venice-video-harness 2.4.0

package/dist/parsers/pdf-parser.js

@@ -0,0 +1,237 @@
+import { readFile } from "node:fs/promises";
+import pdfParse from "pdf-parse";
+// ---------------------------------------------------------------------------
+// Constants
+// ---------------------------------------------------------------------------
+/** Scene heading prefixes (case-insensitive, after trimming). */
+const SCENE_HEADING_RE = /^\s*(INT\.|EXT\.|INT\.\/EXT\.|I\/E\.)\s*/i;
+/**
+ * Common transition patterns.
+ * Matches lines ending with "TO:" and well-known named transitions.
+ */
+const TRANSITION_ENDINGS_RE = /\bTO:$/;
+const NAMED_TRANSITIONS = new Set([
+    "CUT TO:",
+    "SMASH CUT TO:",
+    "MATCH CUT TO:",
+    "JUMP CUT TO:",
+    "DISSOLVE TO:",
+    "FADE IN:",
+    "FADE OUT.",
+    "FADE OUT:",
+    "FADE TO BLACK.",
+    "FADE TO BLACK:",
+    "FADE TO WHITE.",
+    "FADE TO WHITE:",
+    "WIPE TO:",
+    "IRIS IN:",
+    "IRIS OUT.",
+    "IRIS OUT:",
+    "TIME CUT:",
+]);
+/**
+ * Heuristic: a line is a character cue if it is ALL CAPS (ignoring
+ * parenthetical extensions like "(V.O.)" or "(O.S.)"), at least 2
+ * characters long, and not a scene heading or transition.
+ */
+const CHARACTER_CUE_RE = /^[A-Z][A-Z0-9 .'\-/]+(\s*\(.*\))?$/;
+/**
+ * Lines that look like page numbers commonly emitted by PDF renderers.
+ */
+const PAGE_NUMBER_RE = /^\s*\d{1,4}\s*\.?\s*$/;
+// ---------------------------------------------------------------------------
+// Raw text extraction
+// ---------------------------------------------------------------------------
+/**
+ * Extract the full text content from a PDF screenplay file.
+ *
+ * @param filePath - Absolute or relative path to a `.pdf` file.
+ * @returns The concatenated plain-text content of every page.
+ * @throws If the file cannot be read or is not a valid PDF.
+ */
+export async function parsePdf(filePath) {
+    const buffer = await readFile(filePath);
+    const result = await pdfParse(buffer);
+    return result.text;
+}
+// ---------------------------------------------------------------------------
+// Heuristic tokenizer
+// ---------------------------------------------------------------------------
+/**
+ * Determine whether a trimmed, non-empty line is a transition.
+ */
+function isTransition(line) {
+    const upper = line.toUpperCase();
+    if (TRANSITION_ENDINGS_RE.test(upper))
+        return true;
+    if (NAMED_TRANSITIONS.has(upper))
+        return true;
+    return false;
+}
+/**
+ * Determine whether a trimmed, non-empty line looks like a scene heading.
+ */
+function isSceneHeading(line) {
+    return SCENE_HEADING_RE.test(line);
+}
+/**
+ * Determine whether a trimmed, non-empty line looks like a character cue.
+ * Must NOT also be a scene heading or transition.
+ */
+function isCharacterCue(line) {
+    if (isSceneHeading(line))
+        return false;
+    if (isTransition(line))
+        return false;
+    if (PAGE_NUMBER_RE.test(line))
+        return false;
+    // Character names are typically short-ish; reject very long lines.
+    if (line.length > 60)
+        return false;
+    return CHARACTER_CUE_RE.test(line.trim());
+}
+/**
+ * Detect whether a line is a parenthetical (e.g. "(softly)").
+ */
+function isParenthetical(line) {
+    const trimmed = line.trim();
+    return trimmed.startsWith("(") && trimmed.endsWith(")");
+}
+/**
+ * Extract screenplay text from a PDF and apply format-agnostic heuristics
+ * to produce the same `FountainResult` shape that `parseFountain` returns.
