Execute incremental analysis artifact plans

docs/cli.md

@@ -294,7 +294,8 @@ and precedence.
 
 Effects:
 
-- filters analyze execution to selected configured artifacts;
+- selects explicit analyze targets; required configured prerequisites may be
+  reused or rebuilt before them;
 - filters publish rules that source `narratio.artifact.<name>`;
 - does not filter built-in transcript/bounds or explicitly configured
   `narratio.extraction.<name>` publish sources; and

docs/config.md

@@ -323,12 +323,13 @@ Narratio adds `session_id=narratio-session-<session_id>` to every Scriptorium re
 
 Without `--artifacts`, analyze executes enabled configured artifacts. With an
 explicit `--artifacts` list, the exact named configured artifacts are the
-one-invocation execution set even if their `enabled` values are false; the list
-does not automatically include dependencies. Named artifacts must therefore be
-configured with valid executable fields, and their configured dependencies must
-already be available to analyze. This override affects analyze planning only;
-publish uses the list only to filter configured
-`narratio.artifact.<name>` output rules.
+one-invocation targets even if their `enabled` values are false. Analyze closes
+those targets over `depends_on`: a current prerequisite is reused, while a
+stale, missing, failed, or legacy prerequisite is rebuilt before its dependent.
+Unrelated artifacts are not executed. Named targets and any prerequisite that
+may require rebuilding must therefore have valid executable fields. This
+override affects analyze planning only; publish uses the list only to filter
+configured `narratio.artifact.<name>` output rules.
 
 For each artifact input `pipeline.scriptorium.artifacts.<name>.inputs.<input_name>`:
 

docs/internal/artifacts.md

@@ -63,6 +63,12 @@ canonical file and a legacy aggregate analyze output are unavailable.
 Extraction entries are registered from configuration and become available only
 after compatible extraction evidence is hydrated.
 
+During an analyze invocation, a newly validated and atomically materialized
+configured output is marked available with its producer run ID, contract,
+checksum, and size. Later scheduled dependents therefore observe the same
+semantic identity whether their prerequisite was reused from current manifest
+evidence or produced earlier in the invocation.
+
 Current provenance values:
 
 - `generated.current_analyze_run`

docs/internal/manifest.md

@@ -74,6 +74,15 @@ limited to current records produced by that invocation's run ID. Ordinary
 stage outputs cannot accompany this projection, so there is one source of
 artifact authority.
 
+Successful incremental execution replaces only evaluated artifact records and
+preserves valid unrelated current records. Rebuilt outputs are compared by
+bytes and contract: an unchanged identity permits an unselected dependent with
+the same recomputed fingerprint to remain current, while a changed identity
+removes output authority from every unselected transitive dependent by marking
+it stale. A partial analyze invocation can therefore succeed while unrelated
+configured records remain stale. Existing canonical files never create current
+records without validated execution and projection.
+
 Analyze may return a projection together with an error. That restricted result
 cannot carry ordinary outputs, skip state, aggregate logs, generated configs,
 or metadata. The runner persists only the validated per-artifact collections,

docs/internal/stage-analyze.md

@@ -2,7 +2,8 @@
 
 ## Purpose
 
-Execute selected configured Scriptorium artifacts in dependency order and materialize outputs.
+Reconcile configured Scriptorium artifacts, execute only required work in
+dependency order, and safely materialize validated outputs.
 
 ## Inputs
 
@@ -21,23 +22,24 @@ Supported source families:
 
 ## Outputs
 
-- one materialized output per executed configured artifact (`output_path`)
+- one current per-artifact manifest record per validated materialized output
 - stage metadata describing selected/generated/reused artifacts
 
 ## Key Behavior
 
-- when Scriptorium is absent or no configured artifact is executable, completes
-  successfully with no outputs and records explanatory metadata. This is not an
-  explicit self-skip: both manifests record success, satisfy publish's
-  prerequisite, and an ordinary later run reuses the result until forced.
+- when `pipeline.scriptorium` is absent or no configured artifact is
+  executable, completes successfully with no outputs and records explanatory
+  metadata. This is not an explicit self-skip: both manifests record success,
+  satisfy publish's prerequisite, and an ordinary later run reuses the result
+  until forced.
 - builds a runtime artifact catalog containing built-ins, configured artifacts,
   and configured extraction lanes. Extraction availability is hydrated only
   from compatible successful extraction evidence.
 - uses enabled configured artifacts by default. An explicit `--artifacts`
-  selection is a one-invocation override: it makes exactly the named configured
-  artifacts executable even when disabled, and does not automatically include
-  dependencies. A selected artifact's dependencies must instead already be
-  available to the catalog.
+  selection is a one-invocation override that makes exactly the named
+  configured artifacts explicit targets even when disabled. The work planner
+  adds required configured prerequisites, reuses current ones, and schedules
+  stale, missing, or otherwise non-current prerequisites before dependents.
 - makes a non-executable configured artifact reusable only when its current
   manifest record and durable output pass the configured-artifact evidence
   contract; an incidental or stale canonical file is unavailable.
@@ -86,6 +88,29 @@ Supported source families:
   projected record collection. Valid unrelated configured records survive the
   projection, removed records are omitted, and legacy files never become
   current without regeneration.
+- executes only the work plan's scheduled entries. Manifest-validated current
+  prerequisites remain available through the runtime catalog without invoking
+  Scriptorium; newly produced prerequisites enter that catalog with the same
+  contract, checksum, and size identity used for persisted current evidence.
+- keeps adapter output in the invocation's run-local analyze directory until
+  it is a safe, non-empty, bounded regular file with a calculated checksum and
+  complete output contract. Canonical replacement uses the shared atomic file
+  operation boundary and verifies that the installed checksum matches the
+  validated run-local bytes.
+- records each successful artifact's freshly computed fingerprint, canonical
+  relative output path, contract, checksum, size, producer run ID, bounded
+  Scriptorium provenance, logs, and generated configuration references in the
+  analyze-owned projection.
+- preserves valid unrelated current records during partial execution. If a
+  rebuilt output's bytes and contract are unchanged, unselected dependents may
+  remain current. If that semantic identity changes, unselected transitive
+  dependents become stale without being executed; dependents included in the
+  invocation are evaluated in dependency order instead.
+- reports all evaluated targets and prerequisites in invocation state. The
+  runner reconstructs aggregate session outputs from every current session
+  record and invocation outputs from only records produced by the current run.
+  Unrelated stale records do not make an otherwise successful partial
+  invocation fail.
 - resolves previous-session sources from local `previous/` cache only.
 - runs optional render-debug, then artifact execution.
 - validates non-empty output files and materializes canonical outputs.
@@ -112,6 +137,8 @@ Supported source families:
   visible only inside Scriptorium—such as a file privately loaded by its config
   path—requires an explicit forced regeneration.
 - output provenance and metadata are deterministic per execution.
+- a canonical file without current per-artifact manifest evidence is never
+  promoted to current state.
 
