@mengine/medeo-client 2.1.1-dsl.1 → 2.1.1-dsl.11

package/dist/{dsl-EV6kCdr6.js → dsl-Co3RI5F1.js}

@@ -1,4 +1,4 @@
-import { $ as assertRelation, A as hardenCidDescriptors, B as safeStringify, C as applySchemaToInferOptions, D as deepEqual, E as decodeNestedJsonValues, F as isObject, G as valuesEqual, H as stripUndefined, I as isObjectLike, J as getChildSchema, K as CID_KEY, L as isStateAndSchemaOfType, M as inferContainerTypeFromValue, N as insertChildToMap, O as defineCidProperty, P as isArrayLike, Q as assertId, R as isTreeID, S as applyEncode, T as containerIdToContainerType, U as tryInferContainerType, V as schemaToContainerType, W as tryUpdateToContainer, X as assertEntity, Y as getMapFieldSchema, Z as assertEntityPatch, _ as isLoroTreeSchema, at as assertRelationIdentity, b as validateSchema, c as isInvalidAtomic, ct as relationId, d as getDefaultValue, et as entityTables, f as isContainerSchema, g as isLoroMovableListSchema, h as isLoroMapSchema, it as validLanguage, j as hasTransform, k as getRootContainerByType, lt as resourceId, m as isLoroListSchema, n as textStyles, nt as mergeFields, o as medeoDslLoroShape, ot as assertResourceIdentity, p as isListLikeSchema, q as getChildContainerSchema, rt as relationTables, s as captionContentCodec, st as defaultIdFactory, t as DEFAULT_UNIT_TIME_MS, tt as isRecord, u as schema, v as isRootSchemaType, w as cidsEqual, x as applyDecode, y as schemaContainsIgnoreField, z as isValueOfContainerType } from "./video-draft-types-TXRv3wS6.js";
+import { $ as assertEntityPatch, A as getRootContainerByType, B as isValueOfContainerType, C as applyEncode, D as decodeNestedJsonValues, E as containerIdToContainerType, F as isArrayLike, G as tryUpdateToContainer, H as schemaToContainerType, I as isObject, J as getChildContainerSchema, K as valuesEqual, L as isObjectLike, M as hasTransform, N as inferContainerTypeFromValue, O as deepEqual, P as insertChildToMap, Q as assertEntity, R as isStateAndSchemaOfType, S as applyDecode, T as cidsEqual, U as stripUndefined, V as safeStringify, W as tryInferContainerType, X as getMapFieldSchema, Y as getChildSchema, Z as assertCaptionContent, _ as isLoroMovableListSchema, at as atomicFields, b as schemaContainsIgnoreField, c as isInvalidAtomic, ct as mergeFields, d as schema, dt as validLanguage, et as assertId, f as getDefaultValue, ft as assertRelationIdentity, g as isLoroMapSchema, gt as resourceId, h as isLoroListSchema, ht as relationId, it as assertTimeAnchor, j as hardenCidDescriptors, k as defineCidProperty, l as phonemeCodec, lt as patchRow, m as isListLikeSchema, mt as defaultIdFactory, n as textStyles, nt as assertPhoneticScript, o as medeoDslLoroShape, ot as entityTables, p as isContainerSchema, pt as assertResourceIdentity, q as CID_KEY, rt as assertRelation, s as captionContentCodec, st as isRecord, t as DEFAULT_UNIT_TIME_MS, tt as assertPhoneme, ut as relationTables, v as isLoroTreeSchema, w as applySchemaToInferOptions, x as validateSchema, y as isRootSchemaType, z as isTreeID } from "./video-draft-types-BnbfWrBx.js";
 import { Cursor, LoroDoc, LoroList, LoroMap, LoroMovableList, LoroText, LoroTree, isContainer, isContainerId } from "loro-crdt";
 import { Immer, setAutoFreeze } from "immer";
 //#region ../medeo-dsl/src/text.ts
@@ -12,20 +12,15 @@ function range(text, start, end) {
 	assertTextOffset(text, end);
 	if (start >= end) throw new Error("Expected a nonempty text range");
 }
-function assertMark(key, value, selected) {
-	if (key !== "language" && key !== "pauseMs") throw new Error(`Unsupported mark: ${key}`);
-	if (value === null) return;
-	if (key === "language" && !validLanguage(value)) throw new Error("Invalid language");
-	if (key === "pauseMs" && (selected !== " " || typeof value !== "number" || !Number.isSafeInteger(value) || value < 0)) throw new Error("pauseMs requires a nonnegative integer on one ordinary space");
+/** A script marks only what language a run is written in; how to say it belongs to its phonetic script. */
+function assertMark(key, value) {
+	if (key !== "language") throw new Error(`Unsupported mark: ${key}`);
+	if (value !== null && !validLanguage(value)) throw new Error("Invalid language");
 }
-function attributes(attrs, selected) {
+function attributes(attrs) {
 	if (!attrs) return;
-	const result = {};
-	for (const [key, value] of Object.entries(attrs)) {
-		assertMark(key, value, selected);
-		result[key] = key === "pauseMs" && value === 0 ? null : value;
-	}
-	return result;
+	for (const [key, value] of Object.entries(attrs)) assertMark(key, value);
+	return { ...attrs };
 }
 function previewDelta(text, delta) {
 	let offset = 0;
@@ -44,7 +39,7 @@ function previewDelta(text, delta) {
 		].includes(key))) throw new Error("Invalid text Delta operation");
 		if ("insert" in op) {
 			if (typeof op.insert !== "string") throw new Error("Only text insertions are supported");
-			const attrs = attributes(op.attributes, op.insert);
+			const attrs = attributes(op.attributes);
 			text = text.slice(0, offset) + op.insert + text.slice(offset);
 			normalized.push({
 				insert: op.insert,
@@ -56,7 +51,7 @@ function previewDelta(text, delta) {
 			if (!Number.isSafeInteger(length) || length < 0) throw new Error("Invalid Delta length");
 			assertTextOffset(text, offset + length);
 			if ("retain" in op) {
-				const attrs = attributes(op.attributes, text.slice(offset, offset + length));
+				const attrs = attributes(op.attributes);
 				normalized.push({
 					retain: length,
 					...attrs ? { attributes: attrs } : {}
@@ -91,11 +86,11 @@ function prepareTextEdits(text, edits) {
 					break;
 				case "mark":
 					range(text, edit.start, edit.end);
-					assertMark(edit.key, edit.value, text.slice(edit.start, edit.end));
+					assertMark(edit.key, edit.value);
 					break;
 				case "unmark":
 					range(text, edit.start, edit.end);
-					assertMark(edit.key, null, "");
+					assertMark(edit.key, null);
 					break;
 				case "delta": {
 					const result = previewDelta(text, edit.delta);
@@ -109,7 +104,28 @@ function prepareTextEdits(text, edits) {
 			}
 			return edit;
 		}),
-		body: text
+		text
+	};
+}
+/** The language a whole segment carries: one shared mark on every unit, else none. */
+function markedLanguage(text) {
+	if (!text.text.length) return void 0;
+	const first = text.langs[0];
+	for (let i = 1; i < text.text.length; i++) if (text.langs[i] !== first) return void 0;
+	return first;
+}
+/** Read language marks off a rich-text delta. */
+function langsFromDelta(delta) {
+	let text = "";
+	const langs = [];
+	for (const run of delta) {
+		text += run.insert;
+		const language = typeof run.attributes?.language === "string" ? run.attributes.language : void 0;
+		for (let i = 0; i < run.insert.length; i++) langs.push(language);
+	}
+	return {
+		text,
+		langs
 	};
 }
 //#endregion
@@ -122,73 +138,116 @@ function requireEntity(rows, table, id) {
 	return row;
 }
 /**
-* Whether one relation row may stand among these rows: its fields, its derived ID, and what
-* its kind requires of each end. A row being linked need not be in `rows` yet. Caption content
-* selects text, so its range is checked against the body with {@link assertCaptionSelection}.
+* Whether one relation row may stand among these rows: its fields, its derived ID, and what its
+* kind requires of each end. A row being linked need not be in `rows` yet. Relations an entity
+* holds at most one of are fields of that entity; see {@link assertReferenceAdmitted}.
 */
 function assertRelationAdmitted(rows, table, id, value) {
 	assertId(id);
 	assertRelation(table, value);
 	assertRelationIdentity(table, id, value);
 	const v = value;
-	const require = (endpoint, id) => requireEntity(rows, endpoint, id);
 	switch (table) {
-		case "clip-source-relations":
-			require("clip-entities", v.clipId);
-			if (!rows["media-entities"][v.sourceId] && !rows["caption-entities"][v.sourceId]) throw new Error("Invalid Clip source");
-			break;
-		case "time-anchor-relations": {
-			require("clip-entities", v.from);
-			if (rows["timeline-entities"][v.to]) {
-				if (v.offsetMs < 0) throw new Error("Absolute offset must be nonnegative");
-			} else require("clip-entities", v.to);
-			const visited = /* @__PURE__ */ new Set();
-			let from = v.from;
-			while (true) {
-				if (visited.has(from)) throw new Error("Time anchor cycle");
-				visited.add(from);
-				const anchor = from === v.from ? v : rows["time-anchor-relations"][`ta:${from}`];
-				if (!isRecord(anchor) || typeof anchor.to !== "string") break;
-				from = anchor.to;
-			}
-			break;
-		}
-		case "caption-content-relations":
-			assertCaptionEndpoints(rows, v);
-			break;
-		case "caption-alignment-relations": {
-			require("caption-entities", v.captionId);
-			const media = require("media-entities", v.mediaId);
-			if (media.kind === "image" || v.startMs >= v.endMs || v.endMs > media.durationMs) throw new Error("Invalid source alignment");
-			break;
-		}
 		case "voice-timbre-relations":
-			if (require("media-entities", v.recordingId).kind !== "voiceover-recording") throw new Error("Timbre requires a voiceover recording");
-			require("voice-entities", v.voiceId);
+			if (requireEntity(rows, "media-entities", v.recordingId).kind !== "voiceover-recording") throw new Error("Timbre requires a voiceover recording");
+			requireEntity(rows, "voice-entities", v.voiceId);
 			break;
 		case "derived-from-relations":
 			for (const end of [v.entityId, v.sourceId]) if (!entityTables.some((t) => Object.hasOwn(rows[t], end))) throw new Error("Missing provenance endpoint");
 			break;
-		case "from-asset-relations":
-		case "to-asset-relations": {
-			require("audio-script-entities", v.audioScriptId);
-			const asset = require("asset-entities", v.assetId);
-			if (table === "to-asset-relations" && !asset.digest) throw new Error("Export Asset requires digest");
+	}
+}
+/** Entity fields that hold a single-valued relation, by the entity that owns it. */
+const referenceFields = {
+	"clip-entities": ["source", "anchor"],
+	"phonetic-script-entities": ["phoneme"],
+	"caption-entities": ["selection"]
+};
+/** The segment a selection names: refused when its script or the segment is missing. */
+function requireSegment(rows, audioScriptId, segmentId) {
+	const script = requireEntity(rows, "audio-script-entities", audioScriptId);
+	const segment = Array.isArray(script.segments) ? script.segments.find((s) => isRecord(s) && s.id === segmentId) : void 0;
+	if (!segment) throw new Error(`Missing segment ${String(segmentId)} of ${String(audioScriptId)}`);
+	return segment;
+}
+/**
+* Whether a single-valued relation, stored as a field of the entity that owns it, may hold this
+* value among these rows. The owner need not be stored yet; an absent value unlinks it.
+* A selection's run is checked against the segment's live text by whoever holds it.
+*/
+function assertReferenceAdmitted(rows, table, id, field, value) {
+	assertId(id);
+	const key = `${table}.${field}`;
+	if (value === void 0) {
+		if (key === "clip-entities.source") throw new Error(`Missing required relation: ${key}`);
+		return;
+	}
+	switch (key) {
+		case "clip-entities.source":
+			assertId(value);
+			if (!Object.hasOwn(rows["media-entities"], value) && !Object.hasOwn(rows["caption-entities"], value)) throw new Error("Invalid Clip source");
+			break;
+		case "clip-entities.anchor": {
+			assertTimeAnchor(value);
+			const anchor = value;
+			if (Object.hasOwn(rows["timeline-entities"], anchor.to)) {
+				if (anchor.offsetMs < 0) throw new Error("Absolute offset must be nonnegative");
+				break;
+			}
+			requireEntity(rows, "clip-entities", anchor.to);
+			const visited = new Set([id]);
+			for (let next = anchor.to; typeof next === "string";) {
+				if (visited.has(next)) throw new Error("Time anchor cycle");
+				visited.add(next);
+				const clip = rows["clip-entities"][next];
+				next = isRecord(clip) && isRecord(clip.anchor) ? clip.anchor.to : void 0;
+			}
+			break;
+		}
+		case "caption-entities.selection":
+		case "phonetic-script-entities.phoneme": {
+			if (key === "caption-entities.selection") assertCaptionContent(value);
+			else assertPhoneme(value);
+			const { audioScriptId, segmentId } = value;
+			requireSegment(rows, audioScriptId, segmentId);
 			break;
 		}
 	}
 }
-/** Both ends of a caption's selection exist: the caption and the script it selects from. */
+/** Both ends of a caption's selection exist: the caption and the segment it selects. */
 function assertCaptionEndpoints(rows, selection) {
 	requireEntity(rows, "caption-entities", selection.captionId);
-	requireEntity(rows, "audio-script-entities", selection.audioScriptId);
+	requireSegment(rows, selection.audioScriptId, selection.segmentId);
+}
+/** Both ends of a phoneme exist: the PhoneticScript and the segment it is written for. */
+function assertPhonemeEndpoints(rows, selection) {
+	requireEntity(rows, "phonetic-script-entities", selection.phoneticScriptId);
+	requireSegment(rows, selection.audioScriptId, selection.segmentId);
+}
+/**
+* A selection of a segment's current text: the whole segment, or a run of it in UTF-16 offsets
+* with both ends stated. What it selects is never empty.
