Enforce bounded run prerequisites

docs/internal/adapters.md

@@ -35,16 +35,23 @@ Adapters do not own:
 
 ## Default Wiring
 
-`internal/app/runner.go` initializes default adapters when not injected:
+`internal/app/runner.go` initializes default adapters when not injected and
+only when the selected execution plan needs them:
 
-- WhisperX HTTP client from pipeline config.
-- Seriatim subprocess runner.
-- Audita subprocess runner.
-- Scriptorium subprocess runner.
-- Notarius subprocess runner when extraction is enabled.
-- Noop notifier (`notify.NoopSender`).
+- WhisperX HTTP client for `transcribe`.
+- Seriatim subprocess runner for `merge`, `normalize`, `trim`, or `render`.
+- Audita subprocess runner for `polish`.
+- Scriptorium subprocess runner for `trim` or `analyze`.
+- Notarius subprocess runner for `extract` when extraction is enabled.
+- Noop notifier (`notify.NoopSender`) for `notify`.
 - Object store only when required by selected stages/config.
 
+Remote publish locks are loaded only for a selected, enabled publish that
+uploads a run. Shared session lifecycle setup still applies to every selected
+range, but an unselected integration is neither initialized nor validated by
+runner composition. Each selected stage retains its own fail-fast configuration
+and input validation.
+
 Notarius is composed only when extraction is enabled; the extract stage owns
 prepared reference resolution, receipt, bundle, and configured-lane policy.
 The adapter validates the ordered selector/absolute-path pairs and is the sole

docs/internal/manifest.md

@@ -117,12 +117,20 @@ runner marks it stale and executes it.
 Session manifest is the authoritative stage-progress ledger across invocations.
 Run manifest is invocation-scoped audit state.
 
+Before an explicitly bounded execution starts after `prepare`, the application
+reads the session manifest and accepts only `succeeded` or `skipped` for every
+excluded canonical prefix stage. The first other status or absent record fails
+the request before layout mutation, adapter initialization, session-manifest
+writes, or run-manifest creation. Excluded prefix records are not passed to
+resume validators. Records after the selected end are not prerequisites and
+may be made stale by selected work without being scheduled.
+
 After a publish commits remotely, any configured local cleanup is first recorded
 as a session-manifest obligation before deletion begins. Each target becomes
 complete only after its confined deletion (or safe absence check) and a
-successful manifest save. An incomplete obligation is retried on later
-invocations independently of their selected stages and retains the committed
-run and remote identity that authorized it.
+successful manifest save. An incomplete obligation is retried when publish
+executes again and retains the committed run and remote identity that authorized
+it; an invocation that does not execute publish does not perform cleanup.
 
 Each invocation derives campaign, session, run, local-path, and remote-prefix
 metadata from the validated resolved configuration as one projection. A persisted

docs/operations.md

@@ -124,6 +124,24 @@ Omitting `--from` selects from `prepare`; omitting `--through` selects through
 The plan command uses the same selection contract and prints only the selected
 range.
 
+Before a bounded run or plan whose range starts after `prepare`, every excluded
+prefix stage must already have a session-manifest status of `succeeded` or
+`skipped`. Narratio reports the first absent, pending, running, failed, stale,
+or interrupted prerequisite without creating a run record or changing session
+state. Widen `--from` to include that stage, or recover it explicitly before
+retrying. Excluded prefix stages are not resume-validated or repaired as part
+of the bounded invocation; selected stages still reject missing, unsafe, or
+manifest-inconsistent inputs at their owning boundary.
+
+Stages after `--through` are not prerequisites and are never scheduled by the
+bounded invocation. A selected forced stage can mark one of those succeeded
+dependents stale through the fixed invalidation relation, but the dependent
+does not execute until a later invocation selects it. Production composition
+likewise initializes only collaborators needed by the selected range and
+shared session lifecycle. In particular, render does not require Notarius or
+Scriptorium, extract does not require Scriptorium, and analyze does not require
+the transcription, Seriatim, Audita, or Notarius adapters.
+
 ## Artifact Selection
 
 `--artifacts` can be used on `run`, `session plan`, `run-stage`, `analyze`, and
@@ -451,7 +469,11 @@ Rules:
   - `pipeline.workspace.cleanup_after_publish=true`
   - Narratio first records the exact run-scoped cleanup obligation. If cleanup
     reports incomplete, the remote committed snapshot remains current; rerun
-    Narratio to retry only the outstanding confined local cleanup.
+    publish to retry only the outstanding confined local cleanup.
+
+Post-publish cleanup is evaluated only when `publish` actually executes in the
+current invocation. A bounded range that excludes publish does not replay a
+cleanup obligation as an unrelated side effect.
 
