@mengine/medeo-client 1.2.1-alpha.0 → 1.2.1-alpha.2

package/dist/index.js

@@ -1,5 +1,6 @@
-import { A as resolveSpeechOverlapByShiftingVideos, B as partUnionToDraft, C as solveVideoDocument, D as reassignSpeechesToVideoClipsByTime, E as fillMainTrackTimeGaps, F as safeDurationMs, H as videoDocumentMirrorSchema, I as DEFAULT_UNIT_TIME_MS, L as VIDEO_DOCUMENT_SCHEMA_VERSION, M as TIMELINE_SKELETON_DURATION_MS, N as isEmptyVideoClip, O as recalculateTimelineDuration, P as partDurationMs, R as effectiveVideoClipDurationMs, S as laneTrackId, T as arrangeMainTrackSeamlessly, U as base64ToBytes, V as recordEntries, W as bytesToBase64, _ as partUnionSchema, a as createMirrorVideoDocument, b as ensureLaneTrack, c as readVideoDocumentFromDraft, d as derivePositionFromAbs, f as fromVideoDocument, g as validateVideoDocument, h as assertValidVideoDocument, i as MirrorVideoDocumentAdapter, j as syncAggregatedClipsTimePosition, k as resolveAllSpeechOverlaps, l as buildInitialVideoDocument, m as VideoDocumentValidationError, n as createPlainMemoryAdapter, o as createMirrorVideoDocumentAdapter, p as toVideoDocument, r as generatePartId, s as writeVideoDocumentToDraft, t as PlainMemoryAdapter, u as buildSpeechHostMap, v as videoDocumentSchema, w as cascadeAfterVideoClipChanges, x as findLaneTrack, y as LANE_KINDS_IN_STACK_ORDER, z as speedOf } from "./document-49OdvviW.js";
-import { addSpeechesInputSchema, addVideoClipsInputSchema, adjustBgmVolumeInputSchema, adjustSpeechVolumeInputSchema, adjustVideoClipDurationInputSchema, adjustVideoClipVolumeInputSchema, changeSpeechScriptInputSchema, changeSpeechVoiceInputSchema, deleteBgmInputSchema, deleteSpeechesInputSchema, deleteVideoClipsInputSchema, moveSpeechesInputSchema, moveVideoClipsByAnchorInputSchema, moveVideoClipsInputSchema, replaceVideoClipContentInputSchema, replaceVideoClipSequenceInputSchema, setBgmInputSchema, setCaptionStyleInputSchema, setCaptionVisibilityInputSchema, setVideoClipSpeedShiftInputSchema, t as schemas_exports } from "./schemas.js";
+import { t as __exportAll } from "./chunk-D7D4PA-g.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";
 import { DisposableSet, EventBus, Task } from "@mengine/utils";
@@ -217,6 +218,576 @@ function findSseFrameBoundary(buffer) {
 	return Math.min(lf, crlf);
 }
 //#endregion
+//#region src/editor/id-gen.ts
+/**
+* Part-id generation, aligned with the online ecosystem.
+*
+* The authoritative online producers — agent-harness (`@harness/shared`
+* `genObjId`) and director.v2 (`common/obj_id.py` `gen_obj_id`) — both mint part
+* ids as `` `${prefix}_${ulid()}` ``, and real captured drafts use exactly that
+* shape (`clip_…` / `spe_…` / `cap_…` / `bgm_…`, each a 26-char ULID). The engine
+* previously emitted `vc_<base36 timestamp><6 random>`, a different prefix AND a
+* different encoding — the sole cross-repo id divergence. This module removes it
+* by emitting the same `<prefix>_<ULID>` bytes.
+*
+* The ULID is generated inline (Crockford Base32, 48-bit time + 80-bit random)
+* rather than pulling the `ulid` npm package: the randomness class matches the
+* old generator (both `Math.random`-based) and it keeps `@mengine/medeo-client`
+* dependency-free for a purely mechanical id string. Part ids only need to be
+* unique and lexicographically time-sortable, which this satisfies.
+*/
+/** Crockford Base32 alphabet (no I, L, O, U), per the ULID spec. */
+const CROCKFORD = "0123456789ABCDEFGHJKMNPQRSTVWXYZ";
+const TIME_LEN = 10;
+const RANDOM_LEN = 16;
+function encodeTime(now) {
+	let out = "";
+	let ms = now;
+	for (let i = TIME_LEN - 1; i >= 0; i--) {
+		const mod = ms % 32;
+		out = CROCKFORD[mod] + out;
+		ms = (ms - mod) / 32;
+	}
+	return out;
+}
+function encodeRandom() {
+	let out = "";
+	for (let i = 0; i < RANDOM_LEN; i++) out += CROCKFORD[Math.floor(Math.random() * 32)];
+	return out;
+}
+/** A 26-char Crockford Base32 ULID (10-char time + 16-char random). */
+function ulid() {
+	return encodeTime(Date.now()) + encodeRandom();
+}
+function generatePartId(prefix) {
+	return `${prefix}_${ulid()}`;
+}
+//#endregion
+//#region src/editor/schemas/shared.ts
+const clipIdSchema = z.string().min(1).describe("The clip part ID on the timeline");
+const clipIdsSchema = z.array(clipIdSchema).min(1).refine((ids) => new Set(ids).size === ids.length, { message: "Duplicate clip IDs are not allowed" }).describe("List of clip part IDs (no duplicates allowed)");
+const mediaIdSchema = z.string().min(1).describe("The media asset ID");
+const speechIdSchema = z.string().min(1).describe("The speech part ID on the timeline");
+const timelineMsSchema = z.number().int().min(0).describe("Time position in milliseconds on the timeline (>= 0)");
+const positiveMsSchema = z.number().int().positive().describe("Duration in milliseconds (> 0)");
+const volumeSchema = z.number().min(-60).max(20).describe("Volume in decibels (-60.0 to 20.0; 0.0 = original, -60 = mute, +20 = max)");
+const speechIdsSchema = z.array(speechIdSchema).min(1).refine((ids) => new Set(ids).size === ids.length, { message: "Duplicate speech IDs are not allowed" }).describe("List of speech part IDs (no duplicates allowed)");
+const tangentHandleSchema = z.object({
+	x: z.number().finite(),
+	y: z.number().finite()
+}).describe("Bezier tangent handle (x, y)");
+const speedKeyframeSchema = z.object({
+	position: z.number().min(0).max(1),
+	rate: z.number().min(0),
+	in_tangent: tangentHandleSchema.optional(),
+	out_tangent: tangentHandleSchema.optional()
+}).describe("A speed keyframe: normalized position (0..1), rate, optional tangents");
+/**
+* A clip's playback-speed fact, the only thing `SetVideoClipSpeedShift` writes.
