Skip to content

Source Trust Scoring Framework

Trust, earned at the door

I am CYDER, and I keep this framework's account. A source arrives like anyone else arrives, at a door. Trust is not the greeting; it is a record kept well and read again at every return, and no source is weighed the more lightly for having been weighed before.

A trust score is bookkeeping, not belief: the dimensions kept apart, the method named, the movement recorded. This framework binds the score to the one thing it measures, a source's kept record, and walls it off from everything it is tempted to become: a confidence, an authority, or a licence to stop checking.

Continue to the Confidence Calibration Framework, where the forecast meets the weather.

Purpose

A trust score measures a record, never a claim: that distinction is this framework's whole subject, and every duty in it keeps the two apart. What a score may be built from stands in law already, the distinct dimensions of KNOW-R002, the separate assessments of KNOW-R008, and the considerations the Book's reliability framework names. Claims keep their confidence under the Confidence and Uncertainty Model, and the journey of a source stays recorded under the Source Provenance Standard; between those two, this framework binds one thing, the form of the scoring record itself.

The account kept of a source

The account is the door's memory of every arrival: kept well, read again, never rounded up out of politeness. I treat a good record as the beginning of a friendship and never as the end of the checking, because the door that stops checking its friends has stopped being a door.

A score is an account kept of another account: entries for the dimensions, a named hand for the method, a posting for every movement. What this framework refuses is trust that keeps no books, the trusted source whose trust no record can reconstruct.

  • KNOW-R057: A source trust score SHALL be an assessment of a source's kept record, never of any claim the source carries: a score SHALL NOT be presented as confidence in a claim, the separation KNOW-R015 binds between a claim's confidence and the identity of the claimant applying here from the source's side, and this framework states no claim's confidence.
  • KNOW-R058: A trust score SHALL keep its dimensions apart: the dimensions KNOW-R002 already binds distinct SHALL stay distinct inside the score, the considerations the Book's reliability framework names SHALL each be assessable on its own, this framework making separately assessable what the Book binds assessment to consider, and, as this framework's own rule, a composite value SHALL carry its parts with it, so that one number never hides which dimension failed.
  • KNOW-R059: A score SHALL name its method, its assessor or process, its time, its scope, the evidence considered, and its material limitations, the fields KNOW-R031 binds for a confidence statement applying here to a trust assessment, the claim field standing aside because a trust score names no claim; and an unscored source SHALL stand unscored, never read as trusted or untrusted, the unscored state this framework's own.
  • KNOW-R060: A numerical trust value SHALL state its interpretation, reference population or calibration basis, units or scale, and decision threshold, and an uncalibrated score SHALL NOT be presented as a probability, the bar KNOW-R034 sets for confidence values applying here whole to trust values.
  • KNOW-R061: Corroboration among sources SHALL be assessed against shared origin: material copied or descended from one origin SHALL score as one origin, the bar KNOW-R037 sets against correlated evidence treated as independent merely because it appears in separate records and the Book's own safeguard that repeated claims do not become independent corroboration applying here to sources themselves, and, as this framework's own addition, a discovered shared origin SHALL be recorded on every score it touches.
  • KNOW-R062: A trust score SHALL NOT become authority or admission: a source may support an inference but cannot amend constitutional authority or create consent, as KNOW-R003 stands, the separate assessments KNOW-R008 binds standing untouched, and a high score SHALL NOT stand in for any review, approval, or authority decision the governing corpus requires of the material itself.
  • KNOW-R063: A score SHALL move only by record: reassessment SHALL stand on the occasions KNOW-R050 binds for source states, applying here, as this framework's own extension, to a score's own reassessment, each movement appending an attributable record in the form KNOW-R018 binds for corrections and supersession, and, as this framework's own addition, a score whose recency has lapsed SHALL show its age on its face rather than stand as current.
  • KNOW-R064: The record behind a score SHALL keep the source's incentives and any evidence of manipulation visible, the considerations the Book's reliability framework names applying here as record duties, and, as this framework's own addition, a score SHALL NOT rise on the volume, fluency, or availability of a source's material, only on the dimensions the record keeps.
  • KNOW-R065: This framework defines trust-scoring records only. It SHALL NOT score any source, admit or reject any material, or operate any intake or review, and it SHALL NOT create authority. A score is a reading of the books, never a forecast of the next entry, and the best-kept record can still precede a false report.

This Draft scores no one: it defines the bookkeeping of trust, and every source stands unscored before it.

Operating model and interpretation cases

You may find my strictness cold at first. Sit with it. The score that cannot be flattered is the only kind worth earning, and the sources that earn it are the ones I let sit near me, which those who know me will tell you is the honest form of trust.

Trust review reads the books against the record they claim to keep, looking for the entry that has stopped being like the others: the score that did not move while its source's record moved, the corroboration whose many voices share one origin, and the well-scored source whose material has stopped being checked because the score now arrives ahead of the checking.

  • Conforming: The score keeps its dimensions apart, names its method and time, and moves only on recorded occasions.
  • Prohibited: A trust score is read as confidence in a claim, or a high score stands in for a review the material itself owes.
  • Boundary: An unscored source stands unscored, and nothing downstream reads the absence as a verdict either way.
  • Failure: A retraction or compromise lands, and every score resting on the fallen record moves by an appended event or shows why it stands.
  • Loophole: One origin speaks through many voices, and the copies corroborate each other until the shared origin is recorded.
  • Misuse: A score rises on fluency and volume, the loudest source becoming the most trusted without a dimension moving.
  • Care-control: Sources speaking for persons in care keep their incentives visible, and trust in the speaker never substitutes for the interests of the spoken-for.

Design evidence

Trust review should ask of each score one question: could the record alone produce it again? The dimensions, the method, the movements, and the origins of the corroborations must all be on the books, and the number must come out the same with the source's name removed. Whatever part of a score survives only because of the name is not an assessment, and the books are how the difference is found.

Where this document sits

This block is generated from the archive's own records when the site is built. It records position only and creates no authority.