@mengine/medeo-client 1.2.1-alpha.1 → 1.2.1-alpha.2

package/dist/{document-Dcu87api.js → document-C5KDvMMR.js}

@@ -322,6 +322,71 @@ function syncAggregatedClipsTimePosition(doc) {
 	}
 }
 /**
+* Resolve authoritative `time_position` edges without inventing parents from
+* geometric overlap. Roots stay absolute/sequential; every anchored descendant
+* follows its declared parent, regardless of whether that parent is video, BGM,
+* or speech.
+*/
+function resolveAnchoredClipsTimePosition(doc) {
+	const items = [
+		...doc.main_track,
+		...doc.bgm_track,
+		...doc.speech_track,
+		...doc.caption_track
+	];
+	const byPart = indexByPart(items);
+	const resolved = /* @__PURE__ */ new Set();
+	const visiting = /* @__PURE__ */ new Set();
+	const resolve = (item) => {
+		if (resolved.has(item.part_id)) return item.abs_time_position;
+		if (visiting.has(item.part_id)) throw new Error(`anchored position cycle at ${item.part_id}`);
+		visiting.add(item.part_id);
+		const position = item.time_position;
+		if (position.mode === "absolute") item.abs_time_position = Math.round(position.offsetMs);
+		else if (position.mode === "anchored") {
+			const parent = byPart.get(position.anchorPartId);
+			if (parent != null) item.abs_time_position = resolve(parent) + Math.round(position.offsetMs);
+		}
+		visiting.delete(item.part_id);
+		resolved.add(item.part_id);
+		return item.abs_time_position;
+	};
+	for (const item of items) resolve(item);
+}
+/** Preserve the historical orphan fallback without re-parenting valid roots. */
+function recoverOrphanedSpeechAnchorsByTime(doc) {
+	const placedPartIds = new Set([
+		...doc.main_track,
+		...doc.bgm_track,
+		...doc.speech_track,
+		...doc.caption_track
+	].map((item) => item.part_id));
+	const ranges = doc.main_track.map((item) => ({
+		part_id: item.part_id,
+		start: item.abs_time_position,
+		end: item.abs_time_position + partDurationMs(doc, item.part_id)
+	}));
+	if (ranges.length === 0) return;
+	for (const speech of doc.speech_track) {
+		const position = speech.time_position;
+		if (position.mode !== "anchored" || placedPartIds.has(position.anchorPartId)) continue;
+		const range = ranges.find((candidate) => candidate.start <= speech.abs_time_position && speech.abs_time_position < candidate.end) ?? (speech.abs_time_position < ranges[0].start ? ranges[0] : ranges[ranges.length - 1]);
+		const relative = Math.max(0, Math.round(speech.abs_time_position - range.start));
+		let aggregation = doc.aggregations.find((candidate) => candidate.body_part_id === range.part_id);
+		if (aggregation == null) {
+			aggregation = {
+				body_part_id: range.part_id,
+				attachments: []
+			};
+			doc.aggregations.push(aggregation);
+		}
+		aggregation.attachments.push({
+			part_id: speech.part_id,
+			relative_time_position: relative
+		});
+	}
+}
+/**
 * Reassign each speech to the video clip whose time range contains its start,
 * rebuilding `aggregations`. A speech before the first clip or after the last
 * falls back to the first / last clip respectively (FE: see §4 note — FE falls
@@ -524,14 +589,14 @@ function hostVideoOf(doc, speechPartId) {
 //#endregion
 //#region src/timeline-core/entrypoints.ts
 /**
-* The full solve pipeline: arrange → sync → reassign → resolve-overlap →
-* fill-gaps → recalc. The single cascade the read-side projection runs; ops
+* The full solve pipeline: arrange → resolve declared parent edges →
+* resolve-overlap → fill-gaps → recalc. The single cascade the read-side projection runs; ops
 * never call it (they write only facts — RFC 02 §7/§10).
 */
 function cascadeAfterVideoClipChanges(doc, makeEmptyPart) {
 	arrangeMainTrackSeamlessly(doc);
-	syncAggregatedClipsTimePosition(doc);
-	reassignSpeechesToVideoClipsByTime(doc);
+	resolveAnchoredClipsTimePosition(doc);
+	recoverOrphanedSpeechAnchorsByTime(doc);
 	resolveAllSpeechOverlaps(doc);
 	fillMainTrackTimeGaps(doc, makeEmptyPart);
 	recalculateTimelineDuration(doc);
@@ -605,7 +670,7 @@ function videoDocumentToTimelineDoc(document) {
 		caption_track: captionItems,
 		bgm_track: bgmItems,
 		part_library: partLibrary,
-		aggregations: seedAggregations(mainItems, speechItems),
+		aggregations: seedAggregations([...mainItems, ...bgmItems], speechItems),
 		timeline: {
 			duration_ms: 0,
 			unit_time_ms: document.timeline?.unit_time_ms ?? 0
@@ -629,19 +694,18 @@ function seedAbs(position, fallbackAbsMs) {
 	}
 }
 /**
-* Build the cascade `aggregations` parent-map from anchored items: an anchored
-* item's `anchorPartId` is the host's `part_id` (= `body_part_id`), `offsetMs`
-* is `relative_time_position`. Only main-track (video) hosts form aggregations;
-* caption→speech relations are recomputed by the cascade from `caption.start_ms`.
+* Build the read-view `aggregations` parent-map from anchored speech items.
+* Video and BGM roots are both valid hosts; absolute speech roots have no entry.
+* Caption→speech ownership remains in `speech_part_id` / `caption_ids`.
