Allow checkpoints for empty audio accepted by WhisperX

docs/internal/manifest.md

@@ -64,6 +64,8 @@ canonical `transcript_raw` output with its path, SHA-256 and size, and the
 producer run ID. Paths must be canonical and session-relative, checksums must
 be lowercase SHA-256 values, and the record and output producer identities
 must agree. Other stages cannot carry this transcribe-owned state.
+Audio size is nonnegative; WhisperX determines whether that audio can be
+transcribed. Output size remains positive and the transcript must be valid JSON.
 
 Transcribe reconciles this collection against the current prepared audio set
 and observable WhisperX semantics. Missing or corrupt regular outputs become

docs/roadmap/remediation.md

@@ -1,6 +1,6 @@
 # Transcription Checkpoint Review Remediation
 
-Status: Open.
+Status: Complete.
 
 Reviewed commits `7b0bf99` through `96b4e73` against the speaker-checkpoint and
 upload-replay roadmap and the original user request. No production changes were
@@ -56,7 +56,13 @@ correct rebuilt outputs, and no removed speaker reaching merge.
 
 ## 2. Remove the Unrequested Positive Audio-Size Requirement
 
-Priority: P2. Status: Pending.
+Priority: P2. Status: Complete.
+
+Audio evidence now permits zero size while still rejecting negative values.
+Regression tests verify zero-size manifest round trips, successful service
+results being checkpointed and reused, and service rejection being propagated
+without creating a completion record. Nonempty valid-JSON output validation is
+unchanged. Both new success regressions failed before the fix.
 
 Location: `internal/manifest/transcribe_state.go:95`.
 
@@ -90,7 +96,9 @@ ordinary downstream-failure tests.
 
 ## Verification After Remediation
 
-Run focused stage/manifest/application tests, changed-package race tests, and
-the standard repository validation gates from `docs/development.md`. Preserve
-the existing replay and cancellation regression tests; these findings do not
-require redesigning the HTTP adapter.
+Completed on 2026-09-22 with Go 1.26.8: focused stage/application tests,
+changed-package race tests, `go test ./...`, `go test -race ./...`,
+`go vet ./...`, `go build ./...`, `go test ./internal/doccheck`,
+`go test ./internal/config -run '^TestExamplesLoadAndValidate$'`, and
+`git diff --check` all passed. Existing replay and cancellation regression
+tests remain intact; the HTTP adapter required no changes.

internal/app/transcribe_checkpoint_recovery_test.go

@@ -18,6 +18,57 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/stage"
 )
 
+func TestExecuteStagesDelegatesEmptyAudioAcceptanceToWhisperX(t *testing.T) {
+	for _, rejected := range []bool{false, true} {
+		name := "accepted"
+		if rejected {
+			name = "rejected"
+		}
+		t.Run(name, func(t *testing.T) {
+			cfg, store := setupTranscribeRecoveryFixture(t, []string{"alice"})
+			paths := artifacts.NewLocalStore(cfg.Pipeline.Workspace.Root).SessionPathsFor(cfg.Session.Campaign, cfg.Session.SessionID)
+			if err := os.WriteFile(filepath.Join(paths.AudioDir, "alice.flac"), nil, 0o644); err != nil {
+				t.Fatal(err)
+			}
+			fake := &whisperx.FakeClient{}
+			if rejected {
+				fake.Err = errors.New("service rejected audio")
+			}
+			transcribe := mustSelectStage(t, "transcribe")
+			_, err := executeStages(context.Background(), cfg, []stage.Stage{transcribe}, RunOptions{
+				Env: &Env{ManifestStore: store, WhisperX: fake},
+			})
+			if !errors.Is(err, fake.Err) {
+				t.Fatalf("transcribe error = %v, want %v", err, fake.Err)
+			}
+			if len(fake.RequestsSnapshot()) != 1 {
+				t.Fatal("empty audio was not sent to service")
+			}
+			m := loadTranscribeRecoveryManifest(t, store, cfg)
+			if rejected {
+				if len(m.Stages["transcribe"].TranscribeSpeakers) != 0 {
+					t.Fatal("rejected audio was checkpointed")
+				}
+				return
+			}
+			record, ok := m.Stages["transcribe"].TranscribeSpeakers["alice"]
+			if !ok || record.AudioSize != 0 {
+				t.Fatalf("missing zero-size checkpoint: %#v", record)
+			}
+			retry := &whisperx.FakeClient{}
+			summary, err := executeStages(context.Background(), cfg, []stage.Stage{transcribe}, RunOptions{
+				Env: &Env{ManifestStore: store, WhisperX: retry},
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			if len(summary.Skipped) != 1 || len(retry.RequestsSnapshot()) != 0 {
+				t.Fatal("successful empty-audio result was not reused")
+			}
+		})
+	}
+}
+
 func TestExecuteStagesRetriesOnlyUncommittedTranscribeSpeakers(t *testing.T) {
 	cfg, baseStore := setupTranscribeRecoveryFixture(t, []string{"a", "b", "c", "d"})
 	concurrency := 1

