refactor(P3d): source data_format from the ModalityRegistry

tests/test_modality_registry.py

@@ -85,3 +85,24 @@ def test_base_py_imports_the_registry_derived_sets():
     assert base._FILE_BEARING_CATEGORIES is registry.FILE_BEARING_CATEGORIES
     assert base._TABULAR_FAMILY_CATEGORIES is registry.TABULAR_FAMILY_CATEGORIES
     assert base._SELF_SUPERVISED_CATEGORIES is registry.SELF_SUPERVISED_CATEGORIES
+
+
+def test_data_format_valid_and_matches_conventions():
+    """Every spec carries a valid DataFormat (P3d), and conventions'
+    _data_format_for reads it from the registry (single source)."""
+    from tracebloc_ingestor.cli.conventions import _data_format_for
+    from tracebloc_ingestor.utils.constants import DataFormat
+
+    valid = set(DataFormat.get_all_formats())
+    for category, spec in REGISTRY.items():
+        assert (
+            spec.data_format in valid
+        ), f"{category}: bad data_format {spec.data_format!r}"
+        assert _data_format_for(category) == spec.data_format
+
+
+def test_transfer_present_iff_file_bearing():
+    """The sidecar transfer factory is set exactly for file-bearing categories
+    (P3c invariant)."""
+    for category, spec in REGISTRY.items():
+        assert spec.is_file_bearing == (spec.transfer is not None)

tracebloc_ingestor/cli/conventions.py

@@ -27,42 +27,54 @@
 from dataclasses import dataclass, field
 from typing import Any, Dict, FrozenSet, List, Optional
 
-from ..utils.constants import DataFormat, Intent, TaskCategory
-
+from ..modalities.registry import spec_for
+from ..utils.constants import TaskCategory
 
 # ---------------------------------------------------------------------------
 # Category groupings — used both here and by the entrypoint when deciding
 # which sidecar paths matter. Single source of truth.
 # ---------------------------------------------------------------------------
 
-IMAGE_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.IMAGE_CLASSIFICATION,
-    TaskCategory.OBJECT_DETECTION,
-    TaskCategory.KEYPOINT_DETECTION,
-    TaskCategory.SEMANTIC_SEGMENTATION,
-})
+IMAGE_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.IMAGE_CLASSIFICATION,
+        TaskCategory.OBJECT_DETECTION,
+        TaskCategory.KEYPOINT_DETECTION,
+        TaskCategory.SEMANTIC_SEGMENTATION,
+    }
+)
 
-TEXT_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.TEXT_CLASSIFICATION,
-    TaskCategory.TOKEN_CLASSIFICATION,
-})
+TEXT_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.TEXT_CLASSIFICATION,
+        TaskCategory.TOKEN_CLASSIFICATION,
+    }
+)
 
-TABULAR_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.TABULAR_CLASSIFICATION,
-    TaskCategory.TABULAR_REGRESSION,
-})
+TABULAR_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.TABULAR_CLASSIFICATION,
+        TaskCategory.TABULAR_REGRESSION,
+    }
+)
 
-TIME_SERIES_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.TIME_SERIES_FORECASTING,
-})
+TIME_SERIES_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.TIME_SERIES_FORECASTING,
+    }
+)
 
-TIME_TO_EVENT_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.TIME_TO_EVENT_PREDICTION,
-})
+TIME_TO_EVENT_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.TIME_TO_EVENT_PREDICTION,
+    }
+)
 
-MLM_CATEGORIES: FrozenSet[str] = frozenset({
-    TaskCategory.MASKED_LANGUAGE_MODELING,
-})
+MLM_CATEGORIES: FrozenSet[str] = frozenset(
+    {
+        TaskCategory.MASKED_LANGUAGE_MODELING,
+    }
+)
 
 # Categories where the label is a numeric prediction target rather than
 # class metadata. The schema requires `label.policy` for these so the raw
@@ -101,12 +113,18 @@
     # framework's own samples round-trip through the documented happy-path
     # config with zero overrides. Production users with differently-sized data
     # should set `spec.file_options` in their YAML.
-    TaskCategory.IMAGE_CLASSIFICATION:    {"target_size": [256, 256], "extension": ".jpeg"},
-    TaskCategory.SEMANTIC_SEGMENTATION:   {"target_size": [512, 512], "extension": ".jpg"},
-    TaskCategory.OBJECT_DETECTION:        {"target_size": [1920, 1080], "extension": ".jpg"},
+    TaskCategory.IMAGE_CLASSIFICATION: {
+        "target_size": [256, 256],
+        "extension": ".jpeg",
+    },
+    TaskCategory.SEMANTIC_SEGMENTATION: {
+        "target_size": [512, 512],
+        "extension": ".jpg",
+    },
+    TaskCategory.OBJECT_DETECTION: {"target_size": [1920, 1080], "extension": ".jpg"},
     # keypoint_detection: no target_size default — the customer's pose model
     # dictates input resolution, so the schema requires it top-level.
-    TaskCategory.KEYPOINT_DETECTION:      {"extension": ".jpg"},
+    TaskCategory.KEYPOINT_DETECTION: {"extension": ".jpg"},
 }
 
 DEFAULT_TEXT_FILE_OPTIONS: Dict[str, Any] = {
@@ -122,6 +140,7 @@
 # Resolved configuration — what the entrypoint actually consumes.
 # ---------------------------------------------------------------------------
 
+
 @dataclass
 class ResolvedConfig:
     """A fully-resolved ingest configuration.
@@ -183,6 +202,7 @@ class ResolvedConfig:
 # Resolver
 # ---------------------------------------------------------------------------
 
+
 def resolve(config: Dict[str, Any]) -> ResolvedConfig:
     """Translate a validated ingest.yaml dict into a :class:`ResolvedConfig`.
 
