{ "id": "cdTsbS89f6ZoBmfj", "name": "Filter, Deduplicate and Approve News Articles with Embeddings and a GPT Judge", "tags": [ { "id": "WiPfwUZYcO7Er2qC", "name": "Template", "createdAt": "2026-02-10T16:54:20.430Z", "updatedAt": "2026-02-10T16:54:20.430Z" } ], "nodes": [ { "id": "sticky_config", "name": "Sticky config", "type": "n8n-nodes-base.stickyNote", "position": [ -976, -400 ], "parameters": { "color": 4, "width": 480, "height": 764, "content": "## Edit knobs here, nothing else\nAll tuning lives in Configure context:\n- **sheetDocumentId / sourceTab / articlesTab / approvedTab** - the spreadsheet, the incoming-feed tab, the seen-articles ledger tab, and the output tab\n- **rulesKeywords** - term-to-weight map for Step 1; a headline hit counts double a body hit (word-boundary matched)\n- **sourceWeights** - domain-to-weight map (live articles carry their feed's domain as source); unknown domains score 0\n- **rulesRejectThreshold** - score below this = rejected immediately (default 2)\n- **rulesHardWinThreshold** - score at or above this skips the LLM judge (default 12)\n- **clusterSimilarityThreshold** - cosine similarity at which two articles are the same story (default 0.80; raise it and clusters fragment, lower it and distinct stories merge)\n- **embeddingProvider** - huggingface or openai; both branches emit the same shape\n- **hfEmbeddingModel / openaiEmbeddingModel** - embedding model per provider\n- **judgeModel** - OpenAI chat model used by the Step 3 judge\n- **clearBeforeRun** - true wipes the ledger and output tabs (headers always survive) first (clean demo take); false keeps history, and articles already in the ledger are dropped before any money is spent\n- **rssFeedUrls** - JSON array of RSS feed URLs, all fetched every run; add as many as you like\n- **useLiveFetch** - true (default) pulls every feed in rssFeedUrls into the run; false reads only the Source Articles tab\n- **nicheContext** - the judge's definition of relevant; swap it to point the whole pipeline at another niche" }, "typeVersion": 1 }, { "id": "d38ca2d6-5964-49d5-9523-b9423e859d2c", "name": "Sticky Note1", "type": "n8n-nodes-base.stickyNote", "position": [ -464, -160 ], "parameters": { "color": 7, "width": 416, "height": 528, "content": "## Triggers and context\n\nStarts the pipeline manually or on a daily schedule, then sets the spreadsheet, tab, feed, and run configuration used by downstream nodes." }, "typeVersion": 1 }, { "id": "d2faf635-1be5-4ea9-b28e-c6ee7ab86fc7", "name": "Sticky Note2", "type": "n8n-nodes-base.stickyNote", "position": [ -16, -288 ], "parameters": { "color": 7, "width": 864, "height": 480, "content": "## Fresh run cleanup\n\nChecks whether this run should start fresh, clears prior approved packages and the seen ledger when requested, then rejoins the normal flow." }, "typeVersion": 1 }, { "id": "07a4b82e-48d7-4ca7-b447-e01a034fea1a", "name": "Sticky Note3", "type": "n8n-nodes-base.stickyNote", "position": [ 208, 208 ], "parameters": { "color": 7, "width": 864, "height": 304, "content": "## Live RSS intake\n\nOptionally fetches live articles by splitting configured feed URLs, reading each RSS feed, and mapping results into the same article shape as the sheet data." }, "typeVersion": 1 }, { "id": "881a1639-ffd4-4da9-9556-72c9a9b01490", "name": "Sticky Note4", "type": "n8n-nodes-base.stickyNote", "position": [ 880, -352 ], "parameters": { "color": 7, "width": 416, "height": 544, "content": "## Read source records\n\nReads source articles from Google Sheets, appends any mapped live articles, and reads the existing seen ledger for deduplication." }, "typeVersion": 1 }, { "id": "e1df064e-6455-497f-add1-c66a85de908a", "name": "Sticky Note5", "type": "n8n-nodes-base.stickyNote", "position": [ 1328, -80 ], "parameters": { "color": 7, "width": 640, "height": 320, "content": "## Deduplicate new URLs\n\nDrops articles whose URLs are already in the ledger, stamps the remaining items with pending defaults, and records them as seen before scoring." }, "typeVersion": 1 }, { "id": "83f04724-d163-4124-ae37-dd71fa0f4547", "name": "Sticky Note6", "type": "n8n-nodes-base.stickyNote", "position": [ 2000, -80 ], "parameters": { "color": 7, "width": 640, "height": 320, "content": "## Rule score candidates\n\nApplies deterministic rule-based scoring, removes rejected noise, and collects the surviving items into a batch for embedding." }, "typeVersion": 1 }, { "id": "9afc69f8-6401-4ee2-a55e-a46100b50b4c", "name": "Sticky Note7", "type": "n8n-nodes-base.stickyNote", "position": [ 2672, -144 ], "parameters": { "color": 7, "width": 640, "height": 528, "content": "## Generate article embeddings\n\nSelects an embedding provider, sends the batch to Hugging Face or OpenAI, and normalizes either provider\u2019s vector response into a common vectors field." }, "typeVersion": 1 }, { "id": "19e029ba-3d6a-4b93-9c5e-73fcb88d8f28", "name": "Sticky Note8", "type": "n8n-nodes-base.stickyNote", "position": [ 3344, -80 ], "parameters": { "color": 7, "width": 416, "height": 320, "content": "## Cluster and route\n\nClusters embedded articles to group near-duplicate coverage of the same story, then decides which clusters require LLM judgment." }, "typeVersion": 1 }, { "id": "fb8639bf-a6e1-4188-bb2f-072d0d0c834f", "name": "Sticky Note9", "type": "n8n-nodes-base.stickyNote", "position": [ 3792, -144 ], "parameters": { "color": 7, "width": 400, "height": 528, "content": "## GPT relevance judge\n\nRuns the OpenAI-powered LLM judge for