The Proof Ledger
RankEcho re-tests the exact prompts customers fix, on a schedule, and records whether the citation appeared. This page will publish the aggregate outcomes - reportable wins, median days to first citation, and the retrieval-vs-training pathway split. Until the dataset clears its publishing threshold, the methodology below is the page.
Looking for individual stories instead of aggregates? Proof Notes publishes single engagements end to end - including the ones where nothing moved.
Collection context: 22 measured prompts and 7 reportable wins are currently stored. Both thresholds must be met before figures publish.
A tracked prompt counts as an observed win when the exact prompt - re-run on a schedule after a fix shipped - returns at least one citation of the tracked brand. It only graduates to a reportable win when sampling reaches medium confidence (at least 3 post-ship runs) and the post-ship citation rate beats that prompt's own pre-ship baseline.
First citation within 3 days of shipping is classed as a fast retrieval (RAG) effect; 14 days or later is classed as a slow training-cycle effect; in between is recorded as unknown.
- These are correlation-with-controls observations, not causal guarantees. AI answers vary by engine, model version, retrieval state, time, and wording.
- The sample is self-selected: prompts that RankEcho customers chose to fix and track.
- Figures stay withheld until at least 25 measured prompts and 10 reportable wins exist.
- No customer-level results, domains, or prompts are ever published here - aggregate counts and medians only.
Aggregate counts and medians only. This page never publishes customer-level results.