Complete analyze checkpoint recovery

docs/internal/manifest.md

@@ -82,11 +82,12 @@ The session-stage collection is the reconciled authority across invocations.
 These records remain analyze-owned data inside the fixed stage; they are not
 dynamic stages or generic subtasks.
 
-The stage result contract has one analyze-specific projection boundary. On
+Analyze checkpoints one complete validated candidate collection after each
-success, the runner validates and deep-copies the complete reconciled session
+successful artifact and before exposing that artifact to later work. It saves a
-collection. Aggregate session outputs are rebuilt in configured-key order from
+detached manifest candidate and installs its version, collection, and updated
-current session records only. Ordinary stage outputs cannot accompany this
+timestamp into live state only after the store reports success. Aggregate stage
-projection, so there is one source of artifact authority.
+outputs are ordinary result outputs rebuilt in configured-key order from current
+session records only.
 
 Successful incremental execution replaces only evaluated artifact records and
 preserves valid unrelated current records. Rebuilt outputs are compared by
@@ -95,31 +96,29 @@ 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.
+records without validated execution and a successful checkpoint.
 
 Aggregate analyze status is deliberately coarser than this collection. Resume
 validation may skip a succeeded aggregate record when the selected artifact
 closure is current even if unrelated records are stale. Conversely, a stale
 aggregate record may cross the ordinary runner boundary and perform zero
 Scriptorium calls when reconciliation proves every selected artifact current;
-the successful projection then restores the aggregate status.
+ordinary successful stage completion then restores the aggregate status.
 
-Analyze may return a projection together with an error. That restricted result
+On an artifact error, analyze installs a complete retry collection on the live
-cannot carry ordinary outputs, skip state, aggregate logs, generated configs,
+manifest before returning. The active record is failed without output, current
-or metadata. The runner persists only the validated per-artifact collections,
+transitive dependents are stale, unrelated current records survive, and only
-then marks the aggregate analyze state failed and invalidates delivery
+earlier checkpointed completions remain current. The runner's ordinary failure
-dependents conservatively. Unrelated current records survive because the
+path marks aggregate analyze failed, clears aggregate outputs, invalidates
-session projection is complete. A malformed projection is not applied, and a
+delivery dependents, and saves that collection. A terminal session-save failure
-failed session projection save restores the prior per-artifact authority before
+is joined with the original adapter, validation, filesystem, cancellation, or
-terminal failure persistence.
+checkpoint cause.
 
-The incremental executor constructs this restricted projection over its
+If an artifact checkpoint fails, analyze rolls its working collection back to
-scheduled artifact work. The active record is failed without output,
+the last committed collection before constructing the active failed record.
-current transitive dependents are stale, unrelated current records survive, and
+Newly materialized canonical bytes therefore remain incidental and cannot
-only earlier validated and materialized completions remain current in the
+become current authority. A later ordinary run reconciles the durable records,
-reconciled session collection. A terminal session-save failure is joined with
+reuses earlier current artifacts, and resumes with the failed or stale work.
-the original adapter, validation, or filesystem cause; a failed projection save
-does not turn incidental canonical bytes into manifest authority.
 
 ## Remote Commit Manifest
 

docs/internal/stage-analyze.md

@@ -89,9 +89,9 @@ Supported source families:
   rebuilt when required, while unrelated disabled artifacts are excluded.
 - the work plan carries explicit targets, prerequisite-only work, deterministic
   execution and reuse lists, invalidated and removed records, and a cloned
-  projected record collection. Valid unrelated configured records survive the
+  reconciled starting collection. Valid unrelated configured records survive
-  projection, removed records are omitted, and incidental files never become
+  reconciliation, removed records are omitted, and incidental files never
-  current without manifest-backed regeneration.
+  become current without manifest-backed regeneration.
 - implements aggregate resume validation by running the same read-only catalog,
   fingerprint reconciliation, and work planner used by execution. A succeeded
   aggregate record is reusable exactly when the selected closure schedules no
@@ -112,15 +112,16 @@ Supported source families:
 - 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.
+  analyze-owned manifest collection.
 - 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.
-- returns one complete reconciled session collection. The runner reconstructs
+- returns aggregate outputs from every current session record in normalized-key
-  aggregate session outputs from every current session record; unrelated stale
+  order after the complete reconciled collection has been checkpointed;
-  records do not make an otherwise successful partial invocation fail.
+  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.
@@ -135,20 +136,20 @@ Supported source families:
   guidance.
 - dependency cycles or unavailable required dependencies fail.
 - adapter validation failures fail stage.
-- a scheduled artifact failure returns the restricted analyze-state projection
+- a scheduled artifact failure installs a retry collection on the live manifest
   with the active artifact marked `failed`, a bounded error, and no output
   authority. Current transitive dependents become stale without execution.
 - earlier artifacts remain current only after their run-local output passed
-  validation and canonical materialization; unattempted later artifacts retain
+  validation, canonical materialization, and a successful complete-collection
-  their reconciled session state.
+  manifest checkpoint; unattempted later artifacts retain their reconciled
+  session state.
 - unrelated current records survive a partial failure. Old canonical bytes for
-  the failed artifact and newly materialized bytes whose projection cannot be
+  the failed artifact and newly materialized bytes whose checkpoint fails are
-  persisted are incidental, not current evidence.
+  incidental, not current evidence.
-- the runner applies one valid partial projection at the existing end-of-stage
+- the runner's ordinary error path marks aggregate analyze failed, clears its
-  save boundary before it marks aggregate analyze failed and invalidates publish
+  aggregate outputs, invalidates publish and notify through the application
-  and notify through the application dependency relation. Projection-persistence
+  dependency relation, and persists the live retry collection. Terminal-save
-  errors retain the last durable per-artifact authority and are joined with the
+  errors are joined with the original artifact or checkpoint failure.
-  original failure context. Per-artifact checkpoint persistence is future work.
 
 ## Invariants
 

docs/roadmap/implementation.md

@@ -318,7 +318,7 @@ stop using the end-of-stage projection transport for real analyze execution.
 
