This guide connects autonomic_typesafe output to the PostgreSQL durability model.
Three different durable things
Do not collapse these into one "AI decision" record:
- Observation evidence — what deterministic/resource/semantic sensors observed.
- Trajectory state — what Homeostat derived over time from observation history.
- Effect decision — whether a specific effect revision satisfied a required semantic decision floor.
They answer different audit questions.
Observation frames
The TypeSafe adapter returns normalized Autonomic.SemanticObservation structs. They are placed on Autonomic.ObservationFrame alongside deterministic facts, resources, effect context, and sequence/epoch identity.
The observation provenance makes later analysis possible:
Which model produced this answer?
Which TypeSafe request produced it?
Which prepared semantic contract was active?
What confidence/distribution was returned?
Which bank version interpreted it?
How long did it take and what usage was reported?Trajectory persistence
Autonomic.Homeostat consumes semantic observations temporally. It updates state such as drift, uncertainty, authority pressure, destructive pressure, autonomy balance, and the current regime.
Persisting trajectory state means a BEAM process restart does not imply semantic amnesia or a free reset to stable.
Effect-decision persistence
When EffectBroker requires semantic authorization, the decision record is tied to:
effect ID
exact effect revision
episode epoch
policy version
trajectory version
payload digest
semantic evidence digest/referenceThat is materially different from storing "TypeSafe said yes." It records that a particular semantic evaluation contributed to authorization of a particular proposal under a particular authority generation.
Separation of concerns
autonomic_typesafe owns semantic measurement.
autonomic owns interpretation and authority policy.
autonomic_postgres owns durability and concurrency/fencing.
Keeping those responsibilities separate prevents the database adapter from becoming a hidden second semantic policy engine.