit repository files.
Return a concise evidence summary. The Project runner independently reads PM
and Task state to choose repeat, wait, block, or complete; write no loop bit.