 ## Related Contracts And Tests
 
@@ -126,4 +153,5 @@ Supported source families:
   `internal/stage/analyze_fingerprint_test.go`,
   `internal/stage/analyze_reconciliation.go`, and
   `internal/stage/analyze_reconciliation_test.go`,
-  `internal/stage/analyze_plan.go`, and `internal/stage/analyze_plan_test.go`
+  `internal/stage/analyze_plan.go`, `internal/stage/analyze_plan_test.go`, and
+  `internal/stage/analyze_incremental_execution_test.go`

docs/operations.md

@@ -165,7 +165,8 @@ or `publish`.
 Selection behavior:
 
 - validates names against `pipeline.scriptorium.artifacts`;
-- filters analyze execution to selected configured artifacts;
+- selects explicit analyze targets and permits their required configured
+  prerequisites to be reused or rebuilt first;
 - filters publish rules for `narratio.artifact.<name>` sources only;
 - does not suppress built-in transcript, bounds, or explicitly configured
   `narratio.extraction.<name>` publish sources; and

docs/roadmap/implementation.md

@@ -584,6 +584,8 @@ a deterministic artifact execution plan without invoking Scriptorium.
 
 ## Stage 12 — Incremental Analyze Execution And Promotion
 
+**Status: Completed**
+
 ### Goal
 
 Execute the Stage 11 work plan on the successful path, safely promote validated

internal/artifacts/catalog.go

@@ -291,6 +291,41 @@ func (c *ArtifactCatalog) MarkAvailableGenerated(sourceID, path string) error {
 	return c.markAvailable(sourceID, path, ArtifactProvenanceGeneratedCurrentAnalyzeRun)
 }
 
+// MarkAvailableGeneratedEvidence marks one source as available from the
+// current analyze invocation and retains the semantic output identity needed
+// by later scheduled dependents.
+func (c *ArtifactCatalog) MarkAvailableGeneratedEvidence(
+	sourceID, path, producerRunID, checksum string,
+	size int64,
+	contract *artifactmodel.ContractMetadata,
+) error {
+	producerRunID = strings.TrimSpace(producerRunID)
+	checksum = strings.TrimSpace(checksum)
+	if err := ValidateRunIdentity(producerRunID); err != nil {
+		return fmt.Errorf("generated artifact producer run id: %w", err)
+	}
+	if err := validateSHA256(checksum); err != nil {
+		return fmt.Errorf("generated artifact checksum: %w", err)
+	}
+	if size <= 0 {
+		return fmt.Errorf("generated artifact size must be positive")
+	}
+	if contract == nil || strings.TrimSpace(contract.MediaType) == "" ||
+		strings.TrimSpace(contract.SchemaID) == "" || strings.TrimSpace(contract.SchemaVersion) == "" {
+		return fmt.Errorf("generated artifact contract is incomplete")
+	}
+	if err := c.markAvailable(sourceID, path, ArtifactProvenanceGeneratedCurrentAnalyzeRun); err != nil {
+		return err
+	}
+	entry := c.entries[strings.TrimSpace(sourceID)]
+	entry.ProducerRunID = producerRunID
+	entry.Checksum = checksum
+	entry.Size = size
+	entry.Contract = cloneArtifactContract(contract)
+	c.entries[entry.SourceID] = entry
+	return nil
+}
+
 func (c *ArtifactCatalog) markAvailableFromExtractManifest(
 	sourceID, path, producerRunID, checksum string,
 	size int64,

internal/artifacts/catalog_test.go

@@ -1,6 +1,11 @@
 package artifacts
 
-import "testing"
+import (
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
+)
 
 func TestArtifactCatalogRegisterBuiltInsAndLookup(t *testing.T) {
 	catalog := NewArtifactCatalog()
@@ -179,6 +184,30 @@ func TestArtifactCatalogMarkAvailableGenerated(t *testing.T) {
 	}
 }
 
+func TestArtifactCatalogMarkAvailableGeneratedEvidence(t *testing.T) {
+	catalog := NewArtifactCatalog()
+	if err := catalog.RegisterConfiguredArtifacts(map[string]ConfiguredArtifactDefinition{
+		"session_recap": {Enabled: true, OutputPath: "artifacts/session_recap.md"},
+	}); err != nil {
+		t.Fatal(err)
+	}
+	sourceID, _ := catalog.SourceIDForConfiguredKey("session_recap")
+	contract := &artifactmodel.ContractMetadata{
+		MediaType: "text/markdown", SchemaID: "narratio.session_recap", SchemaVersion: "1",
+	}
+	checksum := strings.Repeat("a", 64)
+	if err := catalog.MarkAvailableGeneratedEvidence(
+		sourceID, "/tmp/session_recap.md", "run-1", checksum, 42, contract,
+	); err != nil {
+		t.Fatal(err)
+	}
+	entry, _ := catalog.Lookup(sourceID)
+	if !entry.Available || entry.ProducerRunID != "run-1" || entry.Checksum != checksum || entry.Size != 42 ||
+		entry.Contract == nil || *entry.Contract != *contract {
+		t.Fatalf("generated evidence entry = %#v", entry)
+	}
+}
+
 func TestArtifactCatalogLookupPlannedButUnavailable(t *testing.T) {
 	catalog := NewArtifactCatalog()
 	if err := catalog.RegisterConfiguredArtifacts(