internal/manifest/transcribe_state.go

@@ -92,8 +92,8 @@ func validateTranscribeSpeakerRecord(speakerID string, record TranscribeSpeakerR
 	if err := validateSHA256("audio_sha256", record.AudioSHA256); err != nil {
 		return err
 	}
-	if record.AudioSize <= 0 {
-		return fmt.Errorf("audio_size must be positive")
+	if record.AudioSize < 0 {
+		return fmt.Errorf("audio_size must not be negative")
 	}
 	if record.Output.Kind != "transcript_raw" {
 		return fmt.Errorf("output kind %q must be transcript_raw", record.Output.Kind)

internal/manifest/transcribe_state_test.go

@@ -12,6 +12,8 @@ import (
 func TestTranscribeSpeakerStateRoundTripsPopulatedAndVersionedEmptyCollections(t *testing.T) {
 	now := time.Date(2026, 9, 21, 12, 0, 0, 0, time.UTC)
 	store := &LocalStore{}
+	zeroAudio := validTranscribeSpeakerRecord("Alice Table 1")
+	zeroAudio.AudioSize = 0
 
 	for _, test := range []struct {
 		name    string
@@ -24,6 +26,7 @@ func TestTranscribeSpeakerStateRoundTripsPopulatedAndVersionedEmptyCollections(t
 			},
 		},
 		{name: "versioned empty", records: map[string]TranscribeSpeakerRecord{}},
+		{name: "zero size audio", records: map[string]TranscribeSpeakerRecord{"Alice Table 1": zeroAudio}},
 	} {
 		t.Run(test.name, func(t *testing.T) {
 			m := New("session", now)
@@ -49,6 +52,9 @@ func TestTranscribeSpeakerStateRoundTripsPopulatedAndVersionedEmptyCollections(t
 			}
 			if len(test.records) != 0 {
 				got := stage.TranscribeSpeakers["Alice Table 1"]
+				if got.AudioSize != test.records["Alice Table 1"].AudioSize {
+					t.Fatalf("audio size = %d, want %d", got.AudioSize, test.records["Alice Table 1"].AudioSize)
+				}
 				if got.AudioPath != "inputs/audio/Alice Table 1.flac" || got.Output.LocalPath != "transcripts/raw/Alice Table 1.json" {
 					t.Fatalf("round-tripped record = %#v", got)
 				}
@@ -99,7 +105,7 @@ func TestValidateTranscribeSpeakerCollectionRejectsMalformedEvidence(t *testing.
 		{name: "unsafe audio path", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.AudioPath = "../alice.flac" }, want: "audio_path is unsafe"},
 		{name: "noncanonical audio path", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.AudioPath = "inputs/audio/../alice.flac" }, want: "not canonical"},
 		{name: "bad audio checksum", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.AudioSHA256 = strings.Repeat("G", 64) }, want: "audio_sha256 must be"},
-		{name: "bad audio size", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.AudioSize = 0 }, want: "audio_size must be positive"},
+		{name: "negative audio size", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.AudioSize = -1 }, want: "audio_size"},
 		{name: "bad output kind", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.Output.Kind = "transcript" }, want: "must be transcript_raw"},
 		{name: "unsafe output path", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.Output.LocalPath = "/tmp/alice.json" }, want: "output local_path is unsafe"},
 		{name: "bad output checksum", speakerID: "alice", mutate: func(r *TranscribeSpeakerRecord) { r.Output.Checksum = "bad" }, want: "output checksum must be"},