ambiguous items and parses its decision into a structured verdict." }, "typeVersion": 1 }, { "id": "53cb95eb-a1de-4f35-b2a0-8d9e7f10bf48", "name": "Sticky Note10", "type": "n8n-nodes-base.stickyNote", "position": [ 4224, -80 ], "parameters": { "color": 7, "width": 416, "height": 320, "content": "## Combine approved paths\n\nMerges items that bypassed the judge with judged results, then assembles final approved story packages from the clustered and scored data." }, "typeVersion": 1 }, { "id": "64599911-5ea6-4268-b520-06bbc9fe0984", "name": "Sticky Note11", "type": "n8n-nodes-base.stickyNote", "position": [ 4672, -208 ], "parameters": { "color": 7, "height": 384, "content": "## Write approved packages\n\nWrites the assembled approved story packages to the configured approved output sheet." }, "typeVersion": 1 }, { "id": "58d2365d-d1dc-40e0-8b95-49b2546d9ecb", "name": "Sticky Note12", "type": "n8n-nodes-base.stickyNote", "position": [ 4672, 192 ], "parameters": { "color": 7, "width": 416, "height": 352, "content": "## Update outcome ledger\n\nCompiles every article\u2019s final fate and updates the seen ledger so future runs can skip or report on previously processed URLs." }, "typeVersion": 1 }, { "id": "293a50f6-37ee-471c-9eb4-0a78d598c687", "name": "Manual Pipeline Trigger", "type": "n8n-nodes-base.manualTrigger", "position": [ -416, 0 ], "parameters": {}, "typeVersion": 1 }, { "id": "db541963-7378-4615-b2e9-ebcbc7ae31c1", "name": "Set Workflow Context", "type": "n8n-nodes-base.set", "position": [ -192, 80 ], "parameters": { "options": {}, "assignments": { "assignments": [ { "id": "c01", "name": "sheetDocumentId", "type": "string", "value": "11UAshmcWzrzNZRBqeDSL78r7b6GWxou-UJdxhCXRlUY" }, { "id": "c16", "name": "sourceTab", "type": "string", "value": "Source Articles" }, { "id": "c02", "name": "articlesTab", "type": "string", "value": "Articles" }, { "id": "c03", "name": "approvedTab", "type": "string", "value": "Approved Package" }, { "id": "c04", "name": "rulesKeywords", "type": "object", "value": "{\"ai\": 1, \"artificial intelligence\": 2, \"model\": 2, \"llm\": 2, \"agent\": 2, \"openai\": 2, \"anthropic\": 2, \"chip\": 1, \"compute\": 2, \"inference\": 2, \"open source\": 1, \"lawsuit\": 1, \"regulation\": 1, \"data center\": 1, \"funding\": 1, \"gpu\": 1, \"voice\": 1, \"copyright\": 1}" }, { "id": "c05", "name": "sourceWeights", "type": "object", "value": "{\"techcrunch.com\": 2, \"theverge.com\": 2, \"arstechnica.com\": 2, \"venturebeat.com\": 1, \"wired.com\": 1, \"reuters.com\": 2}" }, { "id": "c06", "name": "rulesRejectThreshold", "type": "number", "value": 2 }, { "id": "c07", "name": "rulesHardWinThreshold", "type": "number", "value": 12 }, { "id": "c08", "name": "clusterSimilarityThreshold", "type": "number", "value": 0.8 }, { "id": "c15", "name": "clearBeforeRun", "type": "boolean", "value": false }, { "id": "c09", "name": "embeddingProvider", "type": "string", "value": "huggingface" }, { "id": "c10", "name": "hfEmbeddingModel", "type": "string", "value": "sentence-transformers/all-MiniLM-L6-v2" }, { "id": "c11", "name": "openaiEmbeddingModel", "type": "string", "value": "text-embedding-3-small" }, { "id": "c12", "name": "judgeModel", "type": "string", "value": "gpt-5-mini" }, { "id": "c17", "name": "rssFeedUrls", "type": "array", "value": "[\"https://techcrunch.com/category/artificial-intelligence/feed/\"]" }, { "id": "c14", "name": "useLiveFetch", "type": "boolean", "value": true }, { "id": "c13", "name": "nicheContext", "type": "string", "value": "News for people who build AI products and automation: model releases, AI infrastructure and compute, AI business deals, AI regulation, and AI engineering practice." } ] } }, "typeVersion": 3.4 }, { "id": "dca7b91a-c863-444e-a444-bdd69a97d085", "name": "Clear Package Data in Sheets", "type": "n8n-nodes-base.googleSheets", "position": [ 256, -112 ], "parameters": { "operation": "clear", "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.approvedTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" }, "keepFirstRow": true }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "7e2b0a7f-6133-4090-94ab-3efbf4cc0eca", "name": "Read Articles from Sheets", "type": "n8n-nodes-base.googleSheets", "position": [ 928, -192 ], "parameters": { "options": {}, "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.sourceTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" } }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "4c4ae13e-726f-4d8d-bf4c-64f87efd6a96", "name": "Apply Scoring Rules", "type": "n8n-nodes-base.code", "position": [ 2048, 80 ], "parameters": { "jsCode": "// Step 1 of 3: deterministic scoring. Free, instant, and it never changes its mind.\n// Code instead of a native IF-chain: the keyword and source weight maps live in\n// 'Configure context', and native nodes cannot iterate a configurable map.\nconst config = $('Set Workflow Context').first().json;\nconst keywordWeights = config.rulesKeywords;\nconst sourceWeights = config.sourceWeights;\n\nconst scoredItems = [];\nfor (const item of $input.all()) {\n const article = item.json;\n const titleText = String(article.title || '').toLowerCase();\n const bodyText = (String(article.summary || '') + ' ' + String(article.content_excerpt || '')).toLowerCase();\n\n // A keyword in the headline is worth double a keyword buried in the body.\n // Word-boundary matching: 'ai' must not match inside 'bargain'.\n let keywordScore = 0;\n for (const [term, weight] of Object.entries(keywordWeights)) {\n const escapedTerm = term.