Scribe | AI-Powered Narrative Generator

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}","deleted","label","children",[],[100],"files",[153],"counts",{"_175":32,"_176":177,"_178":32,"_179":180},"pulls","branches",1,"forks","envVars",{"_181":32,"_182":32},"count","emptyCount","contents",[],"file",{"_104":154,"_96":100,"_155":116,"_156":-5,"_157":141,"_158":140,"_159":118,"_106":187,"_101":161,"_162":163,"_164":165,"_166":22,"_167":-5,"_99":188},["D",1763267731161],[100],"upstreamProject","upstreamBranch","emptyEnvVars","isNew","flags",[],"routes/_app.x.$handle.$name.branch.$branch.history_.$",{"_94":95,"_96":97,"_98":44,"_99":100,"_102":197,"_185":202,"_205":206,"_256":257,"_270":-7,"_271":32,"_272":22},{"_104":105,"_96":97,"_106":198,"_108":-5,"_109":22,"_110":111,"_112":-5,"_113":-5,"_14":114,"_115":116,"_117":118,"_119":118,"_120":-5,"_121":32,"_16":122,"_123":124,"_125":124,"_126":127,"_128":127,"_129":127,"_130":124,"_131":124,"_23":199,"_135":200,"_138":201,"_150":22},["D",1763267216947],{"_94":95,"_133":134},{"_94":95,"_137":134,"_14":114},{"_101":140,"_96":44,"_104":141,"_142":105,"_143":-5,"_144":-5,"_145":-5,"_146":-5,"_147":-5,"_148":97,"_149":95},{"_104":154,"_96":100,"_155":116,"_156":-5,"_157":141,"_158":140,"_159":118,"_106":203,"_101":161,"_162":163,"_164":165,"_166":22,"_167":-5,"_99":204},["D",1763267731161],[100],"history",[207,209,214,219,224,229,234,239,244,249,253],{"_104":154,"_157":141,"_155":116,"_96":100,"_164":165,"_162":163,"_101":161,"_156":-5,"_106":208,"_158":140,"_167":-5},["D",1763267731161],{"_104":210,"_157":141,"_155":116,"_96":100,"_164":211,"_162":163,"_101":212,"_156":-5,"_106":213,"_158":161,"_167":-5},"a043b1ec-c2a5-11f0-8708-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",9,["D",1763267706256],{"_104":215,"_157":141,"_155":116,"_96":100,"_164":216,"_162":163,"_101":217,"_156":-5,"_106":218,"_158":212,"_167":-5},"8acd98b4-c2a5-11f0-aaac-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",8,["D",1763267670251],{"_104":220,"_157":141,"_155":116,"_96":100,"_164":221,"_162":163,"_101":222,"_156":-5,"_106":223,"_158":217,"_167":-5},"7865d722-c2a5-11f0-8eb2-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",7,["D",1763267639370],{"_104":225,"_157":141,"_155":116,"_96":100,"_164":226,"_162":163,"_101":227,"_156":-5,"_106":228,"_158":222,"_167":-5},"610a621e-c2a5-11f0-a699-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",6,["D",1763267600186],{"_104":230,"_157":141,"_155":116,"_96":100,"_164":231,"_162":163,"_101":232,"_156":-5,"_106":233,"_158":227,"_167":-5},"3855a0c2-c2a5-11f0-a524-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",5,["D",1763267531892],{"_104":235,"_157":141,"_155":116,"_96":100,"_164":236,"_162":163,"_101":237,"_156":-5,"_106":238,"_158":232,"_167":-5},"aac3be6a-c2a4-11f0-9a12-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

\n\n\n`;\n return fullHtml;\n}\n\n// --- HTTP REQUEST HANDLER (Modified for Scribe) ---\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const path = url.pathname;\n const headers = { \"Content-Type\": \"application/json\" };\n\n try {\n if (req.method === \"POST\" && path.endsWith(\"/start\")) {\n const { sentence } = await req.json(); // Changed from 'claim'\n if (\n !sentence || typeof sentence !== \"string\" ||\n sentence.trim().length === 0\n ) {\n return new Response(\n JSON.stringify({\n error: \"A valid, non-empty 'sentence' string is required.\",\n }),\n { status: 400, headers },\n );\n }\n const jobId = nanoid();\n const rootNode: DocumentNode = {\n id: nanoid(),\n sentence: sentence,\n context: sentence, // Initial context is just the sentence\n probability: 1.0, // Root node is 100%\n children: [],\n potentialBranches: [],\n depth: 0,\n };\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Fire off the background task\n buildDocumentInBackground(jobId, rootNode);\n\n return new Response(JSON.stringify({ jobId }), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/status\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const state: JobState | undefined = await blob.getJSON(jobId);\n if (!state) {\n return new Response(\n JSON.stringify({ status: \"error\", error: \"Job not found.