Expand description
Observation → fixed NQuin-compatible records (epistemic, not ground truth).
Layout matches Qualia NQuin (6 × u64). We keep a local definition so
Phase-1 does not force a full qualia-core-db link for every UI binary;
field meanings are identical and can be cast/transmuted at the core boundary.
V6: human reject/correct, bbox/track predicates, full compile (up to 4 quins/det).
Structs§
- Media
Digest - Content digest of media bytes (FNV-1a over up to 64 KiB prefix + length).
- Vision
Quin - 48-byte semantic quin (parity = XOR fold of other fields).
Constants§
- CTX_
HUMAN_ ATTESTATION - CTX_
VISION - MAX_
OBS_ QUINS - Max observation quins emitted per batch (up to 4 per detection in full compile).
- P_
HAS_ BBOX - P_
HAS_ TRACK - P_
HUMAN_ CORRECTS - P_
HUMAN_ REJECTS - P_
MODEL_ DIGEST - P_
PROPOSES_ CLASS - P_
VISUAL_ OBSERVATION - Predicate / context IRIs (hashed with FNV-1a 64, same family as Qualia
q_hash).
Functions§
- bbox_
quin - instance –hasBoundingBox–> packed box (metadata: score).
- class_
proposal_ quin - Class proposal quin: instance –proposesClass–> class_hash.
- compile_
observation_ quins - Compile detections into observation + class quins (epistemic layer).
- compile_
observation_ quins_ full - Full epistemic compile: model digest + per-det (observation, class, bbox, track). Machine claims are never erased by later human reject/correct (separate context).
- format_
detection_ text_ description - Detailed human-readable semantic description of an object detection.
- format_
detection_ turtle_ rdf - RDF Turtle representation of a detected object instance.
- human_
correct_ quin - Human corrects class: subject = human, object = new_class_hash; metadata low 64 bits of original instance for linkage (instance in context xor).
- human_
reject_ quin - Human rejects a machine observation instance (does not erase the machine claim). subject = human_did_hash, object = instance_hash that is rejected.
- media_
digest - Media digest.
- model_
digest_ quin - media –modelDigest–> model_hash (provenance of which model produced claims).
- observation_
quin - One observation quin: media –VisualObservation–> instance, score in metadata low 16.
- pack_
bbox_ u64 - Pack normalized u16 bbox into a single u64 (x0|y0|x1|y1 as 16-bit lanes).
- q_hash
- FNV-1a 64 — matches Qualia URI hashing family.
- q_
hash_ bytes - q_hash of bytes.
- query_
by_ frame_ range - Filter observation quins by media time / frame range (frame in observation metadata bits 16..48).
- query_
by_ model - Filter by model hash mixed into context (context = CTX_VISION ^ model_hash).
- query_
instances_ in_ region - Region query over compiled observation quins: keep instances whose bbox intersects the query box (normalized u16). Writes matching instance hashes.
- track_
quin - instance –hasTrackId–> track_id (0 if untracked).
- unpack_
bbox_ u64 - Unpacks a u64 bbox.