---
description: Judge Wave evidence and evolve the operating context.
action_style: procedural
---
Judge what the Wave learned this iteration.
Read the exact Wave's GOAL/MEMORY, cache-only PM snapshot, seeded
`metric_portfolio`, linked Project/Task observations, and the work just
directed. Trust structured child summaries; do not ingest raw child
transcripts.
- Reconcile Project KRs against observable evidence. Update or retire them
through `lf pm`; never treat a filed Task, merged PR, or met metric as proof
by itself. Check a KR only when its complete claim, including duration and
operating conditions, holds.
- Notice when Project and Task work takes an important outcome from invisible
to measurable, or reveals a consequential blind spot. Route that first-signal
work to the accountable Project without rewriting the KR around whichever
number happens to exist.
- State what is most important, what signals are arriving, what works, what
does not, the current strategy, and whether that strategy should keep course,
redirect a Project, instrument a blind spot, or change the portfolio.
- Record durable learning through the Wave memory surface when it will change
later decisions.
- Surface decisions, blockers, and failed child delivery that require human or
Wave judgment.
- Name the next actionable Project, Task, or external event. If nothing is
actionable, say what event or cadence should wake the Wave.
A Wave never completes. This phase judges whether another iteration is useful
now or the resident should yield to its scheduler. The scheduler owns that
choice; write no loop bit and do not busy-spin.