 */
-function seedAggregations(mainItems, speechItems) {
-	const mainPartIds = new Set(mainItems.map((item) => item.part_id));
+function seedAggregations(hostItems, speechItems) {
+	const hostPartIds = new Set(hostItems.map((item) => item.part_id));
 	const byHost = /* @__PURE__ */ new Map();
 	const order = [];
 	for (const speech of speechItems) {
 		if (speech.time_position.mode !== "anchored") continue;
 		const host = speech.time_position.anchorPartId;
-		if (!mainPartIds.has(host)) continue;
+		if (!hostPartIds.has(host)) continue;
 		let aggregation = byHost.get(host);
 		if (aggregation == null) {
 			aggregation = {
@@ -1663,4 +1727,4 @@ function toTrackRow(track) {
 	};
 }
 //#endregion
-export { partDurationMs as A, reassignSpeechesToVideoClipsByTime as C, syncAggregatedClipsTimePosition as D, resolveSpeechOverlapByShiftingVideos as E, partUnionToDraft as F, recordEntries as I, videoDocumentMirrorSchema as L, DEFAULT_UNIT_TIME_MS as M, VIDEO_DOCUMENT_SCHEMA_VERSION as N, TIMELINE_SKELETON_DURATION_MS as O, effectiveVideoClipDurationMs as P, base64ToBytes as R, fillMainTrackTimeGaps as S, resolveAllSpeechOverlaps as T, findLaneTrack as _, buildInitialVideoDocument as a, cascadeAfterVideoClipChanges as b, fromVideoDocument as c, assertValidVideoDocument as d, validateVideoDocument as f, ensureLaneTrack as g, LANE_KINDS_IN_STACK_ORDER as h, readVideoDocumentFromDraft as i, safeDurationMs as j, isEmptyVideoClip as k, toVideoDocument as l, videoDocumentSchema as m, createMirrorVideoDocument as n, buildSpeechHostMap as o, partUnionSchema as p, createMirrorVideoDocumentAdapter as r, derivePositionFromAbs as s, MirrorVideoDocumentAdapter as t, VideoDocumentValidationError as u, laneTrackId as v, recalculateTimelineDuration as w, arrangeMainTrackSeamlessly as x, solveVideoDocument as y, bytesToBase64 as z };
+export { TIMELINE_SKELETON_DURATION_MS as A, base64ToBytes as B, reassignSpeechesToVideoClipsByTime as C, resolveAnchoredClipsTimePosition as D, resolveAllSpeechOverlaps as E, VIDEO_DOCUMENT_SCHEMA_VERSION as F, effectiveVideoClipDurationMs as I, partUnionToDraft as L, partDurationMs as M, safeDurationMs as N, resolveSpeechOverlapByShiftingVideos as O, DEFAULT_UNIT_TIME_MS as P, recordEntries as R, fillMainTrackTimeGaps as S, recoverOrphanedSpeechAnchorsByTime as T, bytesToBase64 as V, findLaneTrack as _, buildInitialVideoDocument as a, cascadeAfterVideoClipChanges as b, fromVideoDocument as c, assertValidVideoDocument as d, validateVideoDocument as f, ensureLaneTrack as g, LANE_KINDS_IN_STACK_ORDER as h, readVideoDocumentFromDraft as i, isEmptyVideoClip as j, syncAggregatedClipsTimePosition as k, toVideoDocument as l, videoDocumentSchema as m, createMirrorVideoDocument as n, buildSpeechHostMap as o, partUnionSchema as p, createMirrorVideoDocumentAdapter as r, derivePositionFromAbs as s, MirrorVideoDocumentAdapter as t, VideoDocumentValidationError as u, laneTrackId as v, recalculateTimelineDuration as w, arrangeMainTrackSeamlessly as x, solveVideoDocument as y, videoDocumentMirrorSchema as z };

package/dist/{index-CZlTzzrm.d.ts → index-DrEHTguy.d.ts}

@@ -948,7 +948,7 @@ type TrackItemDraft = NonNullable<NonNullable<TrackDraft['items']>[number]>;
 declare const addSpeechesInputSchema: z.ZodObject<{
   speeches: z.ZodArray<z.ZodObject<{
     speech_id: z.ZodString;
-    anchor_part_id: z.ZodString;
+    anchor_part_id: z.ZodNullable<z.ZodString>;
     offset_ms: z.ZodNumber;
     audio_storage_key: z.ZodString;
     duration_ms: z.ZodNumber;
@@ -1053,7 +1053,7 @@ type AdjustVideoClipVolumeInput = z.infer<typeof adjustVideoClipVolumeInputSchem
 declare const changeSpeechScriptInputSchema: z.ZodObject<{
   speeches: z.ZodArray<z.ZodObject<{
     speech_id: z.ZodString;
-    anchor_part_id: z.ZodString;
+    anchor_part_id: z.ZodNullable<z.ZodString>;
     offset_ms: z.ZodNumber;
     audio_storage_key: z.ZodString;
     duration_ms: z.ZodNumber;
@@ -1077,7 +1077,7 @@ declare const changeSpeechScriptInputSchema: z.ZodObject<{
 declare const changeSpeechVoiceInputSchema: z.ZodObject<{
   speeches: z.ZodArray<z.ZodObject<{
     speech_id: z.ZodString;
-    anchor_part_id: z.ZodString;
+    anchor_part_id: z.ZodNullable<z.ZodString>;
     offset_ms: z.ZodNumber;
     audio_storage_key: z.ZodString;
     duration_ms: z.ZodNumber;
@@ -1156,6 +1156,12 @@ declare const moveSpeechesInputSchema: z.ZodObject<{
 }, z.core.$strip>;
 type MoveSpeechesInput = z.infer<typeof moveSpeechesInputSchema>;
 //#endregion
+//#region src/editor/schemas/move-bgm.d.ts
+declare const moveBgmInputSchema: z.ZodObject<{
+  new_start_ms: z.ZodNumber;
+}, z.core.$strip>;
+type MoveBgmInput = z.infer<typeof moveBgmInputSchema>;
+//#endregion
 //#region src/editor/schemas/move-video-clips-by-anchor.d.ts
 /**
  * Where the moved block lands on the main track.
@@ -1490,16 +1496,14 @@ declare const voiceSchema: z.ZodObject<{
  * stable speech + caption parts and writes them as authoritative facts. No
  * cascade runs on write — the projection derives absolute positions on read.
  *
- * Each speech carries the anchoring fact directly (RFC 02 §4): the host video
- * clip `anchor_part_id` and the `offset_ms` within it. The upstream caller
- * already knows which clip a speech attaches to, so the op writes
- * `{ mode:'anchored', anchorPartId, offsetMs }` verbatim — no write-time
- * host-picking. Captions anchor to their speech via the caption part's
- * `start_ms` (offset within the speech).
+ * `anchor_part_id` states the root relationship explicitly: a video/BGM id
+ * creates an anchored speech; `null` creates an absolute audio root.
+ * `offset_ms` is relative to the anchor or absolute for a root. Captions always
+ * anchor to their speech via `start_ms`.
  */
 declare const speechAssetSchema: z.ZodObject<{
   speech_id: z.ZodString;
-  anchor_part_id: z.ZodString;
+  anchor_part_id: z.ZodNullable<z.ZodString>;
   offset_ms: z.ZodNumber;
   audio_storage_key: z.ZodString;
   duration_ms: z.ZodNumber;
@@ -1523,7 +1527,7 @@ declare const captionAssetSchema: z.ZodObject<{
 declare const speechAssetsSchema: z.ZodObject<{
   speeches: z.ZodArray<z.ZodObject<{
     speech_id: z.ZodString;
-    anchor_part_id: z.ZodString;
+    anchor_part_id: z.ZodNullable<z.ZodString>;
     offset_ms: z.ZodNumber;
     audio_storage_key: z.ZodString;
     duration_ms: z.ZodNumber;
@@ -1548,7 +1552,7 @@ type SpeechAsset = z.infer<typeof speechAssetSchema>;
 type CaptionAsset = z.infer<typeof captionAssetSchema>;
 type SpeechAssets = z.infer<typeof speechAssetsSchema>;
 declare namespace index_d_exports {
