Observed outcomes require affirmative evidence

Decision

The observation days are a minimum settlement delay after the latest write. Only terminal runs qualify. The result is observed at derivation time, not at a fixed seven-day closing boundary. The outcome row timestamps identify that observation; scheduler delay can intentionally change the evidence.

An old retained history row cannot prove continuous logging or exclude selective deletion. With no affirmative later activity, derive UNKNOWN. ACCEPTED_UNCHANGED remains readable for historical data but has no automatic writer in this protocol. EDITED means subsequent recorded activity; neither human authorship nor modification of the generated fields is established. Only confirmed absence or deletion establishes DISCARDED. Missing extension tables count as absent; transient database exceptions fail for retry.

Replace outcomes at grain (run UUID, source) under a TYPO3 exclusive key lock and a database transaction. Last successful replacement wins, rollback preserves the former value, and the other source remains independent. All writers use the repository. Multi-node installations need a shared lock strategy; default host-local locks do not coordinate separate hosts.

Evidence

ObservedOutcomeDeriverTest and WrittenRecordRepositoryTest check uncertainty and terminal-run eligibility. WrittenRecordReadFailureTest checks transient storage failure. CallOutcomeRepositoryTest injects a real SQLite insertion failure and overlaps independent processes at the locking boundary.