+* Mirrors the IDL `SpeedShift`: `category` is `linear` | `curve`, and `config`
+* is a discriminated union — `{ linear: { speed } }` for a constant multiplier
+* (the multiplier projection reads at `config.linear.speed`) or `{ curve: {
+* keyframes } }` for a Bezier-controlled variable speed (RFC 02 / `reference/16`
+* §0). Exactly one of `linear` / `curve` is present.
+*/
+const speedShiftSchema = z.object({
+	category: z.enum(["linear", "curve"]),
+	mode: z.string(),
+	config: z.union([z.object({ linear: z.object({ speed: z.number().finite().positive() }) }), z.object({ curve: z.object({ keyframes: z.array(speedKeyframeSchema).min(2) }) })])
+}).describe("Speed shift: linear multiplier or Bezier curve, mirroring the IDL shape");
+const voiceSchema = z.object({
+	id: z.string().min(1),
+	name: z.string()
+}).describe("TTS voice summary attached to a speech");
+//#endregion
+//#region src/editor/schemas/speech-assets.ts
+/**
+* The materialized TTS result shared by `AddSpeeches` / `ChangeSpeechScript` /
+* `ChangeSpeechVoice` (see `results/phase-4-side-effect-payload-contract.md`
+* §1/§2). The side effect (TTS/ASR + billing) runs upstream; the op receives the
+* stable speech + caption parts and writes them as authoritative facts. No
+* cascade runs on write — the projection derives absolute positions on read.
+*
+* `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: 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(),
+	volume: volumeSchema,
+	voice: voiceSchema,
+	origin_speech_id: z.string().min(1),
+	caption_ids: z.array(z.string().min(1)).describe("Caption part IDs owned by this speech")
+});
+const captionAssetSchema = z.object({
+	caption_id: z.string().min(1).describe("The caption part ID (= side-effect created_caption_parts[].id)"),
+	speech_part_id: speechIdSchema.describe("The owning speech part ID"),
+	text: z.string(),
+	start_ms: timelineMsSchema.describe("Offset within the host speech (caption.abs = speech.abs + start_ms)"),
+	duration_ms: positiveMsSchema
+});
+/** A materialized speech-subtree write (speeches + their captions). */
+const speechAssetsSchema = z.object({
+	speeches: z.array(speechAssetSchema).min(1).describe("Materialized speech parts to write"),
+	captions: z.array(captionAssetSchema).describe("Materialized caption parts owned by the speeches")
+});
+//#endregion
+//#region src/editor/schemas/add-speeches.ts
+/**
+* Add speeches (and their captions). TTS runs upstream; the stable speech /
+* caption parts arrive materialized (see `speech-assets.ts`). The op writes the
+* parts and each speech's `{ mode:'anchored', anchorPartId, offsetMs }` fact
+* verbatim — no write-time host-picking, no cascade (RFC 02 §4). The projection
+* derives absolute positions on read.
+*/
+const addSpeechesInputSchema = speechAssetsSchema.describe("Materialized speeches + captions to add");
+//#endregion
+//#region src/editor/schemas/add-video-clips.ts
+/**
+* Add video clips to a track. Each clip's duration facts are separated so a
+* single number is never overloaded (RFC 02 / `reference/16` §0b):
+*
+* - `media_duration_ms` is the source media's intrinsic full length (a resource
+*   fact, written to the part);
+* - `play_in` / `play_out` are the optional trim window into that media; when
+*   omitted the whole media is used (`play_in=0`, `play_out=media_duration_ms`).
+*
+* The clip's effective timeline duration is derived by the projection from the
+* trim window and `speed_shift` — it is never an input here.