 ## Stage 4 — Complete Failure Recovery And Remove Projection Machinery
 
-**Status:** Pending
+**Status:** Completed
 
 ### Goal
 

internal/app/analyze_checkpoint_recovery_test.go

@@ -0,0 +1,329 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/scriptorium"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+	"gitea.maximumdirect.net/eric/narratio/internal/stage"
+)
+
+func TestExecuteStagesRetriesFromLastAnalyzeCheckpoint(t *testing.T) {
+	cfg := testConfig(t)
+	cfg.Pipeline.Scriptorium = &config.ScriptoriumConfig{Artifacts: checkpointRecoveryArtifacts("original")}
+	store := &manifest.LocalStore{}
+	seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	seed.Campaign = cfg.Session.Campaign
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim"} {
+		seed.MarkStageSucceeded(name, time.Now().UTC(), nil)
+	}
+	saveBoundedManifest(t, cfg, seed)
+	analyze := mustSelectStage(t, "analyze")
+
+	baselineRunner := &analyzeCheckpointRunner{}
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{analyze}, RunOptions{
+		Env: &Env{ManifestStore: store, Scriptorium: baselineRunner},
+	}); err != nil {
+		t.Fatalf("baseline executeStages() error = %v", err)
+	}
+	baseline, err := store.Load(context.Background(), manifestPathFor(cfg))
+	if err != nil {
+		t.Fatalf("load baseline manifest: %v", err)
+	}
+	baseline.MarkStageSucceeded("publish", time.Now().UTC(), nil)
+	baseline.MarkStageSucceeded("notify", time.Now().UTC(), nil)
+	if err := store.Save(context.Background(), manifestPathFor(cfg), baseline); err != nil {
+		t.Fatalf("save delivery baseline: %v", err)
+	}
+	upstream := cloneStageRecords(baseline, "prepare", "transcribe", "merge", "polish", "normalize", "trim")
+	unrelated := baseline.Stages["analyze"].AnalyzeArtifacts["unrelated"]
+
+	cfg.Pipeline.Scriptorium.Artifacts = checkpointRecoveryArtifacts("revised")
+	failure := errors.New("injected quest-log failure")
+	failingRunner := &analyzeCheckpointRunner{FailPrompt: "quest-log-revised", Err: failure}
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{analyze}, RunOptions{
+		Env: &Env{ManifestStore: store, Scriptorium: failingRunner},
+	}); !errors.Is(err, failure) {
+		t.Fatalf("executeStages() error = %v, want quest-log failure", err)
+	}
+	if want := []string{"chronicle-revised", "encounters-revised", "quest-log-revised"}; !reflect.DeepEqual(failingRunner.Calls, want) {
+		t.Fatalf("failed-run calls = %#v, want %#v", failingRunner.Calls, want)
+	}
+
+	failed := loadAnalyzeRecoveryManifest(t, store, cfg)
+	failedAnalyze := failed.Stages["analyze"]
+	if failedAnalyze.Status != manifest.StatusFailed || len(failedAnalyze.Outputs) != 0 {
+		t.Fatalf("aggregate analyze = %#v, want failed without outputs", failedAnalyze)
+	}
+	assertAnalyzeArtifactStatus(t, failedAnalyze, "chronicle", manifest.AnalyzeArtifactCurrent)
+	assertAnalyzeArtifactStatus(t, failedAnalyze, "encounters", manifest.AnalyzeArtifactCurrent)
+	assertAnalyzeArtifactStatus(t, failedAnalyze, "quest_log", manifest.AnalyzeArtifactFailed)
+	assertAnalyzeArtifactStatus(t, failedAnalyze, "player_handout", manifest.AnalyzeArtifactStale)
+	if got := failedAnalyze.AnalyzeArtifacts["unrelated"]; !reflect.DeepEqual(got, unrelated) {
+		t.Fatalf("unrelated artifact = %#v, want preserved %#v", got, unrelated)
+	}
+	for _, name := range []string{"publish", "notify"} {
+		if got := failed.Stages[name]; got == nil || got.Status != manifest.StatusStale {
+			t.Fatalf("%s stage = %#v, want stale", name, got)
+		}
+	}
+	assertStageRecordsUnchanged(t, failed, upstream)
+
+	retryRunner := &analyzeCheckpointRunner{}
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{analyze}, RunOptions{
+		Env: &Env{ManifestStore: store, Scriptorium: retryRunner},
+	}); err != nil {
+		t.Fatalf("retry executeStages() error = %v", err)
+	}
+	if want := []string{"quest-log-revised", "player-handout-revised"}; !reflect.DeepEqual(retryRunner.Calls, want) {
+		t.Fatalf("retry calls = %#v, want %#v", retryRunner.Calls, want)
+	}
+
+	completed := loadAnalyzeRecoveryManifest(t, store, cfg)
+	if completed.Stages["analyze"].Status != manifest.StatusSucceeded {
+		t.Fatalf("aggregate analyze = %#v, want succeeded", completed.Stages["analyze"])
+	}
+	for _, key := range []string{"chronicle", "encounters", "player_handout", "quest_log", "unrelated"} {
+		assertAnalyzeArtifactStatus(t, completed.Stages["analyze"], key, manifest.AnalyzeArtifactCurrent)
+	}
+	assertStageRecordsUnchanged(t, completed, upstream)
+}
+
+func TestExecuteStagesJoinsAnalyzeFailureWithTerminalSaveFailure(t *testing.T) {
+	cfg := testConfig(t)
+	cfg.Pipeline.Scriptorium = &config.ScriptoriumConfig{
+		Artifacts: map[string]config.ScriptoriumArtifactConfig{
+			"chronicle": {Enabled: true, PromptID: "chronicle", OutputPath: "artifacts/chronicle.md"},
+		},
+	}
+	artifactErr := errors.New("injected analyze failure")
+	terminalErr := errors.New("injected terminal save failure")
+	store := &failedAnalyzeSaveStore{Store: &manifest.LocalStore{}, Err: terminalErr}
+	runner := &analyzeCheckpointRunner{FailPrompt: "chronicle", Err: artifactErr}
+
+	_, err := executeStages(context.Background(), cfg, []stage.Stage{mustSelectStage(t, "analyze")}, RunOptions{
+		Env: &Env{ManifestStore: store, Scriptorium: runner},
+	})
+	if !errors.Is(err, artifactErr) || !errors.Is(err, terminalErr) {
+		t.Fatalf("executeStages() error = %v, want analyze and terminal save failures", err)
+	}
+	if !store.Failed {
+		t.Fatal("terminal save failure was not injected")
+	}
+}
+
+func TestExecuteStagesPersistsAnalyzeCheckpointFailureRecovery(t *testing.T) {
+	cfg := testConfig(t)
+	cfg.Pipeline.Scriptorium = &config.ScriptoriumConfig{
+		Artifacts: map[string]config.ScriptoriumArtifactConfig{
+			"chronicle":  {Enabled: true, PromptID: "chronicle", OutputPath: "artifacts/chronicle.md"},
+			"encounters": {Enabled: true, PromptID: "encounters", OutputPath: "artifacts/encounters.md"},
+			"quest_log":  {Enabled: true, PromptID: "quest-log", OutputPath: "artifacts/quest_log.md"},
+		},
+	}
+	baseStore := &manifest.LocalStore{}
+	seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	seed.Campaign = cfg.Session.Campaign
+	seed.MarkStageSucceeded("publish", time.Now().UTC(), nil)
+	seed.MarkStageSucceeded("notify", time.Now().UTC(), nil)
+	saveBoundedManifest(t, cfg, seed)
+	checkpointErr := errors.New("injected artifact checkpoint failure")
+	store := &currentAnalyzeCheckpointFailingStore{