+*/
+function assertSegmentRange(range, text) {
+	if (range.start === void 0 !== (range.end === void 0)) throw new Error("A run inside a segment states both its start and its end");
+	if (range.start === void 0 || range.end === void 0) {
+		if (!text.length) throw new Error("An empty segment cannot be selected");
+		return;
+	}
+	assertTextOffset(text, range.start);
+	assertTextOffset(text, range.end);
+	if (range.start >= range.end) throw new Error("Expected a nonempty range in the current segment");
 }
-/** A caption selects a nonempty run of its script's current body, in UTF-16 offsets. */
-function assertCaptionSelection(rows, selection, body) {
+/** A caption selects a segment of its script, or a nonempty run of it. */
+function assertCaptionSelection(rows, selection, text) {
 	assertCaptionEndpoints(rows, selection);
-	assertTextOffset(body, selection.start);
-	assertTextOffset(body, selection.end);
-	if (selection.start >= selection.end) throw new Error("Expected a nonempty range in the current body");
+	assertSegmentRange(selection, text);
+}
+/** A PhoneticScript is written for a segment of its script, or a nonempty run of it, as a Caption selects one. */
+function assertPhonemeSelection(rows, selection, text) {
+	assertPhonemeEndpoints(rows, selection);
+	assertSegmentRange(selection, text);
+	if (selection.phonemeScript !== void 0 && typeof selection.phonemeScript !== "string") throw new Error("phonemeScript is a string");
 }
 /** A caption's text comes from a selection of a script or from its own text, never both. */
 function assertCaptionSource(caption, selected) {
@@ -261,8 +320,9 @@ function replacementPatch(before, after) {
 /**
 * Admission of one whole change, asked once before it is published. Only what the change touched
 * is checked, so an unrelated edit still lands in a merged document that holds invalid rows.
-* Containment and caption selections are checked by whoever holds the tree and the script bodies,
-* with `assertContainTree` and `assertCaptionSelection`.
+* Single-valued relations are fields of their owner and are checked when the owner's field is
+* touched. Containment and caption ranges are checked by whoever holds the tree and the script
+* bodies, with `assertContainTree` and `assertCaptionSelection`.
 */
 function assertChangeAdmitted(change) {
 	const { before, after } = change;
@@ -278,18 +338,28 @@ function assertChangeAdmitted(change) {
 			if (!isRecord(value)) throw new Error("Expected an entity object");
 			assertEntityPatch(table, Object.fromEntries([...fields].map((field) => [field, replacementPatch(old[field], value[field])])), old);
 		}
-		const marker = value.seqMarker;
+		const row = value;
+		for (const field of referenceFields[table] ?? []) if (!old || fields.has(field)) assertReferenceAdmitted(after, table, id, field, row[field]);
+		const marker = row.seqMarker;
 		if (table === "clip-entities" && (!old || fields.has("seqMarker")) && marker?.outMs <= marker?.inMs) throw new Error("Clip window must be nonempty");
+		if (table === "clip-entities" && (!old || fields.has("seqMarker") || fields.has("source"))) assertFillsWithWholeSource(after, marker, row.source);
 	}
 	if (change.entities.has("timeline-entities") && Object.keys(after["timeline-entities"]).length > 1) throw new Error("Document already has a Timeline");
-	for (const [table, ids] of change.relations) {
-		if (table === "caption-content-relations") continue;
-		for (const id of ids) {
-			const value = after[table][id];
-			if (value !== void 0) assertRelationAdmitted(after, table, id, value);
-		}
+	for (const [table, ids] of change.relations) for (const id of ids) {
+		const value = after[table][id];
+		if (value !== void 0) assertRelationAdmitted(after, table, id, value);
 	}
 }
+/**
+* A Clip that fills the Timeline plays its whole source at 1×: the player loops a short source and
+* cuts a long one at the Timeline's end (RFC 02 §3.3), so a window or speed of its own has no meaning.
+*/
+function assertFillsWithWholeSource(rows, marker, source) {
+	if (marker?.fillTimeline !== true) return;
+	const media = typeof source === "string" ? rows["media-entities"][source] : void 0;
+	const durationMs = typeof media?.durationMs === "number" ? media.durationMs : void 0;
+	if (marker.inMs !== 0 || marker.rate !== void 0 && marker.rate !== 1 || durationMs !== void 0 && marker.outMs !== durationMs) throw new Error(`A Clip that fills the Timeline plays its whole source at 1×: inMs 0, outMs ${durationMs ?? "its length"}, no rate; the Timeline's end cuts it`);
+}
 //#endregion
 //#region src/dsl/projection/types.ts
 /** Unsupported legacy representations fail explicitly; recoverable CRDT state is returned as diagnostics. */
@@ -331,9 +401,9 @@ function resolveDslLayout(snapshot) {
 		const clips = [];
 		for (const child of node.children) {
 			const id = child.entityId, clip = snapshot["clip-entities"][id];
-			const sourceId = snapshot["clip-source-relations"][`cs:${id}`]?.sourceId;
-			const media = sourceId === void 0 ? void 0 : snapshot["media-entities"][sourceId];
-			const caption = sourceId === void 0 ? void 0 : snapshot["caption-entities"][sourceId];
+			const sourceId = clip.source;
+			const media = snapshot["media-entities"][sourceId];
+			const caption = snapshot["caption-entities"][sourceId];
 			const marker = clip.seqMarker;
 			let inMs = marker.inMs, outMs = marker.outMs;
 			const present = media !== void 0 || caption !== void 0;
@@ -347,7 +417,7 @@ function resolveDslLayout(snapshot) {
 			const rate = marker.rate ?? 1;
 			const duration = present ? Math.max(0, Math.round((outMs - inMs) / rate)) : 0;
 			if (!Number.isSafeInteger(duration)) projectionError("time-overflow", `clip-entities/${id}`, "Duration exceeds safe integer milliseconds");
-			const anchor = snapshot["time-anchor-relations"][`ta:${id}`];
+			const anchor = clip.anchor;
 			const placement = {
 				clipId: id,
 				trackId: node.entityId,
@@ -384,7 +454,7 @@ function resolveDslLayout(snapshot) {
 			}
 			seen.set(current, path.length);
 			path.push(current);
-			current = snapshot["time-anchor-relations"][`ta:${current}`]?.to;
+			current = snapshot["clip-entities"][current]?.anchor?.to;
 		}
 		for (const member of path) done.add(member);
 	}
@@ -393,7 +463,7 @@ function resolveDslLayout(snapshot) {
 	const position = (id) => {
 		const known = positions.get(id);
 		if (known) return known;
-		const clip = result.clips[id], anchor = snapshot["time-anchor-relations"][`ta:${id}`];
+		const clip = result.clips[id], anchor = snapshot["clip-entities"][id]?.anchor;
 		let value;
 		if (anchor && clip.mode === "anchored" && !cycle.has(id) && result.clips[anchor.to]) {
 			const parent = position(anchor.to);
@@ -500,14 +570,6 @@ function resolveDslLayout(snapshot) {
 		const a = track.clips[i];
 		for (const b of track.clips.slice(i + 1)) if (overlap(a, b)) diagnose("unresolved-overlap", b.clipId, `Overlaps ${a.clipId}; fixed positions, anchors and complete windows are retained`);
 	}
-	for (const [id, relation] of Object.entries(snapshot["caption-alignment-relations"])) {
-		const media = snapshot["media-entities"][relation.mediaId];
-		if (!media || media.kind === "image" || relation.endMs <= relation.startMs || relation.endMs > media.durationMs) result.diagnostics.push({
-			code: "invalid-caption-alignment",
-			path: `caption-alignment-relations/${id}`,
-			message: "No valid source alignment; the placed display window is retained"
-		});
-	}
 	return result;
 }
 function overlap(a, b) {
@@ -563,9 +625,10 @@ function timelineId(s) {
 /** Placement facts are authored only by explicit actions; projection never writes them back. */
 function placementMutations(s, clipId, placement) {
 	if (placement.kind === "sequential") return [{
-		type: "delete-relation",
-		table: "time-anchor-relations",
-		id: `ta:${clipId}`
+		type: "patch-entity",
+		table: "clip-entities",
+		id: clipId,
+		patch: { anchor: void 0 }
 	}];
 	const offset = placement.kind === "absolute" ? placement.startMs : placement.offsetMs;
 	if (!Number.isSafeInteger(offset) || placement.kind === "absolute" && offset < 0) throw new Error("Invalid placement time");
@@ -578,27 +641,31 @@ function placementMutations(s, clipId, placement) {
 		while (next && s["clip-entities"][next]) {
 			if (visited.has(next)) throw new Error("Cyclic time anchor");
 			visited.add(next);
-			next = s["time-anchor-relations"][`ta:${next}`]?.to;
+			next = s["clip-entities"][next]?.anchor?.to;
 		}
 		fallback = Math.max(0, resolveDslLayout(s).clips[target].startMs + offset);
 	}
 	if (!Number.isSafeInteger(fallback)) throw new Error("Invalid fallback time");
 	return [{
-		type: "set-relation",
-		table: "time-anchor-relations",
-		value: {
-			from: clipId,
+		type: "patch-entity",
+		table: "clip-entities",
+		id: clipId,
+		patch: { anchor: {
 			to: target,
 			offsetMs: offset,
 			fallback_abs_ms: fallback
-		}
+		} }
 	}];
 }
+/** Clips whose time anchor points at a Clip in `ids`, with that anchor. */
+function anchoredTo(s, ids) {
+	return Object.entries(s["clip-entities"]).flatMap(([id, clip]) => clip.anchor && ids.has(clip.anchor.to) ? [id] : []);
+}
 function detachChildren(s, ids) {
 	const layout = resolveDslLayout(s);
-	return Object.values(s["time-anchor-relations"]).flatMap((r) => ids.has(r.to) && !ids.has(r.from) && layout.clips[r.from] ? placementMutations(s, r.from, {
+	return anchoredTo(s, ids).flatMap((id) => !ids.has(id) && layout.clips[id] ? placementMutations(s, id, {
 		kind: "absolute",
-		startMs: layout.clips[r.from].startMs
+		startMs: layout.clips[id].startMs
 	}) : []);
 }
 function deleteClips(s, ids, policy) {
@@ -608,9 +675,9 @@ function deleteClips(s, ids, policy) {
 	const removed = new Set(ids);
 	if (policy === "cascade") {
 		const queue = [...ids];
-		for (let i = 0; i < queue.length; i++) for (const r of Object.values(s["time-anchor-relations"])) if (r.to === queue[i] && s["clip-entities"][r.from] && !removed.has(r.from)) {
-			removed.add(r.from);
-			queue.push(r.from);
+		for (let i = 0; i < queue.length; i++) for (const id of anchoredTo(s, new Set([queue[i]]))) if (!removed.has(id)) {
+			removed.add(id);
+			queue.push(id);
 		}
 	}
 	const mutations = policy === "detach" ? detachChildren(s, removed) : [];
@@ -620,14 +687,6 @@ function deleteClips(s, ids, policy) {
 			entityId: id
 		});
 		mutations.push({
-			type: "delete-relation",
-			table: "time-anchor-relations",
-			id: `ta:${id}`
-		}, {
-			type: "delete-relation",
-			table: "clip-source-relations",
-			id: `cs:${id}`
-		}, {
 			type: "delete-entity",
 			table: "clip-entities",
 			id
@@ -640,11 +699,10 @@ function deleteClips(s, ids, policy) {
 /** Batch edits validate against preceding prepared facts without touching the live document. */
 function updateClipsPlan(base, updates) {
 	uniqueIds(updates.map((update) => update.clipId));
-	const clips = { ...base["clip-entities"] }, anchors = { ...base["time-anchor-relations"] };
+	const clips = { ...base["clip-entities"] };
 	const preview = {
 		...base,
-		"clip-entities": clips,
-		"time-anchor-relations": anchors
+		"clip-entities": clips
 	};
 	const mutations = [];
 	for (const update of updates) {
@@ -660,9 +718,7 @@ function updateClipsPlan(base, updates) {
 		if (clip.seqMarker.fillTimeline && (windowChanged || Object.hasOwn(marker, "rate"))) throw new Error("BGM trim and speed are not enabled");
 		const next = mergeFields(clip, patch);
 		if (windowChanged) {
-			const sourceId = preview["clip-source-relations"][`cs:${update.clipId}`]?.sourceId;
-			if (!sourceId) throw new Error("Missing Clip source");
-			const source = requireSource(preview, sourceId, requireTrack(preview, trackOf(preview, update.clipId)).role);
+			const source = requireSource(preview, clip.source, requireTrack(preview, trackOf(preview, update.clipId)).role);
 			validateWindow(next.seqMarker, "durationMs" in source ? source.durationMs : void 0);
 		}
 		if (Object.keys(patch).length) mutations.push({
@@ -676,8 +732,7 @@ function updateClipsPlan(base, updates) {
 			trackOf(preview, update.clipId);
 			const changes = placementMutations(preview, update.clipId, update.placement);
 			mutations.push(...changes);
-			for (const change of changes) if (change.type === "set-relation" && change.table === "time-anchor-relations") anchors[`ta:${update.clipId}`] = change.value;
-			else if (change.type === "delete-relation") delete anchors[change.id];
+			for (const change of changes) if (change.type === "patch-entity" && change.table === "clip-entities") clips[update.clipId] = mergeFields(clips[update.clipId], change.patch);
 		}
 	}
 	return mutations;
@@ -724,7 +779,10 @@ function addClipsPlan(s, input, ids) {
 			marker.fillTimeline = true;
 		}
 		validateWindow(marker, duration);
-		const clip = { seqMarker: marker };
+		const clip = {
+			seqMarker: marker,
+			source: source.id
+		};