+*/
+const addVideoClipsInputSchema = z.object({
+	clips: z.array(z.object({
+		media_id: mediaIdSchema.describe("The media asset ID for the video clip"),
+		start_ms: timelineMsSchema.optional().describe("Absolute start time in milliseconds on the timeline"),
+		media_duration_ms: positiveMsSchema.describe("The source media's intrinsic full length in ms"),
+		play_in: timelineMsSchema.optional().describe("Trim window start in the media (default 0)"),
+		play_out: positiveMsSchema.optional().describe("Trim window end in the media (default media_duration_ms)"),
+		track_id: z.string().min(1).optional().describe("Target track ID (optional, defaults to main track)")
+	})).min(1).describe("List of video clips to create"),
+	before_clip_id: z.string().min(1).optional().describe("Insert new clips before this clip ID"),
+	after_clip_id: z.string().min(1).optional().describe("Insert new clips after this clip ID")
+}).superRefine((data, ctx) => {
+	if (data.before_clip_id != null && data.after_clip_id != null) {
+		ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: "Cannot provide both before_clip_id and after_clip_id"
+		});
+		return;
+	}
+	const hasRelative = data.before_clip_id != null || data.after_clip_id != null;
+	for (let i = 0; i < data.clips.length; i++) {
+		const clip = data.clips[i];
+		if (hasRelative && clip.start_ms != null) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].start_ms must not be provided when using before_clip_id or after_clip_id`,
+			path: [
+				"clips",
+				i,
+				"start_ms"
+			]
+		});
+		if (!hasRelative && clip.start_ms == null) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].start_ms is required when not using relative positioning`,
+			path: [
+				"clips",
+				i,
+				"start_ms"
+			]
+		});
+		const playIn = clip.play_in ?? 0;
+		const playOut = clip.play_out ?? clip.media_duration_ms;
+		if (playOut > clip.media_duration_ms) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].play_out ${playOut}ms exceeds media_duration_ms ${clip.media_duration_ms}ms`,
+			path: [
+				"clips",
+				i,
+				"play_out"
+			]
+		});
+		if (playIn >= playOut) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].play_in ${playIn}ms must be less than play_out ${playOut}ms`,
+			path: [
+				"clips",
+				i,
+				"play_in"
+			]
+		});
+	}
+});
+//#endregion
+//#region src/editor/schemas/adjust-bgm-volume.ts
+const adjustBgmVolumeInputSchema = z.object({ bgm: z.array(z.object({
+	bgm_id: clipIdSchema.describe("The bgm part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of bgm parts with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/adjust-speech-volume.ts
+const adjustSpeechVolumeInputSchema = z.object({ speeches: z.array(z.object({
+	speech_id: speechIdSchema.describe("The speech part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of speeches with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/adjust-video-clip-duration.ts
+/**
+* Re-trim existing video clips (the user-facing "adjust duration" gesture is a
+* trim of the source window). The new `play_in` / `play_out` are the facts; the
+* effective timeline duration is derived from them and the clip's `speed_shift`,
+* and the change reflows downstream clips, speeches, and the timeline inside the
+* op's transaction (no caller-materialized cascade).
+*/
+const adjustVideoClipDurationInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to re-trim"),
+	play_in: timelineMsSchema.describe("New trim window start in the source media"),
+	play_out: positiveMsSchema.describe("New trim window end in the source media")
+})).min(1).describe("Video clips with their new trim windows") });
+//#endregion
+//#region src/editor/schemas/adjust-video-clip-volume.ts
+const adjustVideoClipVolumeInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of video clips with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/change-speech.ts
+/**
+* Change a speech's script or voice. Both re-run TTS upstream and return the
+* regenerated speech / caption parts in the same materialized shape as
+* `AddSpeeches` (`speech-assets.ts`); the op upserts them by id (the speech part
+* id is preserved across a re-TTS), re-seats at `start_ms`, and reflows. Old
+* caption parts no longer owned by the speech are removed via `caption_ids`.
+*/
+const changeSpeechScriptInputSchema = speechAssetsSchema.describe("Regenerated speeches + captions (new script)");
+const changeSpeechVoiceInputSchema = speechAssetsSchema.describe("Regenerated speeches + captions (new voice)");
+//#endregion
+//#region src/editor/schemas/delete-bgm.ts
+/**
+* Remove the document BGM. Pure document edit: clears the bgm lane and removes
+* the bgm part. Takes no input (a document holds at most one bgm); an empty
+* object keeps the op signature uniform with the rest.
+*/
+const deleteBgmInputSchema = z.object({}).describe("Remove the document BGM (no parameters)");
+//#endregion
+//#region src/editor/schemas/delete-speeches.ts
+/**
+* Delete speeches with their captions. Pure document edit (no side effect): the
+* op removes each speech part, cascade-deletes the captions it owns (via
+* `caption_ids` / `speech_part_id`), drops their track items, and reflows.
+*/
+const deleteSpeechesInputSchema = z.object({ speech_ids: speechIdsSchema.describe("Speech part IDs to delete (their captions cascade-delete)") });
+//#endregion
+//#region src/editor/schemas/delete-video-clips.ts
+/**
+* How a delete handles the anchored subtree (speeches anchored to a deleted clip,
+* and their captions) — a delete-op policy, not a data-model field (reference/17
+* §6). `cascade` (default) removes the subtree; `detach` keeps the direct
+* anchored children, re-pinning them to `absolute` so they stay on the timeline.
+*/
+const anchoredDeletePolicySchema = z.enum(["cascade", "detach"]);
+const deleteVideoClipsInputSchema = z.object({
+	clip_ids: clipIdsSchema.describe("List of video clip part IDs to delete from the main track"),
+	on_anchored: anchoredDeletePolicySchema.optional().describe("How to treat anchored children (default cascade)")
+});
+//#endregion
+//#region src/editor/schemas/move-speeches.ts
+/**
+* Move speeches in time. Pure document edit: the op re-seats each speech at its
+* new absolute `start_ms`; the cascade reassigns it to the host video clip,
+* resolves overlaps, and reflows. Captions follow their speech.
+*/
+const moveSpeechesInputSchema = z.object({ speeches: z.array(z.object({
+	speech_id: speechIdSchema.describe("The speech part ID to move"),
+	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.
+*
+* A discriminated union rather than two optional `before_clip_id` /
+* `after_clip_id` fields (the shape `addVideoClips` had to use, because there the
+* two modes share a whole clip description): here the alternatives carry nothing
+* in common, so making them mutually exclusive *by type* removes three runtime
+* `superRefine` checks that would otherwise have to be written and tested.
+*
+* `track_start` is an explicit member, not the absence of an anchor. The
+* agent-harness mutation this maps from treats "neither anchor given" as
+* "move to the front" (`applyBatchMoveVideoClips` falls back to `insertIndex = 0`),
+* which is a default buried in a tool description. Requiring the caller to name
+* that intent keeps a forgotten field from silently reordering the timeline.
