Test combat semantics correction flow

docs/roadmap/implementation.md

@@ -308,6 +308,8 @@ This stage is small enough for one implementation prompt.
 
 ## Stage 5: Prove Feedback-Aware Producer Correction
 
+✅ Complete
+
 ### Goal
 
 Prove the assembled D&D path can reject a well-formed wrong scene kind, guide

internal/modules/integration/dnd_npc_grounded_test.go

@@ -28,6 +28,7 @@ import (
 	sceneextract "gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/extract/scenedescriptions"
 	"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/extract/spells"
 	npcnormalize "gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/normalize/npcregistry"
+	combatsemantics "gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/validate/scenedescriptions/combat_semantics"
 )
 
 func TestNPCOutputGroundsSpellAndCombatConsumersThroughOneOperation(t *testing.T) {
@@ -258,6 +259,70 @@ func TestNPCOutputGroundsSpellAndCombatConsumersThroughOneOperation(t *testing.T
 	assertCurrentEvidence(t, combatValue.CombatTurns[0].SourceRefs)
 }
 
+func TestSceneCombatSemanticsCorrectionRetriesAndRevalidates(t *testing.T) {
+	registries := productionNPCRegistries(t)
+	configValue := loadGroundedPipelineConfig(t)
+	profile := configValue.Pipelines["dnd-npc-grounded"]
+	sceneLane := profile.Steps[0].Artifacts["scene-descriptions"]
+	sceneLane.Extract.Retries = 1
+	sceneLane.Extract.Validators = pipeline.ValidatorOverride{Set: true, Validators: []pipeline.ModuleBinding{
+		pipeline.Binding("generic/valid_json"),
+		pipeline.Binding("extract/dnd/scene-descriptions/shape"),
+		pipeline.Binding("extract/dnd/scene-descriptions/source_refs"),
+		pipeline.Binding("generic/valid_json_schema"),
+		pipeline.Binding("extract/dnd/scene-descriptions/source_relatedness"),
+		pipeline.Binding(combatsemantics.Key),
+	}}
+	profile.Steps[0].Artifacts["scene-descriptions"] = sceneLane
+	configValue.Pipelines["dnd-npc-grounded"] = profile
+
+	client := &groundedDNDLLMClient{sceneKinds: []dnd.SceneKind{dnd.SceneKindNarrative, dnd.SceneKindCombat}, combatSemanticsVerdicts: []string{"combat_should_be_added", "approved"}}
+	output := runGroundedPipeline(t, configValue, registries, client, nil)
+	if len(output.Rejected) != 0 {
+		t.Fatalf("rejected = %#v, want corrected acceptance", output.Rejected)
+	}
+	for _, group := range output.Diagnostics.Groups {
+		if group.Origin.ValidatorKey == combatsemantics.Key {
+			t.Fatalf("combat-semantics validator left a diagnostic after correction: %#v", group)
+		}
+	}
+	var scenes dnd.SceneDescriptionList
+	for _, serialized := range output.NormalizeOutputs {
+		if serialized.LaneID == "scene-descriptions" {
+			var err error
+			scenes, err = scenecodec.New().Decode(serialized.Artifact.Content)
+			if err != nil {
+				t.Fatal(err)
+			}
+		}
+	}
+	if len(scenes.Scenes) != 1 || scenes.Scenes[0].Kind != dnd.SceneKindCombat {
+		t.Fatalf("accepted scenes = %#v, want corrected combat scene", scenes)
+	}
+	requests := client.requestsSnapshot()
+	var sceneRequests, validatorRequests []contracts.StructuredCompletionRequest
+	for _, request := range requests {
+		switch request.PromptID {
+		case sceneextract.PromptID:
+			sceneRequests = append(sceneRequests, request)
+		case combatsemantics.PromptID:
+			validatorRequests = append(validatorRequests, request)
+		}
+	}
+	if len(sceneRequests) != 2 || len(validatorRequests) != 2 || sceneRequests[0].Correction != nil || sceneRequests[1].Correction == nil {
+		t.Fatalf("scene/validator requests = %d/%d; want correction and revalidation", len(sceneRequests), len(validatorRequests))
+	}
+	correction := sceneRequests[1].Correction
+	if !strings.Contains(string(correction.AssistantResponse), `"kind":"narrative"`) || !strings.Contains(correction.UserGuidance, "Return kind: combat") || !strings.Contains(correction.UserGuidance, "active encounter") {
+		t.Fatalf("correction = %#v, want latest candidate and useful guidance", correction)
+	}
+	for _, forbidden := range []string{combatsemantics.Key, combatsemantics.ReasonCodeActiveCombatNotClassified, "session:chunk"} {
+		if strings.Contains(correction.UserGuidance, forbidden) {
+			t.Fatalf("correction guidance leaked internal detail %q: %q", forbidden, correction.UserGuidance)
+		}
+	}
+}
+
 func TestProductionDNDOutputPublishesSelectedEvidenceContext(t *testing.T) {
 	registries := productionNPCRegistries(t)
 	configValue := loadGroundedPipelineConfig(t)
@@ -514,6 +579,10 @@ type groundedDNDLLMClient struct {
 	sceneTitle              string
 	firstUnitID             int
 	thirdUnitID             int
+	sceneKinds              []dnd.SceneKind
+	combatSemanticsVerdicts []string
+	sceneCalls              int
+	combatSemanticsCalls    int
 }
 
 func (client *groundedDNDLLMClient) CompleteStructured(ctx context.Context, request contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
@@ -547,6 +616,10 @@ func (client *groundedDNDLLMClient) CompleteStructured(ctx context.Context, requ
 		payload = map[string]any{"duplicate_groups": []any{}}
 	case sceneextract.PromptID:
 		kind := client.sceneKind
+		if client.sceneCalls < len(client.sceneKinds) {
+			kind = client.sceneKinds[client.sceneCalls]
+		}
+		client.sceneCalls++
 		if kind == "" {
 			kind = dnd.SceneKindCombat
 		}
@@ -555,6 +628,13 @@ func (client *groundedDNDLLMClient) CompleteStructured(ctx context.Context, requ
 			title = "A combat encounter"
 		}
 		payload = map[string]any{"kind": kind, "title": title, "summary": "The party faces an active encounter."}
+	case combatsemantics.PromptID:
+		verdict := "approved"
+		if client.combatSemanticsCalls < len(client.combatSemanticsVerdicts) {
+			verdict = client.combatSemanticsVerdicts[client.combatSemanticsCalls]
+		}
+		client.combatSemanticsCalls++
+		payload = map[string]any{"verdict": verdict, "explanation": "The transcript shows an active encounter with hostile action."}
 	case spells.PromptID:
 		payload = map[string]any{"spell_casts": []any{map[string]any{
 			"caster": "Mira Thorn", "spell": "Cure Wounds",