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&');\n const termPattern = new RegExp(`\\\\b${escapedTerm}\\\\b`, 'i');\n if (termPattern.test(titleText)) keywordScore += weight * 2;\n else if (termPattern.test(bodyText)) keywordScore += weight;\n }\n const sourceScore = sourceWeights[article.source] || 0;\n const rulesScore = keywordScore + sourceScore;\n\n let rulesBand = 'pass';\n if (rulesScore < config.rulesRejectThreshold) rulesBand = 'reject';\n else if (rulesScore >= config.rulesHardWinThreshold) rulesBand = 'hardWin';\n\n scoredItems.push({ json: { ...article, rulesScore, rulesBand, embedText: `${article.title}. ${article.summary}` } });\n}\nreturn scoredItems;" }, "typeVersion": 2 }, { "id": "eb988b03-aaef-4bda-be7c-a90eb4548130", "name": "Filter Rejected Articles", "type": "n8n-nodes-base.filter", "position": [ 2272, 80 ], "parameters": { "options": {}, "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "f1", "operator": { "type": "string", "operation": "notEquals" }, "leftValue": "={{ $json.rulesBand }}", "rightValue": "reject" } ] } }, "typeVersion": 2.2 }, { "id": "d9d9093d-4dee-4ef8-974c-404ae1cc4c58", "name": "Aggregate for Embedding", "type": "n8n-nodes-base.aggregate", "position": [ 2496, 80 ], "parameters": { "options": {}, "aggregate": "aggregateAllItemData" }, "typeVersion": 1 }, { "id": "959210e7-d8cb-47cf-8961-a476de01289f", "name": "Select Embedding Provider", "type": "n8n-nodes-base.switch", "position": [ 2720, 80 ], "parameters": { "rules": { "values": [ { "outputKey": "Hugging Face", "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "s1", "operator": { "type": "string", "operation": "equals" }, "leftValue": "={{ $('Set Workflow Context').first().json.embeddingProvider }}", "rightValue": "huggingface" } ] }, "renameOutput": true }, { "outputKey": "OpenAI", "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "s2", "operator": { "type": "string", "operation": "equals" }, "leftValue": "={{ $('Set Workflow Context').first().json.embeddingProvider }}", "rightValue": "openai" } ] }, "renameOutput": true } ] }, "options": {} }, "typeVersion": 3.2 }, { "id": "bcd2c569-151c-40f0-ae4b-0e3b874234c6", "name": "Post to Hugging Face API", "type": "n8n-nodes-base.httpRequest", "position": [ 2944, 16 ], "parameters": { "url": "={{ 'https://router.huggingface.co/hf-inference/models/' + $('Set Workflow Context').first().json.hfEmbeddingModel + '/pipeline/feature-extraction' }}", "method": "POST", "options": { "response": { "response": { "fullResponse": true } } }, "jsonBody": "={{ JSON.stringify({ inputs: $json.data.map(article => article.embedText) }) }}", "sendBody": true, "specifyBody": "json", "authentication": "predefinedCredentialType", "nodeCredentialType": "huggingFaceApi" }, "credentials": { "huggingFaceApi": { "name": "" } }, "typeVersion": 4.2 }, { "id": "2626a24a-f3a2-48b6-baf4-f1052f2106f9", "name": "Post to OpenAI Embeddings", "type": "n8n-nodes-base.httpRequest", "position": [ 2944, 208 ], "parameters": { "url": "https://api.openai.com/v1/embeddings", "method": "POST", "options": { "response": { "response": { "fullResponse": true } } }, "jsonBody": "={{ JSON.stringify({ model: $('Set Workflow Context').first().json.openaiEmbeddingModel, input: $json.data.map(article => article.embedText) }) }}", "sendBody": true, "specifyBody": "json", "authentication": "predefinedCredentialType", "nodeCredentialType": "openAiApi" }, "credentials": { "openAiApi": { "name": "" } }, "typeVersion": 4.2 }, { "id": "d2c9c1d1-ef13-44c9-b8ca-50203ac1e7fe", "name": "Set Hugging Face Vectors", "type": "n8n-nodes-base.set", "position": [ 3168, 16 ], "parameters": { "options": {}, "assignments": { "assignments": [ { "id": "n1", "name": "vectors", "type": "array", "value": "={{ $json.body }}" } ] } }, "typeVersion": 3.4 }, { "id": "be511255-f154-481f-8e44-0bb25b3b47c9", "name": "Set OpenAI Vectors", "type": "n8n-nodes-base.set", "position": [ 3168, 208 ], "parameters": { "options": {}, "assignments": { "assignments": [ { "id": "n2", "name": "vectors", "type": "array", "value": "={{ $json.body.data.map(entry => entry.embedding) }}" } ] } }, "typeVersion": 3.4 }, { "id": "2c097be8-2772-42f9-9030-697206cf2447", "name": "Cluster Similar Stories", "type": "n8n-nodes-base.code", "position": [ 3392, 80 ], "parameters": { "jsCode": "// Step 2 of 3: group near-duplicate coverage of the same story.\n// Code because similarity clustering needs a pairwise cosine matrix,\n// which no native node expresses. Greedy single-link over 50 articles is instant.\n// One item per STORY leaves this node, so the canvas count itself shows the dedup.\nconst config = $('Set Workflow Context').first().json;\nconst threshold = Number(config.clusterSimilarityThreshold);\nconst articles = $('Filter Rejected Articles').all().map(item => item.json);\nconst vectors = $input.first().json.vectors;\nif (!Array.isArray(vectors) || vectors.length !== articles.length) {\n throw new Error(`Vector count (${vectors && vectors.length}) does not match article count (${articles.length})`);\n}\n\nfunction cosineSimilarity(vectorA, vectorB) {\n let dotProduct = 0;\n let magnitudeA = 0;\n let magnitudeB = 0;\n for (let dim = 0; dim < vectorA.length; dim++) {\n dotProduct += vectorA[dim] * vectorB[dim];\n magnitudeA += vectorA[dim] * vectorA[dim];\n magnitudeB += vectorB[dim] * vectorB[dim];\n }\n return dotProduct / (Math.sqrt(magnitudeA) * Math.sqrt(magnitudeB));\n}\n\nconst clusters = [];\nfor (let articleIndex = 0; articleIndex < articles.length; articleIndex++) {\n let joinedExisting = false;\n for (const cluster of clusters) {\n if (cluster.some(memberIndex => cosineSimilarity(vectors[articleIndex], vectors[memberIndex]) >= threshold)) {\n cluster.push(articleIndex);\n joinedExisting = true;\n break;\n }\n }\n if (!joinedExisting) clusters.push([articleIndex]);\n}\n\n// The primary is picked deterministically: highest rules score, longer excerpt on ties.