Module vision_ingest 

Multimodal image ingest — binds assets to active mmproj + WAL (no Ollama).

V6: also accepts native detector epistemic observation quins (no mmproj required) via append_native_observation_quins — machine claims with model digest + optional human reject/correct edges. Dense pixels never enter NQuins.

Structs

NativeDetection

Fixed-layout detection for WAL packing (mirrors qualia_vision::Detection fields).

VisionIngestResult

Enums

VisionError

Functions

append_human_attestation

Append a human reject/correct edge (provenance preserved).

append_native_observation_quins

Append native observation quins to vision_native.wal (no active VLM required).

compile_native_observation_quins

Compile native detections into epistemic NQuins (caller buffer). Writes model digest + per-det observation/class/bbox/track. Returns count.

human_correct_quin

Human class correction — coexists with machine proposesClass.

human_reject_quin

Human reject of a machine instance — does not erase machine quins.

ingest_image_file

ingest_image_with_active_record

resolve_active_multimodal