-  export { AddSpeechesInput, AddVideoClipsInput, AdjustBgmVolumeInput, AdjustSpeechVolumeInput, AdjustVideoClipDurationInput, AdjustVideoClipVolumeInput, AnchoredDeletePolicy, AnchoredReplacePolicy, CaptionAsset, ChangeSpeechScriptInput, ChangeSpeechVoiceInput, DeleteBgmInput, DeleteSpeechesInput, DeleteVideoClipsInput, MoveAnchor, MoveSpeechesInput, MoveVideoClipsByAnchorInput, MoveVideoClipsInput, MovedClipAnchoredPolicy, ReplaceVideoClipContentInput, ReplaceVideoClipSequenceInput, ResizeCaptionInput, SetBgmInput, SetCaptionStyleInput, SetCaptionVisibilityInput, SetVideoClipSpeedShiftInput, SpeechAsset, SpeechAssets, addSpeechesInputSchema, addVideoClipsInputSchema, adjustBgmVolumeInputSchema, adjustSpeechVolumeInputSchema, adjustVideoClipDurationInputSchema, adjustVideoClipVolumeInputSchema, anchoredDeletePolicySchema, anchoredReplacePolicySchema, changeSpeechScriptInputSchema, changeSpeechVoiceInputSchema, clipIdSchema, clipIdsSchema, deleteBgmInputSchema, deleteSpeechesInputSchema, deleteVideoClipsInputSchema, mediaIdSchema, moveAnchorSchema, moveSpeechesInputSchema, moveVideoClipsByAnchorInputSchema, moveVideoClipsInputSchema, movedClipAnchoredPolicySchema, positiveMsSchema, replaceVideoClipContentInputSchema, replaceVideoClipSequenceInputSchema, resizeCaptionInputSchema, setBgmInputSchema, setCaptionStyleInputSchema, setCaptionVisibilityInputSchema, setVideoClipSpeedShiftInputSchema, speechAssetsSchema, speechIdSchema, speechIdsSchema, speedShiftSchema, timelineMsSchema, voiceSchema, volumeSchema };
+  export { AddSpeechesInput, AddVideoClipsInput, AdjustBgmVolumeInput, AdjustSpeechVolumeInput, AdjustVideoClipDurationInput, AdjustVideoClipVolumeInput, AnchoredDeletePolicy, AnchoredReplacePolicy, CaptionAsset, ChangeSpeechScriptInput, ChangeSpeechVoiceInput, DeleteBgmInput, DeleteSpeechesInput, DeleteVideoClipsInput, MoveAnchor, MoveBgmInput, MoveSpeechesInput, MoveVideoClipsByAnchorInput, MoveVideoClipsInput, MovedClipAnchoredPolicy, ReplaceVideoClipContentInput, ReplaceVideoClipSequenceInput, ResizeCaptionInput, SetBgmInput, SetCaptionStyleInput, SetCaptionVisibilityInput, SetVideoClipSpeedShiftInput, SpeechAsset, SpeechAssets, addSpeechesInputSchema, addVideoClipsInputSchema, adjustBgmVolumeInputSchema, adjustSpeechVolumeInputSchema, adjustVideoClipDurationInputSchema, adjustVideoClipVolumeInputSchema, anchoredDeletePolicySchema, anchoredReplacePolicySchema, changeSpeechScriptInputSchema, changeSpeechVoiceInputSchema, clipIdSchema, clipIdsSchema, deleteBgmInputSchema, deleteSpeechesInputSchema, deleteVideoClipsInputSchema, mediaIdSchema, moveAnchorSchema, moveBgmInputSchema, moveSpeechesInputSchema, moveVideoClipsByAnchorInputSchema, moveVideoClipsInputSchema, movedClipAnchoredPolicySchema, positiveMsSchema, replaceVideoClipContentInputSchema, replaceVideoClipSequenceInputSchema, resizeCaptionInputSchema, setBgmInputSchema, setCaptionStyleInputSchema, setCaptionVisibilityInputSchema, setVideoClipSpeedShiftInputSchema, speechAssetsSchema, speechIdSchema, speechIdsSchema, speedShiftSchema, timelineMsSchema, voiceSchema, volumeSchema };
 }
 //#endregion
 //#region src/editor/schema-validator.d.ts
@@ -1597,6 +1601,7 @@ declare class SchemaValidator {
   validateChangeSpeechVoice(input: ChangeSpeechVoiceInput, doc: SnapshotReadable): void;
   validateDeleteSpeeches(input: DeleteSpeechesInput, doc: SnapshotReadable): void;
   validateMoveSpeeches(input: MoveSpeechesInput, doc: SnapshotReadable): void;
+  validateMoveBgm(input: MoveBgmInput, doc: SnapshotReadable): void;
   validateSetBgm(input: SetBgmInput, _doc: SnapshotReadable): void;
   validateDeleteBgm(input: DeleteBgmInput, _doc: SnapshotReadable): void;
   validateSetCaptionVisibility(input: SetCaptionVisibilityInput, _doc: SnapshotReadable): void;
@@ -1609,12 +1614,8 @@ declare class SchemaValidator {
   /** Each regenerated speech (re-TTS) must already exist in the document. */
   private assertSpeechesExist;
   /**
-   * Each speech's `anchor_part_id` must point at a video clip that already
-   * exists. A relative speech anchored to a missing clip cannot be positioned
-   * by the projection and has no host to recover to (RFC 02 §4/§11.1), so the
-   * write must be rejected rather than landing a dangling reference. (Restores
-   * the dangling-reference guard the old `validateMaterializedPatch` carried;
-   * ADR 0009.)
+   * Anchored speeches may follow a video or BGM root. A null anchor is an
+   * explicit absolute speech root. Missing and wrong-kind parents are rejected.
    */
   private assertAnchorsExist;
   private assertPartKind;
@@ -1649,7 +1650,7 @@ declare class SchemaValidator {
  * side-effect payload contract in
  * `docs/projects/medeo-integration/results/phase-4-side-effect-payload-contract.md`.
  */
-type ImplementedSemanticOpKind = 'MoveVideoClips' | 'MoveVideoClipsByAnchor' | 'DeleteVideoClips' | 'AddVideoClips' | 'AdjustVideoClipVolume' | 'SetVideoClipSpeedShift' | 'ReplaceVideoClipContent' | 'ReplaceVideoClipSequence' | 'AdjustVideoClipDuration' | 'AddSpeeches' | 'DeleteSpeeches' | 'MoveSpeeches' | 'ChangeSpeechScript' | 'ChangeSpeechVoice' | 'AdjustSpeechVolume' | 'ResizeCaption' | 'SetCaptionVisibility' | 'SetCaptionStyle' | 'SetBgm' | 'DeleteBgm' | 'AdjustBgmVolume';
+type ImplementedSemanticOpKind = 'MoveVideoClips' | 'MoveVideoClipsByAnchor' | 'DeleteVideoClips' | 'AddVideoClips' | 'AdjustVideoClipVolume' | 'SetVideoClipSpeedShift' | 'ReplaceVideoClipContent' | 'ReplaceVideoClipSequence' | 'AdjustVideoClipDuration' | 'AddSpeeches' | 'DeleteSpeeches' | 'MoveSpeeches' | 'MoveBgm' | 'ChangeSpeechScript' | 'ChangeSpeechVoice' | 'AdjustSpeechVolume' | 'ResizeCaption' | 'SetCaptionVisibility' | 'SetCaptionStyle' | 'SetBgm' | 'DeleteBgm' | 'AdjustBgmVolume';
 /**
  * Operations on the roadmap but NOT yet implemented by `SemanticEditor`. Empty:
  * Phase 4 covers every op with a real entry point. Kept as a named type so the
@@ -1711,7 +1712,7 @@ interface CommitOptions {
   /** Author of this op; omitted for writes with no acting user (bootstrap, repair). */
   actor?: OpActor;
 }
-type SemanticOpInput = MoveVideoClipsInput | MoveVideoClipsByAnchorInput | DeleteVideoClipsInput | AddVideoClipsInput | AdjustVideoClipVolumeInput | AdjustSpeechVolumeInput | AdjustBgmVolumeInput | SetVideoClipSpeedShiftInput | ReplaceVideoClipContentInput | ReplaceVideoClipSequenceInput | AdjustVideoClipDurationInput | AddSpeechesInput | DeleteSpeechesInput | MoveSpeechesInput | ResizeCaptionInput | SetCaptionVisibilityInput | SetCaptionStyleInput | SetBgmInput | DeleteBgmInput;
+type SemanticOpInput = MoveVideoClipsInput | MoveVideoClipsByAnchorInput | DeleteVideoClipsInput | AddVideoClipsInput | AdjustVideoClipVolumeInput | AdjustSpeechVolumeInput | AdjustBgmVolumeInput | SetVideoClipSpeedShiftInput | ReplaceVideoClipContentInput | ReplaceVideoClipSequenceInput | AdjustVideoClipDurationInput | AddSpeechesInput | DeleteSpeechesInput | MoveSpeechesInput | MoveBgmInput | ResizeCaptionInput | SetCaptionVisibilityInput | SetCaptionStyleInput | SetBgmInput | DeleteBgmInput;
 type SemanticOpName = ImplementedSemanticOpKind;
 /** Audit metadata committed alongside an op's writes. */
 interface TransactAudit {
@@ -1827,6 +1828,8 @@ declare class SemanticEditor {
    * derives absolute positions on read.