\n// Every other member rides along on the primary as corroborating coverage.\nconst storyItems = [];\nclusters.forEach((memberIndexes, clusterNumber) => {\n const primaryIndex = memberIndexes.reduce((bestIndex, candidateIndex) => {\n const bestArticle = articles[bestIndex];\n const candidateArticle = articles[candidateIndex];\n if (candidateArticle.rulesScore !== bestArticle.rulesScore) {\n return candidateArticle.rulesScore > bestArticle.rulesScore ? candidateIndex : bestIndex;\n }\n return String(candidateArticle.content_excerpt || '').length > String(bestArticle.content_excerpt || '').length\n ? candidateIndex : bestIndex;\n });\n const corroborating = memberIndexes\n .filter(memberIndex => memberIndex !== primaryIndex)\n .map(memberIndex => ({\n title: articles[memberIndex].title,\n source: articles[memberIndex].source,\n url: articles[memberIndex].url,\n }));\n storyItems.push({ json: {\n ...articles[primaryIndex],\n clusterId: clusterNumber + 1,\n clusterSize: memberIndexes.length,\n isPrimary: true,\n corroborating,\n }});\n});\nreturn storyItems;" }, "typeVersion": 2 }, { "id": "f3d87989-002a-4226-a122-14fb28c57a12", "name": "Check for LLM Judgment", "type": "n8n-nodes-base.if", "position": [ 3616, 80 ], "parameters": { "options": {}, "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "i1", "operator": { "type": "boolean", "operation": "equals" }, "leftValue": "={{ $json.isPrimary }}", "rightValue": true }, { "id": "i2", "operator": { "type": "string", "operation": "notEquals" }, "leftValue": "={{ $json.rulesBand }}", "rightValue": "hardWin" } ] } }, "typeVersion": 2.2 }, { "id": "e85bc260-3647-4351-a51e-1ff687232cea", "name": "Relevance Judgment", "type": "@n8n/n8n-nodes-langchain.chainLlm", "position": [ 3888, 16 ], "parameters": { "text": "=You are the relevance judge in a content pipeline.\nNiche: {{ $('Set Workflow Context').first().json.nicheContext }}\n\nArticle:\nTitle: {{ $json.title }}\nSource: {{ $json.source }}\nSummary: {{ $json.summary }}\nExcerpt: {{ $json.content_excerpt }}\n\nDecide whether this article is relevant enough to become client-facing content in this niche. Reject encyclopedia-style evergreen pages, thin social posts with no substance, and anything off-niche.", "batching": { "batchSize": 10 }, "messages": { "messageValues": [] }, "promptType": "define", "hasOutputParser": true }, "typeVersion": 1.9 }, { "id": "0820d66e-867b-4549-8295-178ff874a6e5", "name": "Set OpenAI Judge Model", "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi", "position": [ 3888, 240 ], "parameters": { "model": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.judgeModel }}" }, "options": {} }, "credentials": { "openAiApi": { "name": "" } }, "typeVersion": 1.2 }, { "id": "8ff317d3-d417-4083-8873-02344192c034", "name": "Parse Judgment Verdict", "type": "@n8n/n8n-nodes-langchain.outputParserStructured", "position": [ 4096, 240 ], "parameters": { "schemaType": "manual", "inputSchema": "{\"type\": \"object\", \"properties\": {\"verdict\": {\"type\": \"string\", \"enum\": [\"approved\", \"rejected\"]}, \"reason\": {\"type\": \"string\", \"description\": \"One sentence explaining the verdict\"}}, \"required\": [\"verdict\", \"reason\"]}" }, "typeVersion": 1.2 }, { "id": "15b20e32-de78-4f2b-90a2-1e49946f9108", "name": "Assemble Approved Articles", "type": "n8n-nodes-base.code", "position": [ 4496, 80 ], "parameters": { "jsCode": "// The payoff: one row per approved story. Code because this aggregates three\n// sources (story primaries, judge verdicts, run counts) into a single package shape.\nconst storyPrimaries = $('Cluster Similar Stories').all().map(item => item.json);\nconst judgedPrimaries = storyPrimaries.filter(story => story.rulesBand !== 'hardWin');\n\nlet judgeOutputs = [];\ntry {\n judgeOutputs = $('Relevance Judgment').all().map(item => item.json);\n} catch (noJudgeRun) {\n judgeOutputs = []; // every story was a hard win; the LLM was never called\n}\n\nconst verdictByCluster = {};\nfor (const story of storyPrimaries) {\n if (story.rulesBand === 'hardWin') {\n verdictByCluster[story.clusterId] = { verdict: 'approved', reason: 'Rules hard win - skipped the LLM entirely' };\n }\n}\njudgedPrimaries.forEach((story, judgeIndex) => {\n const judgeResult = judgeOutputs[judgeIndex] ? (judgeOutputs[judgeIndex].output || judgeOutputs[judgeIndex]) : null;\n verdictByCluster[story.clusterId] = {\n verdict: judgeResult && judgeResult.verdict ? judgeResult.verdict : 'rejected',\n reason: judgeResult && judgeResult.reason ? judgeResult.reason : 'No judge output received',\n };\n});\n\nconst articlesIn = $('Read Articles from Sheets').all().length;\nconst survivedRules = $('Filter Rejected Articles').all().length;\n\nconst runId = $execution.id || 'manual';\nconst runAt = new Date().toISOString();\nconst runSignature = `${articlesIn}-${survivedRules}-${storyPrimaries.length}-${judgedPrimaries.length}`;\n\nconst packageRows = [];\nfor (const story of storyPrimaries) {\n const clusterVerdict = verdictByCluster[story.clusterId] || { verdict: 'rejected', reason: 'Missing verdict' };\n if (clusterVerdict.verdict !