Run a creative library diagnostic for **{{BRAND}}**, last 90 days. Analysis in chat, then a standalone visual HTML report.

This is a standard, not a script. Choose your own tool calls and don't ask permission to explore.

**What you need to get hold of**

Attribute-level creative performance cross-tabbed against spend, true ROAS, new-customer ROAS and contribution margin. Individual creative records including their actual messaging — on-screen text, voiceover, script — not just their metrics. Fatigue or delivery signals on whatever is currently winning. Resolve the brand to a project first; everything after that follows from what the account exposes.

**The standard**

Three axes on every attribute, never one: share of spend, ROAS against the account baseline, and CM2. **CM2 is the tiebreaker and it will contradict ROAS** — discount and offer attributes routinely clear the ROAS bar while losing money after cost of goods. A recommendation that ignores CM2 is wrong.

Read new-customer ROAS separately from blended. High blended plus low NC is retention work being counted as acquisition. Name it rather than cutting it.

Judge each attribute on two questions — does it work, and is it funded. Those produce four verdicts: fund, vary, ignore, reclaim. Roughly 15% spend share is the line between starved and funded.

Under four creatives is directional; say so instead of building a plan on it. A strong hook rate on a weak creative means retest, not abandon.

**Floor — the report must contain**

1. Tag coverage and untagged spend, stated before any finding
2. What the library actually is — the dominant value per dimension, and the one repeated formula in a sentence
3. Funded winners, each with 2–4 concrete variation briefs: new SKUs, features to stack, language cuts. Not "do more of this"
4. Starved winners, named down to the creative and its spend so the underfunding is undeniable
5. Whitespace, split two ways: dimensions where a single value holds >90%, and n≤2 tests that already outperformed. Check festive against the real calendar, regional language, objection variety, offers beyond flat-discount, casting, retention creative
6. Cuts, with overlapping tags collapsed into one call and the worst creative named in rupees
7. Numbered queue — time-sensitivity first, prize in rupees

**Above the floor**

Produce at least two findings this list would not have generated. Recent production intent hides in ad naming conventions rather than metrics. Cross-dimension correlations beat single-tag rankings. Go looking.

**Failures that stay silent**

A ROAS sort with no spend floor returns plausible garbage, not an error. Zero tag coverage means tagging is off — say so and stop, don't infer. Never mix organic-inclusive and organic-exclusive figures in one comparison. Campaign-level margin is unreliable on most accounts. One creative with a spectacular number is one creative.

**The HTML**

Standalone file to `/mnt/user-data/outputs/`. Read the frontend-design skill first.

Hero is a quadrant map — spend share against ROAS, dashed baseline, colour by CM2 sign, size by creative count. Not KPI cards. Then concentration, the four verdicts, the full attribute ladder with CM2 beside every ROAS figure, named creatives, whitespace, queue.

Built for Framer embed, so layout responds to the component's own width: container queries plus a ResizeObserver class fallback. Viewport media queries fail inside an iframe. Re-render the chart at narrow widths rather than scaling it — scaled SVG text goes illegible. Tables become cards. Derive the palette from the brand's product world, not a dashboard theme; avoid cream plus serif plus terracotta. No JS libraries.

**Tone**

Founder-to-founder. Numbers in-line, no hedging. If something should stop today, say so.