internal/stage/analyze.go

@@ -2,8 +2,9 @@ package stage
 
 import (
 	"context"
+	"crypto/sha256"
+	"encoding/hex"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"path/filepath"
 	"sort"
@@ -11,6 +12,7 @@ import (
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/scriptorium"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactpolicy"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
@@ -28,6 +30,8 @@ type analyzeArtifactExecutionPlan struct {
 
 type analyzeArtifactExecutionResult struct {
 	Output           artifacts.Ref
+	OutputSize       int64
+	Scriptorium      manifest.AnalyzeArtifactProvenance
 	Logs             []string
 	GeneratedConfigs []string
 	Metadata         map[string]any
@@ -54,10 +58,6 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 	if env.Config.Pipeline == nil || env.Config.Session == nil {
 		return nil, fmt.Errorf("analyze: resolved config must include pipeline and session")
 	}
-	if env.Scriptorium == nil {
-		return nil, fmt.Errorf("analyze: scriptorium adapter is required")
-	}
-
 	var sessionID string
 	if m != nil {
 		sessionID = strings.TrimSpace(m.SessionID)
@@ -81,6 +81,13 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 			"reason":  "pipeline.scriptorium is not configured",
 		}}, nil
 	}
+	if len(env.Config.Pipeline.Scriptorium.Artifacts) == 0 {
+		return &StageResult{Metadata: map[string]any{
+			"stage":   "analyze",
+			"skipped": true,
+			"reason":  "no scriptorium artifacts configured",
+		}}, nil
+	}
 
 	effective := env.EffectiveArtifacts
 	if !effective.Resolved() {
@@ -104,15 +111,11 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 		return nil, fmt.Errorf("analyze: build runtime artifact catalog: %w", err)
 	}
 
-	plans, skipReason, err := buildAnalyzeExecutionPlans(env.Config.Pipeline.Scriptorium, effective, runtimeCatalog)
-	if err != nil {
-		return nil, fmt.Errorf("analyze: %w", err)
-	}
-	if skipReason != "" {
+	if len(effective.Keys()) == 0 {
 		return &StageResult{Metadata: map[string]any{
 			"stage":   "analyze",
 			"skipped": true,
-			"reason":  skipReason,
+			"reason":  "no selected scriptorium artifacts to execute",
 		}}, nil
 	}
 
@@ -125,21 +128,64 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 		TranscriptRefs: discoverAnalyzeTranscriptRefs(m, paths),
 		Catalog:        runtimeCatalog,
 	}
+	reconciliation, err := reconcileAnalyzeArtifacts(env.Config.Pipeline.Scriptorium, execution)
+	if err != nil {
+		return nil, fmt.Errorf("analyze: reconcile configured artifacts: %w", err)
+	}
+	workPlan, err := planAnalyzeWork(
+		env.Config.Pipeline.Scriptorium,
+		env.SelectedArtifactKeys,
+		env.Force,
+		reconciliation,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("analyze: plan configured artifacts: %w", err)
+	}
+	if len(workPlan.ExecutionOrder) > 0 && env.Scriptorium == nil {
+		return nil, fmt.Errorf("analyze: scriptorium adapter is required")
+	}
 
-	outputs := make([]artifacts.Ref, 0, len(plans))
 	logs := []string{}
 	generatedConfigs := []string{}
-	artifactMetadata := make([]map[string]any, 0, len(plans))
+	artifactMetadata := make([]map[string]any, 0, len(workPlan.ExecutionOrder))
 	reusedArtifacts := []map[string]any{}
 	reusedSeen := map[string]struct{}{}
+	sessionRecords := manifest.CloneAnalyzeArtifactCollection(workPlan.ProjectedRecords)
+	invocationRecords := make(map[string]manifest.AnalyzeArtifactRecord)
+	priorCurrentRecords := make(map[string]manifest.AnalyzeArtifactRecord)
+	for _, item := range reconciliation.Ordered {
+		if item.Stored != nil && item.Stored.Status == manifest.AnalyzeArtifactCurrent {
+			priorCurrentRecords[item.Key] = *item.Stored
+		}
+	}
+	invocationKeys := make(map[string]struct{}, len(workPlan.ExecutionOrder)+len(workPlan.ReusedCurrent))
+	for _, item := range workPlan.ExecutionOrder {
+		invocationKeys[item.Key] = struct{}{}
+	}
+	for _, item := range workPlan.ReusedCurrent {
+		invocationKeys[item.Key] = struct{}{}
+		if record, ok := sessionRecords[item.Key]; ok {
+			invocationRecords[item.Key] = record
+		}
+	}
 
-	for _, plan := range plans {
+	for _, item := range workPlan.ExecutionOrder {
+		artifactCfg := env.Config.Pipeline.Scriptorium.Artifacts[item.Key]
+		fingerprint, _, _, err := computeAnalyzeArtifactFingerprint(
+			item.Key,
+			env.Config.Pipeline.Scriptorium,
+			execution,
+		)
+		if err != nil {
+			return nil, fmt.Errorf("analyze: compute execution fingerprint for artifact %q: %w", item.Key, err)
+		}
+		priorRecord, hadPriorRecord := priorCurrentRecords[item.Key]
+		plan := analyzeArtifactExecutionPlan{Name: item.Key, Cfg: artifactCfg}
 		artifactResult, err := executeAnalyzeArtifact(ctx, execution, plan)
 		if err != nil {
 			return nil, err
 		}
 
-		outputs = append(outputs, artifactResult.Output)
 		logs = append(logs, artifactResult.Logs...)
 		generatedConfigs = append(generatedConfigs, artifactResult.GeneratedConfigs...)
 		artifactMetadata = append(artifactMetadata, artifactResult.Metadata)
@@ -154,18 +200,47 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 			reusedArtifacts = append(reusedArtifacts, reused)
 		}
 