 		if (item.volume !== void 0) clip.volume = item.volume;
 		const id = ids[offset];
 		const result = [
@@ -735,14 +793,6 @@ function addClipsPlan(s, input, ids) {
 				id,
 				value: clip
 			},
-			{
-				type: "set-relation",
-				table: "clip-source-relations",
-				value: {
-					clipId: id,
-					sourceId: source.id
-				}
-			},
 			{
 				type: "insert-node",
 				entityId: id,
@@ -762,10 +812,9 @@ function moveClipsPlan(s, input) {
 	if (!Number.isSafeInteger(input.index) || input.index < 0 || input.index > target.length) throw new Error("Invalid insertion index");
 	for (const id of input.clipIds) {
 		trackOf(s, id);
-		if (s["time-anchor-relations"][`ta:${id}`]) throw new Error("Sequential reorder requires sequential Clips");
-		const source = s["clip-source-relations"][`cs:${id}`];
-		if (!source) throw new Error("Missing Clip source");
-		requireSource(s, source.sourceId, requireTrack(s, input.trackId).role);
+		const clip = requireClip(s, id);
+		if (clip.anchor) throw new Error("Sequential reorder requires sequential Clips");
+		requireSource(s, clip.source, requireTrack(s, input.trackId).role);
 	}
 	const mutations = input.onAnchored === "keepAbsolute" ? detachChildren(s, moved) : [];
 	const order = [...target];
@@ -783,7 +832,7 @@ function replaceSequencePlan(s, input, ids) {
 	uniqueIds(input.clipIds);
 	if (!input.clipIds.length || !input.clips.length) throw new Error("Replacement requires nonempty runs");
 	const trackId = trackOf(s, input.clipIds[0]), order = children(s, trackId), index = order.indexOf(input.clipIds[0]);
-	if (input.clipIds.some((id, i) => order[index + i] !== id || s["time-anchor-relations"][`ta:${id}`])) throw new Error("Expected an ordered contiguous sequential run");
+	if (input.clipIds.some((id, i) => order[index + i] !== id || s["clip-entities"][id]?.anchor)) throw new Error("Expected an ordered contiguous sequential run");
 	if (input.clips.some((c) => c.placement && c.placement.kind !== "sequential")) throw new Error("Replacement Clips must be sequential");
 	const mutations = addClipsPlan(s, {
 		trackId,
@@ -791,7 +840,7 @@ function replaceSequencePlan(s, input, ids) {
 		index
 	}, ids);
 	const removed = new Set(input.clipIds);
-	const anchors = { ...s["time-anchor-relations"] };
+	const clips = { ...s["clip-entities"] };
 	if (input.onAnchored === "remap") {
 		const begin = resolveDslLayout(s).clips[input.clipIds[0]].startMs;
 		const starts = [];
@@ -803,26 +852,30 @@ function replaceSequencePlan(s, input, ids) {
 			const marker = creation.value.seqMarker;
 			end += Math.round((marker.outMs - marker.inMs) / (marker.rate ?? 1));
 		}
-		for (const r of Object.values(s["time-anchor-relations"])) if (removed.has(r.to) && !removed.has(r.from)) {
-			const slot = input.clipIds.indexOf(r.to);
-			if (slot >= ids.length) continue;
+		for (const child of anchoredTo(s, removed)) {
+			const anchor = s["clip-entities"][child].anchor;
+			const slot = input.clipIds.indexOf(anchor.to);
+			if (removed.has(child) || slot >= ids.length) continue;
 			const value = {
-				from: r.from,
 				to: ids[slot],
-				offsetMs: r.offsetMs,
-				fallback_abs_ms: Math.max(0, starts[slot] + r.offsetMs)
+				offsetMs: anchor.offsetMs,
+				fallback_abs_ms: Math.max(0, starts[slot] + anchor.offsetMs)
 			};
 			mutations.push({
-				type: "set-relation",
-				table: "time-anchor-relations",
-				value
+				type: "patch-entity",
+				table: "clip-entities",
+				id: child,
+				patch: { anchor: value }
 			});
-			anchors[`ta:${r.from}`] = value;
+			clips[child] = {
+				...clips[child],
+				anchor: value
+			};
 		}
 	}
 	mutations.push(...deleteClips({
 		...s,
-		"time-anchor-relations": anchors
+		"clip-entities": clips
 	}, input.clipIds, "cascade"));
 	return mutations;
 }
@@ -830,26 +883,18 @@ function replaceSourcePlan(s, clipId, input) {
 	if (requireClip(s, clipId).seqMarker.fillTimeline) throw new Error("Use setBgm to replace a fill Clip");
 	const source = sourcePlan(s, input, trackOf(s, clipId));
 	validateWindow(input, "durationMs" in source.source ? source.source.durationMs : void 0);
-	return [
-		...source.mutations,
-		{
-			type: "patch-entity",
-			table: "clip-entities",
-			id: clipId,
-			patch: { seqMarker: {
+	return [...source.mutations, {
+		type: "patch-entity",
+		table: "clip-entities",
+		id: clipId,
+		patch: {
+			seqMarker: {
 				inMs: input.inMs,
 				outMs: input.outMs
-			} }
-		},
-		{
-			type: "set-relation",
-			table: "clip-source-relations",
-			value: {
-				clipId,
-				sourceId: source.id
-			}
+			},
+			source: source.id
 		}
-	];
+	}];
 }
 const produce = new Immer({
 	autoFreeze: false,
@@ -3868,11 +3913,21 @@ function rootMap(doc, name) {
 	const value = doc.getContainerById(id);
 	return value?.kind() === "Map" ? value : void 0;
 }
-function liveText(doc, scriptId, caption = false) {
-	const row = rootMap(doc, caption ? "caption-entities" : "audio-script-entities")?.get(scriptId);
-	if (!isContainer(row) || row.kind() !== "Map") return;
-	const body = row.get(caption ? "text" : "body");
-	return isContainer(body) && body.kind() === "Text" && !body.isDeleted() ? body : void 0;
+function asMap(value) {
+	return isContainer(value) && value.kind() === "Map" ? value : void 0;
+}
+function asText(value) {
+	return isContainer(value) && value.kind() === "Text" && !value.isDeleted() ? value : void 0;
+}
+/** The live Text a target names: a Caption's own, or a script segment's. */
+function liveText(doc, target) {
+	if ("captionId" in target) return asText(asMap(rootMap(doc, "caption-entities")?.get(target.captionId))?.get("text"));
+	const segments = asMap(rootMap(doc, "audio-script-entities")?.get(target.audioScriptId))?.get("segments");
+	if (!isContainer(segments) || segments.kind() !== "MovableList") return;
+	for (const item of segments.toArray()) {
+		const segment = asMap(item);
+		if (segment?.get("id") === target.segmentId) return asText(segment.get("text"));
+	}
 }
 function readContainNodes(nodes) {
 	const read = (nodes) => nodes.map((node) => ({
@@ -3882,19 +3937,56 @@ function readContainNodes(nodes) {
 	}));
 	return read(nodes);
 }
+/** What a selection holds now: its whole segment, or the run between its Cursors on that segment's Text. */
 function resolveContent(doc, content) {
-	const text = liveText(doc, content.audioScriptId);
-	if (!text || content.start.containerId() !== text.id || content.end.containerId() !== text.id) return;
+	const text = liveText(doc, {
+		audioScriptId: content.audioScriptId,
+		segmentId: content.segmentId
+	});
+	if (!text) return;
+	const whole = text.toString();
+	if (content.start === void 0 || content.end === void 0) return {
+		text: whole,
+		start: 0,
+		end: whole.length,
+		bodyId: text.id
+	};
+	if (content.start.containerId() !== text.id || content.end.containerId() !== text.id) return;
 	const start = doc.getCursorPos(content.start)?.offset;
 	const end = doc.getCursorPos(content.end)?.offset;
 	if (start === void 0 || end === void 0 || start > end) return;
 	return {
-		text: text.toString().slice(start, end),
+		text: whole.slice(start, end),
 		start,
 		end,
 		bodyId: text.id
 	};
 }
+function richTextOf(text) {
+	return {
+		bodyId: text.id,
+		plainText: text.toString(),
+		delta: text.toDelta().flatMap((run) => typeof run.insert === "string" ? [{
+			insert: run.insert,
+			...run.attributes ? { attributes: run.attributes } : {}
+		}] : [])
+	};
+}
+/** Optional relation fields read apart from their row: an unreadable one is dropped, the row is kept. */
+const optionalReferences = {
+	"clip-entities": ["anchor"],
+	"caption-entities": ["selection"]
+};
+/** The entity a single-valued relation field points at, by the field's path. */
+function referenceTargets(table, row) {
+	const nested = (field, key) => isRecord(row[field]) ? row[field][key] : void 0;
+	switch (table) {
+		case "clip-entities": return [["source", row.source], ["anchor/to", nested("anchor", "to")]];
+		case "caption-entities": return [["selection/audioScriptId", nested("selection", "audioScriptId")]];
+		case "phonetic-script-entities": return [["phoneme/audioScriptId", nested("phoneme", "audioScriptId")]];
+		default: return [];
+	}
+}
 function readMedeoDsl(doc, state) {
 	const diagnostics = [];
 	const diagnostic = (code, path, error) => diagnostics.push({
@@ -3907,44 +3999,42 @@ function readMedeoDsl(doc, state) {
 	const richText = {};
 	for (const table of entityTables) for (const [id, raw] of Object.entries(state[table] ?? {})) try {
 		assertId(id);
-		assertEntity(table, raw);
-		const v = raw;
+		let v = raw;
+		for (const field of optionalReferences[table] ?? []) {
+			const value = v[field];
+			if (!isInvalidAtomic(value)) continue;
+			diagnostic("invalid-relation", `${table}/${id}/${field}`, value.reason);
+			const { [field]: _dropped, ...rest } = v;
+			v = rest;
+		}
+		assertEntity(table, v);
 		assertResourceIdentity(table, id, v);
-		entities[table][id] = raw;
-		if (table === "audio-script-entities" || table === "caption-entities" && v.text !== void 0) {
-			const text = liveText(doc, id, table === "caption-entities");
-			if (!text) throw new Error("Missing live LoroText body");
-			richText[id] = {
-				bodyId: text.id,
-				plainText: text.toString(),
-				delta: text.toDelta().flatMap((run) => typeof run.insert === "string" ? [{
-					insert: run.insert,
-					...run.attributes ? { attributes: run.attributes } : {}
-				}] : [])
-			};
-		}
+		if (table === "phonetic-script-entities") assertPhoneticScript(void 0, v);
+		const texts = [];
+		if (table === "audio-script-entities") for (const segment of v.segments) texts.push([segment.id, liveText(doc, {
+			audioScriptId: id,
+			segmentId: segment.id
+		})]);
+		if (table === "caption-entities" && v.text !== void 0) texts.push([id, liveText(doc, { captionId: id })]);
+		for (const [, text] of texts) if (!text) throw new Error("Missing live LoroText");
+		entities[table][id] = v;
+		for (const [key, text] of texts) richText[key] = richTextOf(text);
 	} catch (error) {
 		delete entities[table][id];
 		diagnostic("invalid-entity", `${table}/${id}`, error);
 	}
 	const entityIds = new Set(entityTables.flatMap((t) => Object.keys(entities[t])));
+	for (const table of entityTables) for (const [id, row] of Object.entries(entities[table])) for (const [path, target] of referenceTargets(table, row)) if (typeof target === "string" && !entityIds.has(target)) diagnostic("dangling-reference", `${table}/${id}/${path}`, `Missing entity ${target}`);
 	for (const table of relationTables) for (const [id, value] of Object.entries(state[table] ?? {})) try {
 		if (isInvalidAtomic(value)) throw new Error(value.reason);
 		assertRelation(table, value);
 		assertRelationIdentity(table, id, value);
 		relations[table][id] = value;
 		for (const key of [
-			"clipId",
-			"sourceId",
-			"from",
-			"to",
-			"captionId",
-			"audioScriptId",
-			"mediaId",
 			"recordingId",
 			"voiceId",
 			"entityId",
-			"assetId"
+			"sourceId"
 		]) {
 			const endpoint = value[key];
 			if (typeof endpoint === "string" && !entityIds.has(endpoint)) diagnostic("dangling-reference", `${table}/${id}/${key}`, `Missing entity ${endpoint}`);
@@ -3972,22 +4062,33 @@ function readMedeoDsl(doc, state) {
 		}];
 	});
 	const captionText = {};
-	for (const [id, content] of Object.entries(relations["caption-content-relations"])) try {
-		const resolved = resolveContent(doc, content);
-		if (resolved) captionText[content.captionId] = resolved;
-		else diagnostic("invalid-caption-content", `caption-content-relations/${id}`, "Cursor does not resolve in the current body");
-	} catch (error) {
-		diagnostic("invalid-caption-content", `caption-content-relations/${id}`, error);
-	}
 	for (const [id, caption] of Object.entries(entities["caption-entities"])) {
-		const hasReference = Object.hasOwn(relations["caption-content-relations"], `cc:${id}`) || Object.hasOwn(state["caption-content-relations"] ?? {}, `cc:${id}`);
-		if (hasReference && caption.text !== void 0) diagnostic("multiple-caption-sources", `caption-entities/${id}`, "Content reference takes precedence over inline text");
-		if (!hasReference && caption.text !== void 0 && richText[id]) captionText[id] = {
-			text: caption.text,
-			start: 0,
-			end: caption.text.length,
-			bodyId: richText[id].bodyId
-		};
+		if (caption.text !== void 0) {
+			if (Object.hasOwn(state["caption-entities"]?.[id] ?? {}, "selection")) diagnostic("multiple-caption-sources", `caption-entities/${id}`, "Inline text takes precedence over the content reference");
+			if (richText[id]) captionText[id] = {
+				text: caption.text,
+				start: 0,
+				end: caption.text.length,
+				bodyId: richText[id].bodyId
+			};
+			continue;
+		}
+		if (!caption.selection) continue;
+		try {
+			const resolved = resolveContent(doc, caption.selection);
+			if (resolved) captionText[id] = resolved;