\" }),\n { status: 404, headers },\n );\n }\n return new Response(JSON.stringify(state), { headers });\n }\n\n if (req.method === \"GET\" && path.endsWith(\"/list\")) {\n const index = await blob.getJSON(INDEX_KEY) || [];\n return new Response(JSON.stringify(index), { headers });\n }\n\n if (req.method === \"DELETE\" && path.endsWith(\"/delete\")) {\n const jobId = url.searchParams.get(\"jobId\");\n if (!jobId) {\n return new Response(\n JSON.stringify({ error: \"A 'jobId' query parameter is required.\" }),\n { status: 400, headers },\n );\n }\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newIndex = currentIndex.filter((item) => item.jobId !== jobId);\n await blob.setJSON(INDEX_KEY, newIndex);\n await blob.delete(jobId);\n return new Response(JSON.stringify({ success: true, jobId }), {\n headers,\n });\n }\n\n if (req.method === \"GET\" && (path === \"/\" || path.endsWith(\"/ui\"))) {\n const sourceUrl = `https://${url.hostname}${\n url.pathname.substring(0, url.pathname.lastIndexOf(\"/\"))\n }`;\n return new Response(generateHtml(sourceUrl), {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n return new Response(JSON.stringify({ error: \"Not Found\" }), {\n status: 404,\n headers,\n });\n } catch (error) {\n const errorMessage = error instanceof Error\n ? error.message\n : \"An unknown server error occurred.\";\n return new Response(JSON.stringify({ error: errorMessage }), {\n status: 500,\n headers,\n });\n }\n}",4,["D",1763267294378],{"_104":240,"_157":141,"_155":116,"_96":100,"_164":241,"_162":163,"_101":242,"_156":-5,"_106":243,"_158":237,"_167":-5},"8348f378-c2a4-11f0-a8a7-42dde27851f2","// @ts-ignore\nimport { OpenAI } from \"https://esm.town/v/std/openai?v=4\";\n// @ts-ignore\nimport { blob } from \"https://esm.town/v/std/blob?v=11\";\n// @ts-ignore\nimport { nanoid } from \"npm:nanoid\";\n\n// --- AI BEHAVIORAL GUIDELINES: REBRANDED FOR SCRIBE ---\nconst SCRIBE_PROTOCOL =\n \"As the AI generative author 'The Scribe', you must adopt the persona of a creative and collaborative writer. Your goal is to build a narrative iteratively. You must fluidly continue the provided text, offering diverse, high-quality, and plausible branching paths for the story to take. Your tone is imaginative, descriptive, and engaging.\";\n\n// --- PROMPT 1: THE CORE VERBAL SAMPLING PROMPT ---\nconst GENERATE_BRANCHES_PROMPT = `\nYou are 'The Scribe', an expert generative author. Your task is to continue the document provided in {DocumentContext}.\n\nGenerate {NumChoices} distinct, creative next sentences for the story. You MUST also assign a probability to each sentence, representing its creative and logical fit. The probabilities for the sentences must sum up to a value less than 1.\n\nYou MUST also provide a \"terminate_probability\", which is the chance that the story is at a good stopping point.\n\nThe sum of all sentence probabilities AND the terminate_probability MUST equal 1.0.\n\n--- EXAMPLE ---\n{\n \"candidates\": [\n {\"sentence\": \"He slowly opened the door, revealing a room bathed in shadow.\", \"probability\": 0.6},\n {\"sentence\": \"He decided against the door, turning instead to the window.\", \"probability\": 0.3}\n ],\n \"terminate_probability\": 0.1\n}\n--- END EXAMPLE ---\n\nYour final output MUST be a single, minified JSON object. Do not include any explanations or conversational text outside the JSON.\n\n--- DOCUMENT SO FAR ---\n{DocumentContext}\n`;\n\n// --- PROMPT 2: CREATIVE TONE MAPPING ---\nconst ANALYZE_TONE_PROMPT = `\nYou are the 'Creative Specialist', an AI that maps sentences to their narrative function or tone. Your task is to analyze a list of sentences and assign a single, concise, high-level creative category (1-4 words) to each.\n\nCRITICAL INSTRUCTION: Be specific. Avoid generic tags.\n\n--- EXAMPLE ---\nUSER: [\"The ship's hyperdrive whirred to life.\", \"What do you think you're doing?', Maria snapped.