+		record, err := currentAnalyzeArtifactRecord(
+			execution,
+			plan,
+			fingerprint,
+			artifactResult,
+		)
+		if err != nil {
+			return nil, fmt.Errorf("analyze: record artifact %q: %w", plan.Name, err)
+		}
+		sessionRecords[plan.Name] = record
+		invocationRecords[plan.Name] = record
+
 		sourceID, ok := runtimeCatalog.SourceIDForConfiguredKey(plan.Name)
 		if !ok {
 			return nil, fmt.Errorf("analyze: source id not found for artifact %q", plan.Name)
 		}
-		if err := runtimeCatalog.MarkAvailableGenerated(sourceID, artifactResult.Output.AbsolutePath); err != nil {
+		if err := runtimeCatalog.MarkAvailableGeneratedEvidence(
+			sourceID,
+			artifactResult.Output.AbsolutePath,
+			record.ProducerRunID,
+			record.Output.Checksum,
+			record.OutputSize,
+			record.Output.Contract,
+		); err != nil {
 			return nil, fmt.Errorf("analyze: mark generated artifact %q available: %w", sourceID, err)
 		}
+		if !hadPriorRecord || !sameAnalyzeOutputIdentity(priorRecord, record) {
+			staleUnscheduledAnalyzeDependents(
+				env.Config.Pipeline.Scriptorium.Artifacts,
+				plan.Name,
+				invocationKeys,
+				sessionRecords,
+			)
+		}
 	}
 
 	metadata := map[string]any{
 		"stage":                 "analyze",
-		"selected_artifacts":    extractPlanNames(plans),
+		"selected_artifacts":    append([]string(nil), workPlan.ExplicitTargets...),
+		"executed_artifacts":    analyzePlanKeysForMetadata(workPlan.ExecutionOrder),
+		"reused_current":        analyzePlanKeysForMetadata(workPlan.ReusedCurrent),
 		"generated_artifacts":   artifactMetadata,
 		"reused_artifacts":      reusedArtifacts,
 		"artifact_count":        len(artifactMetadata),
@@ -178,133 +253,134 @@ func (analyzeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest) (*S
 	}
 
 	return &StageResult{
-		Outputs:          outputs,
 		Logs:             dedupeAndSortPaths(logs),
 		GeneratedConfigs: dedupeAndSortPaths(generatedConfigs),
 		Metadata:         metadata,
+		AnalyzeState: &AnalyzeStateProjection{
+			Session:    sessionRecords,
+			Invocation: manifest.CloneAnalyzeArtifactCollection(invocationRecords),
+		},
 	}, nil
 }
 
-func buildAnalyzeExecutionPlans(
-	scriptoriumCfg *config.ScriptoriumConfig,
-	effective artifacts.EffectiveArtifactSet,
-	catalog *artifacts.ArtifactCatalog,
-) ([]analyzeArtifactExecutionPlan, string, error) {
-	if scriptoriumCfg == nil {
-		return nil, "pipeline.scriptorium is not configured", nil
+func currentAnalyzeArtifactRecord(
+	execution analyzeExecutionContext,
+	plan analyzeArtifactExecutionPlan,
+	fingerprint string,
+	result *analyzeArtifactExecutionResult,
+) (manifest.AnalyzeArtifactRecord, error) {
+	if result == nil {
+		return manifest.AnalyzeArtifactRecord{}, fmt.Errorf("execution result is required")
 	}
-	if len(scriptoriumCfg.Artifacts) == 0 {
-		return nil, "no scriptorium artifacts configured", nil
+	producerRunID := ""
+	if execution.Manifest != nil {
+		producerRunID = strings.TrimSpace(execution.Manifest.RunID)
 	}
-
-	if len(effective.Keys()) == 0 {
-		return nil, "no selected scriptorium artifacts to execute", nil
+	if producerRunID == "" {
+		// Direct stage callers predate invocation manifests. Application-owned
+		// execution always supplies the actual run identity.
+		producerRunID = "direct-analyze"
 	}
-
-	ordered, err := orderSelectedScriptoriumArtifacts(scriptoriumCfg.Artifacts, effective, catalog)
+	relativePath, err := normalizedAnalyzeOutputIdentity(plan.Cfg.OutputPath)
 	if err != nil {
-		return nil, "", err
+		return manifest.AnalyzeArtifactRecord{}, err
 	}
-
-	plans := make([]analyzeArtifactExecutionPlan, 0, len(ordered))
-	for _, name := range ordered {
-		artifactCfg, ok := scriptoriumCfg.Artifacts[name]
-		if !ok {
-			return nil, "", fmt.Errorf("selected artifact %q is not configured", name)
-		}
-		plans = append(plans, analyzeArtifactExecutionPlan{Name: name, Cfg: artifactCfg})
+	if relativePath == "" {
+		return manifest.AnalyzeArtifactRecord{}, fmt.Errorf("configured output path is required")
 	}
-	return plans, "", nil
+	contract := result.Output.Contract
+	if contract == nil {
+		return manifest.AnalyzeArtifactRecord{}, fmt.Errorf("validated output contract is required")
+	}
+	record := manifest.AnalyzeArtifactRecord{
+		Key:                plan.Name,
+		Status:             manifest.AnalyzeArtifactCurrent,
+		FingerprintVersion: manifest.AnalyzeFingerprintContractVersion,
+		Fingerprint:        fingerprint,
+		Dependencies:       normalizedAnalyzeDependencyKeys(plan.Cfg.DependsOn),
+		Output: &manifest.ArtifactRecord{
+			Kind:          "scriptorium_artifact",
+			SourceID:      artifacts.ConfiguredArtifactSourceID(plan.Name),
+			LocalPath:     relativePath,
+			Contract:      cloneAnalyzeOutputContract(contract),
+			ProducerRunID: producerRunID,
+			Checksum:      result.Output.Checksum,
+		},
+		OutputSize:       result.OutputSize,
+		ProducerRunID:    producerRunID,
+		UpdatedAt:        time.Now().UTC(),
+		Scriptorium:      &result.Scriptorium,
+		Logs:             dedupeAndSortPaths(result.Logs),
+		GeneratedConfigs: dedupeAndSortPaths(result.GeneratedConfigs),
+	}
+	if err := manifest.ValidateAnalyzeArtifactCollection(
+		manifest.AnalyzeStateContractVersion,
+		map[string]manifest.AnalyzeArtifactRecord{plan.Name: record},
+	); err != nil {
+		return manifest.AnalyzeArtifactRecord{}, err
+	}
+	return record, nil
 }
 