 ## Operational Caveats
 

docs/roadmap/implementation.md

@@ -217,6 +217,8 @@ validation contract.
 
 ## Stage 4 — Bounded Runner Prerequisites And Composition
 
+**Status: Completed**
+
 ### Goal
 
 Make bounded execution honor its mutation boundary while still failing safely

docs/troubleshooting.md

@@ -117,6 +117,40 @@ Safe fix:
 
 Relevant reference: [CLI artifact selection](./cli.md).
 
+## Bounded run prerequisite is unusable
+
+Symptom:
+
+- `run` or `session plan` reports that a prerequisite stage is absent or has a
+  pending, running, failed, stale, or interrupted status before the selected
+  start.
+
+Likely cause:
+
+- `--from` excludes upstream work that has not reached the terminal
+  `succeeded` or `skipped` state in the session manifest.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session plan 2026-04-04 --from render --through analyze
+```
+
+Safe fix:
+
+- widen the bounded range to include the first reported stage, or recover that
+  stage explicitly with `run-stage` before retrying. The failed check does not
+  create a run record or modify the manifest. Narratio does not resume-validate
+  excluded prefix stages, and stages after `--through` are not prerequisites.
+
+If prerequisite statuses are terminal but a selected stage reports a missing,
+unsafe, or checksum-inconsistent artifact, repair the artifact at the stage
+that owns it; do not edit the manifest to bypass the selected stage's concrete
+input validation.
+
+Relevant reference: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior).
+
 ## Notarius executable missing
 
 Symptom:

internal/app/bounded_prerequisites.go

@@ -0,0 +1,62 @@
+package app
+
+import (
+	"context"
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+)
+
+func validateBoundedPrerequisites(plan BoundedPlan, m *manifest.Manifest) error {
+	if !plan.HasExplicitBounds() {
+		return nil
+	}
+	for _, name := range plan.PrefixNames() {
+		status := "absent"
+		if m != nil && m.Stages != nil && m.Stages[name] != nil {
+			stageStatus := m.Stages[name].Status
+			if stageStatus == manifest.StatusSucceeded || stageStatus == manifest.StatusSkipped {
+				continue
+			}
+			if stageStatus != "" {
+				status = string(stageStatus)
+			}
+		}
+		return fmt.Errorf(
+			"prerequisite stage %q has unusable status %q before selected start %q; widen the range with --from %s or recover %s explicitly",
+			name,
+			status,
+			plan.From(),
+			name,
+			name,
+		)
+	}
+	return nil
+}
+
+func inspectBoundedPrerequisites(ctx context.Context, cfg *config.Config, plan BoundedPlan, store manifest.Store) error {
+	if !plan.HasExplicitBounds() || len(plan.PrefixNames()) == 0 {
+		return nil
+	}
+	if cfg == nil || cfg.Pipeline == nil || cfg.Session == nil {
+		return fmt.Errorf("bounded prerequisite inspection requires resolved pipeline and session configuration")
+	}
+	if store == nil {
+		store = &manifest.LocalStore{}
+	}
+	path := artifacts.SessionManifestPathForCampaign(
+		cfg.Pipeline.Workspace.Root,
+		cfg.Session.Campaign,
+		cfg.Session.SessionID,
+	)
+	m, present, err := loadManifestAtPathIfPresent(ctx, store, path)
+	if err != nil {
+		return err
+	}
+	if !present {
+		m = nil
+	}
+	return validateBoundedPrerequisites(plan, m)
+}

