Skip to content
AletheiaAletheiaAletheia docs

How answers cite evidence

An answer you cannot check is worth very little when the question was about production. So every answer carries its evidence.

Citation markers

Section titled “Citation markers”

Claims carry a superscript marker — [1], [2] — pointing at an evidence card below the answer. Each card names the record: what kind of thing it is, which one, and a link straight to it in the product.

Following a citation takes you to the resource, application, or change the claim rests on. You can check any statement without asking a follow-up question.

Evidence is dropped, not invented

Section titled “Evidence is dropped, not invented”

Aletheia builds the evidence list from what its queries actually returned. If a piece of evidence is not cited in the final answer, it is dropped from the list; if a marker survives with nothing behind it, the marker is stripped.

The list you see is therefore the evidence the answer actually used — not everything that was looked at, and never a citation pointing at nothing.

Structured results stay structured

Section titled “Structured results stay structured”

Where an answer is really a list — resources matching a filter, applications without an owner — it is rendered as a table built from the query results themselves rather than retyped into prose. Values that are machine identifiers are shown as such, and link where they can.

This matters for correctness, not just for looks: a table assembled from result rows cannot contain a resource that was not in the results.

Confidence and freshness

Section titled “Confidence and freshness”

Evidence cards carry two qualifiers where they apply:

  • Confidence — for records that are themselves inferred, such as a code mapping or an inferred ownership. A low-confidence mapping cited in an answer is shown as low-confidence.
  • Freshness — when the record was last observed. An answer resting on a scan from last week says so.

Neither is hidden behind a hover. An answer built on weak or old evidence should look like one.

Inference is marked

Section titled “Inference is marked”

Where the assistant draws a conclusion rather than reporting a record, that part is marked as interpretation. “These three resources are unmanaged” is a fact from a query. “This looks like it was created by hand during an incident” is an interpretation, and is labelled as one.

When there is no evidence

Section titled “When there is no evidence”

If Aletheia cannot support a statement with a record, it says so instead of producing something plausible.

An answer with no evidence behind it is a bug, not a feature. If you see a claim with no citation that should have one, that is worth reporting.

Where to go next

Section titled “Where to go next”