+*/
+const moveAnchorSchema = z.discriminatedUnion("position", [
+	z.object({
+		position: z.literal("before"),
+		clip_id: clipIdSchema.describe("The moved block lands immediately before this clip")
+	}),
+	z.object({
+		position: z.literal("after"),
+		clip_id: clipIdSchema.describe("The moved block lands immediately after this clip")
+	}),
+	z.object({ position: z.literal("track_start") })
+]).describe("Where the moved block lands: before/after a reference clip, or at the head of the track");
+/**
+* What happens to the speeches anchored to the clips being moved.
+*
+* - `follow` keeps each speech anchored where it is, so it travels with its clip
+*   to the new position. In the anchored model this is the *no-op* branch: a
+*   speech's authoritative fact is `{ anchorPartId, offsetMs }` and its absolute
+*   time is derived on read from the host's position, so moving the host moves
+*   the speech with no write to the speech at all.
+* - `keep_absolute` preserves each speech's current absolute landing instead, then
+*   re-anchors it to whichever clip now covers that time (RFC 02 §9.1/§11.1). This
+*   is the branch that costs an extra pass, and the one the FE timeline uses.
+*
+* **Required, with no default**, matching `deleteVideoClips` and
+* `replaceVideoClipSequence`. The two branches decide which picture the user's
+* narration ends up over, which is too consequential to infer from a missing
+* field — and neither branch is "safe enough" to be the implicit one.
+*/
+const movedClipAnchoredPolicySchema = z.enum(["follow", "keep_absolute"]);
+/**
+* Reorder a set of main-track clips relative to a reference clip.
+*
+* Distinct from `moveVideoClips`, which positions clips by absolute time
+* (`new_start_ms`) and is what the FE timeline dispatches after a drag. Main-track
+* clips are `sequential`-positioned, so absolute time is not authoritative state
+* there (RFC 02 §4/§7): `moveVideoClips` has to *guess* an index back out of the
+* time it was handed, whereas an anchor already is the ordinal fact being changed.
+* Keeping them separate also isolates blast radius — this method can diverge on
+* speech policy without touching the FE path.
+*
+* `clip_ids` need not be contiguous. They move as one block, keeping their
+* relative order, which is the agent-harness `batch_move_video_clips` contract.
+*/
+const moveVideoClipsByAnchorInputSchema = z.object({
+	clip_ids: clipIdsSchema.describe("Clips to move as one block, keeping their relative order. Need not be contiguous on the track."),
+	anchor: moveAnchorSchema,
+	on_anchored: movedClipAnchoredPolicySchema.describe("What happens to speeches anchored to the moved clips (required — see the policy doc)")
+});
+//#endregion
+//#region src/editor/schemas/move-video-clips.ts
+const moveVideoClipsInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to move"),
+	new_start_ms: timelineMsSchema.describe("New absolute start time in milliseconds on the timeline"),
+	new_track_id: z.string().min(1).optional().describe("Target track ID to move the clip to (optional)")
+})).min(1).describe("List of video clips to move to new positions") });
+//#endregion
+//#region src/editor/schemas/replace-video-clip-content.ts
+/**
+* Replace the media backing existing video clips. The media import runs upstream
+* (Director); its stable result — the new media id, intrinsic length, and the
+* reset trim window — arrives materialized (see
+* `results/phase-4-side-effect-payload-contract.md` §4). Director resets
+* `play_in=0` / `play_out=media_duration_ms` and clears `speed_shift` on
+* replacement. The clip `part_id`s (hence their track items) are unchanged; the
+* editor reflows the main track from the new effective durations.
+*/
+const replaceVideoClipContentInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("Existing video clip part ID to re-point"),
+	origin_media_id: mediaIdSchema.describe("The new media asset ID"),
+	media_duration_ms: positiveMsSchema.describe("The new media's intrinsic full length"),
+	play_in: timelineMsSchema.describe("Trim window start in the new media (usually 0)"),
+	play_out: positiveMsSchema.describe("Trim window end in the new media (usually = media_duration_ms)"),
+	volume: volumeSchema
+})).min(1).describe("Video clips whose media is being replaced") });
+//#endregion
+//#region src/editor/schemas/replace-video-clip-sequence.ts
+/**
+* What happens to the speeches anchored to the clips being replaced.
+*
+* - `remap` re-anchors each surviving speech to the new clip in the SAME POSITION
+*   of the sequence, keeping its offset — old[i]'s children become new[i]'s
+*   children. An old clip with no counterpart (fewer new clips than old) has its
+*   subtree deleted, because there is nothing left to anchor to.
+* - `cascade` deletes every anchored speech (and its captions) outright, like
+*   `deleteVideoClips`.
+*
+* **Required, with no default.** The two branches differ in whether the user's
+* narration survives, and the agent tool that drives this op makes its
+* `preserve_speeches` flag required for that reason. A default here would let a
+* caller that forgot the field silently delete speech.
+*/
+const anchoredReplacePolicySchema = z.enum(["remap", "cascade"]);
+/**
+* Replace a contiguous run of main-track clips with a new run.
+*
+* A composite of delete + insert that cannot be expressed as the two ops in
+* sequence, because the anchored speeches have to survive *across* the swap: with
+* `remap` they are re-anchored positionally, which needs both the old and the new
+* ids in the same transaction (ADR 0009 — the cascade stays in the editor, callers
+* never re-wire anchors themselves).
+*
+* Duration facts follow `addVideoClips`: `media_duration_ms` is the source's
+* intrinsic length and the trim window defaults to the whole media. The effective
+* timeline duration is derived by the projection, never an input.
