Clarify evidence tier authority and proof-of-concept approval gate

2026-07-24 04:37:14 +00:00
parent 1db650227a
commit 63ae28c7ed
1 changed files with 21 additions and 1 deletions
@@ -71,4 +71,24 @@ The company should not overstate indicative findings as proven customer loss.

 Customer Path Leakage may support stronger before/after proof.

-Discovery Failure may support directional before/after comparison, but should be labeled as sampled or indicative when appropriate.
+Discovery Failure may support directional before/after comparison, but should be labeled as sampled or indicative when appropriate.
+
+# Evidence Tier Assignment Authority
+
+Agents may propose an evidence tier for a finding, but the final evidence tier used in any client-facing report must be approved by a human reviewer during Version 1.
+
+A finding may only be classified as Verified Evidence when it is supported by concrete, current, observable, timestamped, or instrument-backed evidence.
+
+If a finding does not clearly meet the standard for Verified Evidence, it defaults to Indicative Evidence until a human reviewer confirms otherwise.
+
+Ambiguous findings must not be presented as confirmed customer loss.
+
+If an agent proposes a finding as Verified Evidence and the human reviewer rejects that classification, the finding defaults to Indicative Evidence unless the reviewer discards the finding entirely or requests additional evidence.
+
+**Examples:**
+
+- A currently visible wrong phone number on a verified business profile may qualify as Verified Evidence.
+- A broken booking link observed and timestamped during review may qualify as Verified Evidence.
+- A stale or unconfirmed citation mismatch defaults to Indicative Evidence until confirmed.
+- Absence from sampled AI answers is Indicative Evidence.
+- Agent-reported findings without supporting artifacts are not sufficient for Verified Evidence.