TAB_AIPORTRAIT_FEEDBACK Class
Decision journal of the AI portrait, append-only: one row per item proposed (written by the ingestion), applied, rejected, modified or refused by the platform. Survives every purge (training corpus and proof of human validation): deleting the entity only clears PARENT_TYPE / PARENT_ID.
FEEDBACK_CONFIDENCE
Decimal(3,2)Confidence declared by the model at extraction (0.00 - 1.00), never displayed
FEEDBACK_DECISION
Tinyint(1)Decision: 0 proposed · 1 applied · 2 rejected · 3 modified before writing · 4 refused by the platform.
Default: 0
FEEDBACK_ITEMKEY
Char(64)Key of the list element (sha256 of the normalized element), empty for a scalar field
Default: ''
FEEDBACK_REASON
Varchar(64)Technical reason when refused by the platform (rights, validation, referential, already filled)
FEEDBACK_VALUE
JSONProposed value (API format). Cleared when the entity or the source action is deleted: the hash is enough to filter rejections.