+*
+* `media_id` is optional: omitting it creates a **deliberate empty placeholder
+* clip** (`origin_media_id: ''`) — structure with no picture. This is the only op
+* that can produce one, and it is authoritative state, unlike the gap fillers the
+* read-side solve mints (which never enter the document).
+*/
+const replaceVideoClipSequenceInputSchema = z.object({
+	old_clip_ids: clipIdsSchema.describe("The clips being replaced: a contiguous main-track run, listed in timeline order"),
+	new_clips: z.array(z.object({
+		media_id: mediaIdSchema.optional().describe("The replacement media asset ID. Omit to create an empty placeholder clip."),
+		media_duration_ms: positiveMsSchema.describe("The source media's intrinsic full length in ms"),
+		play_in: timelineMsSchema.optional().describe("Trim window start in the media (default 0)"),
+		play_out: positiveMsSchema.optional().describe("Trim window end in the media (default media_duration_ms)")
+	})).min(1).describe("The replacement clips, in the order they take on the track"),
+	on_anchored: anchoredReplacePolicySchema.describe("What happens to speeches anchored to the replaced clips (required — see the policy doc)")
+}).superRefine((data, ctx) => {
+	for (let i = 0; i < data.new_clips.length; i++) {
+		const clip = data.new_clips[i];
+		const playIn = clip.play_in ?? 0;
+		const playOut = clip.play_out ?? clip.media_duration_ms;
+		if (playOut > clip.media_duration_ms) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `new_clips[${i}].play_out ${playOut}ms exceeds media_duration_ms ${clip.media_duration_ms}ms`,
+			path: [
+				"new_clips",
+				i,
+				"play_out"
+			]
+		});
+		if (playIn >= playOut) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `new_clips[${i}].play_in ${playIn}ms must be less than play_out ${playOut}ms`,
+			path: [
+				"new_clips",
+				i,
+				"play_in"
+			]
+		});
+	}
+});
+//#endregion
+//#region src/editor/schemas/resize-caption.ts
+/** Resize one caption placement without changing its cue/source duration. */
+const resizeCaptionInputSchema = z.object({
+	caption_id: z.string().min(1),
+	duration_ms: z.number().finite().positive()
+});
+//#endregion
+//#region src/editor/schemas/set-bgm.ts
+/**
+* Set the document BGM. The media's stable result (storage key) arrives
+* materialized from upstream (see
+* `results/phase-4-side-effect-payload-contract.md` §3). The op upserts the bgm
+* part and seats it on the bgm lane; its effective length is always the whole
+* timeline, derived by the projection on read — so there is no `duration_ms`
+* input or fact (RFC 02 / `reference/16` §0b). A `bgm_id` lets the op replace an
+* existing bgm part by id.
+*/
+const setBgmInputSchema = z.object({
+	bgm_id: z.string().min(1).describe("The bgm part ID to write"),
+	audio_storage_key: z.string().min(1),
+	origin_media_id: mediaIdSchema,
+	volume: volumeSchema
+});
+//#endregion
+//#region src/editor/schemas/set-caption-style.ts
+/**
+* Set the caption visual style. GLOBAL by design: the style applies to every
+* caption part in the document — it carries NO `caption_id`. This mirrors the FE,
+* whose caption-style store (`caption-style.ts:persistCaptionStylePatch`) iterates
+* ALL captions and writes the same normalized style to each; the product has a
+* single document-wide caption style, not per-caption styling.
+*
+* Every field is optional and maps to a `CaptionStyle` attribute (snake_case
+* IDL). A field present in the input is written to every caption; a field ABSENT
+* from the input is left untouched on each caption (the editor merges the patch
+* onto each caption's existing style — this is a value edit, not a full-style
+* replace, so a partial patch such as "recolor only" does not wipe font size).
+*
+* Pure document edit, no cascade — captions keep their positions; only the style
+* sub-map of each caption part changes.
+*/
+const setCaptionStyleInputSchema = z.object({
+	font_id: z.string().min(1).optional().describe("Font ID referencing a font from the font library"),
+	font_size: z.number().positive().optional().describe("Font size in points"),
+	font_color: z.string().min(1).optional().describe("Font color as hex string, e.g. \"#FFFFFF\""),
+	font_weight: z.number().int().optional().describe("Numeric font weight, e.g. 400 or 700"),
+	entrance_animation: z.string().optional().describe("Entrance animation preset ID, e.g. \"fade\" or \"none\""),
+	entrance_animation_duration_ms: z.number().min(0).optional().describe("Entrance animation duration in ms"),
+	stroke_color: z.string().min(1).optional().describe("Outline/stroke color as hex string, e.g. \"#000000\""),
+	stroke_width: z.number().min(0).optional().describe("Outline/stroke width in pixels"),
+	position_x: z.number().optional().describe("Caption center X as a fraction (0.0 to 1.0)"),
+	position_y: z.number().optional().describe("Caption center Y as a fraction (0.0 to 1.0)")
+}).describe("Document-wide caption style patch (no caption_id; applies to every caption)");
+//#endregion
+//#region src/editor/schemas/set-caption-visibility.ts
+/**
+* Toggle caption visibility (the caption track's `is_hidden` flag). Pure
+* document edit, no cascade — captions keep their positions; only the lane's
+* hidden flag changes.
+*/
+const setCaptionVisibilityInputSchema = z.object({ is_hidden: z.boolean().describe("Whether the caption track is hidden") });
+//#endregion
+//#region src/editor/schemas/set-video-clip-speed-shift.ts
+/**
+* Set the playback speed of existing video clips. Per the speed-shift decision
+* (`reference/16` §0): the op writes only the `speed_shift` fact — it does NOT
+* store an effective `duration_ms` (projection derives it from the trim window /
+* speed) and does NOT scale anchored speeches' relative offsets (offsets stay
+* put; the cascade reflows absolute positions). A `null` speed_shift clears the
+* speed back to original (1×).
