Confidence Model

Ontoly confidence is deterministic. It is derived from graph evidence, not from language-model judgment.

The confidence object contains:

• score: number from 0 to 1.
• level: none, low, medium, or high.
• explanation: human-readable derivation.
• factors: evidence and diagnostic inputs that affected the score.

Scoring

Capabilities compute confidence from:

• amount of direct graph evidence
• edge provenance kind
• relationship confidence
• diagnostics severity
• missing or ambiguous targets

Warnings reduce confidence. Errors reduce confidence more heavily. No evidence produces score: 0 and level: "none".

Relationship Evidence

Exact syntax and semantic edges carry the strongest confidence. Inferred or low-confidence edges lower the score. This keeps answers honest when a graph contains partial or plugin-derived facts.

Consumer Rules

Consumers should:

• show confidence with the answer
• preserve diagnostics
• avoid presenting low-confidence answers as facts
• ask Ontoly for narrower graph evidence before inspecting source files