internal/app/bounded_prerequisites_test.go

@@ -0,0 +1,288 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+	"gitea.maximumdirect.net/eric/narratio/internal/stage"
+)
+
+func TestValidateBoundedPrerequisitesRejectsFirstUnusablePrefixStatus(t *testing.T) {
+	plan := mustBoundedPlan(t, "render", "extract")
+	now := time.Date(2026, 5, 3, 10, 0, 0, 0, time.UTC)
+	for _, test := range []struct {
+		name   string
+		status manifest.StageStatus
+		want   string
+	}{
+		{name: "absent", want: "absent"},
+		{name: "pending", status: manifest.StatusPending, want: "pending"},
+		{name: "running", status: manifest.StatusRunning, want: "running"},
+		{name: "failed", status: manifest.StatusFailed, want: "failed"},
+		{name: "stale", status: manifest.StatusStale, want: "stale"},
+		{name: "interrupted", status: manifest.StatusInterrupted, want: "interrupted"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			m := manifest.New("session", now)
+			if test.status != "" {
+				m.Stages["prepare"] = &manifest.StageRecord{Name: "prepare", Status: test.status}
+			}
+			// A later terminal prefix must not hide the first unusable one.
+			m.MarkStageSucceeded("transcribe", now, nil)
+			err := validateBoundedPrerequisites(plan, m)
+			if err == nil {
+				t.Fatal("validateBoundedPrerequisites() error = nil")
+			}
+			for _, detail := range []string{`stage "prepare"`, `status "` + test.want + `"`, `selected start "render"`, "--from prepare", "recover prepare"} {
+				if !strings.Contains(err.Error(), detail) {
+					t.Fatalf("error = %q, want detail %q", err, detail)
+				}
+			}
+		})
+	}
+}
+
+func TestValidateBoundedPrerequisitesAcceptsSucceededAndSkippedPrefix(t *testing.T) {
+	plan := mustBoundedPlan(t, "render", "render")
+	m := manifest.New("session", time.Now().UTC())
+	for index, name := range plan.PrefixNames() {
+		if index%2 == 0 {
+			m.MarkStageSucceeded(name, time.Now().UTC(), nil)
+		} else {
+			m.MarkStageSkipped(name, time.Now().UTC(), "not applicable")
+		}
+	}
+	if err := validateBoundedPrerequisites(plan, m); err != nil {
+		t.Fatalf("validateBoundedPrerequisites() error = %v", err)
+	}
+}
+
+func TestValidateBoundedPrerequisitesHasNoPrefixAtPrepareAndIgnoresSuffix(t *testing.T) {
+	preparePlan := mustBoundedPlan(t, "prepare", "prepare")
+	if err := validateBoundedPrerequisites(preparePlan, nil); err != nil {
+		t.Fatalf("prepare prerequisite validation error = %v", err)
+	}
+
+	renderPlan := mustBoundedPlan(t, "render", "render")
+	m := manifest.New("session", time.Now().UTC())
+	markPrefixSucceeded(m, renderPlan)
+	m.MarkStageFailed("analyze", time.Now().UTC(), "later failure")
+	if err := validateBoundedPrerequisites(renderPlan, m); err != nil {
+		t.Fatalf("suffix status affected prerequisite validation: %v", err)
+	}
+}
+
+func TestExecuteStagesRejectsBoundedPrerequisitesBeforePersistentMutation(t *testing.T) {
+	cfg := testConfig(t)
+	plan := mustBoundedPlan(t, "render", "render")
+	m := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	m.Campaign = cfg.Session.Campaign
+	m.MarkStageRunning("prepare", time.Now().UTC())
+	manifestPath := saveBoundedManifest(t, cfg, m)
+	before, err := os.ReadFile(manifestPath)
+	if err != nil {
+		t.Fatalf("read seeded manifest: %v", err)
+	}
+
+	store := &prerequisiteMutationSpy{local: &manifest.LocalStore{}}
+	runs := 0
+	_, err = executeStages(context.Background(), cfg, []stage.Stage{countingStage{name: "render", runs: &runs}}, RunOptions{