@@ -302,25 +322,16 @@ def resolve(config: Dict[str, Any]) -> ResolvedConfig:
 # Helpers
 # ---------------------------------------------------------------------------
 
+
 def _data_format_for(category: str) -> str:
-    """Map ``category`` to the ``DataFormat`` value the framework expects."""
-    if category in IMAGE_CATEGORIES:
-        return DataFormat.IMAGE
-    if category in TEXT_CATEGORIES:
-        return DataFormat.TEXT
-    if (
-        category in TABULAR_CATEGORIES
-        or category in TIME_SERIES_CATEGORIES
-        or category in TIME_TO_EVENT_CATEGORIES
-    ):
-        return DataFormat.TABULAR
-    if category in MLM_CATEGORIES:
-        return DataFormat.TEXT
-    raise ValueError(
-        f"Unknown category {category!r}; cannot derive data_format. "
-        "If this is a new category, add it to the relevant CATEGORY set "
-        "in conventions.py and to the schema enum."
-    )
+    """Map ``category`` to the ``DataFormat`` value the framework expects.
+
+    Reads the single source of truth — the ModalityRegistry — rather than the
+    per-format frozensets in this module (structural refactor backend#796,
+    P3d). ``spec_for`` raises ``ValueError`` on an unknown category, same as
+    the previous ladder did.
+    """
+    return spec_for(category).data_format
 
 
 def _default_file_options_for(category: str) -> Dict[str, Any]:

tracebloc_ingestor/modalities/registry.py

@@ -12,22 +12,24 @@
 
 from typing import Dict
 
-from ..utils.constants import TaskCategory
+from ..utils.constants import DataFormat, TaskCategory
 from . import transfer as t
 from . import validators as v
 from .spec import ModalitySpec
 
 # One entry per supported category — the single source of truth. P3a: the three
 # behavior flags (was three frozensets in ingestors/base.py). P3b: the
 # validator factory (was the map_validators if/elif arm). P3c: the sidecar
-# transfer factory (was the map_file_transfer if/elif arm).
+# transfer factory (was the map_file_transfer if/elif arm). P3d: data_format
+# (was the 6-frozenset ladder in conventions._data_format_for).
 _SPECS = (
     # File-bearing categories (per-row sidecar files under SRC_PATH).
     ModalitySpec(
         TaskCategory.IMAGE_CLASSIFICATION,
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.IMAGE,
         build_validators=v.image_classification,
         transfer=t.image_classification,
     ),
@@ -36,6 +38,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.IMAGE,
         build_validators=v.object_detection,
         transfer=t.object_detection,
     ),
@@ -44,6 +47,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.IMAGE,
         build_validators=v.keypoint_detection,
         transfer=t.keypoint_detection,
     ),
@@ -52,6 +56,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.IMAGE,
         build_validators=v.semantic_segmentation,
         transfer=t.semantic_segmentation,
     ),
@@ -60,6 +65,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.TEXT,
         build_validators=v.text_classification,
         transfer=t.text_classification,
     ),
@@ -68,6 +74,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=False,
+        data_format=DataFormat.TEXT,
         build_validators=v.token_classification,
         transfer=t.token_classification,
     ),
@@ -77,6 +84,7 @@
         is_file_bearing=True,
         is_tabular_family=False,
         is_self_supervised=True,
+        data_format=DataFormat.TEXT,
         build_validators=v.masked_language_modeling,
         transfer=t.masked_language_modeling,
     ),
@@ -86,27 +94,31 @@
         is_file_bearing=False,
         is_tabular_family=True,
         is_self_supervised=False,
+        data_format=DataFormat.TABULAR,
         build_validators=v.tabular_classification,
     ),
     ModalitySpec(
         TaskCategory.TABULAR_REGRESSION,
         is_file_bearing=False,
         is_tabular_family=True,
         is_self_supervised=False,
+        data_format=DataFormat.TABULAR,
         build_validators=v.tabular_regression,
     ),
     ModalitySpec(
         TaskCategory.TIME_SERIES_FORECASTING,
         is_file_bearing=False,
         is_tabular_family=True,
         is_self_supervised=False,
+        data_format=DataFormat.TABULAR,
         build_validators=v.time_series_forecasting,
     ),
     ModalitySpec(
         TaskCategory.TIME_TO_EVENT_PREDICTION,
         is_file_bearing=False,
         is_tabular_family=True,
         is_self_supervised=False,
+        data_format=DataFormat.TABULAR,
         build_validators=v.time_to_event_prediction,
     ),
 )

tracebloc_ingestor/modalities/spec.py

@@ -34,6 +34,9 @@ class ModalitySpec:
 
     Attributes:
         category: the ``TaskCategory`` value this spec describes.
+        data_format: the ``DataFormat`` the framework expects for this category
+            (P3d). Read by ``conventions._data_format_for`` (was a 6-frozenset
+            ladder there).
         build_validators: ``(file_options) -> [validators]`` — the validator
             set this category runs (P3b). Replaces the corresponding
             ``map_validators`` if/elif arm; the factory bodies live in
@@ -61,6 +64,7 @@ class ModalitySpec:
     is_file_bearing: bool
     is_tabular_family: bool
     is_self_supervised: bool
+    data_format: str
     build_validators: Callable[[Dict[str, Any]], List]
     transfer: Optional[
         Callable[[Dict[str, Any], Dict[str, Any]], Optional[Dict[str, Any]]]