== 'approved') continue;\n const corroborating = (story.corroborating || [])\n .map(member => `${member.title} - ${member.source} (${member.url})`)\n .join(' | ');\n packageRows.push({ json: {\n run_id: runId,\n run_at: runAt,\n run_signature: runSignature,\n cluster_id: story.clusterId,\n best_title: story.title,\n best_source: story.source,\n best_url: story.url,\n rules_score: story.rulesScore,\n verdict_reason: clusterVerdict.reason,\n corroborating_sources: corroborating || 'none - single-source story',\n articles_in_cluster: story.clusterSize,\n run_articles_in: articlesIn,\n run_survived_rules: survivedRules,\n run_unique_stories: storyPrimaries.length,\n run_llm_calls: judgedPrimaries.length,\n }});\n}\nreturn packageRows;" }, "typeVersion": 2 }, { "id": "aabb209a-6e59-4017-a42b-8cb996e50580", "name": "Append Approved to Sheets", "type": "n8n-nodes-base.googleSheets", "position": [ 4720, 16 ], "parameters": { "columns": { "value": { "run_at": "={{ $json.run_at }}", "run_id": "={{ $json.run_id }}", "best_url": "={{ $json.best_url }}", "best_title": "={{ $json.best_title }}", "cluster_id": "={{ $json.cluster_id }}", "best_source": "={{ $json.best_source }}", "rules_score": "={{ $json.rules_score }}", "run_llm_calls": "={{ $json.run_llm_calls }}", "run_signature": "={{ $json.run_signature }}", "verdict_reason": "={{ $json.verdict_reason }}", "run_articles_in": "={{ $json.run_articles_in }}", "run_survived_rules": "={{ $json.run_survived_rules }}", "run_unique_stories": "={{ $json.run_unique_stories }}", "articles_in_cluster": "={{ $json.articles_in_cluster }}", "corroborating_sources": "={{ $json.corroborating_sources }}" }, "schema": [ { "id": "run_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_id", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_at", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_at", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_signature", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_signature", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "cluster_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "cluster_id", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "best_title", "type": "string", "display": true, "removed": false, "required": false, "displayName": "best_title", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "best_source", "type": "string", "display": true, "removed": false, "required": false, "displayName": "best_source", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "best_url", "type": "string", "display": true, "removed": false, "required": false, "displayName": "best_url", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "rules_score", "type": "string", "display": true, "removed": false, "required": false, "displayName": "rules_score", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "verdict_reason", "type": "string", "display": true, "removed": false, "required": false, "displayName": "verdict_reason", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "corroborating_sources", "type": "string", "display": true, "removed": false, "required": false, "displayName": "corroborating_sources", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "articles_in_cluster", "type": "string", "display": true, "removed": false, "required": false, "displayName": "articles_in_cluster", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_articles_in", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_articles_in", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_survived_rules", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_survived_rules", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_unique_stories", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_unique_stories", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_llm_calls", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_llm_calls", "defaultMatch": false, "canBeUsedToMatch": false } ], "mappingMode": "defineBelow", "matchingColumns": [] }, "options": { "useAppend": true }, "operation": "append", "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.approvedTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" } }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "41eb7e42-b8b5-4c3e-bf41-4b482d3cc429", "name": "Check for Live Articles", "type": "n8n-nodes-base.if", "position": [ 256, 352 ], "parameters": { "options": {}, "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "lg1", "operator": { "type": "boolean", "operation": "equals" }, "leftValue": "={{ $('Set Workflow Context').first().json.useLiveFetch }}", "rightValue": true } ] } }, "typeVersion": 2.2 }, { "id": "86da3f1f-77e8-4e52-95e5-e1784f2a0219", "name": "Fetch RSS Feeds", "type": "n8n-nodes-base.rssFeedRead", "position": [ 704, 352 ], "parameters": { "url": "={{ $json.rssFeedUrls }}", "options": {} }, "typeVersion": 1.1 }, { "id": "51385fe5-bd16-4d92-8dcd-074c9197636f", "name": "Combine Sheet and Live Articles", "type": "n8n-nodes-base.merge", "position": [ 1152, -160 ], "parameters": {}, "typeVersion": 3.2 }, { "id": "22b0269c-1dfc-4ff7-b023-ae072e225a19", "name": "Set Live Article Fields", "type": "n8n-nodes-base.set", "position": [ 928, 352 ], "parameters": { "options": {}, "assignments": { "assignments": [ { "id": "r1", "name": "id", "type": "string", "value": "={{ 'live-' + ($itemIndex + 1) }}" }, { "id": "r2", "name": "title", "type": "string", "value": "={{ $json.title }}" }, { "id": "r3", "name": "summary", "type": "string", "value": "={{ $json.contentSnippet }}" }, { "id": "r4", "name": "source", "type": "string", "value": "={{ ($json.link || '').replace(/^https?:\\/\\/(www\\.)