+			else diagnostic("invalid-caption-content", `caption-entities/${id}/selection`, "Selection does not resolve in its current segment");
+		} catch (error) {
+			diagnostic("invalid-caption-content", `caption-entities/${id}/selection`, error);
+		}
+	}
+	const phoneticText = {};
+	for (const [id, phonetic] of Object.entries(entities["phonetic-script-entities"])) try {
+		const resolved = resolveContent(doc, phonetic.phoneme);
+		if (resolved) phoneticText[id] = resolved;
+		else diagnostic("invalid-phoneme", `phonetic-script-entities/${id}/phoneme`, "Selection does not resolve in its current segment");
+	} catch (error) {
+		diagnostic("invalid-phoneme", `phonetic-script-entities/${id}/phoneme`, error);
 	}
 	return freezeSnapshot({
 		...entities,
@@ -3995,6 +4096,7 @@ function readMedeoDsl(doc, state) {
 		"contain-relations": validTree(treeNodes),
 		richText,
 		captionText,
+		phoneticText,
 		diagnostics
 	});
 }
@@ -4007,6 +4109,19 @@ function freezeSnapshot(value) {
 }
 //#endregion
 //#region src/dsl/document/draft.ts
+function textKey(target) {
+	return "captionId" in target ? JSON.stringify(["caption", target.captionId]) : JSON.stringify([
+		"segment",
+		target.audioScriptId,
+		target.segmentId
+	]);
+}
+/** Rows holding a segment selection, by the field that holds it (D181). */
+const selectionField = {
+	"caption-entities": "selection",
+	"phonetic-script-entities": "phoneme"
+};
+const isSelectionTable = (table) => Object.hasOwn(selectionField, table);
 const entities = new Set(entityTables);
 const relations = new Set(relationTables);
 const reserved = new Set([
@@ -4039,8 +4154,10 @@ var TransactionDraft = class {
 	entities = /* @__PURE__ */ new Map();
 	relations = /* @__PURE__ */ new Map();
 	deleted = /* @__PURE__ */ new Set();
+	/** Texts this write edits, by {@link textKey}, as its edits so far leave them. */
 	textPreviews = /* @__PURE__ */ new Map();
-	dirtyCaptions = /* @__PURE__ */ new Set();
+	/** Selections the draft wrote, by table then row; they are replayed natively in order with text edits. */
+	dirty = /* @__PURE__ */ new Map();
 	removedCaptionTexts = /* @__PURE__ */ new Set();
 	treeChanged = false;
 	active = true;
@@ -4071,13 +4188,30 @@ var TransactionDraft = class {
 	close() {
 		this.active = false;
 	}
+	dirtySelections() {
+		return [...this.dirty].flatMap(([table, ids]) => [...ids].map((id) => [table, id]));
+	}
+	clearDirtySelections() {
+		this.dirty.clear();
+	}
+	/** Whether this write edits any segment of a script. */
+	usesScript(scriptId) {
+		const prefix = JSON.stringify(["segment", scriptId]).slice(0, -1) + ",";
+		return [...this.textPreviews.keys()].some((key) => key.startsWith(prefix));
+	}
 	wrap(value, path) {
 		if (!value || typeof value !== "object" || value instanceof Cursor) return value;
 		const cached = this.wrappers.get(value);
 		if (cached) return cached;
 		const proxy = new Proxy(value, {
 			get: (target, key) => this.run(() => {
-				if ((path[0] === "audio-script-entities" && key === "body" || path[0] === "caption-entities" && key === "text") && path.length === 2 && this.textPreviews.has(path[1])) return this.textPreviews.get(path[1]);
+				if (key === "text") {
+					const preview = path[0] === "caption-entities" && path.length === 2 ? this.textPreviews.get(textKey({ captionId: path[1] })) : path[0] === "audio-script-entities" && path[2] === "segments" && path.length === 4 ? this.textPreviews.get(textKey({
+						audioScriptId: path[1],
+						segmentId: String(target.id)
+					})) : void 0;
+					if (preview !== void 0) return preview;
+				}
 				if (!path.length && key === "contain-relations" && target[key] === void 0) target[key] = [];
 				return this.wrap(Reflect.get(target, key), [...path, String(key)]);
 			}),
@@ -4132,24 +4266,27 @@ var TransactionDraft = class {
 			if (path.length === 1 && (remove || next === void 0)) this.deleted.add(lifetime);
 			else if (this.deleted.has(lifetime)) throw new Error("Recreate an entity in a separate transaction");
 			if (table === "audio-script-entities") {
-				if (path.length === 2 && key === "body" && (initial || this.textPreviews.has(rowId))) throw new Error("Edit body through the Text adapter");
-				if (path.length === 1 && initial && isRecord(next) && next.body !== initial.body) throw new Error("Edit body through the Text adapter");
-				if (path.length === 1 && this.textPreviews.has(rowId)) throw new Error("Cannot delete or replace a body used in this transaction");
+				if ((path.length === 2 && key === "segments" || path[2] === "segments" || path.length === 1 && initial && isRecord(next) && !sameSegments(next.segments, initial.segments)) && (initial || this.usesScript(rowId))) throw new Error("Change segments with segment mutations and edit their text through the Text adapter");
+				if (path.length === 1 && this.usesScript(rowId)) throw new Error("Cannot delete or replace a Text used in this transaction");
+			}
+			if (isSelectionTable(table)) {
+				const field = selectionField[table];
+				if (path.length === 1 && isRecord(next) && Object.hasOwn(next, field) || path.length === 2 && key === field || path[2] === field) {
+					let ids = this.dirty.get(table);
+					if (!ids) this.dirty.set(table, ids = /* @__PURE__ */ new Set());
+					ids.add(rowId);
+				}
 			}
 			if (table === "caption-entities") {
 				const changesText = path.length === 1 || path.length === 2 && key === "text";
 				const nextText = path.length === 1 ? isRecord(next) ? next.text : void 0 : next;
 				if (changesText) {
-					if (this.textPreviews.has(rowId)) throw new Error("Cannot delete or replace a body used in this transaction");
+					if (this.textPreviews.has(textKey({ captionId: rowId }))) throw new Error("Cannot delete or replace a Text used in this transaction");
 					if (nextText !== void 0 && (this.removedCaptionTexts.has(rowId) || initial?.text !== void 0 && nextText !== initial.text)) throw new Error("Edit Caption text through the Text adapter");
 					if (initial?.text !== void 0 && nextText === void 0) this.removedCaptionTexts.add(rowId);
 				}
 			}
-			if (initial && [
-				"media-entities",
-				"asset-entities",
-				"voice-entities"
-			].includes(table)) {
+			if (initial && ["media-entities", "voice-entities"].includes(table)) {
 				for (const field of [
 					"system",
 					"key",
@@ -4161,7 +4298,7 @@ var TransactionDraft = class {
 			let fields = rows.get(rowId);
 			if (!fields) rows.set(rowId, fields = /* @__PURE__ */ new Set());
 			if (path.length === 1) for (const field of [...Object.keys(initial ?? {}), ...Object.keys(isRecord(next) ? next : {})]) {
-				if ((field === "body" && table === "audio-script-entities" || field === "text" && table === "caption-entities") && initial && isRecord(next) && next[field] === initial[field]) continue;
+				if (initial && isRecord(next) && (field === "segments" && table === "audio-script-entities" && sameSegments(next[field], initial[field]) || field === "text" && table === "caption-entities" && next[field] === initial[field])) continue;
 				fields.add(field);
 			}
 			else fields.add(path[2] ?? key);
@@ -4169,10 +4306,12 @@ var TransactionDraft = class {
 			let ids = this.relations.get(table);
 			if (!ids) this.relations.set(table, ids = /* @__PURE__ */ new Set());
 			ids.add(rowId);
-			if (table === "caption-content-relations") this.dirtyCaptions.add(rowId);
 		}
 	}
 };
+function sameSegments(a, b) {
+	return JSON.stringify(a) === JSON.stringify(b);
+}
 //#endregion
 //#region src/dsl/document/draft-operations.ts
 function nodeIndex(nodes) {
@@ -4224,6 +4363,7 @@ function changeTree(nodes, mutation, rows) {
 }
 //#endregion
 //#region src/dsl/document/mutations.ts
+const resourceTables = ["media-entities", "voice-entities"];
 /** Mutations are a convenience vocabulary over the same guarded draft and native command queue. */
 function applyMutations(tx, mutations) {
 	const rows = tx.draft;
@@ -4233,11 +4373,7 @@ function applyMutations(tx, mutations) {
 			case "create-entity": {
 				assertEntity(m.table, m.value);
 				const old = rows[m.table][m.id];
-				if (old && ![
-					"media-entities",
-					"voice-entities",
-					"asset-entities"
-				].includes(m.table)) throw new Error(`Entity already exists: ${m.id}`);
+				if (old && !resourceTables.includes(m.table)) throw new Error(`Entity already exists: ${m.id}`);
 				if (old) {
 					assertEntityPatch(m.table, m.value, old);
 					applyPatch(old, m.value);
@@ -4247,12 +4383,15 @@ function applyMutations(tx, mutations) {
 			case "patch-entity": {
 				const current = rows[m.table][m.id];
 				if (!current) throw new Error(`Missing entity: ${m.id}`);
-				assertEntityPatch(m.table, m.patch, current);
-				if (m.table === "clip-entities" && Object.hasOwn(m.patch, "seqMarker")) {
-					const { seqMarker, ...rest } = m.patch;
-					current.seqMarker = mergeFields(current.seqMarker, seqMarker ?? {});
-					applyPatch(current, rest);
-				} else applyPatch(current, m.patch);
+				const patch = m.patch;
+				assertEntityPatch(m.table, patch, current);
+				const atomic = atomicFields[m.table] ?? [];
+				const next = patchRow(m.table, current, patch);
+				const ordinary = {};
+				for (const key of Object.keys(patch)) if (!atomic.includes(key)) ordinary[key] = patch[key];
+				else if (next[key] === void 0) delete current[key];
+				else current[key] = next[key];
+				applyPatch(current, ordinary);
 				break;
 			}
 			case "delete-entity":
@@ -4261,18 +4400,24 @@ function applyMutations(tx, mutations) {
 			case "set-relation": {
 				assertRelation(m.table, m.value);
 				const id = relationId(m.table, m.value);
-				if (m.table === "time-anchor-relations" || m.table === "caption-content-relations") tx.draft[m.table][id] = m.value;
-				else {
-					rows[m.table][id] ??= {};
-					applyPatch(rows[m.table][id], m.value);
-				}
+				rows[m.table][id] ??= {};
+				applyPatch(rows[m.table][id], withSourceDigest(rows, m.table, m.value));
 				break;
 			}
 			case "delete-relation":
 				delete rows[m.table][m.id];
 				break;
+			case "insert-segment":
+				tx.insertSegment(m.audioScriptId, m.segmentId, m.text, m.index);
+				break;
+			case "move-segment":
+				tx.moveSegment(m.audioScriptId, m.segmentId, m.index);
+				break;
+			case "delete-segment":
+				tx.deleteSegment(m.audioScriptId, m.segmentId);
+				break;
 			case "edit-text":
-				tx.editText(m.audioScriptId, m.edits, m.expectedBodyId);
+				tx.editText(m.audioScriptId, m.segmentId, m.edits, m.expectedBodyId);
 				break;
 			case "edit-caption-text":
 				tx.editCaptionText(m.captionId, m.edits, m.expectedBodyId);
@@ -4280,10 +4425,32 @@ function applyMutations(tx, mutations) {
 			case "set-caption-content":
 				tx.setCaptionContent(m);
 				break;
+			case "set-phoneme":
+				tx.setPhoneme(m);
+				break;
 			default: changeTree(tx.draft["contain-relations"], m, tx.draft);
 		}
 	}
 }
+/**
+* A derived-from records its source's digest (D177): the one the writer names, otherwise the
+* source's digest as this transaction leaves it. A source with no digest records none.
+*/
+function withSourceDigest(rows, table, value) {
+	if (table !== "derived-from-relations") return value;
+	const relation = value;
+	if (relation.sourceDigest !== void 0) return value;
+	const sourceDigest = digestOf(rows, relation.sourceId);
+	return sourceDigest === void 0 ? value : {
+		...relation,
+		sourceDigest
+	};
+}
+/** A source's digest is its file's: a PhoneticScript's or a medium's Media File `digest`. */
+function digestOf(rows, entityId) {
+	const digest = (rows["phonetic-script-entities"][entityId] ?? rows["media-entities"][entityId])?.digest;
+	return typeof digest === "string" ? digest : void 0;
+}
 //#endregion
 //#region src/dsl/document/native-text.ts
 function applyTextEdits(text, edits) {
@@ -4295,11 +4462,7 @@ function applyTextEdits(text, edits) {
 			text.delete(edit.at, edit.length);
 			break;
 		case "mark":
-			if (edit.key === "pauseMs" && edit.value === 0) text.unmark({
-				start: edit.start,
-				end: edit.end
-			}, edit.key);
-			else text.mark({
+			text.mark({
 				start: edit.start,
 				end: edit.end
 			}, edit.key, edit.value);
@@ -4317,13 +4480,15 @@ function applyTextEdits(text, edits) {
 }
 //#endregion
 //#region src/dsl/document/validation.ts
-/** One relation row by the DSL's rules; caption content must also resolve in the live body. */
-function validateRelation(doc, state, table, id, value) {
-	assertRelationAdmitted(state, table, id, value);
-	if (table !== "caption-content-relations") return;
-	const resolved = resolveContent(doc, value);
-	if (!resolved || resolved.start === resolved.end) throw new Error("Expected a nonempty range in the current body");
-	captionContentCodec.encode(value);
+/**
+* A selection the draft wrote whole, by the DSL's rules: its segment must exist and what it
+* selects must not be empty. A run's Cursors must lie on that segment's live Text.
