Campaign Eval — Orchestrator

Evaluation skill. Converts launched multi-channel campaign evidence into a cycle snapshot + a ledger row + a narrowly-scoped next action inside an existing eval loop. One cycle = the whole campaign across every channel. Aggregate-only — single ads/posts/pages/videos route to their own eval siblings.

Core Question: "Did this campaign cycle, across all channels, create measurable signal strong enough to keep / discard / watch / block — which channels drove it, are the unit economics honest, and what should the next strategy skill know?"

Why, methodology, history: references/playbook.md [PLAYBOOK]. Capability metadata (route triggers, prerequisites, load map): routing.yaml.

Critical Gates

1. Existing eval loop required. program.md + context.md absent → NEEDS_CONTEXT, recommend /run-pipeline. This skill does not create loops.
2. Aggregate-only, whole-campaign scope. Scores campaign-level outcomes from channel-rollup metrics across every channel in one cycle. Per-asset eval artifacts are context only, never re-scored, never drive the verdict. Single-asset OR one-channel-per-cycle splitting → NEEDS_CONTEXT, route to the sibling (evaluate-ad / evaluate-content / evaluate-landing-page / evaluate-shortform).
3. Measurement evidence required + primary metric decides the row. Need ≥1 metric source, window, and current value for the loop's primary metric (operator-pick via program.md: new customers · leads · revenue · blended CAC · ROI). Secondary metrics explain diagnosis; they don't override unless program.md defines an explicit guardrail failure.
4. No fabricated analytics. Unknown values stay unknown. Manual notes only when labeled operator-supplied + tied to date/window/source.
5. Attribution confidence must be explicit. Every verdict includes sample size, baseline comparability, attribution-model honesty (last-click vs multi-touch, assisted conversions, window), confounders (seasonality, concurrent campaign, channel cannibalization, price/promo, organic drift), and confidence: high | medium | low | blocked.
6. Evaluation does not generate strategy. Recommend changes; route re-planning to plan-campaign, ad creative to write-ad, organic copy to write-social, asset-level diagnosis to the asset eval skills.