?/, '').split('/')[0] }}" }, { "id": "r5", "name": "url", "type": "string", "value": "={{ $json.link }}" }, { "id": "r6", "name": "published_at", "type": "string", "value": "={{ $json.isoDate }}" }, { "id": "r7", "name": "content_excerpt", "type": "string", "value": "={{ $json.contentSnippet }}" } ] } }, "typeVersion": 3.4 }, { "id": "51393caa-04df-4be7-9aa5-adbabfdd33a3", "name": "Merge LLM Judgments", "type": "n8n-nodes-base.merge", "position": [ 4272, 80 ], "parameters": {}, "typeVersion": 3.2 }, { "id": "6469bd65-bfcd-4102-8fc7-038e2736749e", "name": "Check for Fresh Start", "type": "n8n-nodes-base.if", "position": [ 32, 16 ], "parameters": { "options": {}, "conditions": { "options": { "version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict" }, "combinator": "and", "conditions": [ { "id": "cg1", "operator": { "type": "boolean", "operation": "equals" }, "leftValue": "={{ $('Set Workflow Context').first().json.clearBeforeRun }}", "rightValue": true } ] } }, "typeVersion": 2.2 }, { "id": "87ed75f1-16cc-42a3-b852-01d2305c4725", "name": "Resume After Clear", "type": "n8n-nodes-base.merge", "position": [ 704, 16 ], "parameters": {}, "typeVersion": 3.2 }, { "id": "e8fbc2b7-759c-45b1-bf31-6cec2d5b7c4f", "name": "Clear Seen Ledger in Sheets", "type": "n8n-nodes-base.googleSheets", "position": [ 480, -112 ], "parameters": { "operation": "clear", "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.articlesTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" }, "keepFirstRow": true }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "5dfd931e-e3a3-47b9-b766-ef98c70f2bde", "name": "Read Seen Ledger from Sheets", "type": "n8n-nodes-base.googleSheets", "position": [ 1152, 32 ], "parameters": { "options": {}, "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.articlesTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" } }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7, "alwaysOutputData": true }, { "id": "45a5a0dd-fc07-428a-926f-0c71dc8006c6", "name": "Remove Duplicate URLs", "type": "n8n-nodes-base.merge", "position": [ 1376, 80 ], "parameters": { "mode": "combine", "options": {}, "joinMode": "keepNonMatches", "outputDataFrom": "input1", "fieldsToMatchString": "url" }, "typeVersion": 3.2 }, { "id": "db28763a-a420-44a7-ba36-8943772936f8", "name": "Log to Seen Ledger", "type": "n8n-nodes-base.googleSheets", "position": [ 1824, 80 ], "parameters": { "columns": { "value": { "id": "={{ $json.id }}", "url": "={{ $json.url }}", "title": "={{ $json.title }}", "reason": "={{ $json.reason }}", "run_id": "={{ $json.run_id }}", "source": "={{ $json.source }}", "status": "={{ $json.status }}", "summary": "={{ $json.summary }}", "cluster_id": "={{ $json.cluster_id }}", "rules_score": "={{ $json.rules_score }}", "published_at": "={{ $json.published_at }}", "content_excerpt": "={{ $json.content_excerpt }}" }, "schema": [ { "id": "id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "id", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "title", "type": "string", "display": true, "removed": false, "required": false, "displayName": "title", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "summary", "type": "string", "display": true, "removed": false, "required": false, "displayName": "summary", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "source", "type": "string", "display": true, "removed": false, "required": false, "displayName": "source", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "url", "type": "string", "display": true, "removed": false, "required": false, "displayName": "url", "defaultMatch": false, "canBeUsedToMatch": true }, { "id": "published_at", "type": "string", "display": true, "removed": false, "required": false, "displayName": "published_at", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "content_excerpt", "type": "string", "display": true, "removed": false, "required": false, "displayName": "content_excerpt", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "status", "type": "string", "display": true, "removed": false, "required": false, "displayName": "status", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "rules_score", "type": "string", "display": true, "removed": false, "required": false, "displayName": "rules_score", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "reason", "type": "string", "display": true, "removed": false, "required": false, "displayName": "reason", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "cluster_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "cluster_id", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_id", "defaultMatch": false, "canBeUsedToMatch": false } ], "mappingMode": "defineBelow", "matchingColumns": [] }, "options": { "useAppend": true }, "operation": "append", "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.articlesTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" } }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "4ed612a6-c350-4134-aa26-df58f572b322", "name": "Set Pending Article Status", "type": "n8n-nodes-base.set", "position": [ 1600, 80 ], "parameters": { "options": {}, "assignments": { "assignments": [ { "id": "p1", "name": "status", "type": "string", "value": "pending" }, { "id": "p2", "name": "rules_score", "type": "string", "value": "" }, { "id": "p3", "name": "reason", "type": "string", "value": "" }, { "id": "p4", "name": "cluster_id", "type": "string", "value": "" }, { "id": "p5", "name": "run_id", "type": "string", "value": "={{ $execution.id }}" } ] }, "includeOtherFields": true }, "typeVersion": 3.4 }, { "id": "c6d5d892-e635-4118-a916-b7ad38cbfa17", "name": "Compile Article Outcomes", "type": "n8n-nodes-base.code", "position": [ 4720, 368 ], "parameters": { "jsCode": "// Writes every article's fate back to the seen ledger, matched by URL.