+*/
+function validateSelection(doc, state, table, id, value) {
+	assertReferenceAdmitted(state, table, id, selectionField[table], value);
+	const selection = value;
+	if (!(selection.start === void 0 ? requireSegment(state, selection.audioScriptId, selection.segmentId).text : resolveContent(doc, selection)?.text)) throw new Error("Expected a nonempty selection of the current segment");
+	(table === "caption-entities" ? captionContentCodec : phonemeCodec).encode(selection);
 }
 /**
 * The DSL's admission of the whole draft, then what only this document can check: containment
@@ -4355,6 +4520,14 @@ function validateDraft(tracked) {
 }
 //#endregion
 //#region src/dsl/document/transact.ts
+const codecs = {
+	"caption-entities": captionContentCodec,
+	"phonetic-script-entities": phonemeCodec
+};
+function entityRow(doc, table, id) {
+	const row = doc.getMap(table).get(id);
+	return isContainer(row) && row.kind() === "Map" ? row : void 0;
+}
 /** Serialization happens before preparing any writes, so invalid/cyclic audit data cannot partly commit. */
 function auditOptions(audit) {
 	if (!audit || typeof audit.semantic_op !== "string" || !audit.semantic_op.length || !Object.hasOwn(audit, "payload")) throw new Error("A structured DSL audit is required");
@@ -4373,8 +4546,8 @@ const nativeHookSupport = /* @__PURE__ */ new WeakMap();
 /**
 * Every write relies on the patched Mirror's `applyNative` (patches/loro-mirror@2.3.3.patch). An
 * unpatched Mirror ignores the option and commits only the draft, silently dropping Text, ranges
-* and Caption relations. The patch runs the hook even for an empty draft without committing, so
-* probe each Mirror class once on a scratch document and refuse before the first real write.
+* and selections. The patch runs the hook even for an empty draft without committing, so probe
+* each Mirror class once on a scratch document and refuse before the first real write.
 */
 function assertNativeHook(mirror) {
 	const Impl = mirror.constructor;
@@ -4408,50 +4581,75 @@ function runTransaction(doc, mirror, edit, options, setReject) {
 	mirror.setState((raw) => {
 		const tracked = new TransactionDraft(raw, before);
 		setReject((error) => tracked.fail(error));
-		const useBody = (scriptId, caption = false) => {
-			assertId(scriptId);
-			const entity = requireEntity(tracked.draft, caption ? "caption-entities" : "audio-script-entities", scriptId);
-			if (tracked.deleted.has(scriptId)) throw new Error("Cannot use a deleted body");
-			const body = entity[caption ? "text" : "body"];
-			if (typeof body !== "string") throw new Error("Missing Text body");
-			if (caption && tracked.removedCaptionTexts.has(scriptId)) throw new Error("Cannot edit a removed Text");
-			tracked.textPreviews.set(scriptId, body);
-			return body;
+		/** The current text of a Caption's own Text or a script segment, as this transaction leaves it so far. */
+		const useText = (target) => {
+			let text;
+			if ("captionId" in target) {
+				assertId(target.captionId);
+				const entity = requireEntity(tracked.draft, "caption-entities", target.captionId);
+				if (tracked.deleted.has(target.captionId)) throw new Error("Cannot use a deleted Text");
+				if (tracked.removedCaptionTexts.has(target.captionId)) throw new Error("Cannot edit a removed Text");
+				text = entity.text;
+			} else {
+				assertId(target.audioScriptId);
+				if (tracked.deleted.has(target.audioScriptId)) throw new Error("Cannot use a deleted Text");
+				text = requireSegment(tracked.draft, target.audioScriptId, target.segmentId).text;
+			}
+			if (typeof text !== "string") throw new Error("Missing Text");
+			tracked.textPreviews.set(textKey(target), text);
+			return text;
+		};
+		const textBefore = (target) => {
+			if ("captionId" in target) return before["caption-entities"][target.captionId]?.text;
+			return before["audio-script-entities"][target.audioScriptId]?.segments?.find((segment) => segment.id === target.segmentId)?.text;
 		};
-		const flushCaptions = () => {
-			for (const id of tracked.dirtyCaptions) {
-				const value = raw["caption-content-relations"][id];
-				if (value !== void 0) {
-					validateRelation(doc, raw, "caption-content-relations", id, value);
-					useBody(value.audioScriptId);
+		const flushSelections = () => {
+			for (const [table, id] of tracked.dirtySelections()) {
+				const row = raw[table][id];
+				if (!row) continue;
+				const field = selectionField[table];
+				const value = row[field];
+				const old = before[table][id]?.[field];
+				if (isInvalidAtomic(value)) {
+					if (isInvalidAtomic(old) && old.raw === value.raw) continue;
+					throw new Error(`Invalid ${field}`);
 				}
+				if (value !== void 0) validateSelection(doc, raw, table, id, value);
 				steps.push({
-					type: "caption",
+					type: "selection",
+					table,
 					id,
 					value: copyInput(value)
 				});
 			}
-			tracked.dirtyCaptions.clear();
+			tracked.clearDirtySelections();
 		};
-		const editBody = (scriptId, edits, expectedBodyId, caption) => {
+		const editBody = (target, edits, expectedBodyId) => {
 			tracked.run(() => {
-				flushCaptions();
-				const body = useBody(scriptId, caption);
-				if (!bodies.has(scriptId)) bodies.set(scriptId, liveText(doc, scriptId, caption));
-				const native = bodies.get(scriptId);
+				flushSelections();
+				const text = useText(target);
+				const key = textKey(target);
+				if (!bodies.has(key)) bodies.set(key, liveText(doc, target));
+				const native = bodies.get(key);
 				if (expectedBodyId !== void 0 && native?.id !== expectedBodyId) throw new Error("Stale or missing Text body");
-				if ((caption ? before["caption-entities"][scriptId]?.text : before["audio-script-entities"][scriptId]?.body) !== void 0 && !native) throw new Error("Missing live Text body");
-				const prepared = prepareTextEdits(body, copyInput(edits));
-				tracked.textPreviews.set(scriptId, prepared.body);
+				if (textBefore(target) !== void 0 && !native) throw new Error("Missing live Text body");
+				const prepared = prepareTextEdits(text, copyInput(edits));
+				tracked.textPreviews.set(key, prepared.text);
 				steps.push({
 					type: "text",
-					scriptId,
-					caption,
+					target,
 					expectedBodyId: native?.id,
 					edits: prepared.edits
 				});
 			});
 		};
+		/** A script's segments as this transaction leaves them; never a Mirror-tracked write. */
+		const segmentsOf = (audioScriptId) => {
+			assertId(audioScriptId);
+			requireEntity(tracked.draft, "audio-script-entities", audioScriptId);
+			if (tracked.deleted.has(audioScriptId)) throw new Error("Cannot change a deleted AudioScript");
+			return raw["audio-script-entities"][audioScriptId].segments;
+		};
 		const tx = Object.freeze({
 			get draft() {
 				return tracked.run(() => tracked.draft);
@@ -4459,29 +4657,95 @@ function runTransaction(doc, mirror, edit, options, setReject) {
 			applyMutations(mutations) {
 				tracked.run(() => applyMutations(tx, mutations));
 			},
-			editText(scriptId, edits, expectedBodyId) {
-				editBody(scriptId, edits, expectedBodyId, false);
+			editText(audioScriptId, segmentId, edits, expectedBodyId) {
+				editBody({
+					audioScriptId,
+					segmentId
+				}, edits, expectedBodyId);
 			},
 			editCaptionText(captionId, edits, expectedBodyId) {
-				editBody(captionId, edits, expectedBodyId, true);
+				editBody({ captionId }, edits, expectedBodyId);
+			},
+			insertSegment(audioScriptId, segmentId, text, index) {
+				tracked.run(() => {
+					assertId(segmentId);
+					if (typeof text !== "string") throw new Error("Segment text is a string");
+					const segments = segmentsOf(audioScriptId);
+					if (segments.some((segment) => segment.id === segmentId)) throw new Error(`Segment already exists: ${segmentId}`);
+					const at = index ?? segments.length;
+					if (!Number.isSafeInteger(at) || at < 0 || at > segments.length) throw new Error("Invalid segment index");
+					segments.splice(at, 0, {
+						id: segmentId,
+						text
+					});
+				});
+			},
+			moveSegment(audioScriptId, segmentId, index) {
+				tracked.run(() => {
+					const segments = segmentsOf(audioScriptId);
+					const from = segments.findIndex((segment) => segment.id === segmentId);
+					if (from < 0) throw new Error(`Missing segment ${segmentId} of ${audioScriptId}`);
+					if (!Number.isSafeInteger(index) || index < 0 || index >= segments.length) throw new Error("Invalid segment index");
+					const [segment] = segments.splice(from, 1);
+					segments.splice(index, 0, segment);
+				});
+			},
+			deleteSegment(audioScriptId, segmentId) {
+				tracked.run(() => {
+					const segments = segmentsOf(audioScriptId);
+					const at = segments.findIndex((segment) => segment.id === segmentId);
+					if (at < 0) throw new Error(`Missing segment ${segmentId} of ${audioScriptId}`);
+					if (tracked.textPreviews.has(textKey({
+						audioScriptId,
+						segmentId
+					}))) throw new Error("Cannot delete a segment whose text this transaction edits");
+					segments.splice(at, 1);
+				});
 			},
 			setCaptionContent(selection) {
 				tracked.run(() => {
-					flushCaptions();
-					const { captionId, audioScriptId, start, end } = selection;
+					flushSelections();
+					const { captionId, audioScriptId, segmentId, start, end } = selection;
 					assertId(captionId);
-					assertCaptionEndpoints(raw, selection);
-					assertCaptionSelection(raw, selection, useBody(audioScriptId));
+					assertCaptionSelection(raw, selection, useText({
+						audioScriptId,
+						segmentId
+					}));
 					steps.push({
 						type: "range",
-						selection: {
-							captionId,
+						table: "caption-entities",
+						id: captionId,
+						range: {
 							audioScriptId,
+							segmentId,
 							start,
 							end
 						}
 					});
 				});
+			},
+			setPhoneme(selection) {
+				tracked.run(() => {
+					flushSelections();
+					const { phoneticScriptId, audioScriptId, segmentId, start, end, phonemeScript } = selection;
+					assertId(phoneticScriptId);
+					assertPhonemeSelection(raw, selection, useText({
+						audioScriptId,
+						segmentId
+					}));
+					steps.push({
+						type: "range",
+						table: "phonetic-script-entities",
+						id: phoneticScriptId,
+						range: {
+							audioScriptId,
+							segmentId,
+							start,
+							end
+						},
+						...phonemeScript === void 0 ? {} : { phonemeScript }
+					});
+				});
 			}
 		});
 		try {
@@ -4494,32 +4758,15 @@ function runTransaction(doc, mirror, edit, options, setReject) {
 				}
 				tracked.assertValid();
 				if (doc.getPendingTxnLength()) throw new Error("Native writes are not allowed during transaction preparation");
-				flushCaptions();
+				flushSelections();
 				for (const id of tracked.entities.get("caption-entities")?.keys() ?? []) {
 					const caption = tracked.draft["caption-entities"][id];
 					if (caption) caption.style ??= {};
 				}
 				validateDraft(tracked);
-				for (const step of steps) if (step.type === "text") requireEntity(raw, step.caption ? "caption-entities" : "audio-script-entities", step.scriptId);
-				const referenceIds = new Set(Object.keys(raw["caption-content-relations"]).map((id) => id.slice(3)));
-				const changedCaptions = /* @__PURE__ */ new Set();
-				for (const [id, fields] of tracked.entities.get("caption-entities") ?? []) if (fields.has("text")) changedCaptions.add(id);
-				for (const step of steps) {
-					if (step.type === "range") {
-						referenceIds.add(step.selection.captionId);
-						changedCaptions.add(step.selection.captionId);
-					}
-					if (step.type === "caption") {
-						const id = step.id.slice(3);
-						changedCaptions.add(id);
-						if (step.value) referenceIds.add(id);
-						else referenceIds.delete(id);
-					}
-					if (step.type === "text" && step.caption) changedCaptions.add(step.scriptId);
-				}
-				for (const id of changedCaptions) assertCaptionSource(raw["caption-entities"][id], referenceIds.has(id));
-				for (const step of steps) if (step.type === "range") assertCaptionEndpoints(raw, step.selection);
-				else if (step.type === "caption" && step.value) assertCaptionEndpoints(raw, step.value);
+				for (const step of steps) if (step.type === "text") if ("captionId" in step.target) requireEntity(raw, "caption-entities", step.target.captionId);
+				else requireSegment(raw, step.target.audioScriptId, step.target.segmentId);
+				assertSelectionsAfter(raw, before, tracked, steps);
 				for (const [id, fields] of tracked.entities.get("caption-entities") ?? []) {
 					const row = raw["caption-entities"][id];
 					if (fields.has("text") && row?.text !== void 0 && before["caption-entities"][id]?.text === void 0) {
@@ -4527,8 +4774,16 @@ function runTransaction(doc, mirror, edit, options, setReject) {
 						delete row.text;
 					}
 				}
-				for (const id of tracked.relations.get("caption-content-relations") ?? []) if (before["caption-content-relations"][id] === void 0) delete raw["caption-content-relations"][id];
-				else raw["caption-content-relations"][id] = before["caption-content-relations"][id];
+				for (const table of Object.keys(selectionField)) {
+					const field = selectionField[table];
+					for (const [id, fields] of tracked.entities.get(table) ?? []) {
+						const row = raw[table][id];
+						if (!row || !fields.has(field)) continue;
+						const old = before[table][id]?.[field];
+						if (old === void 0) delete row[field];
+						else row[field] = old;
+					}
+				}
 			});
 		} finally {
 			tracked.close();
@@ -4537,44 +4792,104 @@ function runTransaction(doc, mirror, edit, options, setReject) {
 	}, {
 		...options,
 		applyNative: (doc) => {
-			const bodyFor = (scriptId, caption = false) => {
-				let text = bodies.get(scriptId);
+			const textFor = (target) => {
+				const key = textKey(target);
+				let text = bodies.get(key);
 				if (!text) {
-					text = liveText(doc, scriptId, caption);
-					bodies.set(scriptId, text);
+					text = liveText(doc, target);
+					bodies.set(key, text);
 				}
-				if (!text || text.isDeleted()) throw new Error("Prepared Text body is missing");
+				if (!text || text.isDeleted()) throw new Error("Prepared Text is missing");
 				return text;
 			};
 			for (const [id, value] of newCaptionTexts) {
-				const body = doc.getMap("caption-entities").get(id).setContainer("text", new LoroText());
-				if (value.length) body.insert(0, value);
-				bodies.set(id, body);
+				const text = doc.getMap("caption-entities").get(id).setContainer("text", new LoroText());
+				if (value.length) text.insert(0, value);
+				bodies.set(textKey({ captionId: id }), text);
 			}
-			for (const step of steps) if (step.type === "text") {
-				const text = bodyFor(step.scriptId, step.caption);
-				if (step.expectedBodyId && text.id !== step.expectedBodyId) throw new Error("Stale Text body");
-				applyTextEdits(text, step.edits);
-			} else if (step.type === "range") {
-				const { captionId, audioScriptId, start, end } = step.selection;
-				const text = bodyFor(audioScriptId);
-				const from = text?.getCursor(start, 0), to = text?.getCursor(end, 0);
-				if (!from || !to) throw new Error("Unable to construct prepared range");
-				doc.getMap("caption-content-relations").set(`cc:${captionId}`, captionContentCodec.encode({
-					captionId,
-					audioScriptId,
-					start: from,
-					end: to
-				}));
-			} else {
-				const relations = doc.getMap("caption-content-relations");
-				if (step.value === void 0) {
-					if (relations.get(step.id) !== void 0) relations.delete(step.id);
-				} else relations.set(step.id, captionContentCodec.encode(step.value));
+			for (const step of steps) {
+				if (step.type === "text") {
+					const text = textFor(step.target);
+					if (step.expectedBodyId && text.id !== step.expectedBodyId) throw new Error("Stale Text body");
+					applyTextEdits(text, step.edits);
+					continue;
+				}
+				const row = entityRow(doc, step.table, step.id);
+				const field = selectionField[step.table];
+				if (step.type === "range") {
+					const { audioScriptId, segmentId, start, end } = step.range;
+					const value = {
+						audioScriptId,
+						segmentId
+					};
+					if (start !== void 0 && end !== void 0) {
+						const text = textFor({
+							audioScriptId,
+							segmentId
+						});
+						value.start = text.getCursor(start, 0);
+						value.end = text.getCursor(end, 0);
+						if (!value.start || !value.end) throw new Error("Unable to construct prepared range");
+					}
+					if (step.phonemeScript !== void 0) value.phonemeScript = step.phonemeScript;
+					if (!row) throw new Error("Unable to construct prepared range");
+					row.set(field, codecs[step.table].encode(value));
+				} else if (step.value !== void 0) {
+					if (!row) throw new Error("Prepared selection owner is missing");
+					row.set(field, codecs[step.table].encode(step.value));
+				} else if (row?.get(field) !== void 0) row.delete(field);
 			}
 		}
 	});
 }
+/**
+* What only the whole write can say about selections, with the native steps counted in order:
+* each Caption keeps one source of words, each selection's segment still exists, and every
+* PhoneticScript the write touched is written for a segment and stores its text as a file.