+ *
+ * Because PDF extraction loses explicit formatting metadata, the heuristics
+ * are necessarily best-effort.  They work well on industry-standard
+ * screenplay PDFs (Final Draft, WriterSolo, Highland, etc.) but may
+ * misclassify some elements in heavily stylised documents.
+ *
+ * @param filePath - Path to a `.pdf` screenplay file.
+ * @returns A `FountainResult` with heuristically classified tokens.
+ */
+export async function parsePdfToTokens(filePath) {
+    const rawText = await parsePdf(filePath);
+    const lines = rawText.split(/\r?\n/);
+    const tokens = [];
+    // Title-page heuristic: the very first non-blank lines before the first
+    // scene heading are treated as the title block.  We try to extract a
+    // title, credit, and author from them.
+    let title = "";
+    let credit = "";
+    let author = "";
+    let titleBlockDone = false;
+    const titleBlockLines = [];
+    // State machine for dialogue grouping.
+    let expectDialogue = false;
+    for (let i = 0; i < lines.length; i++) {
+        const raw = lines[i];
+        const trimmed = raw.trim();
+        // Skip empty lines -- they reset the dialogue expectation.
+        if (trimmed.length === 0) {
+            expectDialogue = false;
+            continue;
+        }
+        // Skip page numbers.
+        if (PAGE_NUMBER_RE.test(trimmed)) {
+            continue;
+        }
+        // --- Title block collection (before first scene heading) ---
+        if (!titleBlockDone) {
+            if (isSceneHeading(trimmed)) {
+                titleBlockDone = true;
+                // Parse collected title block.
+                parseTitleBlock(titleBlockLines, (t, c, a) => {
+                    title = t;
+                    credit = c;
+                    author = a;
+                });
+                // Fall through so this line gets processed as a scene heading.
+            }
+            else {
+                titleBlockLines.push(trimmed);
+                continue;
+            }
+        }
+        // --- Scene heading ---
+        if (isSceneHeading(trimmed)) {
+            expectDialogue = false;
+            tokens.push({ type: "scene_heading", text: trimmed });
+            continue;
+        }
+        // --- Transition ---
+        if (isTransition(trimmed)) {
+            expectDialogue = false;
+            tokens.push({ type: "transition", text: trimmed });
+            continue;
+        }
+        // --- Parenthetical (only valid within dialogue context) ---
+        if (expectDialogue && isParenthetical(trimmed)) {
+            tokens.push({ type: "parenthetical", text: trimmed });
+            // Stay in dialogue mode.
+            continue;
+        }
+        // --- Dialogue (indented text after a character cue) ---
+        if (expectDialogue) {
+            tokens.push({ type: "dialogue", text: trimmed });
+            // Dialogue can span multiple contiguous non-blank lines.
+            continue;
+        }
+        // --- Character cue ---
+        if (isCharacterCue(trimmed)) {
+            tokens.push({ type: "character", text: trimmed });
+            expectDialogue = true;
+            continue;
+        }
+        // --- Everything else is action ---
+        tokens.push({ type: "action", text: trimmed });
+    }
+    // If there was never a scene heading, the title block was never
+    // committed.  Handle that edge case.
+    if (!titleBlockDone && titleBlockLines.length > 0) {
+        parseTitleBlock(titleBlockLines, (t, c, a) => {
+            title = t;
+            credit = c;
+            author = a;
+        });
+    }
+    return { title, credit, author, tokens };
+}
+// ---------------------------------------------------------------------------
+// Title block helpers
+// ---------------------------------------------------------------------------
+/**
+ * Very rough heuristic to pull title / credit / author from the lines
+ * that precede the first scene heading in a PDF screenplay.
+ *
+ * Common patterns:
+ *   Line 1: TITLE
+ *   Line 2: "Written by" | "Screenplay by" | "by"  (credit)
+ *   Line 3: Author Name
+ */
+function parseTitleBlock(lines, cb) {
+    if (lines.length === 0) {
+        cb("", "", "");
+        return;
+    }
+    let title = lines[0] ?? "";
+    let credit = "";
+    let author = "";
+    for (let i = 1; i < lines.length; i++) {
+        const lower = lines[i].toLowerCase();
+        if (lower.startsWith("written by") ||
+            lower.startsWith("screenplay by") ||
+            lower === "by") {
+            credit = lines[i];
+            // The next non-empty line is likely the author.
