Skip to content

0005: Belief-Base Revision Semantics

Date: 2026-06-26

Accepted.

dent8’s contradiction/supersession/retraction/expiry semantics independently reinvent ideas from the formal belief-revision literature. Naming the right theory gives dent8 a defensible vocabulary and pre-empts the obvious reviewer objection (“this isn’t real AGM”). Classical AGM operates on logically-closed belief sets and demands global consistency — neither of which dent8 wants.

Adopt belief-base revision (Hansson), not classical AGM, as dent8’s formal identity:

  • dent8’s “memory” is a fold/projection over an immutable FactEvent base, not a deductively-closed set. The base, and its history, are authoritative.
  • dent8 deliberately does not satisfy the Recovery postulate. Retract-then- reassert must not resurrect dependents; the re-assertion carries fresh provenance and evidence.
  • The contested lifecycle plus preserved contradicted_by edges is a paraconsistent design (inconsistency ≠ triviality): localize a contradiction, keep the store non-trivial, surface it — never silently merge A and ¬A.
  • Authority is an epistemic-entrenchment ordering, kept strictly separate from Confidence (evidential strength). Entrenchment decides what is surrendered first.
  • dent8 makes no fact of logical closure or an entailment engine; it implements the operational spirit of revision operators over opaque triples.

Positive:

  • A precise, citable formal grounding (belief-revision.md).
  • Justifies keeping AuthorityLevel and Confidence as separate fields.
  • The Recovery non-postulate is the correct, defensible answer to “not real AGM.”

Negative:

  • The mapping of asserted/superseded/retracted onto expansion/revision/contraction is inspirational, not rigorous — must be stated as such to avoid overfacting.
  • “Belief-base framing” is principled grounding, not a novel mechanism (bitemporal DBs already provide “history matters”); it must not be sold as a contribution.
  • Implement authority-as-entrenchment arbitration (ADR 0007).
  • Decide JTMS vs ATMS for the debugger’s assumption-environment replay (future ADR).
  • [DONE] Add valid_to (closed valid-time intervals) — built as ADR 0016, together with --as-of / --valid-at time-travel reads.