Candidacy evidence licenses escalation to assessment, not acceptance

Type: kb/types/note.md · Tags: foundations, discovery

Status: conjecture — two worked cases.

In staged evaluation, evidence plays two roles that must not be conflated. Candidacy evidence shows a hypothesis deserves an expensive assessment; verdict evidence decides that assessment. Candidacy evidence can afford to be cheap, defeasible, and merely author-external, because the only thing it puts at risk is assessment cost. The moment it carries verdict authority, the workflow acquires an immunizing slot: claims get accepted on evidence that was calibrated only to justify spending attention on them.

The external anchor is Peirce's economy of research. Peirce's question for abduction was economic — which hypotheses deserve the cost of testing, given explanatory promise and the price of the test — and its output is a pursuit-worthiness judgment, not a truth judgment. The two-role separation restates that: routing decisions and verdicts consume different evidence because they answer different questions. The mechanism behind the prohibition is warrant non-distribution — since theory warrant should be tracked at the finest granularity evidence licenses, evidence for pursuit-worthiness cannot silently become warrant for the assessed conclusion — compounded by selection: candidacy evidence is selected for cheapness, so it systematically underdetermines the verdict.

The first worked witness is the idealization assessment: domain pricing routes an exception to assessment, and the adequacy record decides. The failure mode was also observed, not just predicted — a pricing-gated verdict ("holds as idealization, exceptions priced") was proposed and then defeated in review before adoption, exactly because it would have accepted claims on routing evidence alone.

The second witness is source grounding. Search and thematic fit nominated Pirolli as prior art for an agent-navigation note; only direct comparison could decide which target claims the source established. A combined-context pass labeled three of four compared claims subsumed. An independently decomposed pass tightened all three to needing narrowing or only partial support. The bounded case does not show that role isolation caused the improvement. It does show that the candidacy signal was right to trigger assessment and unfit to substitute for that assessment. Unlike idealization repair, this witness concerns external literature and artifact disposition.

What the claim forbids: no staged workflow soundly accepts on candidacy evidence alone. Its refuter is a staged assessment in which routing-grade evidence deciding the verdict demonstrably opens no immunizing slot, or in which maintaining the separation costs more than the misacceptance it prevents.

Open Questions

  • Do the two witnesses discriminate the same evidence property? Domain pricing and thematic source fit are both cheap relative to assessment, but they differ in author-externality and in what they say about the candidate.
  • What makes evidence candidacy-grade rather than verdict-grade in general? Relative cost, defeasibility, and the low consequence of a false nomination are shared candidates, but their necessity is untested.

Relevant Notes: