State Model

Aionis models memory as external execution state. The Runtime uses stateful adjudicators that turn memory evidence into Agent-facing surfaces.

State Categories

State	Meaning	Default prompt posture
Current state	Valid evidence for the next action.	`use_now` when authority and scope allow.
Failed branch	A route that was tried and rejected.	`do_not_use` or `inspect_before_use`.
Contested memory	Evidence conflicts with newer or stronger evidence.	Inspect-first until resolved.
Stale memory	Previously useful history that may no longer apply.	Inspect-first or blocked.
Procedure memory	Reusable workflow evidence.	Use when scoped, stable, and not contradicted.
Archived memory	Cold history kept outside hot prompt context.	`rehydrate` on demand.

Plain summaries tend to collapse important distinctions:


"We tried path A, then moved to path B."

Aionis keeps the operational difference:


path A = failed branch / counter-evidence
path B = current active path / possible procedure

That difference is what prevents history from becoming accidental instruction.

Execution memory can be represented as an execution-state tree:

Layer	Contains
Raw evidence	Tool calls, attempts, validator results, handoffs, feedback.
Current active path	The branch Aionis believes should carry forward.
Failed branches	Branches retained as counter-evidence instead of primary routes.
Compressed procedures	Reusable workflow candidates derived from successful evidence.
Rehydrate pointers	Links back to raw evidence when compact context needs detail.

The Agent receives the useful state: current path, blocked routes, procedure candidates, and rehydrate hints.

State is always interpreted inside a scope:

Most unexpected “Aionis did not remember” issues are scope mismatch issues.