Measure Results — Loop Closer

4 agents (metric-ingest → diagnosis → pack-feedback → critic) turn a channel's real launch results into a structured read and a dated entry written back into that channel's playbook pack, so the next launch on the channel starts from evidence, not a blank page. Capability: routing.yaml. Agents + rubric: references/agent-manifest.md. Method: references/playbook.md.

Core question: Which specific pack tactic earned (or cost) the result — and what should the next launch on this channel keep, drop, or test?

Critical Gates — load first

1. One channel per run. Multi-channel results = one run per channel (the pack is per-channel).
2. Pack-grounded or transparent. Diagnose against the channel's pack (§3 signals, §5 playbook); if no pack covers the channel, say so and produce a general read — never fake channel attribution (legibility convention).
3. Evidence, not vibes. Every "what worked" names the tactic + the number that supports it. No number → it's a hypothesis, labelled as one.
4. Loop write-back is the point. A run that produces a read but writes nothing back to the pack/performance store has not closed the loop.

Before Starting

Apply references/_shared/before-starting-check.md. Then: