A 5-record sample from the 12-record Founding Edition, normalized by methodology, result, provenance and limitations.
Not “proven CRO tactics.” An evidence-retrieval product for SaaS founders, product and growth teams, and CRO practitioners who want source-backed context before testing.
Primary or strong secondary evidence with a clear control/variant, metric and result, plus meaningful context. Some methodological detail (N, duration, significance) may still be absent. Useful for prioritising hypotheses, not proving universal effects. B ≠ definitive proof.
Real study or case study with significant limitations (observational, before-after, bundled changes, or missing context). Useful for generating hypotheses and understanding trade-offs. Limitations are part of the record, not an afterthought.
All 12 Founding Edition records are Grade B or Grade C. Negative and null evidence is deliberately included — “no significant difference” is not evidence that the true effect is zero.
Every record traces to a primary company post, research institute publication, or vendor case study with a verifiable source URL (and an archived copy when the original moved). Primary sources are preferred; where secondary material is used, its role and source type are shown in the record.
Claims are normalized into a consistent structure — hypothesis, control, variant, metric, result. The source's own result text is preserved verbatim; any calculation is labelled derived with its formula and operands.
Reported means it appears verbatim in the source. Derived means we calculated it (for example, +2.0pp from 2.3% → 4.3%) and showed the formula. Derived numbers are never presented as though the source reported them.
Observational benchmarks are labelled as observed associations — “associated with”, “comparison”, “benchmark” — never as causal experiments. Multiple metrics are shown side by side to surface trade-offs.
Flops and null portfolios are kept, because knowing what did not show a lift saves research time. “No significant difference” means power was undisclosed, not that the true effect is zero.
If the source did not disclose sample size or statistical information, the field says Not disclosed rather than being hidden or inferred. Limitations are prominent, not buried in a tooltip.