+		Store:        baseStore,
+		CurrentLimit: 2,
+		Err:          checkpointErr,
+	}
+	runner := &analyzeCheckpointRunner{}
+
+	_, err := executeStages(context.Background(), cfg, []stage.Stage{mustSelectStage(t, "analyze")}, RunOptions{
+		Env: &Env{ManifestStore: store, Scriptorium: runner},
+	})
+	if !errors.Is(err, checkpointErr) {
+		t.Fatalf("executeStages() error = %v, want checkpoint failure", err)
+	}
+	if want := []string{"chronicle", "encounters"}; !reflect.DeepEqual(runner.Calls, want) {
+		t.Fatalf("adapter calls = %#v, want %#v", runner.Calls, want)
+	}
+
+	persisted := loadAnalyzeRecoveryManifest(t, baseStore, cfg)
+	if persisted.Stages["analyze"].Status != manifest.StatusFailed {
+		t.Fatalf("aggregate analyze = %#v, want failed", persisted.Stages["analyze"])
+	}
+	assertAnalyzeArtifactStatus(t, persisted.Stages["analyze"], "chronicle", manifest.AnalyzeArtifactCurrent)
+	assertAnalyzeArtifactStatus(t, persisted.Stages["analyze"], "encounters", manifest.AnalyzeArtifactFailed)
+	if _, exists := persisted.Stages["analyze"].AnalyzeArtifacts["quest_log"]; exists {
+		t.Fatalf("unattempted quest log appeared: %#v", persisted.Stages["analyze"].AnalyzeArtifacts["quest_log"])
+	}
+	for _, name := range []string{"publish", "notify"} {
+		if got := persisted.Stages[name]; got == nil || got.Status != manifest.StatusStale {
+			t.Fatalf("%s stage = %#v, want stale", name, got)
+		}
+	}
+}
+
+type analyzeCheckpointRunner struct {
+	Calls      []string
+	FailPrompt string
+	Err        error
+}
+
+type failedAnalyzeSaveStore struct {
+	manifest.Store
+	Err    error
+	Failed bool
+}
+
+type currentAnalyzeCheckpointFailingStore struct {
+	manifest.Store
+	CurrentLimit int
+	Err          error
+	Failed       bool
+}
+
+func (s *currentAnalyzeCheckpointFailingStore) Save(ctx context.Context, path string, m *manifest.Manifest) error {
+	if !s.Failed {
+		current := 0
+		if record := m.Stages["analyze"]; record != nil && record.Status == manifest.StatusRunning {
+			for _, artifact := range record.AnalyzeArtifacts {
+				if artifact.Status == manifest.AnalyzeArtifactCurrent {
+					current++
+				}
+			}
+		}
+		if current >= s.CurrentLimit {
+			s.Failed = true
+			return s.Err
+		}
+	}
+	return s.Store.Save(ctx, path, m)
+}
+
+func (s *failedAnalyzeSaveStore) Save(ctx context.Context, path string, m *manifest.Manifest) error {
+	if record := m.Stages["analyze"]; !s.Failed && record != nil && record.Status == manifest.StatusFailed {
+		s.Failed = true
+		return s.Err
+	}
+	return s.Store.Save(ctx, path, m)
+}
+
+func (r *analyzeCheckpointRunner) RunArtifact(
+	ctx context.Context,
+	req scriptorium.RunArtifactRequest,
+) (scriptorium.ArtifactResult, error) {
+	if err := ctx.Err(); err != nil {
+		return scriptorium.ArtifactResult{}, err
+	}
+	r.Calls = append(r.Calls, req.PromptID)
+	if req.PromptID == r.FailPrompt {
+		return scriptorium.ArtifactResult{}, r.Err
+	}
+	if err := os.MkdirAll(filepath.Dir(req.OutputPath), 0o755); err != nil {
+		return scriptorium.ArtifactResult{}, err
+	}
+	if err := os.WriteFile(req.OutputPath, []byte("generated "+req.PromptID+"\n"), 0o644); err != nil {
+		return scriptorium.ArtifactResult{}, err
+	}
+	return scriptorium.ArtifactResult{
+		OutputPath:  req.OutputPath,
+		CommandMode: scriptorium.CommandModeRun,
+		PromptID:    req.PromptID,
+		ProfileID:   req.ProfileID,
+	}, nil
+}
+
+func (*analyzeCheckpointRunner) RenderArtifact(
+	context.Context,
+	scriptorium.RenderArtifactRequest,
+) (scriptorium.ArtifactResult, error) {
+	return scriptorium.ArtifactResult{}, errors.New("unexpected render")
+}
+
+func checkpointRecoveryArtifacts(revision string) map[string]config.ScriptoriumArtifactConfig {
+	return map[string]config.ScriptoriumArtifactConfig{
+		"chronicle": {
+			Enabled: true, PromptID: "chronicle-" + revision, OutputPath: "artifacts/chronicle.md",
+		},
+		"encounters": {
+			Enabled: true, DependsOn: []string{"chronicle"}, PromptID: "encounters-" + revision,
+			OutputPath: "artifacts/encounters.md",
+			Inputs: map[string]config.ScriptoriumInputConfig{
+				"chronicle": {Source: "narratio.artifact.chronicle", Required: true},
+			},
+		},
+		"quest_log": {
+			Enabled: true, DependsOn: []string{"encounters"}, PromptID: "quest-log-" + revision,
+			OutputPath: "artifacts/quest_log.md",
+			Inputs: map[string]config.ScriptoriumInputConfig{
+				"encounters": {Source: "narratio.artifact.encounters", Required: true},
+			},
+		},
+		"player_handout": {
+			Enabled: true, DependsOn: []string{"quest_log"}, PromptID: "player-handout-" + revision,
+			OutputPath: "artifacts/player_handout.md",
+			Inputs: map[string]config.ScriptoriumInputConfig{
+				"quest_log": {Source: "narratio.artifact.quest_log", Required: true},
+			},
+		},
+		"unrelated": {
+			Enabled: true, PromptID: "unrelated", OutputPath: "artifacts/unrelated.md",
+		},
+	}
+}
+
+func mustSelectStage(t *testing.T, name string) stage.Stage {
+	t.Helper()
+	selected, err := stage.Select(name)
+	if err != nil {
+		t.Fatalf("Select(%q) error = %v", name, err)
+	}
+	return selected
+}
+
+func loadAnalyzeRecoveryManifest(t *testing.T, store manifest.Store, cfg *config.Config) *manifest.Manifest {
+	t.Helper()
+	m, err := store.Load(context.Background(), manifestPathFor(cfg))
+	if err != nil {
+		t.Fatalf("load session manifest: %v", err)
+	}
+	return m
+}
+
+func assertAnalyzeArtifactStatus(
+	t *testing.T,
+	record *manifest.StageRecord,
+	key string,
+	want manifest.AnalyzeArtifactStatus,
+) {
+	t.Helper()
+	got, exists := record.AnalyzeArtifacts[key]
+	if !exists || got.Status != want {
+		t.Fatalf("artifact %q = %#v, want status %q", key, got, want)
+	}
+	if want != manifest.AnalyzeArtifactCurrent && got.Output != nil {
+		t.Fatalf("artifact %q retained output authority: %#v", key, got)
+	}
+}
+
+func cloneStageRecords(m *manifest.Manifest, names ...string) map[string]manifest.StageRecord {
+	clones := make(map[string]manifest.StageRecord, len(names))
+	for _, name := range names {
+		clones[name] = *m.Stages[name]
+	}
+	return clones
+}
+
+func assertStageRecordsUnchanged(t *testing.T, m *manifest.Manifest, want map[string]manifest.StageRecord) {
+	t.Helper()
+	for name, expected := range want {
+		if got := m.Stages[name]; got == nil || !reflect.DeepEqual(*got, expected) {
+			t.Fatalf("upstream stage %q = %#v, want unchanged %#v", name, got, expected)
+		}
+	}
+}