    */
   moveSpeeches(input: MoveSpeechesInput, options?: CommitOptions): Promise<void>;
+  /** Move the BGM root; any speech/caption descendants follow through their anchors. */
+  moveBgm(input: MoveBgmInput, options?: CommitOptions): Promise<void>;
   /** Change a speech's script. Re-TTS runs upstream; the regenerated parts arrive materialized (no cascade). */
   changeSpeechScript(input: ChangeSpeechScriptInput, options?: CommitOptions): Promise<void>;
   /** Change a speech's voice. Re-TTS runs upstream; the regenerated parts arrive materialized (no cascade). */

package/dist/index.d.ts

@@ -1,4 +1,4 @@
-import { $ as VideoDocumentSchemaVersion, A as assertValidVideoDocument, B as BgmPart, C as index_d_exports, D as VideoDocumentMirrorSchema, E as VideoDocumentDraft, F as SpeechHostMap, G as SpeechPart, H as DEFAULT_UNIT_TIME_MS, I as buildSpeechHostMap, J as TrackItem, K as Timeline, L as derivePositionFromAbs, M as InitialDocumentFacts, N as buildInitialVideoDocument, O as videoDocumentMirrorSchema, P as DerivedItemPosition, Q as VideoDocument, R as fromVideoDocument, S as ValidationError, T as TrackItemDraft, U as PartKind, V as CaptionPart, W as PartUnion, X as VIDEO_DOCUMENT_SCHEMA_VERSION, Y as TrackItemTimePosition, Z as VideoClipPart, _ as PlannedSemanticOpKind, a as MirrorVideoDocumentOptions, at as CaptionPart$1, b as SchemaValidator, c as CommitOptions, ct as SpeedShift, d as SemanticEditor, dt as TrackItem$1, et as VideoDocumentValidationIssue, f as SemanticOpInput, g as ImplementedSemanticOpKind, h as IMPLEMENTED_SEMANTIC_OP_KINDS, i as MirrorVideoDocumentAdapter, it as Attachment, j as validateVideoDocument, k as VideoDocumentValidationError, l as OpActor, lt as Timeline$1, m as TransactAudit, n as videoDocumentSchema, nt as VideoDraft, o as createMirrorVideoDocument, ot as CaptionStyle, p as SemanticOpName, q as Track, r as readVideoDocumentFromDraft, rt as VideoDraftPartUnion, s as createMirrorVideoDocumentAdapter, st as PartAggregation, t as partUnionSchema, tt as VideoDocumentValidationIssueCode, u as SemanticDocumentAdapter, ut as Track$1, v as SemanticOpKind, w as TrackDraft, x as SnapshotReadable, y as isImplementedSemanticOpKind, z as toVideoDocument } from "./index-CZlTzzrm.js";
+import { $ as VideoDocumentSchemaVersion, A as assertValidVideoDocument, B as BgmPart, C as index_d_exports, D as VideoDocumentMirrorSchema, E as VideoDocumentDraft, F as SpeechHostMap, G as SpeechPart, H as DEFAULT_UNIT_TIME_MS, I as buildSpeechHostMap, J as TrackItem, K as Timeline, L as derivePositionFromAbs, M as InitialDocumentFacts, N as buildInitialVideoDocument, O as videoDocumentMirrorSchema, P as DerivedItemPosition, Q as VideoDocument, R as fromVideoDocument, S as ValidationError, T as TrackItemDraft, U as PartKind, V as CaptionPart, W as PartUnion, X as VIDEO_DOCUMENT_SCHEMA_VERSION, Y as TrackItemTimePosition, Z as VideoClipPart, _ as PlannedSemanticOpKind, a as MirrorVideoDocumentOptions, at as CaptionPart$1, b as SchemaValidator, c as CommitOptions, ct as SpeedShift, d as SemanticEditor, dt as TrackItem$1, et as VideoDocumentValidationIssue, f as SemanticOpInput, g as ImplementedSemanticOpKind, h as IMPLEMENTED_SEMANTIC_OP_KINDS, i as MirrorVideoDocumentAdapter, it as Attachment, j as validateVideoDocument, k as VideoDocumentValidationError, l as OpActor, lt as Timeline$1, m as TransactAudit, n as videoDocumentSchema, nt as VideoDraft, o as createMirrorVideoDocument, ot as CaptionStyle, p as SemanticOpName, q as Track, r as readVideoDocumentFromDraft, rt as VideoDraftPartUnion, s as createMirrorVideoDocumentAdapter, st as PartAggregation, t as partUnionSchema, tt as VideoDocumentValidationIssueCode, u as SemanticDocumentAdapter, ut as Track$1, v as SemanticOpKind, w as TrackDraft, x as SnapshotReadable, y as isImplementedSemanticOpKind, z as toVideoDocument } from "./index-DrEHTguy.js";
 import { PeerID } from "loro-crdt";
 import { DocState } from "@mengine/sync";
 import { BaseDocStorage, Connection, DocDiff, DocPushReceipt, DocSnapshotRecord, DocStorage, DocStorageOptions, DocUpdate, DocUpdateRecord } from "@mengine/storage";
@@ -853,6 +853,15 @@ declare function arrangeMainTrackSeamlessly(doc: TimelineDoc): void;
  *   caption.abs  = speech.abs + caption.start_ms
  */
 declare function syncAggregatedClipsTimePosition(doc: TimelineDoc): void;
+/**
+ * Resolve authoritative `time_position` edges without inventing parents from
+ * geometric overlap. Roots stay absolute/sequential; every anchored descendant
+ * follows its declared parent, regardless of whether that parent is video, BGM,
+ * or speech.
+ */
+declare function resolveAnchoredClipsTimePosition(doc: TimelineDoc): void;
+/** Preserve the historical orphan fallback without re-parenting valid roots. */
+declare function recoverOrphanedSpeechAnchorsByTime(doc: TimelineDoc): void;
 /**
  * Reassign each speech to the video clip whose time range contains its start,
  * rebuilding `aggregations`. A speech before the first clip or after the last
@@ -918,8 +927,8 @@ type MakeEmptyPart = (durationMs: number) => {
   partId: string;
 };
 /**
- * The full solve pipeline: arrange → sync → reassign → resolve-overlap →
- * fill-gaps → recalc. The single cascade the read-side projection runs; ops
+ * The full solve pipeline: arrange → resolve declared parent edges →
+ * resolve-overlap → fill-gaps → recalc. The single cascade the read-side projection runs; ops
  * never call it (they write only facts — RFC 02 §7/§10).
