refactor(taxonomy): pin the CLI registry == schema both ways; drop dead instance_segmentation (#1005)

README.md

@@ -8,7 +8,7 @@ The customer-facing CLI for the tracebloc declarative ingestion path. Wraps the
 
 **v0.3.0 is released** — the latest stable [release](https://github.com/tracebloc/cli/releases/latest), cut from `develop`. It builds on v0.2.0's guided `data ingest` and `dataset rm` with a new `dataset list` command plus home-screen / output polish (clearer copy, guided-first framing). The binary implements `version`, `completion`, `data validate`, `cluster info`, and the full `data ingest` / `dataset list` / `dataset rm` flow — local schema validation, cluster discovery, data staging, submission, and Job watching, end to end.
 
-`data ingest` covers **9 of 10 task categories**: `image_classification`, `object_detection`, `keypoint_detection`, `text_classification`, `masked_language_modeling`, `tabular_classification`, `tabular_regression`, `time_series_forecasting`, and `time_to_event_prediction`. `semantic_segmentation` is pending mask-sidecar support upstream ([data-ingestors#136](https://github.com/tracebloc/data-ingestors/issues/136)); `instance_segmentation` is not yet implemented.
+`data ingest` covers **9 of 10 task categories**: `image_classification`, `object_detection`, `keypoint_detection`, `text_classification`, `masked_language_modeling`, `tabular_classification`, `tabular_regression`, `time_series_forecasting`, and `time_to_event_prediction`. `semantic_segmentation` is pending mask-sidecar support upstream ([data-ingestors#136](https://github.com/tracebloc/data-ingestors/issues/136)).
 
 The release pipeline ships [`v0.3.0`](https://github.com/tracebloc/cli/releases/latest) as **cosign-signed, multi-arch binaries** — Linux (`amd64`, `arm64`, `386`, `arm`), macOS (`amd64`, `arm64`), and Windows (`amd64`, `arm64`) — each with `SHA256SUMS` and the install scripts. Install via [Customer experience](#customer-experience) or [build from source](#building-from-source). (A Homebrew tap and the `install.tracebloc.io` vanity URL are later follow-ups; the GitHub release URL serves installs today.)
 
@@ -117,7 +117,7 @@ All v0.1 phases are merged:
 
 Beyond the original phases, `data ingest` was widened from image-classification-only to 9 of 10 modalities, and the test suite gained unit-coverage wins plus a kind-based integration harness for the real-I/O seams.
 
-**v0.2.0** added a friendlier guided `data ingest` and `dataset rm` on the home screen (#44, #47). **v0.3.0** added the `dataset list` command (#53) plus home-screen / output-spacing polish and feedback-copy refinements (#52, #56). **Next:** cloud-source ingestion (S3/GCS/HTTPS) for datasets above the 1 GiB local cap; `semantic_segmentation` ([data-ingestors#136](https://github.com/tracebloc/data-ingestors/issues/136)) and `instance_segmentation`. Smaller follow-ups are tracked as [open issues](https://github.com/tracebloc/cli/issues).
+**v0.2.0** added a friendlier guided `data ingest` and `dataset rm` on the home screen (#44, #47). **v0.3.0** added the `dataset list` command (#53) plus home-screen / output-spacing polish and feedback-copy refinements (#52, #56). **Next:** cloud-source ingestion (S3/GCS/HTTPS) for datasets above the 1 GiB local cap; `semantic_segmentation` ([data-ingestors#136](https://github.com/tracebloc/data-ingestors/issues/136)). Smaller follow-ups are tracked as [open issues](https://github.com/tracebloc/cli/issues).
 
 Epic: [tracebloc/client#147](https://github.com/tracebloc/client/issues/147).
 

internal/cli/data.go

@@ -449,8 +449,8 @@ other collaborators train against it without ever seeing the raw files.`))
 		// supported
 	case push.IsKnown(a.Spec.Category):
 		// A recognized category data ingest doesn't implement yet — image
-		// (semantic_segmentation / instance_segmentation) or text
-		// (causal_language_modeling). Routed here (not the default branch) so the
+		// (semantic_segmentation) or text (causal_language_modeling, seq2seq,
+		// …). Routed here (not the default branch) so the
 		// user gets the registry's per-category pending-support reason, not a
 		// misleading "unrecognized category". Supported categories were already
 		// caught above, so IsKnown here means known-but-unsupported.