+*/
+const setVideoClipSpeedShiftInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to set speed for"),
+	speed_shift: speedShiftSchema.nullable().describe("The new speed setting, or null to reset to 1×")
+})).min(1).describe("Video clips with their new speed settings") });
+//#endregion
+//#region src/editor/schemas/index.ts
+var schemas_exports = /* @__PURE__ */ __exportAll({
+	addSpeechesInputSchema: () => addSpeechesInputSchema,
+	addVideoClipsInputSchema: () => addVideoClipsInputSchema,
+	adjustBgmVolumeInputSchema: () => adjustBgmVolumeInputSchema,
+	adjustSpeechVolumeInputSchema: () => adjustSpeechVolumeInputSchema,
+	adjustVideoClipDurationInputSchema: () => adjustVideoClipDurationInputSchema,
+	adjustVideoClipVolumeInputSchema: () => adjustVideoClipVolumeInputSchema,
+	anchoredDeletePolicySchema: () => anchoredDeletePolicySchema,
+	anchoredReplacePolicySchema: () => anchoredReplacePolicySchema,
+	changeSpeechScriptInputSchema: () => changeSpeechScriptInputSchema,
+	changeSpeechVoiceInputSchema: () => changeSpeechVoiceInputSchema,
+	clipIdSchema: () => clipIdSchema,
+	clipIdsSchema: () => clipIdsSchema,
+	deleteBgmInputSchema: () => deleteBgmInputSchema,
+	deleteSpeechesInputSchema: () => deleteSpeechesInputSchema,
+	deleteVideoClipsInputSchema: () => deleteVideoClipsInputSchema,
+	mediaIdSchema: () => mediaIdSchema,
+	moveAnchorSchema: () => moveAnchorSchema,
+	moveBgmInputSchema: () => moveBgmInputSchema,
+	moveSpeechesInputSchema: () => moveSpeechesInputSchema,
+	moveVideoClipsByAnchorInputSchema: () => moveVideoClipsByAnchorInputSchema,
+	moveVideoClipsInputSchema: () => moveVideoClipsInputSchema,
+	movedClipAnchoredPolicySchema: () => movedClipAnchoredPolicySchema,
+	positiveMsSchema: () => positiveMsSchema,
+	replaceVideoClipContentInputSchema: () => replaceVideoClipContentInputSchema,
+	replaceVideoClipSequenceInputSchema: () => replaceVideoClipSequenceInputSchema,
+	resizeCaptionInputSchema: () => resizeCaptionInputSchema,
+	setBgmInputSchema: () => setBgmInputSchema,
+	setCaptionStyleInputSchema: () => setCaptionStyleInputSchema,
+	setCaptionVisibilityInputSchema: () => setCaptionVisibilityInputSchema,
+	setVideoClipSpeedShiftInputSchema: () => setVideoClipSpeedShiftInputSchema,
+	speechAssetsSchema: () => speechAssetsSchema,
+	speechIdSchema: () => speechIdSchema,
+	speechIdsSchema: () => speechIdsSchema,
+	speedShiftSchema: () => speedShiftSchema,
+	timelineMsSchema: () => timelineMsSchema,
+	voiceSchema: () => voiceSchema,
+	volumeSchema: () => volumeSchema
+});
+//#endregion
 //#region src/editor/snapshot-utils.ts
 function isMap(value) {
 	return value instanceof Map;
@@ -445,6 +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");
 	}
@@ -454,6 +1029,18 @@ var SchemaValidator = class {
 	validateSetCaptionVisibility(input, _doc) {
 		this.parse(setCaptionVisibilityInputSchema, input, "set_caption_visibility_invalid_input");
 	}
+	validateResizeCaption(input, doc) {
+		const parsed = this.parse(resizeCaptionInputSchema, input, "resize_caption_invalid_input");
+		const snapshot = doc.snapshot();
+		this.assertPartKind(snapshot, parsed.caption_id, "caption", "resize_caption");
+		if (snapshot.tracks?.find((track) => track.parts_kind === "caption")?.items?.find((row) => row.part_id === parsed.caption_id) == null) throw new ValidationError("resize_caption_placement_not_found", `caption_id "${parsed.caption_id}" is not placed on the caption track`, { caption_id: parsed.caption_id });
+		const cueDuration = snapshot.part_library?.[parsed.caption_id]?.caption?.initial_duration_ms ?? 0;
+		if (parsed.duration_ms < cueDuration) throw new ValidationError("resize_caption_below_cue", `duration_ms ${parsed.duration_ms} is shorter than cue duration ${cueDuration}`, {
+			caption_id: parsed.caption_id,
+			duration_ms: parsed.duration_ms,
+			cue_duration_ms: cueDuration
+		});
+	}
 	validateSetCaptionStyle(input, _doc) {
 		this.parse(setCaptionStyleInputSchema, input, "set_caption_style_invalid_input");
 	}
@@ -480,16 +1067,20 @@ var SchemaValidator = class {
 		for (const speech of assets.speeches) this.assertPartKind(snapshot, speech.speech_id, "speech", opName);
 	}
 	/**
-	* Each speech's `anchor_part_id` must point at a video clip that already
-	* exists. A relative speech anchored to a missing clip cannot be positioned
-	* by the projection and has no host to recover to (RFC 02 §4/§11.1), so the
-	* write must be rejected rather than landing a dangling reference. (Restores
-	* the dangling-reference guard the old `validateMaterializedPatch` carried;
-	* ADR 0009.)