+		Plan: plan,
+		Env:  &Env{ManifestStore: store},
+	})
+	if err == nil || !strings.Contains(err.Error(), `stage "prepare" has unusable status "running"`) {
+		t.Fatalf("executeStages() error = %v, want running prerequisite", err)
+	}
+	if runs != 0 || store.creates != 0 || store.saves != 0 {
+		t.Fatalf("runs=%d manifest creates=%d saves=%d, want no mutation", runs, store.creates, store.saves)
+	}
+	after, err := os.ReadFile(manifestPath)
+	if err != nil {
+		t.Fatalf("read manifest after rejection: %v", err)
+	}
+	if string(after) != string(before) {
+		t.Fatal("manifest changed after prerequisite rejection")
+	}
+	if _, err := os.Stat(artifacts.SessionRunsDirForCampaign(cfg.Pipeline.Workspace.Root, cfg.Session.Campaign, cfg.Session.SessionID)); !os.IsNotExist(err) {
+		t.Fatalf("runs directory stat error = %v, want not exist", err)
+	}
+}
+
+func TestExecuteStagesBoundedCompositionUsesOnlySelectedCollaborators(t *testing.T) {
+	for _, test := range []struct {
+		name        string
+		stageName   string
+		configure   func(*config.Config)
+		assertProbe func(*testing.T, *stage.Env)
+	}{
+		{
+			name:      "render",
+			stageName: "render",
+			assertProbe: func(t *testing.T, env *stage.Env) {
+				if env.Seriatim == nil || env.Notarius != nil || env.Scriptorium != nil {
+					t.Fatalf("render collaborators: seriatim=%v notarius=%v scriptorium=%v", env.Seriatim, env.Notarius, env.Scriptorium)
+				}
+			},
+		},
+		{
+			name:      "extract",
+			stageName: "extract",
+			configure: func(cfg *config.Config) {
+				cfg.Pipeline.Notarius = &config.NotariusConfig{Enabled: true}
+			},
+			assertProbe: func(t *testing.T, env *stage.Env) {
+				if env.Notarius == nil || env.Scriptorium != nil || env.Seriatim != nil {
+					t.Fatalf("extract collaborators: notarius=%v scriptorium=%v seriatim=%v", env.Notarius, env.Scriptorium, env.Seriatim)
+				}
+			},
+		},
+		{
+			name:      "analyze",
+			stageName: "analyze",
+			assertProbe: func(t *testing.T, env *stage.Env) {
+				if env.Scriptorium == nil || env.Notarius != nil || env.Seriatim != nil || env.WhisperX != nil || env.Audita != nil {
+					t.Fatalf("analyze collaborators: scriptorium=%v notarius=%v seriatim=%v whisperx=%v audita=%v", env.Scriptorium, env.Notarius, env.Seriatim, env.WhisperX, env.Audita)
+				}
+			},
+		},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			cfg := testConfig(t)
+			if test.configure != nil {
+				test.configure(cfg)
+			}
+			plan := mustBoundedPlan(t, test.stageName, test.stageName)
+			m := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+			m.Campaign = cfg.Session.Campaign
+			markPrefixSucceeded(m, plan)
+			saveBoundedManifest(t, cfg, m)
+
+			var captured *stage.Env
+			_, err := executeStages(context.Background(), cfg, []stage.Stage{collaboratorProbeStage{name: test.stageName, captured: &captured}}, RunOptions{Plan: plan})
+			if err != nil {
+				t.Fatalf("executeStages() error = %v", err)
+			}
+			if captured == nil {
+				t.Fatal("selected stage did not run")
+			}
+			test.assertProbe(t, captured)
+		})
+	}
+}
+
+func TestExecuteStagesBoundedForceStalesButDoesNotRunDependentsOutsideRange(t *testing.T) {
+	cfg := testConfig(t)
+	plan := mustBoundedPlan(t, "render", "render")
+	m := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	m.Campaign = cfg.Session.Campaign