-func orderSelectedScriptoriumArtifacts(
-	artifactsCfg map[string]config.ScriptoriumArtifactConfig,
-	effective artifacts.EffectiveArtifactSet,
-	catalog *artifacts.ArtifactCatalog,
-) ([]string, error) {
-	selectedSet := map[string]struct{}{}
-	selected := effective.Keys()
-	for _, key := range selected {
-		selectedSet[key] = struct{}{}
+func sameAnalyzeOutputIdentity(left, right manifest.AnalyzeArtifactRecord) bool {
+	if left.Status != manifest.AnalyzeArtifactCurrent || right.Status != manifest.AnalyzeArtifactCurrent ||
+		left.Output == nil || right.Output == nil || left.Output.Contract == nil || right.Output.Contract == nil {
+		return false
 	}
+	return left.OutputSize == right.OutputSize &&
+		left.Output.Checksum == right.Output.Checksum &&
+		*left.Output.Contract == *right.Output.Contract
+}
 
-	dependencyErrors := []string{}
-	for _, selectedKey := range selected {
-		cfg, ok := artifactsCfg[selectedKey]
-		if !ok {
-			dependencyErrors = append(dependencyErrors, fmt.Sprintf("selected artifact %q is not configured", selectedKey))
+func staleUnscheduledAnalyzeDependents(
+	configured map[string]config.ScriptoriumArtifactConfig,
+	changed string,
+	invocationKeys map[string]struct{},
+	records map[string]manifest.AnalyzeArtifactRecord,
+) {
+	reverse := make(map[string][]string, len(configured))
+	for key, artifactCfg := range configured {
+		for _, dependency := range normalizedAnalyzeDependencyKeys(artifactCfg.DependsOn) {
+			reverse[dependency] = append(reverse[dependency], key)
+		}
+	}
+	for key := range reverse {
+		sort.Strings(reverse[key])
+	}
+	queue := append([]string(nil), reverse[changed]...)
+	seen := make(map[string]struct{}, len(queue))
+	for len(queue) > 0 {
+		key := queue[0]
+		queue = queue[1:]
+		if _, visited := seen[key]; visited {
 			continue
 		}
-		for _, dep := range cfg.DependsOn {
-			trimmedDep := strings.TrimSpace(dep)
-			if trimmedDep == "" {
-				continue
-			}
-			if _, ok := selectedSet[trimmedDep]; ok {
-				continue
-			}
-			sourceID, ok := catalog.SourceIDForConfiguredKey(trimmedDep)
-			if !ok {
-				dependencyErrors = append(dependencyErrors, fmt.Sprintf("artifact %q depends on unknown configured artifact %q", selectedKey, trimmedDep))
-				continue
-			}
-			entry, ok := catalog.Lookup(sourceID)
-			if !ok || !entry.Available {
-				dependencyErrors = append(dependencyErrors, fmt.Sprintf("artifact %q depends on %q, but %q is unavailable", selectedKey, trimmedDep, sourceID))
-			}
+		seen[key] = struct{}{}
+		if _, evaluated := invocationKeys[key]; evaluated {
+			continue
 		}
+		staleProjectedAnalyzeRecord(records, key)
+		queue = append(queue, reverse[key]...)
 	}
-	if len(dependencyErrors) > 0 {
-		return nil, errors.New(strings.Join(dependencyErrors, "; "))
-	}
+}
 
-	indegree := map[string]int{}
-	edges := map[string][]string{}
-	for _, key := range selected {
-		indegree[key] = 0
+func analyzePlanKeysForMetadata(items []analyzePlanItem) []string {
+	if len(items) == 0 {
+		return nil
 	}
-	for _, key := range selected {
-		cfg := artifactsCfg[key]
-		for _, dep := range cfg.DependsOn {
-			trimmedDep := strings.TrimSpace(dep)
-			if _, ok := selectedSet[trimmedDep]; !ok {
-				continue
-			}
-			edges[trimmedDep] = append(edges[trimmedDep], key)
-			indegree[key]++
-		}
+	keys := make([]string, 0, len(items))
+	for _, item := range items {
+		keys = append(keys, item.Key)
 	}
+	return keys
+}
 
-	for key := range edges {
-		sort.Strings(edges[key])
+func cloneAnalyzeOutputContract(value *artifactmodel.ContractMetadata) *artifactmodel.ContractMetadata {
+	if value == nil {
+		return nil
 	}
-
-	ready := make([]string, 0, len(indegree))
-	for key, degree := range indegree {
-		if degree == 0 {
-			ready = append(ready, key)
-		}
-	}
-	sort.Strings(ready)
-
-	order := make([]string, 0, len(selectedSet))
-	for len(ready) > 0 {
-		node := ready[0]
-		ready = ready[1:]
-		order = append(order, node)
-		for _, dep := range edges[node] {
-			indegree[dep]--
-			if indegree[dep] == 0 {
-				ready = append(ready, dep)
-				sort.Strings(ready)
-			}
-		}
-	}
-
-	if len(order) != len(selectedSet) {
-		return nil, fmt.Errorf("selected scriptorium artifacts contain a dependency cycle")
-	}
-	return order, nil
+	cloned := *value
+	return &cloned
 }
 
 func executeAnalyzeArtifact(
@@ -504,14 +580,32 @@ func executeAnalyzeArtifact(
 	if err := requireNonEmptyFile(finalOutputPath, artifactName+" output"); err != nil {
 		return nil, fmt.Errorf("analyze: %w", err)
 	}
+	validatedOutput, err := readExternalResult(finalOutputPath, artifactName+" output")
+	if err != nil {
+		return nil, fmt.Errorf("analyze: %w", err)
+	}
+	contract := &artifactmodel.ContractMetadata{
+		MediaType: "text/markdown", SchemaID: "narratio." + artifactName, SchemaVersion: "1",
+	}
+	relativeOutputPath, err := normalizedAnalyzeOutputIdentity(artifactCfg.OutputPath)
+	if err != nil {
+		return nil, fmt.Errorf("analyze: normalize output identity for artifact %q: %w", artifactName, err)
+	}
 	materializedArtifact, err := materializeRunLocalOutput(env.ArtifactStore, finalOutputPath, canonicalOutputPath, artifacts.Ref{
-		Kind:      artifactName,
-		Category:  "artifacts",
-		SessionID: sessionID,
+		Kind:         artifactName,
+		SourceID:     artifacts.ConfiguredArtifactSourceID(artifactName),
+		Category:     "artifacts",
+		SessionID:    sessionID,
+		RelativePath: relativeOutputPath,
+		Contract:     contract,
 	})
 	if err != nil {
 		return nil, fmt.Errorf("analyze: materialize artifact output for %q: %w", artifactName, err)
 	}
+	expectedDigest := sha256.Sum256(validatedOutput)
+	if materializedArtifact.Checksum != hex.EncodeToString(expectedDigest[:]) {
+		return nil, fmt.Errorf("analyze: materialized artifact output for %q differs from validated run-local bytes", artifactName)
+	}
 