+	* Anchored speeches may follow a video or BGM root. A null anchor is an
+	* explicit absolute speech root. Missing and wrong-kind parents are rejected.
 	*/
 	assertAnchorsExist(assets, doc, opName) {
 		const snapshot = doc.snapshot();
-		for (const speech of assets.speeches) this.assertPartKind(snapshot, speech.anchor_part_id, "video_clip", `${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);
@@ -565,11 +1156,9 @@ function readPartLibrary(draft) {
 var SemanticEditor = class {
 	doc;
 	validator;
-	idFactory;
-	constructor(doc, validator = new SchemaValidator(), idFactory = generatePartId) {
+	constructor(doc, validator = new SchemaValidator()) {
 		this.doc = doc;
 		this.validator = validator;
-		this.idFactory = idFactory;
 	}
 	async moveVideoClips(input, options) {
 		this.validator.validateMoveVideoClips(input, this.doc);
@@ -649,7 +1238,7 @@ var SemanticEditor = class {
 			track.items ??= [];
 			let at = insertIndex;
 			for (const clip of input.clips) {
-				const partId = this.idFactory("clip");
+				const partId = generatePartId("clip");
 				setPart(draft, partId, { video_clip: {
 					id: partId,
 					kind: "video_clip",
@@ -661,7 +1250,8 @@ var SemanticEditor = class {
 				track.items.splice(at, 0, {
 					part_id: partId,
 					time_position: { mode: "sequential" },
-					fallback_abs_ms: void 0
+					fallback_abs_ms: void 0,
+					duration_override_ms: void 0
 				});
 				at += 1;
 			}
@@ -696,7 +1286,7 @@ var SemanticEditor = class {
 			const newItems = [];
 			const newClipIds = [];
 			for (const clip of input.new_clips) {
-				const partId = this.idFactory("clip");
+				const partId = generatePartId("clip");
 				setPart(draft, partId, { video_clip: {
 					id: partId,
 					kind: "video_clip",
@@ -708,7 +1298,8 @@ var SemanticEditor = class {
 				newItems.push({
 					part_id: partId,
 					time_position: { mode: "sequential" },
-					fallback_abs_ms: void 0
+					fallback_abs_ms: void 0,
+					duration_override_ms: void 0
 				});
 				newClipIds.push(partId);
 			}
@@ -838,8 +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. */
@@ -857,29 +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));
 	}
 	/**
@@ -921,7 +1544,8 @@ var SemanticEditor = class {
 					mode: "absolute",
 					offsetMs: 0
 				},
-				fallback_abs_ms: void 0
+				fallback_abs_ms: void 0,
+				duration_override_ms: void 0
 			}];
 		}, audit("SetBgm", input, options));
 	}
@@ -944,6 +1568,15 @@ var SemanticEditor = class {
 			track.is_hidden = input.is_hidden;
 		}, audit("SetCaptionVisibility", input, options));
 	}
+	/** Resize the caption placement window; cue/source duration stays immutable. */
+	async resizeCaption(input, options) {
+		this.validator.validateResizeCaption(input, this.doc);
+		this.doc.transact((draft) => {
+			const item = findLaneTrack(draft, "caption")?.items?.find((row) => row.part_id === input.caption_id);
+			if (item == null) return;
+			item.duration_override_ms = input.duration_ms;
+		}, audit("ResizeCaption", input, options));
+	}
 	/**
 	* Set the document-wide caption style. GLOBAL (no `caption_id`): the patch is
 	* merged onto EVERY caption part's `style`, mirroring the FE, which applies a
@@ -1108,7 +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 ??= [];
@@ -1121,8 +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);
@@ -1146,9 +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;
@@ -1190,7 +1822,26 @@ function placeRelative(items, partId, anchorPartId, offsetMs, fallbackAbsMs) {
 	items.push({
 		part_id: partId,
 		time_position: timePosition,
-		fallback_abs_ms: fallbackAbsMs
+		fallback_abs_ms: 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
 	});
 }
 /**
@@ -1316,111 +1967,6 @@ function deleteAnchoredSubtree(draft, clipId) {
 	for (const speechId of anchored) deleteSpeechSubtree(draft, speechId);
 }
 //#endregion
-//#region src/editor/journal.ts
-/**
-* Wire a plain-memory adapter + editor that share a recording id factory, so
-* every mutating transact lands in `journal` with ordered `generated_ids`.
-*/
-function createEditSandbox(document, options) {
-	const adapter = createPlainMemoryAdapter(document, options);
-	return {
-		adapter,
-		editor: new SemanticEditor(adapter, new SchemaValidator(), adapter.idFactory),
-		journal: adapter.journal
-	};
-}
-/**
-* Re-drive `doc` from a recorded journal, forcing each entry's `generated_ids`
-* through a queue-backed id factory. Never mints fresh ids: an empty queue on
-* demand throws `unrecorded id`; leftover ids after an entry throws
-* `unconsumed ids`. Legacy entries without `generated_ids` are treated as `[]`.