+*/
+function assertSelectionsAfter(raw, before, tracked, steps) {
+	const selected = new Set(Object.entries(before["caption-entities"]).filter(([, caption]) => typeof caption === "object" && caption.selection !== void 0).map(([id]) => id));
+	const changedCaptions = /* @__PURE__ */ new Set();
+	for (const [id, fields] of tracked.entities.get("caption-entities") ?? []) if (fields.has("text")) changedCaptions.add(id);
+	const phonemes = /* @__PURE__ */ new Map();
+	for (const step of steps) {
+		if (step.type === "text") {
+			if ("captionId" in step.target) changedCaptions.add(step.target.captionId);
+			continue;
+		}
+		const value = step.type === "range" ? {
+			...step.range,
+			...step.phonemeScript === void 0 ? {} : { phonemeScript: step.phonemeScript }
+		} : step.value;
+		if (step.table === "caption-entities") {
+			changedCaptions.add(step.id);
+			if (value) selected.add(step.id);
+			else selected.delete(step.id);
+			if (value) assertCaptionEndpoints(raw, {
+				captionId: step.id,
+				...value
+			});
+		} else {
+			phonemes.set(step.id, value);
+			if (value) assertPhonemeEndpoints(raw, {
+				phoneticScriptId: step.id,
+				...value
+			});
+		}
+	}
+	for (const id of changedCaptions) if (raw["caption-entities"][id]) assertCaptionSource(raw["caption-entities"][id], selected.has(id));
+	const phonetics = new Set([...tracked.entities.get("phonetic-script-entities")?.keys() ?? [], ...phonemes.keys()]);
+	for (const id of phonetics) {
+		const row = raw["phonetic-script-entities"][id];
+		if (!row) continue;
+		const phoneme = phonemes.has(id) ? phonemes.get(id) : row.phoneme;
+		const old = before["phonetic-script-entities"][id];
+		assertPhoneticScript(old, {
+			...row,
+			phoneme: isRecord(phoneme) ? phoneme : void 0
+		});
+	}
+}
 //#endregion
 //#region src/dsl/document/medeo-dsl.ts
 const contentRoots = new Set([
@@ -4665,17 +4980,23 @@ var MedeoDsl = class MedeoDsl {
 	applyMutations(mutations, audit) {
 		this.transact((tx) => tx.applyMutations(mutations), audit);
 	}
-	editText(scriptId, expectedBodyId, edits, audit) {
-		this.transact((tx) => tx.editText(scriptId, edits, expectedBodyId), audit);
+	editText(scriptId, segmentId, expectedBodyId, edits, audit) {
+		this.transact((tx) => tx.editText(scriptId, segmentId, edits, expectedBodyId), audit);
 	}
 	editCaptionText(captionId, expectedBodyId, edits, audit) {
 		this.transact((tx) => tx.editCaptionText(captionId, edits, expectedBodyId), audit);
 	}
-	/** Local ranges exclude insertions at their start and include insertions at their end. K01 remains deferred. */
-	textRange(scriptId, start, end) {
+	/**
+	* A run inside one segment as native Cursors. Local ranges exclude insertions at their start and
+	* include insertions at their end. K01 remains deferred.
+	*/
+	textRange(scriptId, segmentId, start, end) {
 		this.assertActive();
-		const text = liveText(this.doc, scriptId);
-		if (!text) throw new Error("Missing AudioScript body");
+		const text = liveText(this.doc, {
+			audioScriptId: scriptId,
+			segmentId
+		});
+		if (!text) throw new Error("Missing AudioScript segment");
 		assertTextOffset(text.toString(), start);
 		assertTextOffset(text.toString(), end);
 		if (start >= end) throw new Error("Expected a nonempty range");
@@ -4683,6 +5004,7 @@ var MedeoDsl = class MedeoDsl {
 		if (!from || !to) throw new Error("Unable to construct text range");
 		return {
 			audioScriptId: scriptId,
+			segmentId,
 			start: from,
 			end: to
 		};
@@ -4732,6 +5054,132 @@ var MedeoDsl = class MedeoDsl {
 	}
 };
 //#endregion
+//#region src/dsl/projection/audio-scripts.ts
+/**
+* The scripts a recording says. It follows derived-from from the recording; a PhoneticScript is a
+* variant, so its words are the base its phoneme names, and derived-from is not followed past it.
+* When nothing is found that way, a script read off the recording itself is what it says.
+*/
+function spokenScripts(snapshot, recordingId) {
+	const relations = Object.values(snapshot["derived-from-relations"]);
+	const found = /* @__PURE__ */ new Map();
+	const seen = /* @__PURE__ */ new Set();
+	const visit = (id, phoneticScriptId) => {
+		if (seen.has(id)) return;
+		seen.add(id);
+		if (snapshot["audio-script-entities"][id]) {
+			if (!found.has(id)) found.set(id, phoneticScriptId);
+			return;
+		}
+		const phonetic = snapshot["phonetic-script-entities"][id];
+		if (phonetic) {
+			visit(phonetic.phoneme.audioScriptId, id);
+			return;
+		}
+		for (const r of relations) if (r.entityId === id) visit(r.sourceId, phoneticScriptId);
+	};
+	visit(recordingId);
+	if (!found.size) {
+		for (const r of relations) if (r.sourceId === recordingId && snapshot["audio-script-entities"][r.entityId]) found.set(r.entityId, void 0);
+	}
+	return [...found].map(([audioScriptId, phoneticScriptId]) => phoneticScriptId === void 0 ? { audioScriptId } : {
+		audioScriptId,
+		phoneticScriptId
+	});
+}
+/** Every AudioScript in the document with its PhoneticScripts, the Speech Clips saying it and the Caption Clips showing it. */
+function toAudioScriptViews(snapshot) {
+	const views = /* @__PURE__ */ new Map();
+	const segmentIndex = (scriptId, segmentId) => {
+		const index = snapshot["audio-script-entities"][scriptId]?.segments.findIndex((s) => s.id === segmentId) ?? -1;
+		return index < 0 ? void 0 : index;
+	};
+	for (const [id, script] of Object.entries(snapshot["audio-script-entities"])) views.set(id, {
+		id,
+		segments: script.segments.map((segment) => {
+			const rich = snapshot.richText[segment.id];
+			const language = rich ? markedLanguage(langsFromDelta(rich.delta)) : void 0;
+			return language === void 0 ? {
+				id: segment.id,
+				text: segment.text
+			} : {
+				id: segment.id,
+				text: segment.text,
+				language
+			};
+		}),
+		sourceMediaIds: [],
+		phoneticScripts: [],
+		speeches: [],
+		captions: []
+	});
+	for (const r of Object.values(snapshot["derived-from-relations"])) if (snapshot["media-entities"][r.sourceId]) views.get(r.entityId)?.sourceMediaIds.push(r.sourceId);
+	for (const [id, phonetic] of Object.entries(snapshot["phonetic-script-entities"])) {
+		const { audioScriptId, segmentId, phonemeScript } = phonetic.phoneme;
+		const { digest } = phonetic;
+		const segment = segmentIndex(audioScriptId, segmentId);
+		const text = snapshot.phoneticText[id]?.text;
+		views.get(audioScriptId)?.phoneticScripts.push({
+			id,
+			...segment === void 0 ? {} : { segment },
+			...text === void 0 ? {} : { text },
+			...phonemeScript === void 0 ? {} : { phonemeScript },
+			...digest === void 0 ? {} : { digest }
+		});
+	}
+	const timbres = Object.values(snapshot["voice-timbre-relations"]);
+	for (const track of resolveDslLayout(snapshot).tracks) for (const { clipId } of track.clips) {
+		const source = snapshot["clip-entities"][clipId]?.source;
+		if (!source) continue;
+		if (snapshot["media-entities"][source]?.kind === "voiceover-recording") {
+			const voices = timbres.filter((r) => r.recordingId === source).flatMap((r) => {
+				const voice = snapshot["voice-entities"][r.voiceId];
+				if (!voice) return [];
+				return [voice.name === void 0 ? {
+					voiceId: r.voiceId,
+					key: voice.key
+				} : {
+					voiceId: r.voiceId,
+					key: voice.key,
+					name: voice.name
+				}];
+			});
+			for (const spoken of spokenScripts(snapshot, source)) {
+				const applied = appliedSpeech(snapshot, source, spoken);
+				views.get(spoken.audioScriptId)?.speeches.push({
+					clipId,
+					recordingId: source,
+					...spoken.phoneticScriptId === void 0 ? {} : { phoneticScriptId: spoken.phoneticScriptId },
+					voices,
+					...applied === void 0 ? {} : { applied }
+				});
+			}
+			continue;
+		}
+		const caption = snapshot["caption-entities"][source];
+		if (!caption?.selection || caption.text !== void 0 || !snapshot.captionText[source]) continue;
+		const { audioScriptId, segmentId } = caption.selection;
+		const segment = segmentIndex(audioScriptId, segmentId);
+		if (segment !== void 0) views.get(audioScriptId)?.captions.push({
+			clipId,
+			captionId: source,
+			segment
+		});
+	}
+	return [...views.values()];
+}
+/**
+* Whether a recording still matches what it was synthesized from: the digest its derived-from
+* records against the source's file digest now. An AudioScript has no file, so a recording
+* derived from a script directly is not judged.