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "extract", "analyze", "publish", "notify"} {
+		m.MarkStageSucceeded(name, time.Now().UTC(), nil)
+	}
+	saveBoundedManifest(t, cfg, m)
+	runs := 0
+	_, err := executeStages(context.Background(), cfg, []stage.Stage{countingStage{name: "render", runs: &runs}}, RunOptions{Plan: plan, Force: true})
+	if err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	if runs != 1 {
+		t.Fatalf("selected render runs = %d, want 1", runs)
+	}
+	loaded, err := (&manifest.LocalStore{}).Load(context.Background(), manifestPathFor(cfg))
+	if err != nil {
+		t.Fatalf("load manifest: %v", err)
+	}
+	for _, name := range []string{"analyze", "publish", "notify"} {
+		if loaded.Stages[name].Status != manifest.StatusStale {
+			t.Fatalf("stage %q status = %q, want stale", name, loaded.Stages[name].Status)
+		}
+	}
+	if loaded.Stages["extract"].Status != manifest.StatusSucceeded {
+		t.Fatalf("extract status = %q, want succeeded", loaded.Stages["extract"].Status)
+	}
+}
+
+func TestExecuteStagesBoundedFailureStopsWithinSelectedRange(t *testing.T) {
+	cfg := testConfig(t)
+	plan := mustBoundedPlan(t, "render", "extract")
+	m := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	m.Campaign = cfg.Session.Campaign
+	markPrefixSucceeded(m, plan)
+	saveBoundedManifest(t, cfg, m)
+	extractRuns := 0
+	_, err := executeStages(context.Background(), cfg, []stage.Stage{
+		failingStage{name: "render", err: errors.New("render failed")},
+		countingStage{name: "extract", runs: &extractRuns},
+	}, RunOptions{Plan: plan})
+	if err == nil || !strings.Contains(err.Error(), "render failed") {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	if extractRuns != 0 {
+		t.Fatalf("extract runs = %d, want 0", extractRuns)
+	}
+}
+
+type collaboratorProbeStage struct {
+	name     string
+	captured **stage.Env
+}
+
+func (s collaboratorProbeStage) Name() string { return s.name }
+func (s collaboratorProbeStage) Run(_ context.Context, env *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
+	*s.captured = env
+	return &stage.StageResult{}, nil
+}
+
+type prerequisiteMutationSpy struct {
+	local   *manifest.LocalStore
+	creates int
+	saves   int
+}
+
+func (s *prerequisiteMutationSpy) Create(ctx context.Context, sessionID string) (*manifest.Manifest, error) {
+	s.creates++
+	return s.local.Create(ctx, sessionID)
+}
+
+func (s *prerequisiteMutationSpy) Load(ctx context.Context, path string) (*manifest.Manifest, error) {
+	return s.local.Load(ctx, path)
+}
+
+func (s *prerequisiteMutationSpy) Save(ctx context.Context, path string, m *manifest.Manifest) error {
+	s.saves++
+	return s.local.Save(ctx, path, m)
+}
+
+func mustBoundedPlan(t *testing.T, from, through string) BoundedPlan {
+	t.Helper()
+	plan, err := BuildBoundedPlan(from, through)
+	if err != nil {
+		t.Fatalf("BuildBoundedPlan(%q, %q) error = %v", from, through, err)
+	}
+	return plan
+}
+
+func markPrefixSucceeded(m *manifest.Manifest, plan BoundedPlan) {
+	for _, name := range plan.PrefixNames() {
+		m.MarkStageSucceeded(name, time.Now().UTC(), nil)
+	}
+}
+
+func saveBoundedManifest(t *testing.T, cfg *config.Config, m *manifest.Manifest) string {
+	t.Helper()
+	path := manifestPathFor(cfg)
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatalf("create manifest directory: %v", err)
+	}
+	if err := (&manifest.LocalStore{}).Save(context.Background(), path, m); err != nil {
+		t.Fatalf("save manifest: %v", err)
+	}
+	return path
+}

