Campaign analysis

The question this tier answers: is this burning money past the point where you should stop, or is it just early? Nothing here judges which creative wins, how much to scale, or why performance changed — that's this same skill's paid tier. This tier's whole job is loss prevention: catch the campaign that's clearly past its stop-loss threshold, and be honest when the data just isn't there yet to say more.

Before you start

1. Tracking health first. If pixel or conversion tracking is broken, every threshold below is a decision made on garbage data. Route to tracking-health and don't proceed until it comes back clean.
2. Read .agents/profile.md (../contracts/profile-v1.md) for Margin Basis and Target CPA. Either missing → ask the user directly; when you apply the break-even formula in references/stop-loss-thresholds.md, name exactly which inputs you used, not just the output number.
3. Read .agents/data/normalized/ and .agents/data/readouts/ before asking for anything — see Data layer, below. The same export doesn't get asked for twice.

Absolute stop-loss threshold

Compute the break-even CPA/ROAS against Margin Basis — formula and the gross-margin-basis vs net-profit-basis distinction in references/stop-loss-thresholds.md. When mature spend (past the attribution window and any reporting lag — a "zero conversions" reading inside that window isn't a real zero) has run far past what Target CPA implies the conversions should be, recommend stopping.

Fixed wording: "stop-loss on absolute threshold." Never "this angle doesn't work" or "this creative failed" — those are comparative verdicts this tier has no evidence to support (see Boundary, below). The absolute threshold compares spend against an external number — target CPA, tolerable loss — never against another ad or creative, so it doesn't wait on the sample size a comparative verdict would need.

"Wait ≠ not working"

When conversion volume is too thin to compare this ad or creative against anything else, exactly two conclusions are legal: wait (let it keep running to accumulate signal) or restructure to get volume (consolidate audiences, placements, or budget so it accumulates faster). "This isn't working" is a comparative verdict — this tier doesn't have the evidence for it.