Evaluation Data Flow and the PII Boundary

Evaluation Data Flow and the PII Boundary A data-flow diagram generated by Archify. 01 / Caller 02 / Ingress 03 / Snapshot 04 / Decide 05 / Outcome Caller Context · user_id + attributes · 01 / Caller · may be PII Caller Context user_id + attributes may be PII API Key Gate · key + project pin · 02 / Ingress · read-only API Key Gate key + project pin read-only Redis Snapshot · version-matched · 03 / Snapshot · no PII Redis Snapshot version-matched no PII Snapshot Service · resolve or rebuild · 03 / Snapshot · no PII Snapshot Service resolve or rebuild no PII Evaluator · pure function · 04 / Decide · no I/O Evaluator pure function no I/O Flag Decision · value + reason · 05 / Outcome · no attributes Flag Decision value + reason no attributes key + context attributes cross the wire authorized read attributes not retained write-through flag config only version check hit or miss flags + rules no PII value + reason attributes excluded Legend primary data policy / PII async batch data store data flow

Where the attributes stop

  • • The caller's user_id and attributes travel to the API, are matched against rule conditions, and are then written nowhere.
  • • Nothing between the gate and the evaluator persists a caller attribute, so a warm read never handles them at all.

What is actually stored

  • • A snapshot holds keys, default values, kill-switch state, rollout percentages, and rule conditions, and nothing else.
  • • It is tenant-scoped by the Redis key, so one project can never read another project's snapshot.