+            for (let j = i + 1; j < lines.length; j++) {
+                if (lines[j].trim().length > 0) {
+                    author = lines[j].trim();
+                    break;
+                }
+            }
+            break;
+        }
+    }
+    cb(title, credit, author);
+}
+//# sourceMappingURL=pdf-parser.js.map

package/dist/parsers/pdf-parser.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pdf-parser.js","sourceRoot":"","sources":["../../src/parsers/pdf-parser.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,QAAQ,MAAM,WAAW,CAAC;AAMjC,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,iEAAiE;AACjE,MAAM,gBAAgB,GAAG,2CAA2C,CAAC;AAErE;;;GAGG;AACH,MAAM,qBAAqB,GAAG,QAAQ,CAAC;AACvC,MAAM,iBAAiB,GAAG,IAAI,GAAG,CAAC;IAChC,SAAS;IACT,eAAe;IACf,eAAe;IACf,cAAc;IACd,cAAc;IACd,UAAU;IACV,WAAW;IACX,WAAW;IACX,gBAAgB;IAChB,gBAAgB;IAChB,gBAAgB;IAChB,gBAAgB;IAChB,UAAU;IACV,UAAU;IACV,WAAW;IACX,WAAW;IACX,WAAW;CACZ,CAAC,CAAC;AAEH;;;;GAIG;AACH,MAAM,gBAAgB,GAAG,oCAAoC,CAAC;AAE9D;;GAEG;AACH,MAAM,cAAc,GAAG,uBAAuB,CAAC;AAE/C,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,QAAgB;IAC7C,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACxC,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,MAAM,CAAC,CAAC;IACtC,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;GAEG;AACH,SAAS,YAAY,CAAC,IAAY;IAChC,MAAM,KAAK,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IACjC,IAAI,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACnD,IAAI,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IAC9C,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;GAEG;AACH,SAAS,cAAc,CAAC,IAAY;IAClC,OAAO,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACrC,CAAC;AAED;;;GAGG;AACH,SAAS,cAAc,CAAC,IAAY;IAClC,IAAI,cAAc,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACvC,IAAI,YAAY,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACrC,IAAI,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,mEAAmE;IACnE,IAAI,IAAI,CAAC,MAAM,GAAG,EAAE;QAAE,OAAO,KAAK,CAAC;IACnC,OAAO,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;AAC5C,CAAC;AAED;;GAEG;AACH,SAAS,eAAe,CAAC,IAAY;IACnC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;AAC1D,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,QAAgB;IAEhB,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACzC,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAErC,MAAM,MAAM,GAAoB,EAAE,CAAC;IAEnC,wEAAwE;IACxE,qEAAqE;IACrE,uCAAuC;IACvC,IAAI,KAAK,GAAG,EAAE,CAAC;IACf,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,MAAM,eAAe,GAAa,EAAE,CAAC;IAErC,uCAAuC;IACvC,IAAI,cAAc,GAAG,KAAK,CAAC;IAE3B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACrB,MAAM,OAAO,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;QAE3B,2DAA2D;QAC3D,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,cAAc,GAAG,KAAK,CAAC;YACvB,SAAS;QACX,CAAC;QAED,qBAAqB;QACrB,IAAI,cAAc,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;YACjC,SAAS;QACX,CAAC;QAED,8DAA8D;QAC9D,IAAI,CAAC,cAAc,EAAE,CAAC;YACpB,IAAI,cAAc,CAAC,OAAO,CAAC,EAAE,CAAC;gBAC5B,cAAc,GAAG,IAAI,CAAC;gBACtB,+BAA+B;gBAC/B,eAAe,CAAC,eAAe,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE;oBAC3C,KAAK,GAAG,CAAC,CAAC;oBACV,MAAM,GAAG,CAAC,CAAC;oBACX,MAAM,GAAG,CAAC,CAAC;gBACb,CAAC,CAAC,CAAC;gBACH,+DAA+D;YACjE,CAAC;iBAAM,CAAC;gBACN,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBAC9B,SAAS;YACX,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,cAAc,CAAC,OAAO,CAAC,EAAE,CAAC;YAC5B,cAAc