Skip to main content
Evidence

Evidence that says what it establishes
— and what it does not.

HAIEC preserves source, provenance, scope, coverage, and limitations so assurance can distinguish established facts from partial or missing evidence.

Source
Repository / IAM / Policy / Runtime
Evidence
Source + provenance
Coverage
Established / Partial / Unknown
Assurance
Bounded evaluation
Verify
Package / receipt where supported

Explanatory structure — not a claim that every evidence type is always present.

What Evidence Is

Evidence is information connected to an AI system or assurance evaluation. Depending on what is connected and available, evidence may come from:

  • Source analysis of the application code
  • Identity and access evidence
  • Policy evidence
  • Provider IAM evidence where supported
  • Runtime evidence where connected
  • Uploaded or declared evidence where supported

Not every source is always available. What matters is that what is available is clearly distinguished from what is not.

Provenance

Evidence should retain enough information to identify where it came from and what produced it. HAIEC preserves provenance so reviewers can trace evidence back to its source — whether that is a source code scan, an identity record, a policy document, or a runtime observation. Provenance is what makes evidence auditable rather than merely asserted.

Coverage

HAIEC distinguishes between what the evidence establishes, what it partially establishes, and what remains unknown or not evaluated:

Established
Evidence is present and supports the claim within its scope.
Partial
Evidence is present but incomplete or limited in scope.
Unavailable / Unknown
Evidence is not available or has not been connected.
Not Evaluated
No evaluation has been run for this evidence question.

Missing evidence remains missing evidence. HAIEC does not silently convert absence into assurance.

Limitations

Evidence must make gaps visible. A system can have valid source analysis, correct identity configuration, and a clean policy envelope — and still have an unanswered question about what it can actually do at runtime.

Missing evidence remains missing evidence.

HAIEC does not silently convert absence into assurance. Unknown stays unknown. Partial stays partial. Not evaluated stays not evaluated.

Connection to Assurance

Evidence feeds assurance evaluation. Assurance evaluates what the evidence establishes within a defined scope and produces a bounded disposition. Where supported, assurance packages and decision receipts preserve the evaluation output for review.

Evidence does not itself equal Assurance.

Evidence is what you have. Assurance is what the evaluation establishes about it. Evidence ingested is not evidence evaluated — an accepted submission changes the Evaluation only when a new Evaluation runs. Framework mapping is not assurance. An observation that is absent is not the same as a fact that something did not happen.

Integrity

Where supported by the current implementation, HAIEC uses tamper-evident, hash-bound integrity for evidence packages and decision receipts. This means reviewers can verify that artifacts have not been modified after generation. HAIEC does not universally claim that every artifact is permanently immutable or externally notarized unless that is actually established.

What Evidence Is Not

Evidence is not Assurance
Evidence is input. Assurance is the bounded evaluation of that input.
Evidence is not Certification
Framework mapping and evidence packages do not constitute third-party certification.
Source-established is not Safe
Knowing what a system can reach does not mean it is safe to deploy.
Missing evidence is not a Pass
Unknown stays unknown. Absence is never silently converted to approval.
Unknown is not Pass
If the evidence is not available, the disposition reflects that.
Observation absent is not ‘did not happen’
No runtime observation connected means exactly that — not a claim that nothing occurred.

See evidence in context

Explore how evidence connects to AI actions, access, and consequences in the sample AI Action & Access Map.