\n// Recomputes the same verdict mapping the assembler uses - both nodes read\n// identical sources, so they cannot drift.\nconst scoredArticles = $('Apply Scoring Rules').all().map(item => item.json);\nconst storyPrimaries = $('Cluster Similar Stories').all().map(item => item.json);\nconst judgedPrimaries = storyPrimaries.filter(story => story.rulesBand !== 'hardWin');\nlet judgeOutputs = [];\ntry {\n judgeOutputs = $('Relevance Judgment').all().map(item => item.json);\n} catch (noJudgeRun) {\n judgeOutputs = [];\n}\n\nconst config = $('Set Workflow Context').first().json;\nconst runId = String($execution.id || 'manual');\n\nconst verdictByCluster = {};\nfor (const story of storyPrimaries) {\n if (story.rulesBand === 'hardWin') {\n verdictByCluster[story.clusterId] = { verdict: 'approved', reason: 'Rules hard win - skipped the LLM entirely' };\n }\n}\njudgedPrimaries.forEach((story, judgeIndex) => {\n const judgeResult = judgeOutputs[judgeIndex] ? (judgeOutputs[judgeIndex].output || judgeOutputs[judgeIndex]) : null;\n verdictByCluster[story.clusterId] = {\n verdict: judgeResult && judgeResult.verdict ? judgeResult.verdict : 'rejected',\n reason: judgeResult && judgeResult.reason ? judgeResult.reason : 'No judge output received',\n };\n});\n\nconst outcomeByUrl = {};\nfor (const article of scoredArticles) {\n if (article.rulesBand === 'reject') {\n outcomeByUrl[article.url] = {\n status: 'rejected_rules',\n rules_score: article.rulesScore,\n reason: `Rules score ${article.rulesScore} below reject threshold ${config.rulesRejectThreshold}`,\n cluster_id: '',\n };\n }\n}\nfor (const story of storyPrimaries) {\n const clusterVerdict = verdictByCluster[story.clusterId] || { verdict: 'rejected', reason: 'Missing verdict' };\n outcomeByUrl[story.url] = {\n status: clusterVerdict.verdict === 'approved' ? 'approved' : 'rejected_judge',\n rules_score: story.rulesScore,\n reason: clusterVerdict.reason,\n cluster_id: story.clusterId,\n };\n for (const member of story.corroborating || []) {\n const memberScored = scoredArticles.find(candidate => candidate.url === member.url);\n outcomeByUrl[member.url] = {\n status: 'corroborating',\n rules_score: memberScored ? memberScored.rulesScore : '',\n reason: `Duplicate coverage of: ${story.title}`,\n cluster_id: story.clusterId,\n };\n }\n}\n\nreturn Object.entries(outcomeByUrl).map(([url, outcome]) => ({ json: { url, ...outcome, run_id: runId } }));" }, "typeVersion": 2 }, { "id": "c3ceb932-f4cf-4461-a20a-5a4c548c8bc4", "name": "Update Ledger Outcomes", "type": "n8n-nodes-base.googleSheets", "position": [ 4944, 368 ], "parameters": { "columns": { "value": { "url": "={{ $json.url }}", "reason": "={{ $json.reason }}", "run_id": "={{ $json.run_id }}", "status": "={{ $json.status }}", "cluster_id": "={{ $json.cluster_id }}", "rules_score": "={{ $json.rules_score }}" }, "schema": [ { "id": "url", "type": "string", "display": true, "removed": false, "required": false, "displayName": "url", "defaultMatch": false, "canBeUsedToMatch": true }, { "id": "status", "type": "string", "display": true, "removed": false, "required": false, "displayName": "status", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "rules_score", "type": "string", "display": true, "removed": false, "required": false, "displayName": "rules_score", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "reason", "type": "string", "display": true, "removed": false, "required": false, "displayName": "reason", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "cluster_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "cluster_id", "defaultMatch": false, "canBeUsedToMatch": false }, { "id": "run_id", "type": "string", "display": true, "removed": false, "required": false, "displayName": "run_id", "defaultMatch": false, "canBeUsedToMatch": false } ], "mappingMode": "defineBelow", "matchingColumns": [ "url" ] }, "options": {}, "operation": "appendOrUpdate", "sheetName": { "__rl": true, "mode": "name", "value": "={{ $('Set Workflow Context').first().json.articlesTab }}" }, "documentId": { "__rl": true, "mode": "id", "value": "={{ $('Set Workflow Context').first().json.sheetDocumentId }}" } }, "credentials": { "googleSheetsOAuth2Api": { "name": "" } }, "typeVersion": 4.7 }, { "id": "70e487fc-3392-4532-a3b5-e65c3c18d6b7", "name": "Split RSS Feed URLs", "type": "n8n-nodes-base.splitOut", "position": [ 480, 352 ], "parameters": { "options": {}, "fieldToSplitOut": "rssFeedUrls" }, "typeVersion": 1 }, { "id": "69828c23-2118-48f6-ab6a-6116d465e729", "name": "Scheduled Midnight Trigger", "type": "n8n-nodes-base.scheduleTrigger", "position": [ -416, 208 ], "parameters": { "rule": { "interval": [ {} ] } }, "typeVersion": 1.3 }, { "id": "dfd0152a-2eb0-42a3-9c2a-970f011fcb30", "name": "Sticky howitworks", "type": "n8n-nodes-base.stickyNote", "position": [ -1632, -784 ], "parameters": { "width": 620, "height": 1152, "content": "# How It Works \ud83e\udde0\nTurns a noisy news feed into a deduplicated, approved content package - spending real money only on stories that earn it.