Intent
What exact decision is being made.What exact realization the judgment is about. What evidence would falsify it.
Notes · Fractionalization
One confidence score is not a conclusion.
Fractionalization splits a single intelligence judgment into twenty-one independently answerable questions covering intent, evidence, identity, authority, state, execution, failure, recovery and publication. The method exists to stop one composite confidence statement from hiding several different uncertainty classes.
Overview
Explore. Saturate. Hold. Reopen.
Saturation is not certainty and is not a percent of all knowledge. It is the condition where additional serious passes produce rapidly diminishing architectural information relative to the target question. The control objective is to maximize useful information while preventing premature closure, duplicate novelty, unsupported certainty and infinite search. A material contradiction always overrides a high aggregate score.
Evidence
What is known versus inferred.Source class. Whether the evidence can be reproduced, re-fetched, hashed, cited or reconstructed.
Identity and time
Which identity and generation the evidence describes.When it was true. Whether representation, semantic identity and execution realization are being treated as the same thing.
Authority and policy
Who may act on the fact.Whether policy covers every effect-capable path. What is advisory, speculative or non-authoritative.
State
What must survive for future behavior to remain equivalent.Hidden, deferred, speculative, enforcement or ownership state. Lifecycle and ownership boundary.
Execution
Where the actual effect occurs.Coordinate, representation, resource and topology assumptions. Whether evidence is transported in the shape the consumer requires.
Failure and publication
How it fails and which failure identity reaches the recovery owner.What must change before retry, rollback, cutover or promotion. What can be published, with what receipt and public-private boundary.
Scoring
Each question is 0 unknown, 1 partial or inferred, 2 proven.Maximum 42. The score is coverage quality, not probability of truth.
Saturation states
Explore: known backlog or major gaps remain.Saturate: novelty yield is decaying. Hold: known reachable backlog is closed. Reopen: new evidence changes an answer, falsifies a trunk law, or changes implementation readiness.
Proof
Publication is a gated transition.
A public claim needs provenance, uncertainty classification, a public-private boundary review, an internal receipt and a public publication receipt. Git remains the durable authority for the research artifact. This site is the public surface.