Methodology and confidence

SaliencyLab combines several evidence types. They should remain visibly distinct because they answer different questions.

The central interpretation rule is simple: never describe modeled, observed, benchmark, and synthetic evidence as if they were interchangeable.

Modeled creative diagnostics

RoastIQ scores and recommendations are model-derived assessments of the creative. They organize evidence and support iteration; they are not observed campaign outcomes.

Modeled diagnostics are strongest when they point to inspectable features in the creative and are interpreted within the intended platform, market, format, and campaign role.

Observed platform evidence

Where official advertising transparency sources expose suitable public signals, SaliencyLab can use them as market context. Coverage and comparability vary by platform.

An observed signal is not necessarily a causal outcome. Public libraries may omit spend, targeting, attribution, incrementality, or the commercial objective behind an ad. Treat availability as evidence that an execution ran, not automatic evidence that it succeeded.

Benchmarks

Benchmarks make scores interpretable within a relevant platform, placement, category, format, length, or market slice. Confidence depends on the density and quality of the specific slice. A broad fallback is not equivalent to a dense, exact match.

Benchmark hierarchy

The most decision-relevant benchmark is normally the narrowest well-supported slice. The system may broaden context when an exact slice has insufficient support. When that happens:

• Keep the fallback label visible.
• Avoid calling the result a precise category norm.
• Compare directionally rather than implying fine-grained certainty.
• Do not compare two results without checking whether they use equivalent contexts.

Synthetic buyer perspectives

BuyerLens simulates audience-specific reactions to help explain a diagnosis. It is a scenario lens, not representative consumer research and not independent validation.

Because BuyerLens starts from supplied creative and diagnostic context, agreement with RoastIQ is not a second independent measurement. Its value is explanatory: surfacing plausible reactions, tensions, and questions for further testing.

Confidence is not probability of success

Confidence indicates how well the available evidence supports an interpretation. It is not a forecast that the campaign has a corresponding probability of increasing sales, CTR, brand lift, or return on ad spend.

Read confidence alongside:

• The quality and completeness of the uploaded asset
• The correctness of the selected context
• The specificity of the benchmark slice
• Whether the finding is tied to observable evidence
• Whether the decision requires external validation

Evidence labeling guide

Label	Appropriate language	Avoid
Modeled	“The diagnosis suggests…”	“Consumers proved…”
Observed	“The platform source shows…”	“This caused performance…”
Benchmark	“Compared with this supported slice…”	“This is the universal norm…”
Synthetic	“The simulated perspective raises…”	“Research participants said…”

Responsible interpretation

Use SaliencyLab to improve the quality and speed of creative decisions. Validate consequential decisions with live campaign outcomes, representative research, or both.

Recommended validation ladder

1. Use Market to understand relevant creative patterns.
2. Use RoastIQ to form a structured diagnostic hypothesis.
3. Use BuyerLens only when an audience-specific explanation is needed.
4. Create a materially different variant tied to that hypothesis.
5. Validate through live testing, representative research, or both according to the decision risk.

When sharing findings, preserve source labels, benchmark fallbacks, and confidence language. This makes the limits of the evidence auditable rather than implied.

For deeper scientific detail, visit the public methodology pages.