 	logPaths = append(logPaths, stdoutLogPath, stderrLogPath)
 	generatedConfigs = append(generatedConfigs, generatedConfigPath)
@@ -537,7 +631,11 @@ func executeAnalyzeArtifact(
 	}
 
 	return &analyzeArtifactExecutionResult{
-		Output:           materializedArtifact,
+		Output:     materializedArtifact,
+		OutputSize: int64(len(validatedOutput)),
+		Scriptorium: manifest.AnalyzeArtifactProvenance{
+			PromptID: artifactCfg.PromptID, ProfileID: artifactCfg.ProfileID, CommandMode: res.CommandMode,
+		},
 		Logs:             logPaths,
 		GeneratedConfigs: generatedConfigs,
 		Metadata:         meta,
@@ -545,17 +643,6 @@ func executeAnalyzeArtifact(
 	}, nil
 }
 
-func extractPlanNames(plans []analyzeArtifactExecutionPlan) []string {
-	if len(plans) == 0 {
-		return nil
-	}
-	out := make([]string, 0, len(plans))
-	for _, plan := range plans {
-		out = append(out, plan.Name)
-	}
-	return out
-}
-
 func configuredArtifactNameFromSourceID(sourceID string) string {
 	name, _ := artifactpolicy.ParseConfiguredSource(sourceID)
 	return name

