🚩 Evidence Flags: More trust signal, fewer words
Evidence Flags is a Copilot skill that adds brief labels to claims to distinguish what is backed by evidence, reported by users, or inferred by the model. It helps when explicitly requested via /ef commands to clarify trust levels in responses without cluttering simple questions.
post Gerardo Rentería Blog
"Instead of filling a response with qualifiers such as maybe, I think, or it seems, a brief label is added to relevant claims."
Key points
- Uses seven labeled tags across two families: provenance (Supported, Reported, Unchecked) and AI analysis (Inferred, Hypothesis, Forecast, Judgment), plus optional warnings for contested or time-sensitive claims
- Activated only when explicitly requested with /ef commands like /ef on, /ef check, /ef trail, keeping normal conversations unaffected
- Helps distinguish between what Copilot verified in files, what users stated, and what the model deduced, without subjective qualifiers like maybe or seems
- Does not verify truth, assign confidence scores, or replace technical review; Supported means claim is backed by evidence inspected in that conversation only
Quotes
- “Instead of filling a response with qualifiers such as maybe, I think, or it seems, a brief label is added to relevant claims.” (Explains the core design principle: using structured labels instead of hedging language makes trust levels explicit and scannable.)
- “Supported does not mean absolute truth: it means the claim is supported by evidence the model inspected during that conversation.” (Critical boundary on what the skill does: it clarifies context-specific verification, not universal truth.)
AL objects mentioned
As named in the post. A name that matches one object page by exact type and name links to it; the others stay as named.
- other “Customer”
Context
- Features: Evidence Flags skill, Claim labeling by provenance, AI analysis tagging, Time-sensitive warning, Contested source warning, /ef command interface
Source: Gerardo Renteria, community blog. Summary, key points and quotes are derived (CONTENT-NOTICE.md); read the original for the full text.