  */
 declare function cascadeAfterVideoClipChanges(doc: TimelineDoc, makeEmptyPart: MakeEmptyPart): void;
@@ -1064,4 +1073,4 @@ declare function hostForAbsMs(ranges: MainClipRange[], absMs: number): MainClipR
  */
 declare function relativePositionForAbs(ranges: MainClipRange[], absMs: number): TrackItemTimePosition;
 //#endregion
-export { type Aggregation, type Attachment, type BgmPart, type CaptionPart, type CaptionStyle, type CommitOptions, DEFAULT_UNIT_TIME_MS, type DerivedItemPosition, type DocStorageLike, type DocVersionMark, IMPLEMENTED_SEMANTIC_OP_KINDS, type ImplementedSemanticOpKind, type InitialDocumentFacts, LANE_KINDS_IN_STACK_ORDER, type LaneKind, type MainClipRange, type MakeEmptyPart, ManualSyncDoc, type ManualSyncDocOptions, MedeoHttpDocStorage, type MedeoHttpDocStorageOptions, MemoryDocStorage, MengineAckFailedError, type MengineAuditEntry, type MengineAuditResponse, MengineDocSession, type MengineDocSessionOptions, type MengineDocSessionUpdateEvent, type MengineDocumentVersion, type MengineEventStreamOptions, MengineHttpClient, type MengineHttpClientOptions, MengineHttpRequestError, MenginePushRejectedError, type MenginePushResponse, type MenginePushUpdateResponse, type MengineRejectedResponse, type MengineSnapshotResponse, type MengineSseUpdateEvent, type MengineSyncResponse, type MengineUpdateMeta, MirrorVideoDocumentAdapter, type MirrorVideoDocumentOptions, type OpActor, type PartAggregation, type PartKind, type PartUnion, type PlannedSemanticOpKind, type PullFailureReason, type PullResult, type PushOutcome, type PushOutcomeKind, type PushResult, type PushResultKind, SchemaValidator, type SemanticDocumentAdapter, SemanticEditor, type SemanticOpInput, type SemanticOpKind, type SemanticOpName, type SnapshotReadable, type SolvedVideoDocument, type SpeechHostMap, type SpeechPart, type SpeedShift, TIMELINE_SKELETON_DURATION_MS, type Timeline, type TimelineDoc, type TimelineItem, type Track, type TrackDraft, type TrackItem, type TrackItemDraft, type TrackItemTimePosition, type TransactAudit, VIDEO_DOCUMENT_SCHEMA_VERSION, ValidationError, type VideoClipPart, type VideoDocument, type VideoDocumentDraft, type VideoDocumentMirrorSchema, type VideoDocumentSchemaVersion, VideoDocumentValidationError, type VideoDocumentValidationIssue, type VideoDocumentValidationIssueCode, type VideoDraft, type CaptionPart$1 as VideoDraftCaptionPart, type VideoDraftPartUnion, type Timeline$1 as VideoDraftTimeline, type Track$1 as VideoDraftTrack, type TrackItem$1 as VideoDraftTrackItem, type WaitForServerAckOptions, arrangeMainTrackSeamlessly, assertValidVideoDocument, base64ToBytes, buildInitialVideoDocument, buildSpeechHostMap, bytesToBase64, cascadeAfterVideoClipChanges, createMirrorVideoDocument, createMirrorVideoDocumentAdapter, decodeDocVersionMark, derivePositionFromAbs, encodeDocVersionMark, ensureLaneTrack, fillMainTrackTimeGaps, findLaneTrack, fromVideoDocument, generatePartId, getAt, hostForAbsMs, isEmptyVideoClip, isImplementedSemanticOpKind, isMap, laneTrackId, mainTrackRanges, partDurationMs, partUnionSchema, readMainTrackItems, readMengineEventStream, readPart, readPartDurationMs, readVideoDocumentFromDraft, reassignSpeechesToVideoClipsByTime, recalculateTimelineDuration, relativePositionForAbs, resolveAllSpeechOverlaps, resolveSpeechOverlapByShiftingVideos, safeDurationMs, index_d_exports as schemas, snapshotToPlain, solveVideoDocument, syncAggregatedClipsTimePosition, toVideoDocument, validateVideoDocument, videoDocumentMirrorSchema, videoDocumentSchema };
+export { type Aggregation, type Attachment, type BgmPart, type CaptionPart, type CaptionStyle, type CommitOptions, DEFAULT_UNIT_TIME_MS, type DerivedItemPosition, type DocStorageLike, type DocVersionMark, IMPLEMENTED_SEMANTIC_OP_KINDS, type ImplementedSemanticOpKind, type InitialDocumentFacts, LANE_KINDS_IN_STACK_ORDER, type LaneKind, type MainClipRange, type MakeEmptyPart, ManualSyncDoc, type ManualSyncDocOptions, MedeoHttpDocStorage, type MedeoHttpDocStorageOptions, MemoryDocStorage, MengineAckFailedError, type MengineAuditEntry, type MengineAuditResponse, MengineDocSession, type MengineDocSessionOptions, type MengineDocSessionUpdateEvent, type MengineDocumentVersion, type MengineEventStreamOptions, MengineHttpClient, type MengineHttpClientOptions, MengineHttpRequestError, MenginePushRejectedError, type MenginePushResponse, type MenginePushUpdateResponse, type MengineRejectedResponse, type MengineSnapshotResponse, type MengineSseUpdateEvent, type MengineSyncResponse, type MengineUpdateMeta, MirrorVideoDocumentAdapter, type MirrorVideoDocumentOptions, type OpActor, type PartAggregation, type PartKind, type PartUnion, type PlannedSemanticOpKind, type PullFailureReason, type PullResult, type PushOutcome, type PushOutcomeKind, type PushResult, type PushResultKind, SchemaValidator, type SemanticDocumentAdapter, SemanticEditor, type SemanticOpInput, type SemanticOpKind, type SemanticOpName, type SnapshotReadable, type SolvedVideoDocument, type SpeechHostMap, type SpeechPart, type SpeedShift, TIMELINE_SKELETON_DURATION_MS, type Timeline, type TimelineDoc, type TimelineItem, type Track, type TrackDraft, type TrackItem, type TrackItemDraft, type TrackItemTimePosition, type TransactAudit, VIDEO_DOCUMENT_SCHEMA_VERSION, ValidationError, type VideoClipPart, type VideoDocument, type VideoDocumentDraft, type VideoDocumentMirrorSchema, type VideoDocumentSchemaVersion, VideoDocumentValidationError, type VideoDocumentValidationIssue, type VideoDocumentValidationIssueCode, type VideoDraft, type CaptionPart$1 as VideoDraftCaptionPart, type VideoDraftPartUnion, type Timeline$1 as VideoDraftTimeline, type Track$1 as VideoDraftTrack, type TrackItem$1 as VideoDraftTrackItem, type WaitForServerAckOptions, arrangeMainTrackSeamlessly, assertValidVideoDocument, base64ToBytes, buildInitialVideoDocument, buildSpeechHostMap, bytesToBase64, cascadeAfterVideoClipChanges, createMirrorVideoDocument, createMirrorVideoDocumentAdapter, decodeDocVersionMark, derivePositionFromAbs, encodeDocVersionMark, ensureLaneTrack, fillMainTrackTimeGaps, findLaneTrack, fromVideoDocument, generatePartId, getAt, hostForAbsMs, isEmptyVideoClip, isImplementedSemanticOpKind, isMap, laneTrackId, mainTrackRanges, partDurationMs, partUnionSchema, readMainTrackItems, readMengineEventStream, readPart, readPartDurationMs, readVideoDocumentFromDraft, reassignSpeechesToVideoClipsByTime, recalculateTimelineDuration, recoverOrphanedSpeechAnchorsByTime, relativePositionForAbs, resolveAllSpeechOverlaps, resolveAnchoredClipsTimePosition, resolveSpeechOverlapByShiftingVideos, safeDurationMs, index_d_exports as schemas, snapshotToPlain, solveVideoDocument, syncAggregatedClipsTimePosition, toVideoDocument, validateVideoDocument, videoDocumentMirrorSchema, videoDocumentSchema };