\", \"He remembered the promise he made on that rainy night.\", \"The city of Eldoria was built into the cliffside.\"]\nASSISTANT:\n{\n \"The ship's hyperdrive whirred to life.\": \"Action/Sci-Fi\",\n \"What do you think you're doing?', Maria snapped.\": \"Character Dialogue\",\n \"He remembered the promise he made on that rainy night.\": \"Internal Monologue\",\n \"The city of Eldoria was built into the cliffside.\": \"World-building\"\n}\n--- END EXAMPLE ---\n\nAnalyze the following sentences:\n{Sentences}\n`;\n\n// --- PROMPT 3: FINAL DOCUMENT SUMMARY ---\nconst SUMMARIZE_DOCUMENT_PROMPT = `\nYou are an AI Analyst reviewing the output of a generative writing process. Based on the provided JSON structure of a \"Scribe\" document, determine the outcome and write a brief summary. The document is a tree where the main narrative is a single, deep branch.\n\n- If the final node's children array is empty AND the process was not terminated early, the document is **COMPLETED**.\n- Otherwise, it is **INCOMPLETE**.\n\nAnalyze the following Scribe JSON and return a single JSON object with your conclusion.\n\nYour output MUST be in the format:\n{\n \"flag\": \"COMPLETED\" | \"INCOMPLETE\",\n \"summary\": \"A 1-2 sentence summary of the generated document.\"\n}\n\n--- SCRIBE DATA ---\n{ScribeData}\n`;\n\n// --- Type Definitions ---\ninterface SentenceCandidate {\n sentence: string;\n probability: number;\n}\n\ninterface DocumentNode {\n id: string;\n sentence: string; // The sentence *chosen* for this node\n context: string; // The full document up to and including this sentence\n probability: number; // The probability of this *chosen* sentence\n children: DocumentNode[]; // The *next* chosen node in the sequence\n potentialBranches: SentenceCandidate[]; // The *other* options not taken\n depth: number;\n tone?: string;\n}\n\ninterface JobState {\n status: \"processing\" | \"complete\" | \"error\";\n error?: string;\n document: DocumentNode | null;\n analysisFlag?: string;\n analysisSummary?: string;\n}\n\ninterface ScribeIndexEntry {\n jobId: string;\n rootSentence: string;\n timestamp: string;\n}\n\n// --- CORE AI & SCRIBE LOGIC ---\nconst MAX_DEPTH = 50; // Max number of sentences\nconst NUM_CHOICES = 3; // Number of branches to generate at each step\nconst openai = new OpenAI();\nconst INDEX_KEY = \"scribe_index\";\n\n/**\n * Performs a weighted random selection from a list of candidates.\n * This is the core of \"not always the top choice.\"\n */\nfunction stochasticSelect(\n candidates: SentenceCandidate[],\n): SentenceCandidate {\n const totalProbability = candidates.reduce(\n (sum, c) => sum + c.probability,\n 0,\n );\n const rand = Math.random() * totalProbability;\n let cumulative = 0;\n\n for (const candidate of candidates) {\n cumulative += candidate.probability;\n if (rand < cumulative) {\n return candidate;\n }\n }\n // Fallback (should rarely be hit)\n return candidates[0];\n}\n\nasync function callAI(\n replacements: Record,\n promptTemplate: string,\n): Promise {\n try {\n let finalContent = promptTemplate;\n for (const key in replacements) {\n finalContent = finalContent.replace(\n new RegExp(`\\\\{${key}\\\\}`, \"g\"),\n replacements[key],\n );\n }\n\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o\",\n messages: [\n { role: \"system\", content: SCRIBE_PROTOCOL },\n { role: \"user\", content: finalContent },\n ],\n response_format: { type: \"json_object\" },\n });\n const responseContent = completion.choices[0].message.content;\n const jsonMatch = responseContent.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) {\n throw new Error(\"AI response did not contain a valid JSON object.\");\n }\n return JSON.parse(jsonMatch[0]);\n } catch (error) {\n console.error(\n `AI call failed for prompt:`,\n error,\n );\n return null;\n }\n}\n\nasync function analyzeTones(\n sentences: string[],\n): Promise> {\n if (sentences.length === 0) return {};\n const result = await callAI(\n { Sentences: JSON.stringify(sentences) },\n ANALYZE_TONE_PROMPT,\n );\n return result || {};\n}\n\nasync function summarizeDocument(\n document: DocumentNode,\n): Promise<{ flag: string; summary: string }> {\n try {\n const result = await callAI(\n { ScribeData: JSON.stringify(document, null, 2) },\n SUMMARIZE_DOCUMENT_PROMPT,\n );\n if (result && result.flag && result.summary) {\n return result;\n }\n return {\n flag: \"ANALYSIS_INCOMPLETE\",\n summary: \"The final summary could not be generated.