Observations

What an observation carries

An observation targets an entity and carries fields, timing, schema context, and provenance. Its source may be a supplied file or another recorded input. A structured agent write or correction can have no separate source file, so the absence of a document link is not by itself evidence that the observation is invalid. The observation's attribution and write context still matter.

Fields are interpreted under a declared entity schema. A submitted key that the schema does not declare may be retained as a raw fragment rather than appearing in the entity's current fields. After a write, retrieve the entity and verify the exact field, not merely the response status.

An omitted field says less than an explicit value. For example, omitting a status from a new observation leaves the previous status claim available for the reducer; submitting a concrete status creates another claim to consider. Null, zero and an empty string also have their own meanings under the schema and should not be used as a generic spelling for "unknown."

Corrections add evidence

A correction creates another observation. The earlier observation remains inspectable, which lets a reader understand why the current value differs from the first recorded value. Corrections receive a distinct priority in the configured reduction path, but the effective result depends on the field's merge policy. Do not assume the same winner rule for every field.

For a field whose policy is last_write, the later observed timestamp decides. For a field configured as highest_priority, the source priority and then defined tie breakers decide. Array merge policies have their own contribution rules. Read the policy for the entity type before predicting the winner.

Reinterpretation and history

Reinterpreting a source creates new observations with its own configuration context; it does not silently alter the older interpretation. AI-assisted extraction may produce a different result on a later run, so inspect the new observation and its configuration. Deterministic snapshot computation begins after the observation set and schema are fixed.

The observation architecture describes the lifecycle. Conflict resolution defines the field policies. Continue to Current snapshots to see how a present answer is computed.