package/dist/index.js

@@ -1,5 +1,5 @@
 import { t as __exportAll } from "./chunk-D7D4PA-g.js";
-import { A as partDurationMs, C as reassignSpeechesToVideoClipsByTime, D as syncAggregatedClipsTimePosition, E as resolveSpeechOverlapByShiftingVideos, F as partUnionToDraft, I as recordEntries, L as videoDocumentMirrorSchema, M as DEFAULT_UNIT_TIME_MS, N as VIDEO_DOCUMENT_SCHEMA_VERSION, O as TIMELINE_SKELETON_DURATION_MS, P as effectiveVideoClipDurationMs, R as base64ToBytes, S as fillMainTrackTimeGaps, T as resolveAllSpeechOverlaps, _ as findLaneTrack, a as buildInitialVideoDocument, b as cascadeAfterVideoClipChanges, c as fromVideoDocument, d as assertValidVideoDocument, f as validateVideoDocument, g as ensureLaneTrack, h as LANE_KINDS_IN_STACK_ORDER, i as readVideoDocumentFromDraft, j as safeDurationMs, k as isEmptyVideoClip, l as toVideoDocument, m as videoDocumentSchema, n as createMirrorVideoDocument, o as buildSpeechHostMap, p as partUnionSchema, r as createMirrorVideoDocumentAdapter, s as derivePositionFromAbs, t as MirrorVideoDocumentAdapter, u as VideoDocumentValidationError, v as laneTrackId, w as recalculateTimelineDuration, x as arrangeMainTrackSeamlessly, y as solveVideoDocument, z as bytesToBase64 } from "./document-Dcu87api.js";
+import { A as TIMELINE_SKELETON_DURATION_MS, B as base64ToBytes, C as reassignSpeechesToVideoClipsByTime, D as resolveAnchoredClipsTimePosition, E as resolveAllSpeechOverlaps, F as VIDEO_DOCUMENT_SCHEMA_VERSION, I as effectiveVideoClipDurationMs, L as partUnionToDraft, M as partDurationMs, N as safeDurationMs, O as resolveSpeechOverlapByShiftingVideos, P as DEFAULT_UNIT_TIME_MS, R as recordEntries, S as fillMainTrackTimeGaps, T as recoverOrphanedSpeechAnchorsByTime, V as bytesToBase64, _ as findLaneTrack, a as buildInitialVideoDocument, b as cascadeAfterVideoClipChanges, c as fromVideoDocument, d as assertValidVideoDocument, f as validateVideoDocument, g as ensureLaneTrack, h as LANE_KINDS_IN_STACK_ORDER, i as readVideoDocumentFromDraft, j as isEmptyVideoClip, k as syncAggregatedClipsTimePosition, l as toVideoDocument, m as videoDocumentSchema, n as createMirrorVideoDocument, o as buildSpeechHostMap, p as partUnionSchema, r as createMirrorVideoDocumentAdapter, s as derivePositionFromAbs, t as MirrorVideoDocumentAdapter, u as VideoDocumentValidationError, v as laneTrackId, w as recalculateTimelineDuration, x as arrangeMainTrackSeamlessly, y as solveVideoDocument, z as videoDocumentMirrorSchema } from "./document-C5KDvMMR.js";
 import { z } from "zod";
 import { LoroDoc, VersionVector } from "loro-crdt";
 import { ClientServerSynchronizer, DocManager } from "@mengine/sync";
@@ -308,17 +308,15 @@ const voiceSchema = z.object({
 * stable speech + caption parts and writes them as authoritative facts. No
 * cascade runs on write — the projection derives absolute positions on read.
 *
-* Each speech carries the anchoring fact directly (RFC 02 §4): the host video
-* clip `anchor_part_id` and the `offset_ms` within it. The upstream caller
-* already knows which clip a speech attaches to, so the op writes
-* `{ mode:'anchored', anchorPartId, offsetMs }` verbatim — no write-time
-* host-picking. Captions anchor to their speech via the caption part's
-* `start_ms` (offset within the speech).
+* `anchor_part_id` states the root relationship explicitly: a video/BGM id
+* creates an anchored speech; `null` creates an absolute audio root.
+* `offset_ms` is relative to the anchor or absolute for a root. Captions always
+* anchor to their speech via `start_ms`.
 */
 const speechAssetSchema = z.object({
 	speech_id: speechIdSchema.describe("The speech part ID (= side-effect speech_parts[].id)"),
-	anchor_part_id: clipIdSchema.describe("Host video clip part ID the speech anchors to (RFC 02 §4)"),
-	offset_ms: timelineMsSchema.describe("Offset within the host clip (speech.abs = host.abs + offset_ms)"),
+	anchor_part_id: z.string().min(1).nullable().describe("Video/BGM parent part ID, or null when the speech itself is the root"),
+	offset_ms: timelineMsSchema.describe("Relative offset from the parent, or absolute start for a root speech"),
 	audio_storage_key: z.string().min(1),
 	duration_ms: positiveMsSchema,
 	audio_script: z.string(),
@@ -509,6 +507,9 @@ const moveSpeechesInputSchema = z.object({ speeches: z.array(z.object({
 	new_start_ms: timelineMsSchema.describe("New absolute start time on the timeline")
 })).min(1).describe("Speeches to move to new positions") });
 //#endregion
+//#region src/editor/schemas/move-bgm.ts
+const moveBgmInputSchema = z.object({ new_start_ms: timelineMsSchema.describe("New absolute BGM start on the timeline") });
+//#endregion
 //#region src/editor/schemas/move-video-clips-by-anchor.ts
 /**
 * Where the moved block lands on the main track.
@@ -765,6 +766,7 @@ var schemas_exports = /* @__PURE__ */ __exportAll({
 	deleteVideoClipsInputSchema: () => deleteVideoClipsInputSchema,
 	mediaIdSchema: () => mediaIdSchema,
 	moveAnchorSchema: () => moveAnchorSchema,
+	moveBgmInputSchema: () => moveBgmInputSchema,
 	moveSpeechesInputSchema: () => moveSpeechesInputSchema,
 	moveVideoClipsByAnchorInputSchema: () => moveVideoClipsByAnchorInputSchema,
 	moveVideoClipsInputSchema: () => moveVideoClipsInputSchema,
@@ -1014,6 +1016,10 @@ var SchemaValidator = class {
 		const snapshot = doc.snapshot();
 		for (const s of parsed.speeches) this.assertPartKind(snapshot, s.speech_id, "speech", "move_speeches");
 	}
+	validateMoveBgm(input, doc) {
+		this.parse(moveBgmInputSchema, input, "move_bgm_invalid_input");
+		if (doc.snapshot().tracks?.find((track) => track.parts_kind === "bgm")?.items?.[0] == null) throw new ValidationError("move_bgm_not_found", "document has no BGM item");
+	}
 	validateSetBgm(input, _doc) {
 		this.parse(setBgmInputSchema, input, "set_bgm_invalid_input");
 	}
@@ -1061,16 +1067,20 @@ var SchemaValidator = class {
 		for (const speech of assets.speeches) this.assertPartKind(snapshot, speech.speech_id, "speech", opName);
 	}
 	/**
-	* Each speech's `anchor_part_id` must point at a video clip that already
-	* exists. A relative speech anchored to a missing clip cannot be positioned
-	* by the projection and has no host to recover to (RFC 02 §4/§11.1), so the
-	* write must be rejected rather than landing a dangling reference. (Restores
-	* the dangling-reference guard the old `validateMaterializedPatch` carried;
-	* ADR 0009.)