internal/app/analyze_projection.go

@@ -1,110 +0,0 @@
-package app
-
-import (
-	"fmt"
-	"sort"
-
-	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
-	"gitea.maximumdirect.net/eric/narratio/internal/stage"
-)
-
-type validatedAnalyzeProjection struct {
-	session map[string]manifest.AnalyzeArtifactRecord
-}
-
-type analyzeStateSnapshot struct {
-	version int
-	records map[string]manifest.AnalyzeArtifactRecord
-}
-
-func captureAnalyzeState(manifestValue *manifest.Manifest, stageName string) analyzeStateSnapshot {
-	if manifestValue == nil || stageName != "analyze" || manifestValue.Stages["analyze"] == nil {
-		return analyzeStateSnapshot{}
-	}
-	record := manifestValue.Stages["analyze"]
-	return analyzeStateSnapshot{
-		version: record.AnalyzeStateVersion,
-		records: manifest.CloneAnalyzeArtifactCollection(record.AnalyzeArtifacts),
-	}
-}
-
-func restoreAnalyzeState(manifestValue *manifest.Manifest, snapshot analyzeStateSnapshot) {
-	if manifestValue == nil || manifestValue.Stages["analyze"] == nil {
-		return
-	}
-	record := manifestValue.Stages["analyze"]
-	record.AnalyzeStateVersion = snapshot.version
-	record.AnalyzeArtifacts = manifest.CloneAnalyzeArtifactCollection(snapshot.records)
-}
-
-func validateSuccessfulAnalyzeProjection(stageName string, result *stage.StageResult) (*validatedAnalyzeProjection, error) {
-	if result == nil || result.AnalyzeState == nil {
-		return nil, nil
-	}
-	if stageName != "analyze" {
-		return nil, fmt.Errorf("stage %q returned analyze-owned state projection", stageName)
-	}
-	if result.Disposition == stage.StageDispositionSkipped {
-		return nil, fmt.Errorf("skipped analyze result cannot contain analyze-owned state projection")
-	}
-	if len(result.Outputs) != 0 {
-		return nil, fmt.Errorf("analyze result with state projection cannot contain ordinary outputs")
-	}
-	return validateAndCloneAnalyzeProjection(result.AnalyzeState)
-}
-
-func validateFailedAnalyzeProjection(stageName string, result *stage.StageResult) (*validatedAnalyzeProjection, error) {
-	if result == nil || result.AnalyzeState == nil {
-		return nil, nil
-	}
-	if stageName != "analyze" {
-		return nil, fmt.Errorf("stage %q returned analyze-owned state projection with an error", stageName)
-	}
-	if result.Disposition != stage.StageDispositionSucceeded || result.SkipReason != "" || len(result.Outputs) != 0 || len(result.Logs) != 0 || len(result.GeneratedConfigs) != 0 || len(result.Metadata) != 0 {
-		return nil, fmt.Errorf("analyze result with an error may contain only analyze-owned state projection")
-	}
-	return validateAndCloneAnalyzeProjection(result.AnalyzeState)
-}
-
-func validateAndCloneAnalyzeProjection(projection *stage.AnalyzeStateProjection) (*validatedAnalyzeProjection, error) {
-	session := manifest.CloneAnalyzeArtifactCollection(projection.Session)
-	if err := manifest.ValidateAnalyzeArtifactCollection(manifest.AnalyzeStateContractVersion, session); err != nil {
-		return nil, fmt.Errorf("validate reconciled session analyze state: %w", err)
-	}
-	return &validatedAnalyzeProjection{session: session}, nil
-}
-
-func applyAnalyzeProjection(
-	sessionManifest *manifest.Manifest,
-	projection *validatedAnalyzeProjection,
-) {
-	if projection == nil {
-		return
-	}
-	if sessionManifest != nil && sessionManifest.Stages["analyze"] != nil {
-		record := sessionManifest.Stages["analyze"]
-		record.AnalyzeStateVersion = manifest.AnalyzeStateContractVersion
-		record.AnalyzeArtifacts = manifest.CloneAnalyzeArtifactCollection(projection.session)
-	}
-}
-
-func analyzeProjectionOutputs(records map[string]manifest.AnalyzeArtifactRecord) []manifest.ArtifactRecord {
-	keys := make([]string, 0, len(records))
-	for key, record := range records {
-		if record.Status != manifest.AnalyzeArtifactCurrent || record.Output == nil {
-			continue
-		}
-		keys = append(keys, key)
-	}
-	sort.Strings(keys)
-	outputs := make([]manifest.ArtifactRecord, 0, len(keys))
-	for _, key := range keys {
-		record := manifest.CloneAnalyzeArtifactCollection(map[string]manifest.AnalyzeArtifactRecord{key: records[key]})[key]
-		output := *record.Output
-		if output.ProducerRunID == "" {
-			output.ProducerRunID = record.ProducerRunID
-		}
-		outputs = append(outputs, output)
-	}
-	return outputs
-}

