1 targets.protocol
targets.protocol
Target-input protocol and training-admission contract.
This leaf keeps protocol spelling, target-source ownership, and descriptor provenance validation together. Persisted rollout readers may display any raw legacy value; callers cross this interface only when admitting new generation or training data.
1.1 Attributes
| Name | Description |
|---|---|
| ORACLE_GT_TARGET_SOURCE | Canonical source label for descriptors constructed from Oracle GT OBBs. |
| STRICT_GT_IOU_THRESHOLD | Strict lower bound for an unambiguous observed-target GT match. |
1.2 Classes
| Name | Description |
|---|---|
| ActorVisibleTargetSource | Canonical persisted source names admitted for observed targets. |
| TargetDescriptorProvenance | How the descriptor supplied to the actor was constructed. |
| TargetInputProtocol | Canonical target-input protocols admitted by new configs and corpora. |
| TargetLabelEvidence | Persisted target facts used to admit oracle labels for training. |
1.2.1 ActorVisibleTargetSource
targets.protocol.ActorVisibleTargetSource()Canonical persisted source names admitted for observed targets.
1.2.1.1 Attributes
| Name | Description |
|---|---|
| DETECTED_OBBS | Actor-visible OBB detections produced by the VIN source store. |
1.2.2 TargetDescriptorProvenance
targets.protocol.TargetDescriptorProvenance()How the descriptor supplied to the actor was constructed.
1.2.2.1 Attributes
| Name | Description |
|---|---|
| ORACLE_GT | Descriptor constructed from privileged GT target geometry. |
| ACTOR_VISIBLE_DETECTOR | Descriptor constructed from actor-visible detector output. |
| ACTOR_VISIBLE_PREDICTOR | Descriptor constructed from actor-visible predicted target state. |
1.2.3 TargetInputProtocol
targets.protocol.TargetInputProtocol()Canonical target-input protocols admitted by new configs and corpora.
1.2.3.1 Attributes
| Name | Description |
|---|---|
| V0_GT_INPUT | GT-derived target input for non-deployable Oracle experiments. |
| V1_OBSERVED | Actor-visible target input produced by an observed detector or predictor. |
| is_actor_visible_target_protocol | Whether the protocol requires actor-visible target evidence. |
1.2.4 TargetLabelEvidence
targets.protocol.TargetLabelEvidence(
protocol,
target_source,
gt_match_status,
matched_gt_target_row_id,
matched_gt_target_id,
gt_match_iou,
target_valid,
descriptor_source=None,
descriptor_provenance=None,
descriptor_hash=None,
explicit_target_hash=None,
)Persisted target facts used to admit oracle labels for training.
The mapping deliberately keeps target-task validity separate from store provenance. A caller must validate the latter independently before using this predicate for a training view.
1.3 Functions
| Name | Description |
|---|---|
| target_label_is_trainable | Return whether one target’s GT evidence is trainable. |
| validate_target_protocol_admission | Validate one target descriptor before generation or training admission. |
1.3.1 target_label_is_trainable
targets.protocol.target_label_is_trainable(evidence)Return whether one target’s GT evidence is trainable.
V0 retains its historical row-based admission. V1 requires the actor visible protocol, an exact admitted match with both stable identifiers, and a finite oriented IoU strictly above the campaign threshold. Malformed evidence is rejected rather than interpreted as a weak label.
1.3.2 validate_target_protocol_admission
targets.protocol.validate_target_protocol_admission(
protocol,
*,
target_source,
descriptor_source=None,
descriptor_provenance=None,
)Validate one target descriptor before generation or training admission.
1.3.2.1 Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| protocol | TargetInputProtocol | str | Raw persisted value or canonical protocol. | required |
| target_source | str | None | Source that selected the target task. | required |
| descriptor_source | str | None | Source block that constructed the actor descriptor. V0 may omit this because its Oracle source is fixed; V1 must name it. | None |
| descriptor_provenance | TargetDescriptorProvenance | str | None | Construction class for the actor descriptor. V1 accepts only a provenance registered for target_source. |
None |
1.3.2.2 Returns
| Name | Type | Description |
|---|---|---|
| TargetInputProtocol | The canonical admitted protocol. |
1.3.2.3 Notes
Rejection never mutates persisted artifacts. Legacy and unknown strings remain available to audit readers but require corpus regeneration before use in training.