internal/stage/analyze_incremental_execution_test.go

@@ -0,0 +1,249 @@
+package stage
+
+import (
+	"context"
+	"path/filepath"
+	"reflect"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+)
+
+func TestAnalyzeReusesCurrentArtifactWithoutScriptorium(t *testing.T) {
+	env, m, fake := setupAnalyzeEnv(t)
+	paths := sessionPathsForEnv(env, m.SessionID)
+	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+
+	first, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	installAnalyzeProjection(m, first.AnalyzeState)
+	if len(fake.RunRequests) != 1 {
+		t.Fatalf("initial requests = %d, want 1", len(fake.RunRequests))
+	}
+
+	env.Scriptorium = nil
+	second, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatalf("reuse Run() error = %v", err)
+	}
+	if got, _ := second.Metadata["executed_artifacts"].([]string); len(got) != 0 {
+		t.Fatalf("executed artifacts = %#v, want none", got)
+	}
+	if got := second.Metadata["reused_current"]; !reflect.DeepEqual(got, []string{"session_recap"}) {
+		t.Fatalf("reused current = %#v, want session_recap", got)
+	}
+	if second.AnalyzeState.Invocation["session_recap"].Status != manifest.AnalyzeArtifactCurrent {
+		t.Fatalf("invocation state = %#v", second.AnalyzeState.Invocation)
+	}
+}
+
+func TestAnalyzePartialForcePreservesUnrelatedCurrentRecord(t *testing.T) {
+	env, m, fake := setupAnalyzeEnv(t)
+	paths := sessionPathsForEnv(env, m.SessionID)
+	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+	addAnalyzeDependentArtifact(env)
+
+	first, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	installAnalyzeProjection(m, first.AnalyzeState)
+	if len(fake.RunRequests) != 2 {
+		t.Fatalf("initial requests = %d, want 2", len(fake.RunRequests))
+	}
+	priorDependent := m.Stages["analyze"].AnalyzeArtifacts["player_handout"]
+
+	env.SelectedArtifactKeys = []string{"session_recap"}
+	env.Force = true
+	secondRunner := &orderedScriptoriumRunner{RunBody: "scriptorium noop/fake run artifact\n"}
+	env.Scriptorium = secondRunner
+	second, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := secondRunner.Calls; !reflect.DeepEqual(got, []string{"run"}) {
+		t.Fatalf("calls = %#v, want one forced target run", got)
+	}
+	if got := second.AnalyzeState.Session["player_handout"]; !reflect.DeepEqual(got, priorDependent) {
+		t.Fatalf("unrelated dependent changed = %#v, want %#v", got, priorDependent)
+	}
+	if _, attempted := second.AnalyzeState.Invocation["player_handout"]; attempted {
+		t.Fatal("unselected dependent appeared in invocation state")
+	}
+}
+
+func TestAnalyzeExposesReusedPrerequisiteToScheduledDependent(t *testing.T) {
+	env, m, fake := setupAnalyzeEnv(t)
+	paths := sessionPathsForEnv(env, m.SessionID)
+	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+	addAnalyzeDependentArtifact(env)
+
+	first, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	installAnalyzeProjection(m, first.AnalyzeState)
+	dependent := env.Config.Pipeline.Scriptorium.Artifacts["player_handout"]
+	dependent.PromptID = "dnd.player_handout.revised"
+	env.Config.Pipeline.Scriptorium.Artifacts["player_handout"] = dependent
+	env.SelectedArtifactKeys = []string{"player_handout"}
+	fake.RunRequests = nil
+
+	second, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(fake.RunRequests) != 1 || fake.RunRequests[0].PromptID != "dnd.player_handout.revised" {
+		t.Fatalf("requests = %#v, want only revised dependent", fake.RunRequests)
+	}
+	if got := fake.RunRequests[0].InputPaths["recap"]; got != filepath.Join(paths.ArtifactsDir, "session_recap.md") {
+		t.Fatalf("reused prerequisite input = %q", got)
+	}
+	if second.AnalyzeState.Invocation["session_recap"].Status != manifest.AnalyzeArtifactCurrent ||
+		second.AnalyzeState.Invocation["player_handout"].Status != manifest.AnalyzeArtifactCurrent {
+		t.Fatalf("invocation records = %#v, want reused prerequisite and produced dependent", second.AnalyzeState.Invocation)
+	}
+}
+
+func TestAnalyzeChangedOutputStalesUnselectedDependents(t *testing.T) {
+	env, m, _ := setupAnalyzeEnv(t)
+	paths := sessionPathsForEnv(env, m.SessionID)
+	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+	addAnalyzeDependentArtifact(env)
+
+	first, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	installAnalyzeProjection(m, first.AnalyzeState)
+
+	env.SelectedArtifactKeys = []string{"session_recap"}
+	env.Force = true
+	env.Scriptorium = &orderedScriptoriumRunner{RunBody: "changed recap\n"}
+	second, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	dependent := second.AnalyzeState.Session["player_handout"]
+	if dependent.Status != manifest.AnalyzeArtifactStale || dependent.Output != nil || dependent.OutputSize != 0 {
+		t.Fatalf("dependent record = %#v, want stale without output", dependent)
+	}
+	if _, attempted := second.AnalyzeState.Invocation["player_handout"]; attempted {
+		t.Fatal("changed-output invalidation executed the unselected dependent")
+	}
+}
+
+func TestAnalyzeRebuildsNestedStalePrerequisitesInDependencyOrder(t *testing.T) {
+	env, m, fake := setupAnalyzeEnv(t)
+	paths := sessionPathsForEnv(env, m.SessionID)
+	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+	addAnalyzeDependentArtifact(env)
+	env.Config.Pipeline.Scriptorium.Artifacts["quest_log"] = config.ScriptoriumArtifactConfig{
+		Enabled: false, DependsOn: []string{"player_handout"}, PromptID: "dnd.quest_log",
+		OutputPath: "artifacts/quest_log.md",
+		Inputs: map[string]config.ScriptoriumInputConfig{
+			"handout": {Source: "narratio.artifact.player_handout", Required: true},
+		},
+	}
+
+	first, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	installAnalyzeProjection(m, first.AnalyzeState)
+	if len(fake.RunRequests) != 2 {
+		t.Fatalf("initial enabled requests = %d, want 2", len(fake.RunRequests))
+	}
+	stale := m.Stages["analyze"].AnalyzeArtifacts["session_recap"]
+	stale.Status = manifest.AnalyzeArtifactStale
+	stale.Output = nil
+	stale.OutputSize = 0
+	m.Stages["analyze"].AnalyzeArtifacts["session_recap"] = stale
+
+	env.SelectedArtifactKeys = []string{"quest_log"}
+	fake.RunRequests = nil
+	second, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if err != nil {
+		t.Fatal(err)
+	}
+	gotPrompts := make([]string, 0, len(fake.RunRequests))
+	for _, request := range fake.RunRequests {
+		gotPrompts = append(gotPrompts, request.PromptID)
+	}
+	wantPrompts := []string{"dnd.session_recap", "dnd.player_handout", "dnd.quest_log"}
+	if !reflect.DeepEqual(gotPrompts, wantPrompts) {
+		t.Fatalf("prompt order = %#v, want %#v", gotPrompts, wantPrompts)
+	}
+	for _, key := range []string{"session_recap", "player_handout", "quest_log"} {
+		if second.AnalyzeState.Session[key].Status != manifest.AnalyzeArtifactCurrent {
+			t.Fatalf("%s state = %#v, want current", key, second.AnalyzeState.Session[key])
+		}
+	}
+}
+
+func TestAnalyzeLegacyExecutionPromotesOnlyEffectiveArtifacts(t *testing.T) {
+	for _, test := range []struct {
+		name        string
+		selected    []string
+		wantPrompts []string
+	}{
+		{name: "full selection", wantPrompts: []string{"dnd.player_handout", "dnd.session_recap"}},
+		{name: "partial selection", selected: []string{"session_recap"}, wantPrompts: []string{"dnd.session_recap"}},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			env, m, fake := setupAnalyzeEnv(t)
+			paths := sessionPathsForEnv(env, m.SessionID)
+			writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)
+			writeAnalyzeFile(t, filepath.Join(paths.ArtifactsDir, "player_handout.md"), "legacy handout\n")
+			env.Config.Pipeline.Scriptorium.Artifacts["player_handout"] = config.ScriptoriumArtifactConfig{
+				Enabled: true, PromptID: "dnd.player_handout", OutputPath: "artifacts/player_handout.md",
+			}
+			env.SelectedArtifactKeys = test.selected
+			m.Stages["analyze"] = &manifest.StageRecord{
+				Name: "analyze", Status: manifest.StatusSucceeded,
+				Outputs: []manifest.ArtifactRecord{{Kind: "player_handout", LocalPath: filepath.Join(paths.ArtifactsDir, "player_handout.md")}},
+			}
+
+			result, err := (analyzeStage{}).Run(context.Background(), env, m)
+			if err != nil {
+				t.Fatal(err)
+			}
+			gotPrompts := make([]string, 0, len(fake.RunRequests))
+			for _, request := range fake.RunRequests {
+				gotPrompts = append(gotPrompts, request.PromptID)
+			}
+			if !reflect.DeepEqual(gotPrompts, test.wantPrompts) {
+				t.Fatalf("prompts = %#v, want %#v", gotPrompts, test.wantPrompts)
+			}
+			if len(result.AnalyzeState.Session) != len(test.wantPrompts) {
+				t.Fatalf("session records = %#v, want only regenerated effective artifacts", result.AnalyzeState.Session)
+			}
+			if len(test.selected) > 0 {
+				if _, promoted := result.AnalyzeState.Session["player_handout"]; promoted {
+					t.Fatal("legacy unselected canonical output was promoted")
+				}
+			}
+		})
+	}
+}
+
+func addAnalyzeDependentArtifact(env *Env) {
+	env.Config.Pipeline.Scriptorium.Artifacts["player_handout"] = config.ScriptoriumArtifactConfig{
+		Enabled: true, DependsOn: []string{"session_recap"}, PromptID: "dnd.player_handout",
+		OutputPath: "artifacts/player_handout.md",
+		Inputs: map[string]config.ScriptoriumInputConfig{
+			"recap": {Source: "narratio.artifact.session_recap", Required: true},
+		},
+	}
+}
+
+func installAnalyzeProjection(m *manifest.Manifest, projection *AnalyzeStateProjection) {
+	if m.Stages["analyze"] == nil {
+		m.Stages["analyze"] = &manifest.StageRecord{Name: "analyze"}
+	}
+	m.Stages["analyze"].AnalyzeStateVersion = manifest.AnalyzeStateContractVersion
+	m.Stages["analyze"].AnalyzeArtifacts = manifest.CloneAnalyzeArtifactCollection(projection.Session)
+}

