← Evidence standardsMeasurement before expansion

Accuracy study and publication gates

DoctorOwned will publish measured performance only after a reproducible, independently reviewed study. Until then, record counts are inventory, not proof of accuracy.

Study protocol

  1. Freeze a deterministic 300-record sample stratified across no-known-DSO, unresolved, possible-signal, and affirmative-signal tiers.
  2. Have two reviewers independently classify each record from publishable evidence.
  3. Require a source URL and observation date for every non-unsure label.
  4. Adjudicate disagreements without exposing either reviewer's first label to the other.
  5. Publish per-tier precision, DSO recall, abstention rate, sample counts, and 95% confidence intervals.
  6. Version the labeled corpus and rerun it against every material classifier release.

Expansion is gated

Corporate relationship graphs, public APIs, alerts, and bulk exports remain deferred until the study is published and the high-risk affiliation tiers meet the approved precision threshold.

Swipe the table horizontally to read each gate's meaning.

GateCurrent stateMeaning
Correction channel liveReadyTurnstile and the monitored fallback are configured.
Verified tier fails closedEnforcedIdentity, ownership evidence, source coverage, and conflict handling are database-enforced.
Operator conflict disclosedDisclosedThe relationship to Smile Science and the recusal rule are stated publicly.
Affirmative DSO labels have reviewed citationsIn progressUnreviewed automated signals render as ownership unclear rather than affirmative affiliation.
Stratified accuracy study publishedBlockedThe 300-record sample still requires two independent reviews and adjudication.

What is not claimed today

No site-wide precision, recall, or ownership-certification rate has been published. A “no known DSO” result remains an automated negative screen. An unresolved result is an explicit abstention, not a failure hidden from the visitor.

Report an evidence problem