internal/cli/data_test.go

@@ -94,8 +94,8 @@ func execDataIngest(t *testing.T, args []string) (exitCode int, stdout, stderr s
 func TestDataIngest_UnsupportedCategory_ExitsTwo(t *testing.T) {
 	root := imgcLayout(t)
 	for _, badCategory := range []string{
-		"semantic_segmentation",     // blocked on the ingestor (data-ingestors#136)
-		"instance_segmentation",     // not implemented
+		"semantic_segmentation",     // known but blocked on the ingestor (data-ingestors#136)
+		"instance_segmentation",     // dead — removed from the registry (#1005), now unrecognized
 		"definitely-not-a-category", // nonsense; gate catches this too
 	} {
 		t.Run(badCategory, func(t *testing.T) {

internal/cli/interactive.go

@@ -19,8 +19,8 @@ import (
 // category picker. It derives from the push registry's CLI-supported
 // set — the exact categories runDataIngest's gate accepts — so the
 // picker can't drift from what `data ingest` actually supports.
-// semantic_/instance_segmentation are excluded (CLISupported=false)
-// until they're implemented.
+// semantic_segmentation is excluded (CLISupported=false) until it's
+// implemented.
 var promptCategories = push.SupportedCategoryIDs()
 
 // prompter is the narrow seam over the interactive library. Production

internal/push/category.go

@@ -26,8 +26,8 @@ type CategorySpec struct {
 	// never ships to the central backend by default.
 	RegressionClass bool
 	// CLISupported reports whether `dataset push` implements the category
-	// today. semantic_/instance_segmentation are known (the schema
-	// defines them) but not yet pushable.
+	// today. semantic_segmentation is known (the schema defines it) but
+	// not yet pushable.
 	CLISupported bool
 	// UnsupportedNote explains why a known-but-unimplemented category
 	// isn't available yet; surfaced by the push gate. Empty when supported.
@@ -49,11 +49,15 @@ const (
 	FamilyText
 )
 
-// categoryRegistry is the ordered, authoritative list of every category
-// the ingest.v1 schema defines. Order is the display order for help text
-// and the interactive picker (CLI-supported first, in workflow order;
-// the not-yet-implemented ones last). Adding a category to the schema
-// means adding it here — the parity test pins the set.
+// categoryRegistry is the ordered list of every category the ingest.v1
+// schema defines — nothing more, nothing less. Order is the display order for
+// help text and the interactive picker (CLI-supported first, in workflow
+// order; the not-yet-implemented ones last). Adding a category to the schema
+// means adding it here; TestRegistryCoversSchemaCategories +
+// TestRegistryWithinSchema pin the set equal to the schema enum both ways, so
+// it can neither fall behind (a schema category rejected as "unrecognized")
+// nor carry an extra the ingestor won't accept (the instance_segmentation
+// half-ingest class — data-ingestors #240/#99, #1005).
 var categoryRegistry = []CategorySpec{
 	{ID: "image_classification", Family: FamilyImage, Label: "Image classification", CLISupported: true},
 	{ID: "object_detection", Family: FamilyImage, Label: "Object detection", CLISupported: true},
@@ -66,8 +70,6 @@ var categoryRegistry = []CategorySpec{
 	{ID: "time_to_event_prediction", Family: FamilyTabular, Label: "Time-to-event prediction", RegressionClass: true, CLISupported: true},
 	{ID: "semantic_segmentation", Family: FamilyImage, Label: "Semantic segmentation", CLISupported: false,
 		UnsupportedNote: "blocked on the ingestor's mask-sidecar support (data-ingestors#136)"},
-	{ID: "instance_segmentation", Family: FamilyImage, Label: "Instance segmentation", CLISupported: false,
-		UnsupportedNote: "not implemented"},
 	{ID: "causal_language_modeling", Family: FamilyText, Label: "Causal language modeling", CLISupported: false,
 		UnsupportedNote: "schema-recognized (data-ingestors#805); `tracebloc ingest` discover/build for its raw-.txt / prompt\\tcompletion `texts` layout is pending"},
 	{ID: "seq2seq", Family: FamilyText, Label: "Sequence-to-sequence", CLISupported: false,