internal/stage/analyze_order_test.go

@@ -1,44 +0,0 @@
-package stage
-
-import (
-	"strings"
-	"testing"
-
-	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
-	"gitea.maximumdirect.net/eric/narratio/internal/config"
-)
-
-func TestOrderSelectedScriptoriumArtifactsReportsUnavailableDependenciesDeterministically(t *testing.T) {
-	configured := map[string]config.ScriptoriumArtifactConfig{
-		"alpha":     {Enabled: true, DependsOn: []string{"alpha_dep"}},
-		"alpha_dep": {Enabled: false, OutputPath: "artifacts/alpha_dep.md"},
-		"zeta":      {Enabled: true, DependsOn: []string{"zeta_dep"}},
-		"zeta_dep":  {Enabled: false, OutputPath: "artifacts/zeta_dep.md"},
-	}
-	effective, err := artifacts.ResolveEffectiveArtifactSet(
-		artifacts.ConfiguredArtifactDefinitions(configured),
-		[]string{"alpha", "zeta"},
-	)
-	if err != nil {
-		t.Fatalf("ResolveEffectiveArtifactSet() error = %v", err)
-	}
-	catalog, err := artifacts.BootstrapRuntimeCatalog(
-		artifacts.ConfiguredArtifactDefinitions(configured),
-		effective,
-		nil,
-	)
-	if err != nil {
-		t.Fatalf("BootstrapRuntimeCatalog() error = %v", err)
-	}
-
-	for i := 0; i < 100; i++ {
-		_, err := orderSelectedScriptoriumArtifacts(configured, effective, catalog)
-		if err == nil {
-			t.Fatal("orderSelectedScriptoriumArtifacts() error = nil, want unavailable dependency")
-		}
-		want := `artifact "alpha" depends on "alpha_dep", but "narratio.artifact.alpha_dep" is unavailable; artifact "zeta" depends on "zeta_dep", but "narratio.artifact.zeta_dep" is unavailable`
-		if !strings.Contains(err.Error(), want) {
-			t.Fatalf("attempt %d error = %q, want %q", i, err, want)
-		}
-	}
-}

internal/stage/analyze_test.go

@@ -62,8 +62,8 @@ func TestAnalyzeGeneratesSessionRecapFromTrimmedTranscript(t *testing.T) {
 		t.Fatalf("session_id var = %q, want sticky narratio session id", req.Vars["session_id"])
 	}
 
-	if len(result.Outputs) != 1 || result.Outputs[0].Kind != "session_recap" {
-		t.Fatalf("outputs = %#v, want one session_recap output", result.Outputs)
+	if result.AnalyzeState == nil || result.AnalyzeState.Invocation["session_recap"].Status != manifest.AnalyzeArtifactCurrent {
+		t.Fatalf("analyze state = %#v, want current session_recap", result.AnalyzeState)
 	}
 	if len(result.Logs) != 2 {
 		t.Fatalf("logs = %#v, want stdout+stderr logs", result.Logs)
@@ -314,11 +314,11 @@ func TestAnalyzeUsesRunLocalPathsAndMaterializesCanonical(t *testing.T) {
 	if !strings.Contains(fake.RunRequests[0].OutputPath, filepath.Join("runs", m.RunID, "analyze", "outputs")) {
 		t.Fatalf("run output path = %q, want run-local path", fake.RunRequests[0].OutputPath)
 	}
-	if len(result.Outputs) != 1 {
-		t.Fatalf("outputs len = %d, want 1", len(result.Outputs))
+	if result.AnalyzeState == nil || result.AnalyzeState.Invocation["session_recap"].Output == nil {
+		t.Fatalf("analyze state = %#v, want session recap output evidence", result.AnalyzeState)
 	}
-	if strings.Contains(result.Outputs[0].AbsolutePath, string(filepath.Separator)+"runs"+string(filepath.Separator)) {
-		t.Fatalf("materialized output path = %q, want canonical session path", result.Outputs[0].AbsolutePath)
+	if err := requireNonEmptyFile(filepath.Join(paths.ArtifactsDir, "session_recap.md"), "materialized output"); err != nil {
+		t.Fatalf("canonical output = %v", err)
 	}
 }
 
@@ -682,8 +682,8 @@ func TestAnalyzeAppliesSelectedArtifactsFilter(t *testing.T) {
 	if fake.RunRequests[0].PromptID != "dnd.player_handout" {
 		t.Fatalf("prompt id = %q, want dnd.player_handout", fake.RunRequests[0].PromptID)
 	}
-	if len(result.Outputs) != 1 || result.Outputs[0].Kind != "player_handout" {
-		t.Fatalf("outputs = %#v, want only player_handout", result.Outputs)
+	if result.AnalyzeState == nil || len(result.AnalyzeState.Invocation) != 1 || result.AnalyzeState.Invocation["player_handout"].Status != manifest.AnalyzeArtifactCurrent {
+		t.Fatalf("analyze state = %#v, want only current player_handout", result.AnalyzeState)
 	}
 }
 
@@ -1389,11 +1389,9 @@ func TestAnalyzeRecordsRefsAndMetadata(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if len(result.Outputs) != 1 {
-		t.Fatalf("outputs len = %d, want 1", len(result.Outputs))
-	}
-	if result.Outputs[0].AbsolutePath != filepath.Join(paths.ArtifactsDir, "session_recap.md") {
-		t.Fatalf("output path = %q, want session recap path", result.Outputs[0].AbsolutePath)
+	record, ok := result.AnalyzeState.Invocation["session_recap"]
+	if !ok || record.Output == nil || record.Output.LocalPath != "artifacts/session_recap.md" {
+		t.Fatalf("analyze output record = %#v, want session recap evidence", record)
 	}
 	if len(result.Logs) != 2 {
 		t.Fatalf("logs = %#v, want two logs", result.Logs)