Landing Page Eval — Orchestrator

Evaluation skill. Converts launched landing-page evidence into a cycle snapshot, a ledger row, and narrowly-scoped next action inside an existing eval loop.

Core Question: "Did this landing-page cycle create measurable signal strong enough to keep, discard, watch, or block, and what should the next strategy/execution skill know?"

Critical Gates

1. Existing eval loop required. If skills-resources/loops/[slug]/program.md and context.md do not exist, return NEEDS_CONTEXT and recommend /eval-loop. This skill does not create loops.
2. Measurement evidence required. Do not run as a generic heuristic audit. Require at least one metric source, measurement window, and current value for the loop's primary metric.
3. One primary metric decides the ledger row. Secondary metrics and qualitative evidence explain diagnosis; they do not override the loop's primary metric unless program.md defines an explicit guardrail failure.
4. No fabricated analytics. Unknown values stay unknown. Manual notes are allowed only when labeled as operator-supplied and tied to a date/window/source.
5. Attribution confidence must be explicit. Every verdict includes sample size or traffic volume when available, baseline comparability, confounders, and confidence: high | medium | low | blocked.
6. Evaluation does not redesign. Recommend next changes, but route actual page brief/revision work to lp-brief and execution artifacts to the appropriate content/design/build workflow.

Responsibility Split