Skip to content
PrismaLens Docs

Ordered evidence

Every PrismaLens investigation ends in the same structure, and one choice shapes all of it: the report orders hypotheses by evidence rather than attaching a numeric confidence score to each. This is a deliberate design decision (ADR-0002), not an omission.

A number like “87% confident” reads as precise and earns trust it hasn’t necessarily earned. For an LLM-driven tool it’s worse than imprecise — it’s a fabricated statistic. The model has no calibrated basis for “87 vs 72,” so a score invites you to act on a digit instead of on the reasoning behind it.

PrismaLens refuses to manufacture that number. Instead it gives you the thing a score is a lossy summary of: the evidence, ranked.

  • Hypotheses, ordered most to least plausible. The order is the ranking — there’s no separate rank field. Each hypothesis carries the evidence that supports or contradicts it, and where that evidence came from (the exact command or source), so you can check it yourself.
  • Ruled out. Candidate causes that were considered and dropped, with the contradicting evidence that eliminated them — so you can see the paths already walked, not just the conclusion.
  • Coverage. What was queried and what wasn’t. An investigation that couldn’t reach a data source says so, rather than letting silence read as “all clear.”
  • Next steps. Concrete checks to confirm the top hypothesis or dig further.

Because there’s no score to hide behind, the report puts its reasoning in the open — which is exactly how you should read it. A PrismaLens report is a strong, evidence-linked lead, not a verdict; the same run can differ across attempts, and a confidently-worded top hypothesis can still be wrong. The ordering tells you where to look first; the linked evidence lets you confirm it fast. See How PrismaLens works for more on treating a report as a lead.