+*/
+function appliedSpeech(snapshot, recordingId, spoken) {
+	const sourceId = spoken.phoneticScriptId ?? spoken.audioScriptId;
+	const recorded = Object.values(snapshot["derived-from-relations"]).find((r) => r.entityId === recordingId && r.sourceId === sourceId)?.sourceDigest;
+	const current = spoken.phoneticScriptId === void 0 ? void 0 : snapshot["phonetic-script-entities"][sourceId]?.digest;
+	return recorded === void 0 || current === void 0 ? void 0 : recorded === current;
+}
+//#endregion
 //#region src/dsl/voiceover-actions.ts
 function voiceoverPlan(s, input, result) {
 	if (requireTrack(s, input.trackId).role !== "speech" || input.media.kind !== "voiceover-recording") throw new Error("Voiceover requires a speech Track and a recording");
@@ -4744,16 +5192,30 @@ function voiceoverPlan(s, input, result) {
 		id: result.mediaId,
 		value: input.media
 	}];
-	if (input.script.body !== void 0) mutations.push({
-		type: "create-entity",
-		table: "audio-script-entities",
-		id: result.audioScriptId,
-		value: {
-			body: input.script.body,
-			..."defaultLanguage" in input.script && input.script.defaultLanguage !== void 0 ? { defaultLanguage: input.script.defaultLanguage } : {}
-		}
-	});
-	else if (!s["audio-script-entities"][result.audioScriptId]) throw new Error("Missing AudioScript");
+	let segmentIds;
+	if (input.script.segments !== void 0) {
+		if (result.segmentIds.length !== input.script.segments.length) throw new Error("Expected one id per segment");
+		const segments = input.script.segments;
+		segmentIds = result.segmentIds;
+		mutations.push({
+			type: "create-entity",
+			table: "audio-script-entities",
+			id: result.audioScriptId,
+			value: {
+				segments: segments.map((text, i) => ({
+					id: segmentIds[i],
+					text
+				})),
+				..."defaultLanguage" in input.script && input.script.defaultLanguage !== void 0 ? { defaultLanguage: input.script.defaultLanguage } : {}
+			}
+		});
+	} else {
+		const script = s["audio-script-entities"][result.audioScriptId];
+		if (!script) throw new Error("Missing AudioScript");
+		segmentIds = script.segments.map((segment) => segment.id);
+	}
+	const phoneticScriptId = input.script.phoneticScriptId;
+	if (phoneticScriptId !== void 0 && s["phonetic-script-entities"][phoneticScriptId]?.phoneme.audioScriptId !== result.audioScriptId) throw new Error("The PhoneticScript must name this AudioScript as its base");
 	validateWindow({
 		inMs: 0,
 		outMs: input.media.durationMs
@@ -4763,7 +5225,8 @@ function voiceoverPlan(s, input, result) {
 		table: "derived-from-relations",
 		value: {
 			entityId: result.mediaId,
-			sourceId: result.audioScriptId
+			sourceId: phoneticScriptId ?? result.audioScriptId,
+			...input.sourceDigest === void 0 ? {} : { sourceDigest: input.sourceDigest }
 		}
 	});
 	if (input.voice) {
@@ -4793,6 +5256,7 @@ function voiceoverPlan(s, input, result) {
 				rate: void 0,
 				fillTimeline: void 0
 			},
+			source: result.mediaId,
 			...input.volume === void 0 ? {} : { volume: input.volume }
 		}
 	} : {
@@ -4804,15 +5268,9 @@ function voiceoverPlan(s, input, result) {
 				inMs: 0,
 				outMs: input.media.durationMs
 			},
+			source: result.mediaId,
 			...input.volume === void 0 ? {} : { volume: input.volume }
 		}
-	}, {
-		type: "set-relation",
-		table: "clip-source-relations",
-		value: {
-			clipId: result.clipId,
-			sourceId: result.mediaId
-		}
 	});
 	if (!existing || trackOf(s, result.clipId) !== input.trackId) mutations.push({
 		type: existing ? "move-node" : "insert-node",
@@ -4822,13 +5280,14 @@ function voiceoverPlan(s, input, result) {
 	mutations.push(...placementMutations(s, result.clipId, input.placement));
 	uniqueIds(result.captionClipIds);
 	const retained = new Set(result.captionClipIds);
-	const oldCaptions = Object.values(s["time-anchor-relations"]).filter((r) => r.to === result.clipId && s["clip-entities"][r.from] && requireTrack(s, trackOf(s, r.from)).role === "caption").map((r) => r.from);
+	const oldCaptions = existing ? captionsShowing(s, existing.source) : [];
 	mutations.push(...deleteClips(s, oldCaptions.filter((id) => !retained.has(id)), "cascade"));
-	const anchor = mutations.find((m) => m.type === "set-relation" && m.table === "time-anchor-relations" && m.value.from === result.clipId);
+	const placement = mutations.find((m) => m.type === "patch-entity" && m.table === "clip-entities" && m.id === result.clipId && "anchor" in m.patch);
+	const anchor = placement?.type === "patch-entity" && placement.table === "clip-entities" ? placement.patch.anchor : void 0;
 	const trackClips = resolveDslLayout(s).tracks.find((track) => track.trackId === input.trackId).clips;
 	const slot = trackClips.findIndex((clip) => clip.clipId === result.clipId);
 	const predecessor = trackClips.slice(0, slot < 0 ? trackClips.length : slot).findLast((clip) => clip.mode === "sequential" && !clip.fillTimeline);
-	const speechStart = anchor?.type === "set-relation" && anchor.table === "time-anchor-relations" ? anchor.value.fallback_abs_ms : predecessor ? predecessor.startMs + predecessor.durationMs : 0;
+	const speechStart = anchor?.fallback_abs_ms !== void 0 ? anchor.fallback_abs_ms : predecessor ? predecessor.startMs + predecessor.durationMs : 0;
 	input.captions.forEach((caption, i) => {
 		const clipId = result.captionClipIds[i], captionId = result.captionIds[i];
 		validateWindow({
@@ -4837,7 +5296,9 @@ function voiceoverPlan(s, input, result) {
 		}, input.media.durationMs);
 		const previous = s["clip-entities"][clipId];
 		if (previous && !oldCaptions.includes(clipId)) throw new Error("Cannot overwrite another use’s Caption Clip");
-		const previousSource = s["clip-source-relations"][`cs:${clipId}`]?.sourceId;
+		const segmentId = segmentIds[caption.segment];
+		if (segmentId === void 0) throw new Error(`Missing segment ${caption.segment} of the voiceover's script`);
+		const previousSource = previous?.source;
 		const style = caption.style ?? (previousSource ? s["caption-entities"][previousSource]?.style : void 0);
 		mutations.push({
 			type: "create-entity",
@@ -4848,48 +5309,38 @@ function voiceoverPlan(s, input, result) {
 			type: "set-caption-content",
 			captionId,
 			audioScriptId: result.audioScriptId,
-			start: caption.range.start,
-			end: caption.range.end
-		}, {
-			type: "set-relation",
-			table: "caption-alignment-relations",
-			value: {
-				captionId,
-				mediaId: result.mediaId,
-				startMs: caption.startMs,
-				endMs: caption.startMs + caption.durationMs
+			segmentId,
+			...caption.range === void 0 ? {} : {
+				start: caption.range.start,
+				end: caption.range.end
 			}
-		}, previous ? {
+		});
+		const anchor = {
+			to: result.clipId,
+			offsetMs: caption.startMs,
+			fallback_abs_ms: speechStart + caption.startMs
+		};
+		const seqMarker = {
+			inMs: 0,
+			outMs: caption.durationMs
+		};
+		mutations.push(previous ? {
 			type: "patch-entity",
 			table: "clip-entities",
 			id: clipId,
-			patch: { seqMarker: {
-				inMs: 0,
-				outMs: caption.durationMs
-			} }
+			patch: {
+				seqMarker,
+				source: captionId,
+				anchor
+			}
 		} : {
 			type: "create-entity",
 			table: "clip-entities",
 			id: clipId,
-			value: { seqMarker: {
-				inMs: 0,
-				outMs: caption.durationMs
-			} }
-		}, {
-			type: "set-relation",
-			table: "clip-source-relations",
-			value: {
-				clipId,
-				sourceId: captionId
-			}
-		}, {
-			type: "set-relation",
-			table: "time-anchor-relations",
 			value: {
-				from: clipId,
-				to: result.clipId,
-				offsetMs: caption.startMs,
-				fallback_abs_ms: speechStart + caption.startMs
+				seqMarker,
+				source: captionId,
+				anchor
 			}
 		});
 		if (!previous || trackOf(s, clipId) !== input.captionTrackId) mutations.push({
@@ -4900,15 +5351,44 @@ function voiceoverPlan(s, input, result) {
 	});
 	return mutations;
 }
+/**
+* Caption Clips showing what a recording says (D183): their Caption selects the run of the segment
+* the PhoneticScript it was spoken from is written for, or any segment of the script it was read
+* from directly.
+*/
+function captionsShowing(s, recordingId) {
+	const said = spokenScripts(s, recordingId).map(({ audioScriptId, phoneticScriptId }) => {
+		if (phoneticScriptId === void 0) return { audioScriptId };
+		const { segmentId } = s["phonetic-script-entities"][phoneticScriptId].phoneme;
+		return {
+			audioScriptId,
+			segmentId,
+			run: s.phoneticText[phoneticScriptId]
+		};
+	});
+	return Object.entries(s["clip-entities"]).flatMap(([id, clip]) => {
+		const selection = s["caption-entities"][clip.source]?.selection;
+		const shown = s.captionText[clip.source];
+		if (!selection) return [];
+		return said.some((words) => {
+			if (selection.audioScriptId !== words.audioScriptId) return false;
+			if (!("segmentId" in words)) return true;
+			if (selection.segmentId !== words.segmentId || !shown || !words.run) return false;
+			return shown.start < words.run.end && words.run.start < shown.end;
+		}) && requireTrack(s, trackOf(s, id)).role === "caption" ? [id] : [];
+	});
+}
 //#endregion
 //#region src/dsl/dsl-editor.ts
 /** Business edits write DSL facts in one audited transaction; layout is derived on read. */
 var DslEditor = class {
 	dsl;
 	idFactory;
-	constructor(dsl, idFactory = defaultIdFactory) {
+	uploadText;
+	constructor(dsl, idFactory = defaultIdFactory, uploadText) {
 		this.dsl = dsl;
 		this.idFactory = idFactory;
+		this.uploadText = uploadText;
 	}
 	createTrack(role, options) {
 		const id = this.idFactory("track");
@@ -4950,11 +5430,10 @@ var DslEditor = class {
 	/** index is measured in the target's children after removing the moved Clip. */
 	moveClip(clipId, trackId, index, options) {
 		const snapshot = this.dsl.getSnapshot();
-		requireClip(snapshot, clipId);
+		const clip = requireClip(snapshot, clipId);
 		const track = snapshot["track-entities"][trackId];
-		const source = snapshot["clip-source-relations"][`cs:${clipId}`];
-		if (!track || !source) throw new Error("Missing target Track or Clip source");
-		requireSource(snapshot, source.sourceId, track.role);
+		if (!track) throw new Error("Missing target Track or Clip source");
+		requireSource(snapshot, clip.source, track.role);
 		this.write("moveClip", {
 			clipId,
 			trackId,
@@ -4969,8 +5448,7 @@ var DslEditor = class {
 	trimClip(clipId, window, options) {
 		const snapshot = this.dsl.getSnapshot(), clip = requireClip(snapshot, clipId);
 		if (clip.seqMarker.fillTimeline) throw new Error("BGM trim is not enabled");
-		const sourceId = snapshot["clip-source-relations"][`cs:${clipId}`]?.sourceId;
-		const source = sourceId ? snapshot["media-entities"][sourceId] ?? snapshot["caption-entities"][sourceId] : void 0;
+		const source = snapshot["media-entities"][clip.source] ?? snapshot["caption-entities"][clip.source];
 		if (!source) throw new Error("Cannot trim a Clip with a missing source");
 		const patch = {};
 		if (window.inMs !== void 0) patch.inMs = window.inMs;
@@ -5013,30 +5491,76 @@ var DslEditor = class {
 			patch: { volume }
 		}], options);
 	}
-	createAudioScript(body, defaultLanguage, options) {
+	/** A script of these segments, in order; each segment gets its own id. */
+	createAudioScript(segments, defaultLanguage, options) {
 		const id = this.idFactory("audio-script");
+		const value = {
+			segments: segments.map((text) => ({
+				id: this.idFactory("segment"),
+				text
+			})),
+			...defaultLanguage === void 0 ? {} : { defaultLanguage }
+		};
 		this.write("createAudioScript", {
 			id,
-			body,
-			defaultLanguage
+			...value
 		}, [{
 			type: "create-entity",
 			table: "audio-script-entities",
 			id,
-			value: {
-				body,
-				...defaultLanguage === void 0 ? {} : { defaultLanguage }
-			}
+			value
 		}], options);
 		return id;
 	}
-	editAudioScript(scriptId, expectedBodyId, edits, options) {
-		this.dsl.editText(scriptId, expectedBodyId, edits, audit("editAudioScript", {
+	editAudioScript(scriptId, segmentId, expectedBodyId, edits, options) {
+		this.dsl.editText(scriptId, segmentId, expectedBodyId, edits, audit("editAudioScript", {
 			scriptId,
+			segmentId,
 			expectedBodyId,
 			edits
 		}, options));
 	}
+	/** Adds a segment at `index`, the end by default, and returns its id. */
+	insertScriptSegment(scriptId, text, index, options) {
+		const segmentId = this.idFactory("segment");
+		this.write("insertScriptSegment", {
+			scriptId,
+			segmentId,
+			text,
+			index
+		}, [{
+			type: "insert-segment",
+			audioScriptId: scriptId,
+			segmentId,
+			text,
+			...index === void 0 ? {} : { index }
+		}], options);
+		return segmentId;
+	}
+	/** `index` is measured after removing the segment. */
+	moveScriptSegment(scriptId, segmentId, index, options) {
+		this.write("moveScriptSegment", {
+			scriptId,
+			segmentId,
+			index
+		}, [{
+			type: "move-segment",
+			audioScriptId: scriptId,
+			segmentId,
+			index
+		}], options);
+	}
+	/** Captions and PhoneticScripts selecting the segment keep their selection, which then shows nothing (D72). */
+	deleteScriptSegment(scriptId, segmentId, options) {
+		this.write("deleteScriptSegment", {
+			scriptId,
+			segmentId
+		}, [{
+			type: "delete-segment",
+			audioScriptId: scriptId,
+			segmentId
+		}], options);
+	}
 	createCaption(input, options) {
 		const id = this.idFactory("caption");
 		this.write("createCaption", {
@@ -5053,18 +5577,16 @@ var DslEditor = class {
 		}, ...input.text !== void 0 ? [] : [{
 			type: "set-caption-content",
 			captionId: id,
-			audioScriptId: input.audioScriptId,
-			start: input.start,
-			end: input.end
+			...segmentRange(input)
 		}]], options);
 		return id;
 	}
-	setCaptionContent(captionId, audioScriptId, start, end, options) {
+	/** Show a segment of a script, or a run of it, in place of the Caption's own text. */
+	setCaptionContent(captionId, selection, options) {
+		const range = segmentRange(selection);
 		this.write("setCaptionContent", {
 			captionId,
-			audioScriptId,
-			start,
-			end
+			...range
 		}, [{
 			type: "patch-entity",
 			table: "caption-entities",
@@ -5073,9 +5595,7 @@ var DslEditor = class {
 		}, {
 			type: "set-caption-content",
 			captionId,
-			audioScriptId,
-			start,
-			end
+			...range
 		}], options);
 	}
 	setCaptionText(captionId, text, options) {
@@ -5083,9 +5603,10 @@ var DslEditor = class {
 		if (!caption) throw new Error("Missing Caption");
 		this.dsl.transact((tx) => {
 			tx.applyMutations([{
-				type: "delete-relation",
-				table: "caption-content-relations",
-				id: `cc:${captionId}`
+				type: "patch-entity",
+				table: "caption-entities",
+				id: captionId,
+				patch: { selection: void 0 }
 			}]);
 			if (caption.text === void 0) tx.draft["caption-entities"][captionId].text = text;
 			else if (caption.text !== text) tx.editCaptionText(captionId, [...caption.text.length ? [{
@@ -5199,7 +5720,8 @@ var DslEditor = class {
 		const result = {
 			clipId: input.clipId ?? this.idFactory("clip"),
 			mediaId: resourceId("media", input.media.system, input.media.key),
-			audioScriptId: input.script.audioScriptId ?? this.idFactory("audio-script"),
+			audioScriptId: this.voiceoverScriptId(input.script),
+			segmentIds: input.script.segments?.map(() => this.idFactory("segment")) ?? [],
 			captionClipIds: input.captions.map((c) => c.clipId ?? this.idFactory("clip")),
 			captionIds: input.captions.map(() => this.idFactory("caption"))
 		};
@@ -5209,6 +5731,68 @@ var DslEditor = class {
 		}, voiceoverPlan(this.dsl.getSnapshot(), input, result), options);
 		return result;
 	}
+	voiceoverScriptId(script) {
+		if (script.audioScriptId !== void 0) return script.audioScriptId;
+		if (script.phoneticScriptId === void 0) return this.idFactory("audio-script");
+		const phonetic = this.dsl.getSnapshot()["phonetic-script-entities"][script.phoneticScriptId];
+		if (!phonetic) throw new Error("Missing PhoneticScript");
+		return phonetic.phoneme.audioScriptId;
+	}
+	/**
+	* Write a PhoneticScript: create one for a segment of a script, or a run of it, or, given
+	* `phoneticScriptId`, replace its selection and what to synthesize together (D181). A
+	* `phonemeScript` is stored as a Memota text file first, and the file's facts are written with
+	* it; an unchanged one keeps its file. Recordings already spoken from it keep their derived-from.