,GAAG,KAAK,CAAC;YACvB,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACtD,SAAS;QACX,CAAC;QAED,qBAAqB;QACrB,IAAI,YAAY,CAAC,OAAO,CAAC,EAAE,CAAC;YAC1B,cAAc,GAAG,KAAK,CAAC;YACvB,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACnD,SAAS;QACX,CAAC;QAED,6DAA6D;QAC7D,IAAI,cAAc,IAAI,eAAe,CAAC,OAAO,CAAC,EAAE,CAAC;YAC/C,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACtD,yBAAyB;YACzB,SAAS;QACX,CAAC;QAED,yDAAyD;QACzD,IAAI,cAAc,EAAE,CAAC;YACnB,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACjD,yDAAyD;YACzD,SAAS;QACX,CAAC;QAED,wBAAwB;QACxB,IAAI,cAAc,CAAC,OAAO,CAAC,EAAE,CAAC;YAC5B,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YAClD,cAAc,GAAG,IAAI,CAAC;YACtB,SAAS;QACX,CAAC;QAED,oCAAoC;QACpC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;IACjD,CAAC;IAED,gEAAgE;IAChE,qCAAqC;IACrC,IAAI,CAAC,cAAc,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClD,eAAe,CAAC,eAAe,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE;YAC3C,KAAK,GAAG,CAAC,CAAC;YACV,MAAM,GAAG,CAAC,CAAC;YACX,MAAM,GAAG,CAAC,CAAC;QACb,CAAC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;AAC3C,CAAC;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;;;GAQG;AACH,SAAS,eAAe,CACtB,KAAe,EACf,EAA2D;IAE3D,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,EAAE,CAAC,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC;QACf,OAAO;IACT,CAAC;IAED,IAAI,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC3B,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,MAAM,GAAG,EAAE,CAAC;IAEhB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;QACrC,IACE,KAAK,CAAC,UAAU,CAAC,YAAY,CAAC;YAC9B,KAAK,CAAC,UAAU,CAAC,eAAe,CAAC;YACjC,KAAK,KAAK,IAAI,EACd,CAAC;YACD,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;YAClB,gDAAgD;YAChD,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC1C,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBAC/B,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;oBACzB,MAAM;gBACR,CAAC;YACH,CAAC;YACD,MAAM;QACR,CAAC;IACH,CAAC;IAED,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;AAC5B,CAAC"}

package/dist/parsers/scene-extractor.d.ts

@@ -0,0 +1,41 @@
+import type { FountainResult, FountainToken } from "./fountain-parser.js";
+export type { FountainResult, FountainToken };
+export interface DialogueLine {
+    character: string;
+    parenthetical?: string;
+    text: string;
+}
+export interface Scene {
+    number: number;
+    heading: string;
+    location: string;
+    timeOfDay: string;
+    characters: string[];
+    action: string[];
+    dialogue: DialogueLine[];
+    transitions: string[];
+    mood: string;
+}
+/**
+ * Walk through a flat token stream produced by either `parseFountain` or
+ * `parsePdfToTokens` and group them into structured `Scene` objects.
+ *
+ * Tokens appearing before the first `scene_heading` token are discarded
+ * (they typically represent title-page metadata that is already captured
+ * in `FountainResult.title` etc.).
+ *
+ * @param parsed - The `FountainResult` from any parser.
+ * @returns An ordered array of `Scene` objects, numbered from 1.
+ */
+export declare function extractScenes(parsed: FountainResult): Scene[];
+/**
+ * Collect every unique character name that appears across all scenes.
+ *
+ * Names are normalised (parenthetical extensions removed) and de-duplicated.
+ * The returned array preserves first-appearance order.
+ *
+ * @param scenes - Array of `Scene` objects produced by `extractScenes`.
