Event Taxonomy
Good memory starts with clean event semantics. Keep event names stable and meaningful across your product.
user_questionassistant_responselearning_progressassessment_resultuser_attemptpreference_statedInferred memory types
Orbit can also create inferred memories when evidence is strong enough.
inferred_learning_patternGenerated from repeated semantically similar behavior for the same entity. Example: a learner repeatedly asks loop-control questions.
inferred_preferenceGenerated from sustained feedback trends. Example: concise answers consistently receive higher helpfulness scores.
inferred_user_factGenerated from explicit natural-language statements. Example: user constraint or family preference extracted into structured fact metadata.
inferred_user_fact_conflictGenerated when critical fact statements conflict. Orbit marks these as clarification-required before relying on them.
Taxonomy rule of thumb
If two events represent meaningfully different behaviors, give them different event types. If not, keep one type and use metadata fields for detail.