internal/app/analyze_projection_test.go

@@ -1,309 +0,0 @@
-package app
-
-import (
-	"context"
-	"errors"
-	"reflect"
-	"strings"
-	"testing"
-	"time"
-
-	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
-	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
-	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
-	"gitea.maximumdirect.net/eric/narratio/internal/stage"
-)
-
-type projectionStage struct {
-	name string
-	run  func(*stage.Env, *manifest.Manifest) (*stage.StageResult, error)
-}
-
-func (s projectionStage) Name() string { return s.name }
-func (s projectionStage) Run(_ context.Context, env *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-	return s.run(env, m)
-}
-
-func TestExecuteStagesProjectsReconciledSessionAnalyzeState(t *testing.T) {
-	cfg := testConfig(t)
-	oldAt := time.Date(2026, 5, 3, 9, 0, 0, 0, time.UTC)
-	oldRecord := appAnalyzeRecord("player_handout", manifest.AnalyzeArtifactCurrent, "old-run", oldAt)
-
-	stageToRun := projectionStage{name: "analyze", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		newRecord := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		staleRecord := appAnalyzeRecord("quest_log", manifest.AnalyzeArtifactStale, m.RunID, time.Now().UTC())
-		session := map[string]manifest.AnalyzeArtifactRecord{
-			"player_handout": oldRecord,
-			"quest_log":      staleRecord,
-			"session_recap":  newRecord,
-		}
-		return &stage.StageResult{
-			Logs: []string{"aggregate-analyze.log"},
-			AnalyzeState: &stage.AnalyzeStateProjection{
-				Session: session,
-			},
-		}, nil
-	}}
-
-	summary, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{})
-	if err != nil {
-		t.Fatalf("executeStages() error = %v", err)
-	}
-	store := &manifest.LocalStore{}
-	sessionManifest, err := store.Load(context.Background(), summary.ManifestPath)
-	if err != nil {
-		t.Fatalf("Load(session) error = %v", err)
-	}
-	analyze := sessionManifest.Stages["analyze"]
-	if analyze.AnalyzeStateVersion != manifest.AnalyzeStateContractVersion || len(analyze.AnalyzeArtifacts) != 3 {
-		t.Fatalf("session analyze state = %#v", analyze)
-	}
-	if got := analyzeArtifactOutputKeys(analyze.Outputs); !reflect.DeepEqual(got, []string{"player_handout", "session_recap"}) {
-		t.Fatalf("session aggregate outputs = %#v, want current records only", got)
-	}
-	if len(analyze.Logs) != 1 || analyze.Logs[0] != "aggregate-analyze.log" {
-		t.Fatalf("session aggregate logs = %#v", analyze.Logs)
-	}
-
-}
-
-func TestExecuteStagesPersistsRestrictedAnalyzeStateOnPartialError(t *testing.T) {
-	cfg := testConfig(t)
-	now := time.Date(2026, 5, 3, 9, 0, 0, 0, time.UTC)
-	seed := manifest.New(cfg.Session.SessionID, now)
-	seed.Campaign = cfg.Session.Campaign
-	seed.MarkStageSucceeded("analyze", now, nil)
-	seed.Stages["analyze"].AnalyzeStateVersion = manifest.AnalyzeStateContractVersion
-	seed.Stages["analyze"].AnalyzeArtifacts = map[string]manifest.AnalyzeArtifactRecord{
-		"player_handout": appAnalyzeRecord("player_handout", manifest.AnalyzeArtifactCurrent, "old-run", now),
-	}
-	seed.MarkStageSucceeded("publish", now, nil)
-	saveBoundedManifest(t, cfg, seed)
-
-	stageToRun := projectionStage{name: "analyze", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		unrelated := seed.Stages["analyze"].AnalyzeArtifacts["player_handout"]
-		completed := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		failed := appAnalyzeRecord("quest_log", manifest.AnalyzeArtifactFailed, m.RunID, time.Now().UTC())
-		session := map[string]manifest.AnalyzeArtifactRecord{
-			"player_handout": unrelated,
-			"quest_log":      failed,
-			"session_recap":  completed,
-		}
-		return &stage.StageResult{AnalyzeState: &stage.AnalyzeStateProjection{
-			Session: session,
-		}}, errors.New("quest log failed")
-	}}
-
-	_, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{Force: true})
-	if err == nil || !strings.Contains(err.Error(), "quest log failed") {
-		t.Fatalf("executeStages() error = %v", err)
-	}
-	store := &manifest.LocalStore{}
-	loaded, err := store.Load(context.Background(), manifestPathFor(cfg))
-	if err != nil {
-		t.Fatalf("Load(session) error = %v", err)
-	}
-	analyze := loaded.Stages["analyze"]
-	if analyze.Status != manifest.StatusFailed || len(analyze.Outputs) != 0 {
-		t.Fatalf("aggregate analyze state = %#v, want failed without outputs", analyze)
-	}
-	if analyze.AnalyzeArtifacts["player_handout"].Status != manifest.AnalyzeArtifactCurrent ||
-		analyze.AnalyzeArtifacts["session_recap"].Status != manifest.AnalyzeArtifactCurrent ||
-		analyze.AnalyzeArtifacts["quest_log"].Status != manifest.AnalyzeArtifactFailed {
-		t.Fatalf("partial session projection = %#v", analyze.AnalyzeArtifacts)
-	}
-	if loaded.Stages["publish"].Status != manifest.StatusStale {
-		t.Fatalf("publish status = %q, want stale", loaded.Stages["publish"].Status)
-	}
-
-}
-
-func TestExecuteStagesRejectsInvalidAnalyzeProjectionWithoutReplacingPriorState(t *testing.T) {
-	cfg := testConfig(t)
-	now := time.Date(2026, 5, 3, 9, 0, 0, 0, time.UTC)
-	prior := appAnalyzeRecord("player_handout", manifest.AnalyzeArtifactCurrent, "old-run", now)
-	seed := manifest.New(cfg.Session.SessionID, now)
-	seed.Campaign = cfg.Session.Campaign
-	seed.MarkStageSucceeded("analyze", now, nil)
-	seed.Stages["analyze"].AnalyzeStateVersion = manifest.AnalyzeStateContractVersion
-	seed.Stages["analyze"].AnalyzeArtifacts = map[string]manifest.AnalyzeArtifactRecord{"player_handout": prior}
-	saveBoundedManifest(t, cfg, seed)
-
-	stageToRun := projectionStage{name: "analyze", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		invalid := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		invalid.Output.Checksum = "invalid"
-		return &stage.StageResult{AnalyzeState: &stage.AnalyzeStateProjection{
-			Session: map[string]manifest.AnalyzeArtifactRecord{"session_recap": invalid},
-		}}, errors.New("analysis failed")
-	}}
-
-	_, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{Force: true})
-	if err == nil || !strings.Contains(err.Error(), "checksum") {
-		t.Fatalf("executeStages() error = %v, want projection validation failure", err)
-	}
-	loaded, loadErr := (&manifest.LocalStore{}).Load(context.Background(), manifestPathFor(cfg))
-	if loadErr != nil {
-		t.Fatalf("Load() error = %v", loadErr)
-	}
-	if len(loaded.Stages["analyze"].AnalyzeArtifacts) != 1 || !reflect.DeepEqual(loaded.Stages["analyze"].AnalyzeArtifacts["player_handout"], prior) {
-		t.Fatalf("prior state replaced by invalid projection: %#v", loaded.Stages["analyze"].AnalyzeArtifacts)
-	}
-}
-
-func TestExecuteStagesRollsBackAnalyzeAuthorityWhenProjectionSaveFails(t *testing.T) {
