← Glossary
Evidence surfaces
Evidence surfaces are the “why this” layer of an experience: the curated evidence behind a recommendation, decision, or agent action. They are designed to be readable under time pressure and strong enough to support dispute.
Definition
- An evidence surface is a UI/UX component (or operator view) that exposes: key inputs, constraints, provenance, and reasoning signals behind an outcome.
- It is not raw data. It is a curated, minimal basis for confidence.
- Evidence surfaces can support users (“why was I asked this?”) and operators (“why did the agent do that?”).
Why it matters
- Without evidence surfaces, AI decisions feel arbitrary. Users either over-trust (danger) or under-trust (no adoption).
- Evidence surfaces improve governance: they make policy constraints visible and enforceable.
- They enable contestability: users and operators can challenge decisions with concrete reference points.
- Evidence surfaces lower the cost of oversight: when signals are visible, humans can supervise many decisions without micromanaging each one.
- In practice, this is where many digital programs fail: the concept is understood, but the operating discipline is missing.
Common failure modes
- Explainability theatre: vague statements with no actionable evidence.
- Data dumping: showing everything, so the signal is lost.
- No provenance: users can’t tell whether evidence came from user input, system data, or external sources.
- No link to action: evidence is shown, but the user can’t correct inputs or appeal outcomes.
- Inconsistent surfaces: evidence appears sometimes, disappears when it matters most.
How I design it
- Define the minimal evidence set per decision type (typically 3–7 items).
- Show constraints explicitly: policy rules, thresholds, and required conditions.
- Provide correction routes: edit inputs, attach supporting evidence, or request review.
- Differentiate certainty levels: confidence signals, disagreement flags, or “human review required”.
- Log the evidence surface in the audit trail so investigations can reconstruct the decision basis.
- Separate user and operator evidence: the same truth, different depth. Don’t overload users with operational detail.
- Treat it as a repeatable pattern: define it, test it in production, measure it, and evolve it with evidence.
Related work
Proof map claims
Case studies
See also
Contact
Let’s discuss a leadership role, advisory work, or a complex product challenge.