Evidence methodology · current

Methodology v0.3

Current immutable method for uncertainty-aware claim decisions and versioned calculated-record revisions.

Published

August 4, 2026 · 2026-08-04T00:00:00.000Z. This route is an immutable public method record.

Binary decisions

Binary estimator: qualifying successes divided by qualifying observations. The interval is a two-sided 95% Wilson score interval. A threshold is supported when the lower bound is at least the predefined threshold, not supported when the upper bound is below it, and inconclusive when the interval includes it.

Continuous decisions

For continuous or paired claims, the Claim Protocol locks the estimator_id and allowed estimator family. Paired bootstrap resamples participant pairs and recomputes that statistic in every sample; it is a two-sided 95% percentile interval with 10,000 resamples and is unavailable without required raw pairs.

Verification and completion

Endpoint purchase-verification coverage = endpoint completers who are purchase verified ÷ all endpoint completers. Endpoint completion = endpoint completers ÷ endpoint-eligible participants. Qualifying cohort excludes missing primary measurements, prespecified exclusions, and manipulation-review failures.

Missing intervals

When a required interval is unavailable, threshold certainty is not evaluable and final decision is inconclusive. A point estimate may still be described as above or below threshold, but it is not an interval-certainty classification.

Revision rule

A substantive recalculation creates a new immutable calculated-record revision with methodology, protocol, engine, and dataset-snapshot versions. Earlier decisions remain visible and are never relabelled as a later method.

Qualification matrix · deterministic

Evidence strength is separate from threshold certainty.

Insufficient

Any protocol requirement missing, manipulation review not passed, or qualifying cohort below 25.

Emerging

At least 25 qualifying records; ≥60% endpoint purchase verification; ≥40% endpoint completion; required method fields present; manipulation review passed.

Established

At least 100 qualifying records; ≥70% endpoint purchase verification; ≥55% endpoint completion; valid primary measurement and manipulation review passed.

Strong

At least 500 qualifying records; ≥80% endpoint purchase verification; ≥65% endpoint completion; valid primary measurement, manipulation review passed, and either a longitudinal replication or independent replication.

Each assessable Claim Card links to a versioned protocol that fixes its estimator, comparator, endpoint, threshold, eligibility, exclusions, verification, missing-data handling, manipulation review, uncertainty method, and evidence-strength rule before interpretation.

Browse Claim Protocols

Demonstration evidence is fictional and does not constitute a live consumer outcome network.