Walkthrough · 4 min read · real ad, real score

One ad, scored on the record, with the trace open.

This is the exact output a buyer sees when they score a creative through RoastIQ. Same five KPIs, same weights, same verdict ladder, same Zod-validated decision trace. Nothing has been smoothed for the demo.

Ad Coca-Cola, Holidays are coming, 2024 Surface Meta Reels Region US Scored 2026-05-18

Score your own ad → Jump to the trace

// 01, the ad

The opening, on the record.

Eight-second Meta-Feed cut. Holiday sentiment narrative, brand cue late. The diagnostic moment is the gap between the emotional payoff and the pack arriving on screen.

Coca-Cola · Meta Reels · US

Holidays are coming.

Pulled from Meta Ad Library, scored against benchmark_pool_version v.2026-05-05 (n=2,047). benchmark_only = true, customer not contacted.

8.4s 1080×1920 Brand in first 3s · NO

// 02, the verdict

Composite 62 / 100, SHARPEN.

The composite is a weighted sum of the five frozen KPIs. 62 lands in the 55, 69 Sharpen band. No KPI drops below 45, so the Rebuild double-rule doesn't trip. The diagnosis is unambiguous.

// Verdict SHARPEN Composite 55, 69 AND no KPI below 45. Action: fix the bottleneck KPI, re-score, re-decide. Single weak signal, not a structural problem.

// 03, five KPIs, frozen weights

The breakdown, per KPI.

The names are frozen. The weights are frozen. Strong ≥ 70, Moderate 45, 69, Weak below 45. The Brand Impact row is the bottleneck, the trace points to a 6.8s logo entry on an 8-second cut.

Beat the Skip

25%

Strong

Get Noticed

20%

Moderate

Brand Impact

20%

Moderate

Sell Proposition

20%

Moderate

Build Brand

15%

Moderate

Composite math. (70 × 0.25) + (65 × 0.20) + (45 × 0.20) + (68 × 0.20) + (60 × 0.15) = 17.5 + 13.0 + 9.0 + 13.6 + 9.0 = 62.1 → rounds to 62. The 45 on Brand Impact bleeds 9 points out of a possible 20. Fix that one row and the verdict moves to Scale.

// 04, evidence

Frames the model actually looked at.

Every KPI score points at one or more timestamps the multimodal pipeline pulled as evidence. No model is asking you to trust it on vibes.

0.4s · Beat the Skip Strong opening attention signal, hand-over-glass.

2.1s · Get Noticed Composition holds. No brand signal yet.

6.8s · Brand Impact First distinctive pack appearance, late.

8.4s · Build Brand Flat end card. Memorable pattern incomplete.

// 05, recommended edits

What an editor can fix this week.

Specific enough to act on. The verdict moves from Sharpen to Scale if all three land.

Move the first distinctive pack into 0, 1.5s.

The 1.5-second rule. The pack does not have to be the hero, but it has to be on screen, held by talent. Brand Impact moves Moderate → Strong without changing the script.

+9 Brand Impact

Extend the hero shot of the bottle by ~0.4s.

Currently the pack reads but doesn't dwell. A short extension gives memory encoding room without breaking pace.

+4 Build Brand

Re-cut the end card with a second pack appearance.

Replace the flat logo card with a held bottle plus the seasonal claim. Two pack touches > one logo card every time on Brand Impact.

+3 Build Brand

// 06, where 62 lands in the pool

Quintile context, 2,047 ads.

The composite percentile against the v.2026-05-05 benchmark pool. Quintile 4 of 5 on the public engagement signal we validate against. Top quintile outperforms bottom by 6.5×.

Q10, 35×1.0 engagement

Q235, 50×1.7

Q350, 60×2.6

Q4 ← this ad60, 75×4.1

Q575+×6.5

// 07, BuyerLens

Same ad, thirty-six buyers.

A RoastIQ verdict tells you the creative is Sharpen. BuyerLens tells you why a segment would reject it, in the segment's own register. Thirty-six LLM-modelled personas walked through the cut. The verbatims below are what they said, not what we'd want them to.

36personas

6segments

4m 22spanel runtime

68%would skip

Urban mum, 32, UK Would skip · 7/8

"It's pretty but I don't know what they're selling me. Half a bottle would have done it. I scrolled before the music landed."

Objection: late brand · low intent

Early-career planner, 26, US Would skip · 6/8

"The hand-and-glass shot is the moment. I'd cut everything before second three. Christmas Coca-Cola is already a story I know, you don't need the wind-up."

Objection: pacing · slow opening

Senior buyer, 48, FR Would watch · 5/8

"I like that the brand is not screaming at me. But the bottle is so late, I don't remember whether it was Coca-Cola or Pepsi. Hold the pack longer at the end."

Objection: brand recall fade

Gen-Z student, 21, DE Would skip · 8/8

"This is the same Coke ad as ten years ago. I don't have a reason to not skip. Show me something I haven't seen, or pay the influencer."

Objection: pattern fatigue · low novelty

Family driver, 41, ES Would watch · 6/8

"Reminds me of being a kid. I'd watch it once at Christmas. But I see this on Reels at 11pm in May, it does not land. Wrong season, wrong scroll."

Objection: surface-mismatch · seasonal

Performance marketer, 35, AE Would skip · 7/8

"As a buyer I'd look at the hook rate at 1.5 and stop here. Beautiful, expensive, dead on Reels. There is no thumb-stop trigger in frame one."

Objection: no scroll-stop signal

What BuyerLens is, and isn't. A persona-based buyer-rejection diagnosis. Useful for verbatims that feed straight into a re-brief. Not a probabilistic sample of a real population, not a replacement for ethnographic depth, not a predictor of in-market sales. Thirty-six personas in under five minutes. The honest ceiling is on the methodology page.

// 08, the re-brief

The brief, written for you.

The RoastIQ verdict, the BuyerLens verbatims and the recommendation list collapse into one creative brief. Five fields. Hand it to the editor. Re-score the recut against the same composite.

// Brief brief_rq_2026_05_18_a4f1

Objective

Move the Sharpen verdict to Scale by lifting Brand Impact from 45 to ≥ 60 without changing the script or the spend.

Audience

Holiday-moment Meta-Reels buyer, primarily 28-44, mixed urban / family. BuyerLens flagged two latent segments: pattern-fatigued Gen-Z (deprioritise) and family-driver mid-evening scroll (the verbatims to write for).

The change, on-screen

Hero pack arrives at 1.2s, held by talent (not as a sting).
Held bottle extended ~0.4s mid-cut for memory encoding.
End card replaced with a second pack appearance + seasonal claim; logo card retired.

What to keep

The hand-and-glass opening shot (Beat the Skip 70 is the ad's strongest signal). Don't recut the audio bed.

Success criteria

Re-scored composite ≥ 70 AND no KPI < 55, verdict moves to Scale. Brand Impact lifts from Moderate to Strong. Otherwise iterate again. Re-score is free; the budget is editor time, not panel time.

What this report does not claim. RoastIQ scores predict engagement and click intent against public outcome data, TikTok engagement, TikTok CTR percentile, YouTube view counts. Held-out OOS Spearman ρ as of 2026-05-05: +0.30 to +0.32. BuyerLens is a buyer-rejection diagnosis, not a probabilistic sample. We do not predict sales lift, ROAS, attributed conversion or brand recall, those are downstream business outcomes a pre-spend report cannot read.