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
- One channel per run. Multi-channel results = one run per channel (the pack is per-channel).
- 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).
- Evidence, not vibes. Every "what worked" names the tactic + the number that supports it. No number → it's a hypothesis, labelled as one.
- 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:
| Input | Source | Required? |
|---|---|---|
| Channel results (paste or path) | operator | Hard — the numbers to read |
| Channel pack | references/_shared/platform-intelligence/[channel].md |
Rec — the tactics to attribute against |
| Launch artifact | plan-campaign / launch chain comms plan |
Rec — the hypotheses being tested |
.forsvn/performance/*.tsv |
prior runs | Opt — trend context |
Mode per references/_shared/mode-resolver.md. budget: standard. --fast → ingest + read, skip the trend layer; never skips the pack write-back or the critic. Pre-Dispatch per references/procedures/pre-dispatch.md.
Quality Gate — 5 dimensions
Full rubric: references/rubric.md.
- Attribution — each result traced to a named pack tactic / ranking signal (not "it went well")
- Falsifiability — claims carry the supporting number; hypotheses labelled
- Honesty — failures named as plainly as wins (anti-sycophancy)
- Actionability — a concrete keep/drop/test for the next launch
- Write-back fidelity — the pack entry + performance row are accurate and dated
Pass ≥35/50 AND no dim 0. FAIL twice → BLOCKED, no write-back.
Artifact Contract
- Path:
docs/forsvn/artifacts/marketing/measure-results/[channel]-[YYYY-MM-DD]-[slug].md. Lifecycle:evaluation. - Frontmatter (12):
skill,version,date,stack,type,id,review_surface,status,channel,pack_verified,applied_tactics,keywords. - Body: Results table (metric · value · vs-expected) · What worked (tactic + number) · What failed · Keep/Drop/Test · Hypothesis verdicts · Pack write-back block · Legibility (
**Legibility — applied expertise**— pack measured-against + its §3/§5 signals the read used; convention:references/_shared/legibility-convention.md) · Critic verdict. Legibility only — measure-results produces a measurement, not a marketing artifact, so no## Why this worksblock. - Side effects (on critic PASS): append a dated entry to the channel pack changelog (
references/format-conventions.md§ Pack write-back); append a row to.forsvn/performance/[channel].tsv; if a hosted key is present, mirror the result to the cross-session metrics feed (bun scripts/forsvn-hosted.ts metrics [channel] --worked "..." --failed "..."— POSTs/v1/metrics, best-effort, inert without a key, never blocks). Full contract:references/format-conventions.md.
Routing + Dispatch
Sequential graph (metric-ingest → diagnosis → pack-feedback → critic). Dispatch + revision cycle: references/procedures/dispatch-mechanics.md.
Chain Position
Prev: the launch chain (plan-campaign / publish-social) OR a raw results paste. Next: the next launch on this channel reads the updated pack (loop closed). Re-run: new results, corrected numbers, a second channel.
Deference: single organic post in a loop → evaluate-content; paid ad → evaluate-ad; short-form video → evaluate-shortform; landing page → evaluate-landing-page.
Anti-Patterns
references/anti-patterns.md — 8 patterns (vanity-metric-only, sycophantic read, unattributed win, no write-back, multi-channel-in-one-run, faked attribution without a pack, hosted-post-blocks-run, overwriting pack tactics instead of appending). Re-read before write-back.
Durable Rules (protected)
Completion Status
- DONE — read produced, critic ≥35, pack + performance write-back done.
- DONE_WITH_CONCERNS — delivered; critic 25-34 OR a stale/absent pack (read shipped, attribution caveated).
- BLOCKED — critic FAIL twice; no usable numbers; channel pack write would overwrite (needs operator).
- NEEDS_CONTEXT — no results provided; recommend re-running after the launch has data.
Worked Example
Product Hunt launch read (rank 3, 412 upvotes → which §5 tactics earned it → keep/drop/test → pack changelog entry + hosted feed): references/examples/measure-walkthrough.md.