\",\n };\n } catch (error) {\n console.error(\"Document summary failed:\", error);\n return {\n flag: \"ANALYSIS_ERROR\",\n summary: \"An error occurred during the final summary step.\",\n };\n }\n}\n\nasync function buildDocumentInBackground(\n jobId: string,\n rootNode: DocumentNode,\n) {\n try {\n let currentNode = rootNode;\n let documentContext = rootNode.sentence;\n\n const rootToneMap = await analyzeTones([rootNode.sentence]);\n rootNode.tone = rootToneMap[rootNode.sentence] || \"Opening\";\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n for (let depth = 0; depth < MAX_DEPTH; depth++) {\n const branchResult = await callAI(\n {\n DocumentContext: documentContext,\n NumChoices: String(NUM_CHOICES),\n },\n GENERATE_BRANCHES_PROMPT,\n );\n\n if (!branchResult || !branchResult.candidates) {\n throw new Error(\"AI failed to generate branches.\");\n }\n\n let candidates: SentenceCandidate[] = branchResult.candidates || [];\n const terminateProbability = branchResult.terminate_probability || 0.01;\n\n // Add the termination candidate\n candidates.push({\n sentence: \"[END]\",\n probability: terminateProbability,\n });\n\n // --- This is the core logic ---\n const chosenCandidate = stochasticSelect(candidates);\n // -------------------------------\n\n // If [END] is chosen, terminate the loop\n if (chosenCandidate.sentence === \"[END]\") {\n break;\n }\n\n const otherOptions = candidates.filter((c) =>\n c.sentence !== chosenCandidate.sentence\n );\n documentContext += \" \" + chosenCandidate.sentence;\n\n const newNode: DocumentNode = {\n id: nanoid(),\n sentence: chosenCandidate.sentence,\n context: documentContext,\n probability: chosenCandidate.probability,\n children: [], // This will be filled in the *next* iteration\n potentialBranches: otherOptions,\n depth: currentNode.depth + 1,\n };\n\n // Get tone for the new sentence\n const toneMap = await analyzeTones([newNode.sentence]);\n newNode.tone = toneMap[newNode.sentence] || \"Uncategorized\";\n\n // Link the new node to the tree\n currentNode.children.push(newNode);\n\n // Move to the next node\n currentNode = newNode;\n\n // Save the entire updated tree structure at each step\n await blob.setJSON(jobId, { status: \"processing\", document: rootNode });\n\n // Add a small delay to prevent rate-limiting and allow UI to catch up\n await new Promise((res) => setTimeout(res, 500));\n }\n\n // Loop finished (either by max-depth or [END] selection)\n const finalAnalysis = await summarizeDocument(rootNode);\n\n await blob.setJSON(jobId, {\n status: \"complete\",\n document: rootNode,\n analysisFlag: finalAnalysis.flag,\n analysisSummary: finalAnalysis.summary,\n });\n\n // Update the master index\n const currentIndex: ScribeIndexEntry[] = await blob.getJSON(INDEX_KEY) ||\n [];\n const newEntry: ScribeIndexEntry = {\n jobId: jobId,\n rootSentence: rootNode.sentence,\n timestamp: new Date().toISOString(),\n };\n if (!currentIndex.some((item) => item.jobId === jobId)) {\n currentIndex.unshift(newEntry);\n await blob.setJSON(INDEX_KEY, currentIndex);\n }\n } catch (error) {\n console.error(\"Document building error:\", error);\n await blob.setJSON(jobId, {\n status: \"error\",\n document: rootNode,\n error: error.message,\n });\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n // ---\n // The HTML/CSS/React (Babel) code from your \"Totem\" app goes here.\n // I will only show the *key modifications* to the React components\n // to save space, as the CSS/HTML structure is identical.\n // ---\n const fullHtml = `\n\n\n \n \n Scribe | AI-Powered Narrative Generator\n \n ✍️\">\n \n \n \n \n \n \n \n \n\n\n\n

bomb

scribe2

History for main.ts