Skip to content

Latest commit

 

History

History
130 lines (104 loc) · 5.97 KB

File metadata and controls

130 lines (104 loc) · 5.97 KB

EXPLAINME — VeriSimDB Orientation

VeriSimDB does not merely store records. It maintains identity consonance across modal witnesses and federation boundaries. VCL expresses propositions and epistemic requests over that consonance state. VCL-total validates whether proposed identity transitions are admissible.

This is the human-readable orientation. For the feature/benefit framing see README.adoc; for the normative VCL grammar see the VCL specification (file still historically named VCL-SPEC.adoc).

The one idea to hold onto

If you take away a single thing: a VeriSimDB "entity" is not a row. It is a consonance subject — an identity under continuous maintenance across up to eight modal witnesses (the octad: graph, vector, tensor, semantic, document, temporal, provenance, spatial). The system’s job is not storage; it is deciding, continuously, whether those witnesses still cohere as one identity — and what to do when they do not.

So the verbs are not CRUD. They are lifecycle transitions on identity claims: declare, assert, inspect, verify, transform, retract, merge, split, normalise, tombstone.

Why "consonance", not "consistency"

Classic databases enforce consistency up front (transactions, locks) and treat divergence as an error to prevent. VeriSimDB instead expects modal witnesses to drift — a document is edited before its embedding is recomputed; a spatial coordinate is refined independently of the text — and treats that drift as a first-class, typed condition to be measured, classified, and repaired. The maintained property is consonance: the witnesses still refer to the same identity, within policy.

A modality failure is therefore not automatically an identity failure. These are distinct states with distinct proof obligations and repair paths:

Observation State

document changed, vector stale

modal drift (repairable by normalisation)

hash chain broken

provenance / integrity failure

remote node asserts an incompatible type

federation conflict (needs arbitration)

all live witnesses retracted

inactive / tombstoned identity

The identity lifecycle

Candidate
  -> Declared
  -> Asserted
  -> Consonant
  -> Drifting
  -> Normalised
  -> Merged / Split
  -> Retracted
  -> Tombstoned
  -> Expunged  (only under explicit policy)

Retraction is not deletion. A retracted claim usually stays visible to provenance, audit, federation, and proof systems — it is no longer live. Tombstoning durably retires an identity or claim without pretending it never existed. Expungement is a separate, policy-governed operation.

Core terms

Term Meaning

Consonance Subject

The lifecycle-managed identity under consideration.

Identity Claim

A claim that some subject exists, has a property, has a relation, or survives a transition.

Modality Witness

One modal presentation of a subject: graph, vector, tensor, semantic, document, temporal, provenance, or spatial.

Drift Report

Evidence that one or more modal witnesses no longer cohere.

Transition

A proposed identity event: declare, assert, retract, normalise, merge, split, federate, tombstone.

Proof Obligation

What must be shown before a transition is admissible.

Federation Attestation

A remote or local authority claim about identity, provenance, validity, or trust.

Tombstone

A durable marker that an identity claim is no longer live, without pretending it never existed.

VCL, VCL-total, and VeriSimDB

Layer Role

VeriSimDB

The federated consonance engine. Stores and maintains identity consonance across the octad of modal witnesses; detects drift; repairs under policy.

VCL

The production consonance language. Statements are propositions (DECLARE, ASSERT, RETRACT) or epistemic requests (INSPECT, VERIFY), plus MERGE / SPLIT / NORMALISE — directed at the engine, not queries against a passive store.

VCL-total

The proof-bearing safety substrate behind VCL (repo: vcl-ut). It decides whether a proposed identity transition is admissible under schema, modality, provenance, authority, freshness, resource, effect, federation, and proof constraints.

The proof goal is deliberately narrow and useful — not "prove the whole database correct", but:

prove that THIS identity transition is admissible,
traceable, authorised, and consonant enough to affect live state.

What is implemented vs design intent (honest)

  • Implemented today: the octad stores (graph, vector, tensor, semantic, document, temporal, provenance hash-chain, spatial R-tree); cross-modal drift detection + self-normalisation; the federation coordinator; and a VCL surface whose current in-repo grammar (spec/grammar.ebnf v3.0) is a read-style, SQL-shaped convenience (SELECT … FROM HEXAD …, with HEXAD the legacy keyword for the octad source) plus PROOF-carrying statements.

  • The model / direction: the lifecycle/consonance verbs (DECLARENORMALISE) and full proof-bearing admissibility are the semantic target, carried by VCL-total (vcl-ut). Read the SQL-ish read-path as epistemic inspection of consonance state, not row retrieval.

  • Formal core: invariants are mechanised in Coq under formal/ (octad modality algebra, provenance hash-chain, drift bounds, planner equivalence, WAL/normaliser idempotence, VCL preservation) — see the cross-repo map and the trusted-base ledger.

Where to go next