-	cfg := testConfig(t)
-	now := time.Date(2026, 5, 3, 9, 0, 0, 0, time.UTC)
-	prior := appAnalyzeRecord("player_handout", manifest.AnalyzeArtifactCurrent, "old-run", now)
-	seed := manifest.New(cfg.Session.SessionID, now)
-	seed.Campaign = cfg.Session.Campaign
-	seed.MarkStageSucceeded("analyze", now, nil)
-	seed.Stages["analyze"].AnalyzeStateVersion = manifest.AnalyzeStateContractVersion
-	seed.Stages["analyze"].AnalyzeArtifacts = map[string]manifest.AnalyzeArtifactRecord{"player_handout": prior}
-	saveBoundedManifest(t, cfg, seed)
-
-	stageReturned := false
-	store := &analyzeProjectionFailingStore{delegate: &manifest.LocalStore{}, shouldFail: func(m *manifest.Manifest) bool {
-		return stageReturned && m.Stages["analyze"] != nil && m.Stages["analyze"].Status == manifest.StatusSucceeded && m.Stages["analyze"].AnalyzeArtifacts["session_recap"].Status == manifest.AnalyzeArtifactCurrent
-	}}
-	stageToRun := projectionStage{name: "analyze", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		stageReturned = true
-		newRecord := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		return &stage.StageResult{AnalyzeState: &stage.AnalyzeStateProjection{
-			Session: map[string]manifest.AnalyzeArtifactRecord{
-				"player_handout": prior,
-				"session_recap":  newRecord,
-			},
-		}}, nil
-	}}
-
-	_, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{
-		Force: true,
-		Env:   &Env{ManifestStore: store},
-	})
-	if err == nil || !strings.Contains(err.Error(), "injected analyze projection save failure") {
-		t.Fatalf("executeStages() error = %v", err)
-	}
-	if !store.failed {
-		t.Fatal("projection persistence failure was not injected")
-	}
-	loaded, loadErr := store.delegate.Load(context.Background(), manifestPathFor(cfg))
-	if loadErr != nil {
-		t.Fatalf("Load() error = %v", loadErr)
-	}
-	analyze := loaded.Stages["analyze"]
-	if analyze.Status != manifest.StatusFailed || len(analyze.AnalyzeArtifacts) != 1 || !reflect.DeepEqual(analyze.AnalyzeArtifacts["player_handout"], prior) {
-		t.Fatalf("durable analyze state after rollback = %#v", analyze)
-	}
-}
-
-func TestExecuteStagesRejectsAnalyzeProjectionFromOtherStage(t *testing.T) {
-	cfg := testConfig(t)
-	stageToRun := projectionStage{name: "prepare", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		record := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		return &stage.StageResult{AnalyzeState: &stage.AnalyzeStateProjection{
-			Session: map[string]manifest.AnalyzeArtifactRecord{"session_recap": record},
-		}}, nil
-	}}
-	_, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{})
-	if err == nil || !strings.Contains(err.Error(), "returned analyze-owned state projection") {
-		t.Fatalf("executeStages() error = %v", err)
-	}
-}
-
-func TestExecuteStagesRejectsContradictoryAnalyzeResultWithError(t *testing.T) {
-	cfg := testConfig(t)
-	stageToRun := projectionStage{name: "analyze", run: func(_ *stage.Env, m *manifest.Manifest) (*stage.StageResult, error) {
-		record := appAnalyzeRecord("session_recap", manifest.AnalyzeArtifactCurrent, m.RunID, time.Now().UTC())
-		return &stage.StageResult{
-			Outputs: []artifacts.Ref{{Kind: "session_recap", RelativePath: "artifacts/session-recap.md"}},
-			AnalyzeState: &stage.AnalyzeStateProjection{
-				Session: map[string]manifest.AnalyzeArtifactRecord{"session_recap": record},
-			},
-		}, errors.New("analysis failed")
-	}}
-	_, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{})
-	if err == nil || !strings.Contains(err.Error(), "may contain only analyze-owned state projection") {
-		t.Fatalf("executeStages() error = %v", err)
-	}
-}
-
-func TestExecuteStagesExposesSelectedForceDecisionToStage(t *testing.T) {
-	for _, force := range []bool{false, true} {
-		t.Run(strings.ToLower(strings.TrimSpace(map[bool]string{false: "ordinary", true: "forced"}[force])), func(t *testing.T) {
-			cfg := testConfig(t)
-			captured := !force
-			stageToRun := projectionStage{name: "prepare", run: func(env *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
-				captured = env.Force
-				return &stage.StageResult{}, nil
-			}}
-			if _, err := executeStages(context.Background(), cfg, []stage.Stage{stageToRun}, RunOptions{Force: force}); err != nil {
-				t.Fatalf("executeStages() error = %v", err)
-			}
-			if captured != force {
-				t.Fatalf("stage env force = %v, want %v", captured, force)
-			}
-		})
-	}
-}
-
-func appAnalyzeRecord(key string, status manifest.AnalyzeArtifactStatus, producerRunID string, at time.Time) manifest.AnalyzeArtifactRecord {
-	record := manifest.AnalyzeArtifactRecord{
-		Key:           key,
-		Status:        status,
-		ProducerRunID: producerRunID,
-		UpdatedAt:     at,
-	}
-	if status == manifest.AnalyzeArtifactFailed {
-		record.Error = "scriptorium failed"
-		return record
-	}
-	if status != manifest.AnalyzeArtifactCurrent {
-		record.FingerprintVersion = manifest.AnalyzeFingerprintContractVersion
-		record.Fingerprint = strings.Repeat("b", 64)
-		return record
-	}
-	record.FingerprintVersion = manifest.AnalyzeFingerprintContractVersion
-	record.Fingerprint = strings.Repeat("a", 64)
-	record.OutputSize = 42
-	record.Output = &manifest.ArtifactRecord{
-		Kind:          key,
-		SourceID:      artifacts.ConfiguredArtifactSourceID(key),
-		LocalPath:     "artifacts/" + strings.ReplaceAll(key, "_", "-") + ".md",
-		ProducerRunID: producerRunID,
-		Checksum:      strings.Repeat("c", 64),
-		Contract: &artifactmodel.ContractMetadata{
-			MediaType: "text/markdown", SchemaID: "narratio." + key, SchemaVersion: "1",
-		},
-	}
-	return record
-}
-
-func analyzeArtifactOutputKeys(outputs []manifest.ArtifactRecord) []string {
-	keys := make([]string, 0, len(outputs))
-	for _, output := range outputs {
-		keys = append(keys, strings.TrimPrefix(output.SourceID, "narratio.artifact."))
-	}
-	return keys
-}
-
-type analyzeProjectionFailingStore struct {
-	delegate   *manifest.LocalStore
-	shouldFail func(*manifest.Manifest) bool
-	failed     bool
-}
-
-func (s *analyzeProjectionFailingStore) Create(ctx context.Context, sessionID string) (*manifest.Manifest, error) {
-	return s.delegate.Create(ctx, sessionID)
-}
-
-func (s *analyzeProjectionFailingStore) Load(ctx context.Context, path string) (*manifest.Manifest, error) {
-	return s.delegate.Load(ctx, path)
-}
-
-func (s *analyzeProjectionFailingStore) Save(ctx context.Context, path string, m *manifest.Manifest) error {
-	if !s.failed && s.shouldFail != nil && s.shouldFail(m) {
-		s.failed = true
-		return errors.New("injected analyze projection save failure")
-	}
-	return s.delegate.Save(ctx, path, m)
-}