internal/push/category_registry_test.go

@@ -15,7 +15,7 @@ import (
 func TestRegistryKnownCategories(t *testing.T) {
 	want := []string{
 		"image_classification", "object_detection", "keypoint_detection",
-		"semantic_segmentation", "instance_segmentation",
+		"semantic_segmentation",
 		"text_classification", "token_classification",
 		"masked_language_modeling", "causal_language_modeling", "seq2seq",
 		"sentence_pair_classification", "embeddings",
@@ -45,9 +45,9 @@ func TestSupportedCategories(t *testing.T) {
 			t.Errorf("SupportedCategoryIDs returned %q but IsCLISupported is false", id)
 		}
 	}
-	// segmentation + the self-supervised text categories (CLM, seq2seq) +
-	// token_classification are known but not yet pushable, and must explain why.
-	for _, id := range []string{"semantic_segmentation", "instance_segmentation", "causal_language_modeling", "seq2seq", "token_classification", "sentence_pair_classification", "embeddings"} {
+	// semantic_segmentation + the self-supervised text categories (CLM, seq2seq)
+	// + token_classification are known but not yet pushable, and must explain why.
+	for _, id := range []string{"semantic_segmentation", "causal_language_modeling", "seq2seq", "token_classification", "sentence_pair_classification", "embeddings"} {
 		if !IsKnown(id) {
 			t.Errorf("%s should be known", id)
 		}
@@ -87,16 +87,12 @@ func TestPredicatesDeriveFromRegistry(t *testing.T) {
 	}
 }
 
-// TestRegistryCoversSchemaCategories pins registry⇄schema parity: every
-// category the ingest schema accepts must be known to the registry, or a
-// schema-valid `dataset push --category=X` is wrongly rejected as
-// "unrecognized" (the token_classification drift, Bugbot v0.4.0 RC). The
-// existing tests only pin the registry against a hand-written list, which
-// stays internally consistent while drifting from the schema — this closes
-// that gap. The reverse direction isn't required: the registry may carry a
-// known-but-unsupported alias the v1 schema doesn't list yet (e.g.
-// instance_segmentation), which is gated out before schema validation.
-func TestRegistryCoversSchemaCategories(t *testing.T) {
+// schemaCategoryEnum returns the category enum from the embedded ingest.v1
+// schema — the single source of truth the registry is pinned against (#1005).
+// The schema is vendored + drift-checked against data-ingestors by
+// scripts/sync-schema.sh, so this ties the registry transitively to upstream.
+func schemaCategoryEnum(t *testing.T) []string {
+	t.Helper()
 	var doc struct {
 		Properties struct {
 			Category struct {
@@ -110,14 +106,52 @@ func TestRegistryCoversSchemaCategories(t *testing.T) {
 	if len(doc.Properties.Category.Enum) == 0 {
 		t.Fatal("no category enum found in the embedded schema (parse path wrong?)")
 	}
-	for _, id := range doc.Properties.Category.Enum {
+	return doc.Properties.Category.Enum
+}
+
+// registryAliases are registry category IDs deliberately NOT in the ingest.v1
+// schema enum — declared placeholders. Empty today: instance_segmentation used
+// to sit here unchecked, but it's dead (it half-ingested with no validators or
+// file transfer — data-ingestors #240/#99) and was removed, not kept. A future
+// known-but-unschema'd placeholder must be DECLARED here, so TestRegistryWithinSchema
+// flags undeclared drift while allowing an intentional superset (#1005).
+var registryAliases = map[string]bool{}
+
+// TestRegistryCoversSchemaCategories pins schema ⊆ registry: every category the
+// ingest schema accepts must be known to the registry, or a schema-valid
+// `dataset push --category=X` is wrongly rejected as "unrecognized" (the
+// token_classification drift, Bugbot v0.4.0 RC).
+func TestRegistryCoversSchemaCategories(t *testing.T) {
+	for _, id := range schemaCategoryEnum(t) {
 		if !IsKnown(id) {
 			t.Errorf("schema category %q missing from the registry — `dataset push --category=%s` "+
 				"would be rejected as unrecognized despite passing schema validation", id, id)
 		}
 	}
 }
 
+// TestRegistryWithinSchema pins registry ⊆ schema (+ declared aliases): the
+// registry must not carry a category the ingest schema — and therefore the
+// ingestor — doesn't accept. An undeclared extra is exactly the
+// instance_segmentation half-ingest class: the backend/CLI would accept a
+// `--category` the pipeline can't handle, and the config half-ingests (DB rows
+// + API records, zero files staged; #1005, data-ingestors #240/#99). Together
+// with TestRegistryCoversSchemaCategories this pins registry == schema, modulo
+// explicitly declared placeholders in registryAliases.
+func TestRegistryWithinSchema(t *testing.T) {
+	inSchema := make(map[string]bool)
+	for _, id := range schemaCategoryEnum(t) {
+		inSchema[id] = true
+	}
+	for _, id := range AllCategoryIDs() {
+		if !inSchema[id] && !registryAliases[id] {
+			t.Errorf("registry category %q is not in the ingest.v1 schema enum and not a declared "+
+				"alias — add it to the schema (data-ingestors) if it's real, or declare it in "+
+				"registryAliases if it's an intentional placeholder", id)
+		}
+	}
+}
+
 func equalSet(a, b []string) bool {
 	if len(a) != len(b) {
 		return false