+	*/
+	async setPhoneticScript(input, options) {
+		const id = input.phoneticScriptId ?? this.idFactory("phonetic-script");
+		const range = segmentRange(input);
+		const file = input.phonemeScript === void 0 ? {} : await this.fileFor(id, input.phonemeScript);
+		const exists = Object.hasOwn(this.dsl.getSnapshot()["phonetic-script-entities"], id);
+		const cleared = {
+			system: void 0,
+			key: void 0,
+			storageKey: void 0,
+			digest: void 0
+		};
+		const phonemeScript = input.phonemeScript === void 0 ? {} : { phonemeScript: input.phonemeScript };
+		this.write("setPhoneticScript", {
+			id,
+			...range,
+			...phonemeScript,
+			...cleared,
+			...file
+		}, [exists ? {
+			type: "patch-entity",
+			table: "phonetic-script-entities",
+			id,
+			patch: {
+				...cleared,
+				...file
+			}
+		} : {
+			type: "create-entity",
+			table: "phonetic-script-entities",
+			id,
+			value: file
+		}, {
+			type: "set-phoneme",
+			phoneticScriptId: id,
+			...range,
+			...phonemeScript
+		}], options);
+		return id;
+	}
+	/** The file holding `text`: the one the PhoneticScript keeps while its text is unchanged, else a new upload. */
+	async fileFor(id, text) {
+		const current = this.dsl.getSnapshot()["phonetic-script-entities"][id];
+		if (current?.phoneme.phonemeScript === text) {
+			const { phoneme: _, ...file } = current;
+			return file;
+		}
+		if (!this.uploadText) throw new Error("This editor stores no files");
+		return this.uploadText(text);
+	}
 	setBgm(input, options) {
 		const s = this.dsl.getSnapshot();
 		if (requireTrack(s, input.trackId).role !== "bgm") throw new Error("Expected a BGM Track");
@@ -5227,37 +5811,28 @@ var DslEditor = class {
 			fillTimeline: true
 		};
 		const volume = input.volume === void 0 ? {} : { volume: input.volume };
-		const mutations = [
-			...source.mutations,
-			old ? {
-				type: "patch-entity",
-				table: "clip-entities",
-				id,
-				patch: {
-					seqMarker: {
-						...seqMarker,
-						rate: void 0
-					},
-					...volume
-				}
-			} : {
-				type: "create-entity",
-				table: "clip-entities",
-				id,
-				value: {
-					seqMarker,
-					...volume
-				}
-			},
-			{
-				type: "set-relation",
-				table: "clip-source-relations",
-				value: {
-					clipId: id,
-					sourceId: source.id
-				}
+		const mutations = [...source.mutations, old ? {
+			type: "patch-entity",
+			table: "clip-entities",
+			id,
+			patch: {
+				seqMarker: {
+					...seqMarker,
+					rate: void 0
+				},
+				source: source.id,
+				...volume
 			}
-		];
+		} : {
+			type: "create-entity",
+			table: "clip-entities",
+			id,
+			value: {
+				seqMarker,
+				source: source.id,
+				...volume
+			}
+		}];
 		if (!old) mutations.push({
 			type: "insert-node",
 			entityId: id,
@@ -5273,6 +5848,15 @@ var DslEditor = class {
 		this.dsl.applyMutations(mutations, audit(action, payload, options));
 	}
 };
+/** A selection as a write states it: a run only when both its ends are given. */
+function segmentRange(input) {
+	return {
+		audioScriptId: input.audioScriptId,
+		segmentId: input.segmentId,
+		...input.start === void 0 ? {} : { start: input.start },
+		...input.end === void 0 ? {} : { end: input.end }
+	};
+}
 function audit(action, payload, options) {
 	return {
 		semantic_op: action,
@@ -5485,25 +6069,18 @@ function toVideoDraft(snapshot) {
 function requireWholeAudio(media, clip, fail) {
 	if (media.kind === "image" || clip.inMs !== 0 || clip.outMs !== media.durationMs || clip.rate !== 1) fail("VideoDraft audio has no trim or speed fields");
 }
+/**
+* What a recording says: what the selection of the PhoneticScript it was spoken from holds now,
+* or, for one derived from a script directly, the whole script, one segment per line.
+*/
 function spokenText(snapshot, recordingId) {
-	const relations = Object.values(snapshot["derived-from-relations"]);
-	const scripts = /* @__PURE__ */ new Set(), seen = /* @__PURE__ */ new Set();
-	const visit = (id) => {
-		if (seen.has(id)) return;
-		seen.add(id);
-		if (snapshot["audio-script-entities"][id]) {
-			scripts.add(id);
-			return;
-		}
-		for (const r of relations) if (r.entityId === id) visit(r.sourceId);
-	};
-	visit(recordingId);
-	if (!scripts.size) {
-		for (const r of relations) if (r.sourceId === recordingId && snapshot["audio-script-entities"][r.entityId]) scripts.add(r.entityId);
-	}
-	if (scripts.size > 1) projectionError("ambiguous-script", `media-entities/${recordingId}`, "Multiple scripts cannot be silently concatenated into audio_script");
-	const id = [...scripts][0];
-	return id ? snapshot.richText[id]?.plainText ?? "" : "";
+	const spoken = spokenScripts(snapshot, recordingId);
+	if (new Set(spoken.map((s) => s.phoneticScriptId ?? s.audioScriptId)).size > 1) projectionError("ambiguous-script", `media-entities/${recordingId}`, "Multiple scripts cannot be silently concatenated into audio_script");
+	const [first] = spoken;
+	if (!first) return "";
+	if (first.phoneticScriptId !== void 0) return snapshot.phoneticText[first.phoneticScriptId]?.text ?? "";
+	const script = snapshot["audio-script-entities"][first.audioScriptId];
+	return script ? script.segments.map((segment) => segment.text).join("\n") : "";
 }
 //#endregion
 //#region src/dsl/projection/from-video-draft.ts
@@ -5645,16 +6222,16 @@ function prepare(input, options) {
 		let script = scriptByRecording.get(recordingId);
 		if (script && script.text !== speech.audio_script) fail(`part_library/${id}/audio_script`, "One recording cannot acquire conflicting source scripts during import");
 		if (!script) {
-			script = {
-				id: claim(mint("audio-script")),
-				text: speech.audio_script
-			};
+			script = importedScript(claim(mint("audio-script")), speech.audio_script, () => mint("segment"));
 			scriptByRecording.set(recordingId, script);
 			mutations.push({
 				type: "create-entity",
 				table: "audio-script-entities",
 				id: script.id,
-				value: { body: script.text }
+				value: { segments: script.segments.map(({ id, text }) => ({
+					id,
+					text
+				})) }
 			}, {
 				type: "set-relation",
 				table: "derived-from-relations",
@@ -5702,12 +6279,18 @@ function prepare(input, options) {
 			const text = caption.text;
 			if (typeof text !== "string") fail(`part_library/${captionId}/text`, "Expected Caption text");
 			const start = text.length ? script.text.indexOf(text, cursor) : -1;
-			if (start >= 0) {
-				cursor = start + text.length;
-				selections.set(captionId, {
+			const end = start + text.length;
+			const segment = script.segments.find((s) => start >= s.start && end <= s.start + s.text.length);
+			if (start >= 0 && segment) {
+				cursor = end;
+				selections.set(captionId, text === segment.text ? {
+					audioScriptId: script.id,
+					segmentId: segment.id
+				} : {
 					audioScriptId: script.id,
-					start,
-					end: cursor
+					segmentId: segment.id,
+					start: start - segment.start,
+					end: end - segment.start
 				});
 			} else diagnostics.push({
 				code: "caption-text-unmatched",
@@ -5770,28 +6353,19 @@ function prepare(input, options) {
 				captionId: sourceId,
 				...selected
 			});
-			const speechId = caption.speech_part_id, mediaId = speechId ? mediaByPart.get(speechId) : void 0;
-			if (mediaId && caption.start_ms !== void 0) {
-				const startMs = milliseconds(caption.start_ms, `part_library/${id}/start_ms`), endMs = startMs + (caption.initial_duration_ms ?? outMs);
-				const media = imported.get(mediaId);
-				if ("durationMs" in media && endMs > startMs && endMs <= media.durationMs) mutations.push({
-					type: "set-relation",
-					table: "caption-alignment-relations",
-					value: {
-						captionId: sourceId,
-						mediaId,
-						startMs,
-						endMs
-					}
-				});
-				else diagnostics.push({
-					code: "invalid-caption-alignment",
-					path: `part_library/${id}`,
-					message: "Source interval is invalid; display window is retained without alignment"
-				});
-			}
 		}
 		if (outMs <= inMs || !sourceId) fail(`part_library/${id}`, "Expected a nonempty source window");
+		let anchor;
+		if (placement.role !== "video_clip") {
+			const host = hosts.get(id);
+			const speechId = part.caption?.speech_part_id;
+			const offsetMs = host?.offset ?? (speechId ? milliseconds(part.caption?.start_ms ?? 0, `part_library/${id}/start_ms`) : placement.abs);
+			anchor = {
+				to: host?.to ?? speechId ?? timelineId,
+				offsetMs,
+				fallback_abs_ms: placement.abs
+			};
+		}
 		mutations.push({
 			type: "create-entity",
 			table: "clip-entities",
@@ -5803,36 +6377,15 @@ function prepare(input, options) {
 					...speed ? { rate: speed.rate } : {},
 					...part.bgm ? { fillTimeline: true } : {}
 				},
+				source: sourceId,
+				...anchor === void 0 ? {} : { anchor },
 				...volume === void 0 ? {} : { volume }
 			}
-		}, {
-			type: "set-relation",
-			table: "clip-source-relations",
-			value: {
-				clipId: id,
-				sourceId
-			}
 		}, {
 			type: "insert-node",
 			entityId: id,
 			parentId: placement.trackId
 		});
-		if (placement.role !== "video_clip") {
-			const host = hosts.get(id);
-			const speechId = part.caption?.speech_part_id;
-			const to = host?.to ?? speechId ?? timelineId;
-			const offsetMs = host?.offset ?? (speechId ? milliseconds(part.caption?.start_ms ?? 0, `part_library/${id}/start_ms`) : placement.abs);
-			mutations.push({
-				type: "set-relation",
-				table: "time-anchor-relations",
-				value: {
-					from: id,
-					to,
-					offsetMs,
-					fallback_abs_ms: placement.abs
-				}
-			});
-		}
 	}
 	return {
 		timelineId,
@@ -5848,5 +6401,22 @@ function milliseconds(value, path, positive = false, signed = false) {
 	if (typeof value !== "number" || !Number.isSafeInteger(value) || !signed && value < 0 || positive && value <= 0) projectionError("invalid-time", path, "Expected safe integer milliseconds in the valid range");
 	return value;
 }
+function importedScript(id, text, mintSegment) {
+	const segments = [];
+	let start = 0;
+	if (text.length) for (const line of text.split("\n")) {
+		segments.push({
+			id: mintSegment(),
+			text: line,
+			start
+		});
+		start += line.length + 1;
+	}
+	return {
+		id,
+		text,
+		segments
+	};
+}
 //#endregion
-export { MedeoDsl as a, DslEditor as i, toVideoDraft as n, resolveDslLayout as o, createInitialMedeoDsl as r, DslProjectionError as s, fromVideoDraft as t };
+export { toAudioScriptViews as a, DslProjectionError as c, DslEditor as i, toVideoDraft as n, MedeoDsl as o, createInitialMedeoDsl as r, resolveDslLayout as s, fromVideoDraft as t };