+ * @returns Ordered, de-duplicated array of character names.
+ */
+export declare function extractAllCharacters(scenes: Scene[]): string[];
+//# sourceMappingURL=scene-extractor.d.ts.map

package/dist/parsers/scene-extractor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"scene-extractor.d.ts","sourceRoot":"","sources":["../../src/parsers/scene-extractor.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAG1E,YAAY,EAAE,cAAc,EAAE,aAAa,EAAE,CAAC;AAM9C,MAAM,WAAW,YAAY;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,KAAK;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,QAAQ,EAAE,YAAY,EAAE,CAAC;IACzB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;CACd;AA2PD;;;;;;;;;;GAUG;AACH,wBAAgB,aAAa,CAAC,MAAM,EAAE,cAAc,GAAG,KAAK,EAAE,CAiI7D;AAMD;;;;;;;;GAQG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,EAAE,CAc9D"}

package/dist/parsers/scene-extractor.js

@@ -0,0 +1,400 @@
+// ---------------------------------------------------------------------------
+// Heading parsing helpers
+// ---------------------------------------------------------------------------
+/**
+ * Extract the location portion from a scene heading.
+ *
+ * Examples:
+ *   "INT. JAZZ CLUB - NIGHT"          -> "JAZZ CLUB"
+ *   "EXT. ROOFTOP - DOWNTOWN - DUSK"  -> "ROOFTOP - DOWNTOWN"
+ *   "INT./EXT. CAR"                   -> "CAR"
+ *   "I/E. APARTMENT"                  -> "APARTMENT"
+ */
+function extractLocation(heading) {
+    // Remove the INT./EXT./INT./EXT./I/E. prefix.
+    const withoutPrefix = heading
+        .replace(/^\s*(INT\.\s*\/\s*EXT\.|I\s*\/\s*E\.|INT\.|EXT\.)\s*/i, "")
+        .trim();
+    // Everything before the *last* " - " delimiter is location.
+    // If there is no delimiter the entire remainder is the location.
+    const lastDash = withoutPrefix.lastIndexOf(" - ");
+    if (lastDash === -1)
+        return withoutPrefix;
+    return withoutPrefix.substring(0, lastDash).trim();
+}
+/**
+ * Extract the time-of-day portion from a scene heading.
+ *
+ * Looks for the text after the last " - " separator.  If the result
+ * matches a known time-of-day keyword it is returned; otherwise the
+ * raw text after the last dash is returned (to preserve non-standard
+ * time indicators like "LATER" or "MOMENTS LATER").
+ */
+const TIME_OF_DAY_KEYWORDS = new Set([
+    "DAY",
+    "NIGHT",
+    "DAWN",
+    "DUSK",
+    "MORNING",
+    "AFTERNOON",
+    "EVENING",
+    "SUNRISE",
+    "SUNSET",
+    "LATER",
+    "CONTINUOUS",
+    "MOMENTS LATER",
+    "SAME",
+    "SAME TIME",
+]);
+function extractTimeOfDay(heading) {
+    const lastDash = heading.lastIndexOf(" - ");
+    if (lastDash === -1)
+        return "";
+    const candidate = heading.substring(lastDash + 3).trim().toUpperCase();
+    if (TIME_OF_DAY_KEYWORDS.has(candidate))
+        return candidate;
+    // Partial match: heading may say "NIGHT (FLASHBACK)" etc.