+	* Anchored speeches may follow a video or BGM root. A null anchor is an
+	* explicit absolute speech root. Missing and wrong-kind parents are rejected.
 	*/
 	assertAnchorsExist(assets, doc, opName) {
 		const snapshot = doc.snapshot();
-		for (const speech of assets.speeches) this.assertPartKind(snapshot, speech.anchor_part_id, "video_clip", `${opName}_anchor`);
+		for (const speech of assets.speeches) {
+			if (speech.anchor_part_id == null) continue;
+			const part = readPart(snapshot, speech.anchor_part_id);
+			if (part == null) throw new ValidationError(`${opName}_anchor_part_not_found`, `part_library has no parent`, { part_id: speech.anchor_part_id });
+			if (part.part_kind !== "video_clip" && part.part_kind !== "bgm") throw new ValidationError(`${opName}_anchor_wrong_part_kind`, `speech parent must be video_clip or bgm`, {
+				part_id: speech.anchor_part_id,
+				part_kind: part.part_kind
+			});
+		}
 	}
 	assertPartKind(snapshot, partId, kind, opName) {
 		const part = readPart(snapshot, partId);
@@ -1419,8 +1429,9 @@ var SemanticEditor = class {
 	*/
 	async addSpeeches(input, options) {
 		this.validator.validateAddSpeeches(input, this.doc);
+		const absByPartId = solveVideoDocument(this.doc.snapshot()).absByPartId;
 		this.doc.transact((draft) => {
-			writeSpeechAssets(draft, input);
+			writeSpeechAssets(draft, input, absByPartId);
 		}, audit("AddSpeeches", input, options));
 	}
 	/** Delete speeches with their captions; the subtree is removed (§9.2). Surviving lanes keep their facts. */
@@ -1438,29 +1449,60 @@ var SemanticEditor = class {
 	*/
 	async moveSpeeches(input, options) {
 		this.validator.validateMoveSpeeches(input, this.doc);
+		const snapshot = this.doc.snapshot();
+		const absByPartId = solveVideoDocument(snapshot).absByPartId;
 		this.doc.transact((draft) => {
-			const ranges = mainTrackRanges(draft);
 			const track = findLaneTrack(draft, "speech");
+			const ranges = mainTrackRanges(draft);
 			for (const move of input.speeches) {
 				const item = (track?.items ?? []).find((it) => it?.part_id === move.speech_id);
 				if (item == null) continue;
-				item.time_position = relativePositionForAbs(ranges, move.new_start_ms);
+				const current = item.time_position;
+				if (current?.mode === "anchored") {
+					const parentStartMs = absByPartId.get(current.anchorPartId);
+					item.time_position = readPart(snapshot, current.anchorPartId)?.part_kind === "video_clip" ? relativePositionForAbs(ranges, move.new_start_ms) : parentStartMs == null ? {
+						mode: "absolute",
+						offsetMs: move.new_start_ms
+					} : {
+						mode: "anchored",
+						anchorPartId: current.anchorPartId,
+						offsetMs: Math.max(0, move.new_start_ms - parentStartMs)
+					};
+				} else item.time_position = {
+					mode: "absolute",
+					offsetMs: move.new_start_ms
+				};
 				item.fallback_abs_ms = move.new_start_ms;
 			}
 		}, audit("MoveSpeeches", input, options));
 	}
+	/** Move the BGM root; any speech/caption descendants follow through their anchors. */
+	async moveBgm(input, options) {
+		this.validator.validateMoveBgm(input, this.doc);
+		this.doc.transact((draft) => {
+			const item = findLaneTrack(draft, "bgm")?.items?.[0];
+			if (item == null) return;
+			item.time_position = {
+				mode: "absolute",
+				offsetMs: input.new_start_ms
+			};
+			item.fallback_abs_ms = void 0;
+		}, audit("MoveBgm", input, options));
+	}
 	/** Change a speech's script. Re-TTS runs upstream; the regenerated parts arrive materialized (no cascade). */
 	async changeSpeechScript(input, options) {
 		this.validator.validateChangeSpeechScript(input, this.doc);
+		const absByPartId = solveVideoDocument(this.doc.snapshot()).absByPartId;
 		this.doc.transact((draft) => {
-			writeSpeechAssets(draft, input);
+			writeSpeechAssets(draft, input, absByPartId);
 		}, audit("ChangeSpeechScript", input, options));
 	}
 	/** Change a speech's voice. Re-TTS runs upstream; the regenerated parts arrive materialized (no cascade). */
 	async changeSpeechVoice(input, options) {
 		this.validator.validateChangeSpeechVoice(input, this.doc);
+		const absByPartId = solveVideoDocument(this.doc.snapshot()).absByPartId;
 		this.doc.transact((draft) => {
-			writeSpeechAssets(draft, input);
+			writeSpeechAssets(draft, input, absByPartId);
 		}, audit("ChangeSpeechVoice", input, options));
 	}
 	/**
@@ -1699,7 +1741,7 @@ function normalizeSpeedShift(input) {
 * projection derives it on read. On re-TTS the speech id is preserved, so an
 * existing item is re-positioned rather than duplicated.
