WS-CAND-001 — Authority cannot confer truth

Wellspring · hosted from the CENTL repository

Research library · Wellspring

Wellspring

A reusable computational principle: a producer of bytes, prose, or proposals never becomes authority over mathematical meaning, artifact identity, or assurance.

Source in the repository

Status: candidate

Date identified: 2026-08-14

Investigators / contributors: CENTL/FCF research expedition

Originating expedition: secret-oasis-2026-08-14

Source / commit / artifact identity: docs/NUMERICS.md, docs/CENTL-MIRAGE.md, docs/CARAVAN.md

Core finding

A reusable computational principle: a producer of bytes, prose, or proposals never becomes authority over mathematical meaning, artifact identity, or assurance.

Exactness is independent of transformation success. Model output is untrusted input. Carriers provide availability, not trust. Generated code does not inherit verified-core status. Admission requires an evidence lattice rather than a reward score.

Assumptions and scope

  • The claim is a unifying principle across CENTL surfaces, not a claim that each constituent technique is unprecedented.
  • Prior art exists independently in interval arithmetic, proof-carrying code, and The Update Framework.

Evidence

  • Implemented numerical contract: exact values, enclosures, and unresolved results remain distinguishable.
  • F* validates dyadic endpoints before decimal rendering.
  • CENTL-SCi treats local models as interpreters of intent.
  • MIRAGE admission is hard-gated before any scoring.
  • CARAVAN carriers cannot define trusted artifact identity.

Independent review: not performed; this expedition records a candidate only.

Known counterexamples / failure modes

  • If a scientific result required manufacturing an unqualified digit, the principle would be too strong.
  • If model output could be shown to confer mathematical authority without a deterministic checker, the interpreter/authority split would fail.

Downstream avenues opened

  • stronger numerical and rendering contracts
  • self-development admissibility without model authority
  • preservation networks that separate availability from trust
  • machine interfaces that cannot promote generated assurance

Oasis impact

None by itself. Any implementation still requires ordinary Oasis qualification.

Security or publication constraints

Does not authorize publication of sensitive expedition materials.

Falsifiers / demotion conditions

  • Independent review shows the unification is only a rename of one existing technique.
  • A required CENTL result cannot be stated without collapsing exactness, transformation resolution, or assurance.
  • Reproduction fails to locate the principle in more than one independent surface.

References