+    for (const kw of TIME_OF_DAY_KEYWORDS) {
+        if (candidate.startsWith(kw))
+            return kw;
+    }
+    return candidate;
+}
+// ---------------------------------------------------------------------------
+// Mood inference
+// ---------------------------------------------------------------------------
+const MOOD_KEYWORDS = {
+    dark: [
+        "dark",
+        "shadow",
+        "dim",
+        "gloomy",
+        "ominous",
+        "sinister",
+        "bleak",
+        "eerie",
+        "foreboding",
+        "dread",
+    ],
+    tense: [
+        "tense",
+        "nervous",
+        "anxious",
+        "sweat",
+        "grip",
+        "clench",
+        "trembl",
+        "panic",
+        "frantic",
+        "urgent",
+        "desperate",
+        "suspense",
+    ],
+    violent: [
+        "punch",
+        "kick",
+        "gun",
+        "shot",
+        "blood",
+        "fight",
+        "stab",
+        "explo",
+        "crash",
+        "smash",
+        "attack",
+        "scream",
+        "wound",
+        "kill",
+        "dead",
+    ],
+    romantic: [
+        "kiss",
+        "embrace",
+        "love",
+        "tender",
+        "gentle",
+        "caress",
+        "intimate",
+        "passion",
+        "heart",
+        "gaze",
+    ],
+    warm: [
+        "warm",
+        "cozy",
+        "comfort",
+        "smile",
+        "laugh",
+        "joy",
+        "happy",
+        "bright",
+        "sun",
+        "glow",
+        "golden",
+        "cheerful",
+    ],
+    bright: [
+        "vivid",
+        "colorful",
+        "radiant",
+        "sparkl",
+        "gleam",
+        "shine",
+        "brilliant",
+        "luminous",
+    ],
+    quiet: [
+        "quiet",
+        "silent",
+        "still",
+        "calm",
+        "peace",
+        "serene",
+        "hushed",
+        "soft",
+        "whisper",
+        "murmur",
+    ],
+    chaotic: [
+        "chaos",
+        "crowd",
+        "rush",
+        "noise",
+        "loud",
+        "frenzy",
+        "mayhem",
+        "hectic",
+        "commotion",
+        "scramble",
+        "disorder",
+    ],
+    melancholy: [
+        "sad",
+        "tear",
+        "cry",
+        "weep",
+        "sorrow",
+        "grief",
+        "mourn",
+        "lonely",
+        "desolat",
+        "wistful",
+        "melan",
+        "somber",
+    ],
+    mysterious: [
+        "mystery",
+        "strange",
+        "odd",
+        "peculiar",
+        "enigma",
+        "unknown",
+        "secret",
+        "hidden",
+        "fog",
+        "mist",
+        "vanish",
+    ],
+};
+/**
+ * Score the combined action text of a scene against mood keyword buckets
+ * and return the mood with the highest hit count.  Falls back to "neutral"
+ * when no keywords are matched.
+ */
+function inferMood(actionLines) {
+    const blob = actionLines.join(" ").toLowerCase();
+    if (blob.length === 0)
+        return "neutral";
+    let bestMood = "neutral";
+    let bestScore = 0;
+    for (const [mood, keywords] of Object.entries(MOOD_KEYWORDS)) {
+        let score = 0;
+        for (const kw of keywords) {
+            // Use indexOf for substring matching so partial stems work
+            // (e.g. "trembl" matches "trembling").
+            let idx = 0;
+            while ((idx = blob.indexOf(kw, idx)) !== -1) {
+                score++;
+                idx += kw.length;
+            }
+        }
+        if (score > bestScore) {
+            bestScore = score;
+            bestMood = mood;
+        }
+    }
+    return bestMood;
+}
+// ---------------------------------------------------------------------------
+// Character name normalisation
+// ---------------------------------------------------------------------------
+/**
+ * Normalise a character name by stripping parenthetical extensions
+ * like "(V.O.)", "(O.S.)", "(CONT'D)" and trimming whitespace.
+ */
+function normaliseCharacterName(raw) {
+    return raw.replace(/\s*\(.*?\)\s*/g, "").trim();
+}
+// ---------------------------------------------------------------------------
+// Scene extraction
+// ---------------------------------------------------------------------------
+/**
+ * Walk through a flat token stream produced by either `parseFountain` or
+ * `parsePdfToTokens` and group them into structured `Scene` objects.
+ *
+ * Tokens appearing before the first `scene_heading` token are discarded
+ * (they typically represent title-page metadata that is already captured
+ * in `FountainResult.title` etc.).
+ *
+ * @param parsed - The `FountainResult` from any parser.
+ * @returns An ordered array of `Scene` objects, numbered from 1.
+ */
+export function extractScenes(parsed) {
+    const scenes = [];
+    let current = null;
+    let sceneNumber = 0;
+    // Dialogue accumulation state.