internal/app/runner_stage_execution.go

@@ -55,7 +55,6 @@ func executeSelectedStage(runtime *invocationRuntime, selected stage.Stage) (sta
 	}
 
 	priorOutcome := capturePriorStageOutcome(m, selected.Name())
-	priorAnalyzeState := captureAnalyzeState(m, selected.Name())
 	attempted := stageExecutionOutcome{executed: true}
 	startedAt := runtime.dependencies.now()
 	if _, err := beginStageAttempt(m, selected.Name(), decision, startedAt); err != nil {
@@ -75,23 +74,10 @@ func executeSelectedStage(runtime *invocationRuntime, selected stage.Stage) (sta
 	env.Logger.Debug("manifest saved", "stage", selected.Name(), "transition", "running", "path", runtime.manifestPath)
 
 	result, err := selected.Run(runtime.ctx, env, m)
-	var analyzeProjection *validatedAnalyzeProjection
 	if err == nil {
 		err = validateStageResult(result)
-		if err == nil {
-			analyzeProjection, err = validateSuccessfulAnalyzeProjection(selected.Name(), result)
-		}
-	} else if result != nil && result.AnalyzeState != nil {
-		var projectionErr error
-		analyzeProjection, projectionErr = validateFailedAnalyzeProjection(selected.Name(), result)
-		if projectionErr != nil {
-			err = errors.Join(err, projectionErr)
-		}
 	}
 	if err != nil {
-		if analyzeProjection != nil {
-			applyAnalyzeProjection(m, analyzeProjection)
-		}
 		failedAt := runtime.dependencies.now()
 		m.MarkStageFailed(selected.Name(), failedAt, err.Error())
 		if _, invalidationErr := invalidateDependentSucceededStagesWithReason(m, selected.Name(), failedAt, staleReasonFailure); invalidationErr != nil {
@@ -119,22 +105,14 @@ func executeSelectedStage(runtime *invocationRuntime, selected stage.Stage) (sta
 	}
 
