Reconstruct operational episodes instead of trusting mutable rows

Operational support tables often store the latest state of a reusable issue row while immutable events contain several distinct episodes. When rows can reopen or mutate, sample them reproducibly, reconstruct each detection episode from immutable events, attribute evidence only within bounded episode windows, and keep the initiating cause separate from the eventual pathway or outcome.

Centralize system-actor identity as a second guardrail. Multiple sentinel identities can make automated events appear human and corrupt labor or automation metrics. Missing browser or conversation evidence remains unknown, and timing correlation must not be converted into causality.