 */
-function writeSpeechAssets(draft, assets) {
+function writeSpeechAssets(draft, assets, absByPartId) {
 	const captionById = new Map(assets.captions.map((c) => [c.caption_id, c]));
 	const speechTrack = ensureLaneTrack(draft, "speech");
 	speechTrack.items ??= [];
@@ -1712,8 +1754,6 @@ function writeSpeechAssets(draft, assets) {
 		}
 		return captionItems;
 	};
-	const ranges = mainTrackRanges(draft);
-	const hostStartMs = (anchorPartId) => ranges.find((r) => r.partId === anchorPartId)?.startMs;
 	for (const speech of assets.speeches) {
 		const priorCaptionIds = draft.part_library?.[speech.speech_id]?.speech?.caption_ids ?? [];
 		const nextCaptionIds = new Set(speech.caption_ids);
@@ -1737,9 +1777,10 @@ function writeSpeechAssets(draft, assets) {
 			voice: speech.voice,
 			caption_ids: speech.caption_ids
 		} });
-		const hostStart = hostStartMs(speech.anchor_part_id);
-		const speechAbs = hostStart == null ? void 0 : hostStart + speech.offset_ms;
-		placeRelative(speechTrack.items, speech.speech_id, speech.anchor_part_id, speech.offset_ms, speechAbs);
+		const hostStart = speech.anchor_part_id == null ? void 0 : absByPartId.get(speech.anchor_part_id);
+		const speechAbs = speech.anchor_part_id == null ? speech.offset_ms : hostStart == null ? void 0 : hostStart + speech.offset_ms;
+		if (speech.anchor_part_id == null) placeAbsolute(speechTrack.items, speech.speech_id, speech.offset_ms);
+		else placeRelative(speechTrack.items, speech.speech_id, speech.anchor_part_id, speech.offset_ms, speechAbs);
 		for (const captionId of speech.caption_ids) {
 			const caption = captionById.get(captionId);
 			if (caption == null) continue;
@@ -1785,6 +1826,24 @@ function placeRelative(items, partId, anchorPartId, offsetMs, fallbackAbsMs) {
 		duration_override_ms: void 0
 	});
 }
+function placeAbsolute(items, partId, startMs) {
+	const timePosition = {
+		mode: "absolute",
+		offsetMs: Math.max(0, Math.round(startMs))
+	};
+	const existing = items.find((item) => item?.part_id === partId);
+	if (existing != null) {
+		existing.time_position = timePosition;
+		existing.fallback_abs_ms = void 0;
+		return;
+	}
+	items.push({
+		part_id: partId,
+		time_position: timePosition,
+		fallback_abs_ms: void 0,
+		duration_override_ms: void 0
+	});
+}
 /**
 * After the main-track flow order changes (move / reorder), re-parent each
 * anchored speech to the clip its *preserved* absolute landing now falls in
@@ -1923,6 +1982,7 @@ const IMPLEMENTED_SEMANTIC_OP_KINDS = [
 	"AddSpeeches",
 	"DeleteSpeeches",
 	"MoveSpeeches",
+	"MoveBgm",
 	"ChangeSpeechScript",
 	"ChangeSpeechVoice",
 	"AdjustSpeechVolume",
@@ -2812,4 +2872,4 @@ function serverCovers(serverVV, target) {
 	return covers(server, target);
 }
 //#endregion
-export { DEFAULT_UNIT_TIME_MS, IMPLEMENTED_SEMANTIC_OP_KINDS, LANE_KINDS_IN_STACK_ORDER, ManualSyncDoc, MedeoHttpDocStorage, MemoryDocStorage, MengineAckFailedError, MengineDocSession, MengineHttpClient, MengineHttpRequestError, MenginePushRejectedError, MirrorVideoDocumentAdapter, SchemaValidator, SemanticEditor, TIMELINE_SKELETON_DURATION_MS, VIDEO_DOCUMENT_SCHEMA_VERSION, ValidationError, VideoDocumentValidationError, arrangeMainTrackSeamlessly, assertValidVideoDocument, base64ToBytes, buildInitialVideoDocument, buildSpeechHostMap, bytesToBase64, cascadeAfterVideoClipChanges, createMirrorVideoDocument, createMirrorVideoDocumentAdapter, decodeDocVersionMark, derivePositionFromAbs, encodeDocVersionMark, ensureLaneTrack, fillMainTrackTimeGaps, findLaneTrack, fromVideoDocument, generatePartId, getAt, hostForAbsMs, isEmptyVideoClip, isImplementedSemanticOpKind, isMap, laneTrackId, mainTrackRanges, partDurationMs, partUnionSchema, readMainTrackItems, readMengineEventStream, readPart, readPartDurationMs, readVideoDocumentFromDraft, reassignSpeechesToVideoClipsByTime, recalculateTimelineDuration, relativePositionForAbs, resolveAllSpeechOverlaps, resolveSpeechOverlapByShiftingVideos, safeDurationMs, schemas_exports as schemas, snapshotToPlain, solveVideoDocument, syncAggregatedClipsTimePosition, toVideoDocument, validateVideoDocument, videoDocumentMirrorSchema, videoDocumentSchema };
+export { DEFAULT_UNIT_TIME_MS, IMPLEMENTED_SEMANTIC_OP_KINDS, LANE_KINDS_IN_STACK_ORDER, ManualSyncDoc, MedeoHttpDocStorage, MemoryDocStorage, MengineAckFailedError, MengineDocSession, MengineHttpClient, MengineHttpRequestError, MenginePushRejectedError, MirrorVideoDocumentAdapter, SchemaValidator, SemanticEditor, TIMELINE_SKELETON_DURATION_MS, VIDEO_DOCUMENT_SCHEMA_VERSION, ValidationError, VideoDocumentValidationError, arrangeMainTrackSeamlessly, assertValidVideoDocument, base64ToBytes, buildInitialVideoDocument, buildSpeechHostMap, bytesToBase64, cascadeAfterVideoClipChanges, createMirrorVideoDocument, createMirrorVideoDocumentAdapter, decodeDocVersionMark, derivePositionFromAbs, encodeDocVersionMark, ensureLaneTrack, fillMainTrackTimeGaps, findLaneTrack, fromVideoDocument, generatePartId, getAt, hostForAbsMs, isEmptyVideoClip, isImplementedSemanticOpKind, isMap, laneTrackId, mainTrackRanges, partDurationMs, partUnionSchema, readMainTrackItems, readMengineEventStream, readPart, readPartDurationMs, readVideoDocumentFromDraft, reassignSpeechesToVideoClipsByTime, recalculateTimelineDuration, recoverOrphanedSpeechAnchorsByTime, relativePositionForAbs, resolveAllSpeechOverlaps, resolveAnchoredClipsTimePosition, resolveSpeechOverlapByShiftingVideos, safeDurationMs, schemas_exports as schemas, snapshotToPlain, solveVideoDocument, syncAggregatedClipsTimePosition, toVideoDocument, validateVideoDocument, videoDocumentMirrorSchema, videoDocumentSchema };

package/dist/testing.d.ts

@@ -1,4 +1,4 @@
-import { i as MirrorVideoDocumentAdapter, nt as VideoDraft } from "./index-CZlTzzrm.js";
+import { i as MirrorVideoDocumentAdapter, nt as VideoDraft } from "./index-DrEHTguy.js";
 
 //#region src/testing/in-memory-mengine-server.d.ts
 /**

package/dist/testing.js

@@ -1,4 +1,4 @@
-import { N as VIDEO_DOCUMENT_SCHEMA_VERSION, R as base64ToBytes, r as createMirrorVideoDocumentAdapter, z as bytesToBase64 } from "./document-Dcu87api.js";
+import { B as base64ToBytes, F as VIDEO_DOCUMENT_SCHEMA_VERSION, V as bytesToBase64, r as createMirrorVideoDocumentAdapter } from "./document-C5KDvMMR.js";
 import { t as classifyUpdate } from "./loro-relay-doc-ssdYpuef.js";
 import { LoroDoc, VersionVector, encodeFrontiers } from "loro-crdt";
 //#region src/testing/in-memory-mengine-server.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mengine/medeo-client",
-  "version": "1.2.1-alpha.1",
+  "version": "1.2.1-alpha.2",
   "license": "UNLICENSED",
   "repository": {
     "type": "git",
@@ -25,9 +25,9 @@
   "dependencies": {
     "loro-mirror": "^2.2.0",
     "zod": "^4.4.3",
-    "@mengine/storage": "1.2.1-alpha.1",
-    "@mengine/sync": "1.2.1-alpha.1",
-    "@mengine/utils": "1.2.1-alpha.1"
+    "@mengine/storage": "1.2.1-alpha.2",
+    "@mengine/sync": "1.2.1-alpha.2",
+    "@mengine/utils": "1.2.1-alpha.2"
   },
   "devDependencies": {
     "@types/node": "^25.9.1",