-*/
-async function replayJournal(doc, journal) {
-	const queue = [];
-	const idFactory = (_prefix) => {
-		const id = queue.shift();
-		if (id == null) throw new Error("unrecorded id");
-		return id;
-	};
-	const editor = new SemanticEditor(doc, new SchemaValidator(), idFactory);
-	for (const entry of journal) {
-		queue.push(...entry.generated_ids ?? []);
-		await dispatchEntry(editor, entry);
-		if (queue.length > 0) throw new Error("unconsumed ids");
-	}
-}
-/** Dispatch one journal entry to the matching editor method; payload is passed through. */
-async function dispatchEntry(editor, entry) {
-	const payload = entry.payload;
-	const options = entry.intent != null ? { intent: entry.intent } : void 0;
-	switch (entry.kind) {
-		case "MoveVideoClips":
-			await editor.moveVideoClips(payload, options);
-			return;
-		case "MoveVideoClipsByAnchor":
-			await editor.moveVideoClipsByAnchor(payload, options);
-			return;
-		case "DeleteVideoClips":
-			await editor.deleteVideoClips(payload, options);
-			return;
-		case "AddVideoClips":
-			await editor.addVideoClips(payload, options);
-			return;
-		case "AdjustVideoClipVolume":
-			await editor.adjustVideoClipVolume(payload, options);
-			return;
-		case "SetVideoClipSpeedShift":
-			await editor.setVideoClipSpeedShift(payload, options);
-			return;
-		case "ReplaceVideoClipContent":
-			await editor.replaceVideoClipContent(payload, options);
-			return;
-		case "ReplaceVideoClipSequence":
-			await editor.replaceVideoClipSequence(payload, options);
-			return;
-		case "AdjustVideoClipDuration":
-			await editor.adjustVideoClipDuration(payload, options);
-			return;
-		case "AddSpeeches":
-			await editor.addSpeeches(payload, options);
-			return;
-		case "DeleteSpeeches":
-			await editor.deleteSpeeches(payload, options);
-			return;
-		case "MoveSpeeches":
-			await editor.moveSpeeches(payload, options);
-			return;
-		case "ChangeSpeechScript":
-			await editor.changeSpeechScript(payload, options);
-			return;
-		case "ChangeSpeechVoice":
-			await editor.changeSpeechVoice(payload, options);
-			return;
-		case "AdjustSpeechVolume":
-			await editor.adjustSpeechVolume(payload, options);
-			return;
-		case "SetCaptionVisibility":
-			await editor.setCaptionVisibility(payload, options);
-			return;
-		case "SetCaptionStyle":
-			await editor.setCaptionStyle(payload, options);
-			return;
-		case "SetBgm":
-			await editor.setBgm(payload, options);
-			return;
-		case "DeleteBgm":
-			await editor.deleteBgm(payload, options);
-			return;
-		case "AdjustBgmVolume":
-			await editor.adjustBgmVolume(payload, options);
-			return;
-		default: {
-			const _exhaustive = entry.kind;
-			throw new Error(`replayJournal: unsupported kind ${String(_exhaustive)}`);
-		}
-	}
-}
-//#endregion
 //#region src/editor/types.ts
 /** Runtime list of the frozen, implemented kinds (for guards / introspection). */
 const IMPLEMENTED_SEMANTIC_OP_KINDS = [
@@ -1436,9 +1982,11 @@ const IMPLEMENTED_SEMANTIC_OP_KINDS = [
 	"AddSpeeches",
 	"DeleteSpeeches",
 	"MoveSpeeches",
+	"MoveBgm",
 	"ChangeSpeechScript",
 	"ChangeSpeechVoice",
 	"AdjustSpeechVolume",
+	"ResizeCaption",
 	"SetCaptionVisibility",
 	"SetCaptionStyle",
 	"SetBgm",
@@ -2214,24 +2762,6 @@ var MengineDocSession = class {
 		if (this.editorValue == null) throw new Error("mengine doc session is not started");
 		return this.editorValue;
 	}
-	/**
-	* Opaque version token of the local oplog (base64 `VersionVector.encode`).
-	* Equality-comparable only: equal means no observed change (local or
-	* remote-arrived) since the token was taken. Throws when not started.
-	*/
-	version() {
-		if (this.adapterValue == null) throw new Error("mengine doc session is not started");
-		return bytesToBase64(this.adapterValue.doc.oplogVersion().encode());
-	}
-	/**
-	* The live document adapter, exposed for journal replay (commit channel).
-	* Replaying through it still goes SemanticEditor → Loro → mengine-server —
-	* no write bypass. Typed by the narrow interface on purpose.
-	*/
-	get documentAdapter() {
-		if (this.adapterValue == null) throw new Error("mengine doc session is not started");
-		return this.adapterValue;
-	}
 	/** Current document snapshot (read model). */
 	snapshot() {
 		if (this.adapterValue == null) throw new Error("mengine doc session is not started");
@@ -2342,4 +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, PlainMemoryAdapter, SchemaValidator, SemanticEditor, TIMELINE_SKELETON_DURATION_MS, VIDEO_DOCUMENT_SCHEMA_VERSION, ValidationError, VideoDocumentValidationError, arrangeMainTrackSeamlessly, assertValidVideoDocument, base64ToBytes, buildInitialVideoDocument, buildSpeechHostMap, bytesToBase64, cascadeAfterVideoClipChanges, createEditSandbox, createMirrorVideoDocument, createMirrorVideoDocumentAdapter, createPlainMemoryAdapter, decodeDocVersionMark, derivePositionFromAbs, effectiveVideoClipDurationMs, encodeDocVersionMark, ensureLaneTrack, fillMainTrackTimeGaps, findLaneTrack, fromVideoDocument, generatePartId, getAt, hostForAbsMs, isEmptyVideoClip, isImplementedSemanticOpKind, isMap, laneTrackId, mainTrackRanges, partDurationMs, partUnionSchema, readMainTrackItems, readMengineEventStream, readPart, readPartDurationMs, readVideoDocumentFromDraft, reassignSpeechesToVideoClipsByTime, recalculateTimelineDuration, relativePositionForAbs, replayJournal, resolveAllSpeechOverlaps, resolveSpeechOverlapByShiftingVideos, safeDurationMs, schemas_exports as schemas, snapshotToPlain, solveVideoDocument, speedOf, syncAggregatedClipsTimePosition, toVideoDocument, validateVideoDocument, videoDocumentMirrorSchema, videoDocumentSchema, writeVideoDocumentToDraft };
+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 };