 	outputs := mapResultOutputs(selected.Name(), result, runtime.identity.RunID)
-	if analyzeProjection != nil {
-		outputs = analyzeProjectionOutputs(analyzeProjection.session)
-	}
 	succeededAt := runtime.dependencies.now()
 	if _, err := applyCompletedStageOutcome(m, selected.Name(), result, outputs, semanticConfig, priorOutcome, succeededAt); err != nil {
 		return stageExecutionOutcome{}, runtime.persistTerminalFailure(
 			fmt.Errorf("invalidate dependents after changed stage %q result: %w", selected.Name(), err),
 		)
 	}
-	applyAnalyzeProjection(m, analyzeProjection)
-
 	if err := runtime.manifestStore.Save(runtime.ctx, runtime.manifestPath, m); err != nil {
 		operationErr := fmt.Errorf("save manifest after stage %q: %w", selected.Name(), err)
-		if analyzeProjection != nil {
-			restoreAnalyzeState(m, priorAnalyzeState)
-		}
 		failedAt := runtime.dependencies.now()
 		m.MarkStageFailed(selected.Name(), failedAt, operationErr.Error())
 		if _, invalidationErr := invalidateDependentSucceededStagesWithReason(m, selected.Name(), failedAt, staleReasonFailure); invalidationErr != nil {

internal/manifest/analyze_state.go

@@ -87,7 +87,7 @@ func ValidateAnalyzeArtifactCollection(version int, records map[string]AnalyzeAr
 }
 
 // CloneAnalyzeArtifactCollection returns a deep copy in canonical dependency
-// order so projections cannot be mutated after application.
+// order so candidate or installed authority cannot share mutable record data.
 func CloneAnalyzeArtifactCollection(records map[string]AnalyzeArtifactRecord) map[string]AnalyzeArtifactRecord {
 	if records == nil {
 		return nil

internal/stage/analyze_failure_test.go

@@ -48,10 +48,55 @@ func TestAnalyzeAdapterFailureRecordsCompletedAndFailedArtifacts(t *testing.T) {
 					}
 				}
 			}
+
+			manifestPath := artifacts.SessionManifestPathForCampaign(
+				env.Config.Pipeline.Workspace.Root, m.Campaign, m.SessionID,
+			)
+			durable, loadErr := env.ManifestStore.Load(context.Background(), manifestPath)
+			if loadErr != nil {
+				t.Fatalf("load durable manifest: %v", loadErr)
+			}
+			for index, key := range keys {
+				record, exists := durable.Stages["analyze"].AnalyzeArtifacts[key]
+				if index < failAt-1 {
+					if !exists || record.Status != manifest.AnalyzeArtifactCurrent {
+						t.Fatalf("durable completed %s = %#v", key, record)
+					}
+				} else if exists && record.Status == manifest.AnalyzeArtifactCurrent {
+					t.Fatalf("durable uncommitted %s = %#v", key, record)
+				}
+			}
 		})
 	}
 }
 
+func TestAnalyzeCancellationPreservesEarlierCheckpoint(t *testing.T) {
+	env, m, _ := setupAnalyzeEnv(t)
+	env.Config.Pipeline.Scriptorium.Artifacts = independentAnalyzeArtifacts()
+	env.Scriptorium = &indexedAnalyzeFailureRunner{FailAt: 2, Err: context.Canceled}
+
+	_, err := (analyzeStage{}).Run(context.Background(), env, m)
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("Run() error = %v, want cancellation", err)
+	}
+	manifestPath := artifacts.SessionManifestPathForCampaign(
+		env.Config.Pipeline.Workspace.Root, m.Campaign, m.SessionID,
+	)
+	durable, loadErr := env.ManifestStore.Load(context.Background(), manifestPath)
+	if loadErr != nil {
+		t.Fatalf("load durable manifest: %v", loadErr)
+	}
+	if got := durable.Stages["analyze"].AnalyzeArtifacts["alpha"]; got.Status != manifest.AnalyzeArtifactCurrent {
+		t.Fatalf("durable alpha = %#v, want current", got)
+	}
+	if _, exists := durable.Stages["analyze"].AnalyzeArtifacts["beta"]; exists {
+		t.Fatalf("durable beta should be absent after cancellation: %#v", durable.Stages["analyze"].AnalyzeArtifacts["beta"])
+	}
+	if got := m.Stages["analyze"].AnalyzeArtifacts["beta"]; got.Status != manifest.AnalyzeArtifactFailed || got.Output != nil {
+		t.Fatalf("live beta = %#v, want failed without output", got)
+	}
+}
+
 func TestAnalyzeMiddleFailureStalesDependentsAndPreservesUnrelatedCurrent(t *testing.T) {
 	env, m, _ := setupAnalyzeEnv(t)
 	env.Config.Pipeline.Scriptorium.Artifacts = chainedAnalyzeArtifacts()

internal/stage/analyze_incremental_execution_test.go

@@ -39,7 +39,7 @@ func TestAnalyzeReusesCurrentArtifactWithoutScriptorium(t *testing.T) {
 	}
 }
 
-func TestAnalyzePartialForcePreservesUnrelatedCurrentRecord(t *testing.T) {
+func TestAnalyzeIdenticalReplacementPreservesUnselectedDependent(t *testing.T) {
 	env, m, fake := setupAnalyzeEnv(t)
 	paths := sessionPathsForEnv(env, m.SessionID)
 	writeAnalyzeFile(t, filepath.Join(paths.TranscriptsDir, "final.trimmed.json"), `{"segments":[]}`)

internal/stage/stage.go

@@ -129,11 +129,4 @@ type StageResult struct {
 	Logs             []string
 	GeneratedConfigs []string
 	Metadata         map[string]any
-	AnalyzeState     *AnalyzeStateProjection
-}
-
-// AnalyzeStateProjection carries the reconciled session authority proposed for
-// the runner's existing end-of-stage session-manifest save.
-type AnalyzeStateProjection struct {
-	Session map[string]manifest.AnalyzeArtifactRecord
 }