+    let currentCharacter = null;
+    let currentParenthetical = undefined;
+    /** Flush a completed dialogue line into the current scene. */
+    function flushDialogue(text) {
+        if (current && currentCharacter) {
+            const line = {
+                character: currentCharacter,
+                text,
+            };
+            if (currentParenthetical) {
+                line.parenthetical = currentParenthetical;
+            }
+            current.dialogue.push(line);
+            // Add character to the scene's unique set.
+            const normalised = normaliseCharacterName(currentCharacter);
+            if (normalised && !current.characters.includes(normalised)) {
+                current.characters.push(normalised);
+            }
+        }
+        currentParenthetical = undefined;
+    }
+    for (const token of parsed.tokens) {
+        switch (token.type) {
+            case "scene_heading": {
+                // Finalise previous scene mood before starting a new one.
+                if (current) {
+                    current.mood = inferMood(current.action);
+                }
+                sceneNumber++;
+                const heading = token.text ?? "";
+                current = {
+                    number: sceneNumber,
+                    heading,
+                    location: extractLocation(heading),
+                    timeOfDay: extractTimeOfDay(heading),
+                    characters: [],
+                    action: [],
+                    dialogue: [],
+                    transitions: [],
+                    mood: "neutral",
+                };
+                scenes.push(current);
+                currentCharacter = null;
+                currentParenthetical = undefined;
+                break;
+            }
+            case "character": {
+                if (!current)
+                    break;
+                currentCharacter = token.text ?? "";
+                currentParenthetical = undefined;
+                // Track the character even if they have no dialogue lines.
+                const normalised = normaliseCharacterName(currentCharacter);
+                if (normalised && !current.characters.includes(normalised)) {
+                    current.characters.push(normalised);
+                }
+                break;
+            }
+            case "parenthetical": {
+                if (!current)
+                    break;
+                currentParenthetical = token.text ?? "";
+                break;
+            }
+            case "dialogue": {
+                if (!current)
+                    break;
+                flushDialogue(token.text ?? "");
+                break;
+            }
+            case "action": {
+                if (!current)
+                    break;
+                currentCharacter = null;
+                currentParenthetical = undefined;
+                const text = token.text ?? "";
+                if (text.length > 0) {
+                    current.action.push(text);
+                }
+                break;
+            }
+            case "transition": {
+                if (!current)
+                    break;
+                currentCharacter = null;
+                currentParenthetical = undefined;
+                const text = token.text ?? "";
+                if (text.length > 0) {
+                    current.transitions.push(text);
+                }
+                break;
+            }
+            // Sections, synopses, notes, centered text, lyrics -- carry useful
+            // context but are treated as action for mood/content purposes.
+            case "section":
+            case "synopsis":
+            case "note":
+            case "centered":
+            case "lyrics": {
+                if (!current)
+                    break;
+                const text = token.text ?? "";
+                if (text.length > 0) {
+                    current.action.push(text);
+                }
+                break;
+            }
+            default:
+                // Unknown or structural token types are silently ignored.
+                break;
+        }
+    }
+    // Finalise mood for the last scene.
+    if (current) {
+        current.mood = inferMood(current.action);
+    }
+    return scenes;
+}
+// ---------------------------------------------------------------------------
+// Character extraction
+// ---------------------------------------------------------------------------
+/**
+ * Collect every unique character name that appears across all scenes.
+ *
+ * Names are normalised (parenthetical extensions removed) and de-duplicated.
+ * The returned array preserves first-appearance order.
+ *
+ * @param scenes - Array of `Scene` objects produced by `extractScenes`.
+ * @returns Ordered, de-duplicated array of character names.
+ */
+export function extractAllCharacters(scenes) {
+    const seen = new Set();
+    const result = [];
+    for (const scene of scenes) {
+        for (const name of scene.characters) {
+            if (!seen.has(name)) {
+                seen.add(name);
+                result.push(name);
+            }
+        }
+    }
+    return result;
+}
+//# sourceMappingURL=scene-extractor.js.map