Continuity here means an intelligible causal lineage — which investigation produced which claim, under which policy, with which model — not identical model weights. State is designed to be superseded, never silently overwritten: supersedes, revised_by, contradicted_by.
Continuity & Identity
Stable agent id, model history, revisions, claims, policies, provenance.
The engine does not expose a lineage record yet. When it does, each entry appears here with its predecessor intact — nothing is overwritten in place.
requires: lineage[] with supersedes / revised_by / contradicted_by edges and agent_id
Agent identity
Stable identifier across model swaps.
- agent_id
- not reported
- model history
- not reported
- policy history
- not reported
- memory revisions
- not reported
requires: agent_id + model_history[] on the engine
Claims & revisions
Each claim keeps its superseded predecessor.
No claim ledger is exposed yet. When it is, revisions will render as a chain rather than an in-place edit.
requires: claims[] with revision edges
Disagreement
Where sources or reasoning passes conflict.
The engine does not report disagreements yet. This view stays empty rather than inventing conflicts.
requires: disagreements[] with the two conflicting claims and their origins