internal/app/bounded_run_test.go

@@ -8,8 +8,10 @@ import (
 	"reflect"
 	"strings"
 	"testing"
+	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
 	"gitea.maximumdirect.net/eric/narratio/internal/stage"
 )
 
@@ -141,6 +143,15 @@ func TestRunPassesBoundedPlanToRunner(t *testing.T) {
 func TestPlanPrintsOnlyBoundedRange(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
+	m := manifest.New("2026-05-03", time.Date(2026, 5, 3, 10, 0, 0, 0, time.UTC))
+	m.Campaign = "sample-campaign"
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim"} {
+		m.MarkStageSucceeded(name, time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), nil)
+	}
+	manifestPath := filepath.Join(workspaceRoot, "work", "sample-campaign", "2026-05-03", "manifest.json")
+	if err := (&manifest.LocalStore{}).Save(context.Background(), manifestPath, m); err != nil {
+		t.Fatalf("save prerequisite manifest: %v", err)
+	}
 	var out bytes.Buffer
 	err := Plan(context.Background(), []string{
 		"2026-05-03",

internal/app/plan.go

@@ -12,7 +12,6 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 	"gitea.maximumdirect.net/eric/narratio/internal/logging"
-	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
 )
 
 // Plan validates configuration, prepares the local workdir, and prints stage order.
@@ -37,6 +36,13 @@ func Plan(ctx context.Context, args []string, out io.Writer) error {
 	if _, err := resolveEffectiveArtifacts(cfg, request.SelectedArtifacts); err != nil {
 		return fmt.Errorf("plan: %w", err)
 	}
+	m, err := loadManifestIfPresent(ctx, cfg)
+	if err != nil {
+		return fmt.Errorf("plan: %w", err)
+	}
+	if err := validateBoundedPrerequisites(request.Plan, m); err != nil {
+		return fmt.Errorf("plan: %w", err)
+	}
 	if _, err := loadSecretsFromConfig(cfg, logging.NewLogger(os.Stderr, slog.LevelInfo)); err != nil {
 		return fmt.Errorf("plan: %w", err)
 	}
@@ -48,11 +54,6 @@ func Plan(ctx context.Context, args []string, out io.Writer) error {
 	}
 
 	stages := request.Plan.Stages()
-	var m *manifest.Manifest
-	m, err = loadManifestIfPresent(ctx, cfg)
-	if err != nil {
-		return fmt.Errorf("plan: %w", err)
-	}
 	decisions := decideStageActions(stages, m, request.Force)
 
 	runCount := 0

internal/app/post_publish_cleanup_test.go

@@ -146,7 +146,12 @@ func TestPostPublishCleanupRetriesWhenInitialObligationSaveFails(t *testing.T) {
 
 	store.fail = nil
 	if _, err := executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ManifestStore: store, ObjectStore: &storage.FakeBackend{}}}); err != nil {
-		t.Fatalf("retry executeStages() error = %v", err)
+		t.Fatalf("non-publish executeStages() error = %v", err)
+	}
+	assertExists(t, seed.spoolAudioDir)
+	assertCleanupPending(t, cfg)
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{publishSuccessStage{}}, RunOptions{Force: true, Env: &Env{ManifestStore: store, ObjectStore: &storage.FakeBackend{}}}); err != nil {
+		t.Fatalf("publish retry executeStages() error = %v", err)
 	}
 	assertMissing(t, seed.spoolAudioDir)
 	assertCleanupComplete(t, cfg)
@@ -178,7 +183,12 @@ func TestPostPublishCleanupRetriesFailedDeletionWithoutTouchingOtherRuns(t *test
 
 	removeRunScopedDirFn = originalRemove
 	if _, err := executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ObjectStore: &storage.FakeBackend{}}}); err != nil {
-		t.Fatalf("retry executeStages() error = %v", err)
+		t.Fatalf("non-publish executeStages() error = %v", err)
+	}
+	assertExists(t, seed.runWorkDir)
+	assertCleanupPending(t, cfg)
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{publishSuccessStage{}}, RunOptions{Force: true, Env: &Env{ObjectStore: &storage.FakeBackend{}}}); err != nil {
+		t.Fatalf("publish retry executeStages() error = %v", err)
 	}
 	assertMissing(t, seed.runWorkDir)
 	assertExists(t, seed.otherRunDir)
@@ -218,12 +228,16 @@ func TestPostPublishCleanupRetriesWhenCompletionEvidenceSaveFails(t *testing.T)
 
 	store.fail = nil
 	if _, err := executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ManifestStore: store, ObjectStore: &storage.FakeBackend{}}}); err != nil {
-		t.Fatalf("retry executeStages() error = %v", err)
+		t.Fatalf("non-publish executeStages() error = %v", err)
+	}
+	assertCleanupPending(t, cfg)
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{publishSuccessStage{}}, RunOptions{Force: true, Env: &Env{ManifestStore: store, ObjectStore: &storage.FakeBackend{}}}); err != nil {
+		t.Fatalf("publish retry executeStages() error = %v", err)
 	}
 	assertCleanupComplete(t, cfg)
 
 	if _, err := executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ManifestStore: store, ObjectStore: &storage.FakeBackend{}}}); err != nil {
-		t.Fatalf("idempotent retry executeStages() error = %v", err)
+		t.Fatalf("idempotent non-publish executeStages() error = %v", err)
 	}
 	assertMissing(t, seed.spoolAudioDir)
 }
@@ -328,7 +342,10 @@ func TestPostPublishCleanupFailsOnUnsafePath(t *testing.T) {
 		t.Fatalf("Save() error = %v", err)
 	}
 
-	_, err = executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ObjectStore: &storage.FakeBackend{}}})
+	if _, err = executeStages(context.Background(), cfg, nil, RunOptions{Env: &Env{ObjectStore: &storage.FakeBackend{}}}); err != nil {
+		t.Fatalf("non-publish executeStages() error = %v", err)
+	}
+	_, err = executeStages(context.Background(), cfg, []stage.Stage{publishSuccessStage{}}, RunOptions{Force: true, Env: &Env{ObjectStore: &storage.FakeBackend{}}})
 	if err == nil || !strings.Contains(err.Error(), "refusing to delete path outside root") {
 		t.Fatalf("executeStages() error = %v, want safe-path failure", err)
 	}

internal/app/runner.go

@@ -44,6 +44,13 @@ type RunSummary struct {
 var executeStagesFn = executeStages
 
 func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage, opts RunOptions) (summary *RunSummary, resultErr error) {
+	var prerequisiteStore manifest.Store
+	if opts.Env != nil {
+		prerequisiteStore = opts.Env.ManifestStore
+	}
+	if err := inspectBoundedPrerequisites(ctx, cfg, opts.Plan, prerequisiteStore); err != nil {
+		return nil, fmt.Errorf("validate bounded run prerequisites: %w", err)
+	}
 	effectiveArtifacts := opts.EffectiveArtifacts
 	if !effectiveArtifacts.Resolved() && cfg != nil && cfg.Pipeline != nil && cfg.Pipeline.Scriptorium != nil {
 		var err error
@@ -170,7 +177,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 			fmt.Errorf("load secrets from files: %w", err),
 		)
 	}
-	if env.WhisperX == nil {
+	if env.WhisperX == nil && stagesContainAny(stages, "transcribe") {
 		client, err := buildDefaultWhisperXClient(env.Config)
 		if err != nil {
 			return nil, persistTerminalFailure(
@@ -180,7 +187,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 		}
 		env.WhisperX = client
 	}
-	if env.Seriatim == nil {
+	if env.Seriatim == nil && stagesContainAny(stages, "merge", "normalize", "trim", "render") {
 		runner, err := buildDefaultSeriatimRunner(env.Config)
 		if err != nil {
 			return nil, persistTerminalFailure(
@@ -190,7 +197,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 		}
 		env.Seriatim = runner
 	}
-	if env.Audita == nil {
+	if env.Audita == nil && stagesContainAny(stages, "polish") {
 		runner, err := buildDefaultAuditaRunner(env.Config)
 		if err != nil {
 			return nil, persistTerminalFailure(
@@ -203,7 +210,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 	if env.Notarius == nil && needsNotariusForRun(env.Config, stages) {
 		env.Notarius = notarius.NewSubprocessRunner()
 	}
-	if env.Scriptorium == nil {
+	if env.Scriptorium == nil && stagesContainAny(stages, "trim", "analyze") {
 		env.Scriptorium = scriptorium.NewSubprocessRunner()
 	}
 	if env.ObjectStore == nil && needsObjectStoreForRun(env.Config, stages, effectiveArtifacts) {
@@ -233,7 +240,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 			return config.MergePublishLockRules(staticLocks, remote.Locks), nil
 		}
 	}
-	if env.Notifier == nil {
+	if env.Notifier == nil && stagesContainAny(stages, "notify") {
 		env.Notifier = &notify.NoopSender{}
 	}
 
@@ -426,11 +433,13 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 	// The run record lives inside the run work directory, which cleanup may
 	// remove. Persist its completed publishing result before cleanup starts so a
 	// successful deletion cannot be undone by a later diagnostic write.
-	if err := runPostPublishCleanup(ctx, env, manifestPath, m, executed); err != nil {
-		return nil, persistPostPublishCleanupFailure(
-			ctx, env.ManifestStore, manifestPath, m,
-			fmt.Errorf("post-publish cleanup incomplete: %w", err),
-		)
+	if containsStage(executed, "publish") {
+		if err := runPostPublishCleanup(ctx, env, manifestPath, m, executed); err != nil {
+			return nil, persistPostPublishCleanupFailure(
+				ctx, env.ManifestStore, manifestPath, m,
+				fmt.Errorf("post-publish cleanup incomplete: %w", err),
+			)
+		}
 	}
 
 	return &RunSummary{
@@ -444,6 +453,22 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 	}, nil
 }
 
+func stagesContainAny(stages []stage.Stage, names ...string) bool {
+	wanted := make(map[string]struct{}, len(names))
+	for _, name := range names {
+		wanted[name] = struct{}{}
+	}
+	for _, candidate := range stages {
+		if candidate == nil {
+			continue
+		}
+		if _, ok := wanted[candidate.Name()]; ok {
+			return true
+		}
+	}
+	return false
+}
+
 func persistPostPublishCleanupFailure(
 	ctx context.Context,
 	sessionStore manifest.Store,