\n\nThree escalating-cost steps:\n1. **Rules tier (free)** - keyword + source scoring kills the obvious noise; articles above the hard-win threshold skip the LLM entirely\n2. **Semantic dedup (pennies)** - embeddings + cosine similarity cluster outlets covering the same story; one primary per cluster proceeds\n3. **LLM judge (paid, and only here)** - the model rules only on unique, ambiguous stories; every clustered duplicate inherits its primary's verdict for free. **The LLM never sees the same story twice.**\n\nOne row per approved story lands in the Approved Package tab: best article, every other outlet's angle as corroborating sources, and the run counts. Every article's fate is written back to the Articles ledger with the reason - and already-seen URLs are dropped before a penny is spent, so repeat runs are free.\n\n## Setup (~5 minutes)\n1. **Credentials** - create Google Sheets (OAuth), Hugging Face (free token) and OpenAI credentials, then attach them to the Sheets, embedding and judge nodes\n2. **Sheet** - copy the starter sheet linked in the template description (File \u2192 Make a copy), or run the starter seed workflow from the GitHub repo\n3. **Configure context** - paste your spreadsheet ID into `sheetDocumentId` and put your feeds in `rssFeedUrls`\n4. **Run** - live headlines flow through the three steps into the Approved Package tab\n\n**Adapt it to your niche** by editing three config values: `nicheContext` (what the judge considers relevant), `rulesKeywords`, and `sourceWeights`.\n\n## Starter Google Sheet: https://docs.google.com/spreadsheets/d/11UAshmcWzrzNZRBqeDSL78r7b6GWxou-UJdxhCXRlUY/edit?usp=sharing\n\n## Data sources\nTwo ways to feed the pipeline:\n- **Live RSS (default)** - every URL in `rssFeedUrls` is fetched each run; an article's `source` is its domain, so key `sourceWeights` by domain (e.g. `techcrunch.com`)\n- **Source Articles tab** - populate it from another workflow (or any tool that writes rows in its seven-column shape: id, title, summary, source, url, published_at, content_excerpt) and the pipeline categorises whatever lands there\n\nThe Articles ledger drops anything already seen, so repeat runs cost nothing." }, "typeVersion": 1 } ], "active": true, "settings": { "binaryMode": "separate", "callerPolicy": "workflowsFromSameOwner", "timeSavedMode": "fixed", "availableInMCP": true, "executionOrder": "v1" }, "versionId": "bbbc4b56-c411-4d05-a4fc-ae925a52e1cc", "nodeGroups": [], "connections": { "Fetch RSS Feeds": { "main": [ [ { "node": "Set Live Article Fields", "type": "main", "index": 0 } ] ] }, "Log to Seen Ledger": { "main": [ [ { "node": "Apply Scoring Rules", "type": "main", "index": 0 } ] ] }, "Relevance Judgment": { "main": [ [ { "node": "Merge LLM Judgments", "type": "main", "index": 0 } ] ] }, "Resume After Clear": { "main": [ [ { "node": "Read Articles from Sheets", "type": "main", "index": 0 }, { "node": "Read Seen Ledger from Sheets", "type": "main", "index": 0 } ] ] }, "Set OpenAI Vectors": { "main": [ [ { "node": "Cluster Similar Stories", "type": "main", "index": 0 } ] ] }, "Apply Scoring Rules": { "main": [ [ { "node": "Filter Rejected Articles", "type": "main", "index": 0 } ] ] }, "Merge LLM Judgments": { "main": [ [ { "node": "Assemble Approved Articles", "type": "main", "index": 0 } ] ] }, "Split RSS Feed URLs": { "main": [ [ { "node": "Fetch RSS Feeds", "type": "main", "index": 0 } ] ] }, "Set Workflow Context": { "main": [ [ { "node": "Check for Fresh Start", "type": "main", "index": 0 }, { "node": "Check for Live Articles", "type": "main", "index": 0 } ] ] }, "Check for Fresh Start": { "main": [ [ { "node": "Clear Package Data in Sheets", "type": "main", "index": 0 } ], [ { "node": "Resume After Clear", "type": "main", "index": 1 } ] ] }, "Remove Duplicate URLs": { "main": [ [ { "node": "Set Pending Article Status", "type": "main", "index": 0 } ] ] }, "Check for LLM Judgment": { "main": [ [ { "node": "Relevance Judgment", "type": "main", "index": 0 } ], [ { "node": "Merge LLM Judgments", "type": "main", "index": 1 } ] ] }, "Parse Judgment Verdict": { "ai_outputParser": [ [ { "node": "Relevance Judgment", "type": "ai_outputParser", "index": 0 } ] ] }, "Set OpenAI Judge Model": { "ai_languageModel": [ [ { "node": "Relevance Judgment", "type": "ai_languageModel", "index": 0 } ] ] }, "Aggregate for Embedding": { "main": [ [ { "node": "Select Embedding Provider", "type": "main", "index": 0 } ] ] }, "Check for Live Articles": { "main": [ [ { "node": "Split RSS Feed URLs", "type": "main", "index": 0 } ] ] }, "Cluster Similar Stories": { "main": [ [ { "node": "Check for LLM Judgment", "type": "main", "index": 0 } ] ] }, "Manual Pipeline Trigger": { "main": [ [ { "node": "Set Workflow Context", "type": "main", "index": 0 } ] ] }, "Set Live Article Fields": { "main": [ [ { "node": "Combine Sheet and Live Articles", "type": "main", "index": 1 } ] ] }, "Compile Article Outcomes": { "main": [ [ { "node": "Update Ledger Outcomes", "type": "main", "index": 0 } ] ] }, "Filter Rejected Articles": { "main": [ [ { "node": "Aggregate for Embedding", "type": "main", "index": 0 } ] ] }, "Post to Hugging Face API": { "main": [ [ { "node": "Set Hugging Face Vectors", "type": "main", "index": 0 } ] ] }, "Set Hugging Face Vectors": { "main": [ [ { "node": "Cluster Similar Stories", "type": "main", "index": 0 } ] ] }, "Post to OpenAI Embeddings": { "main": [ [ { "node": "Set OpenAI Vectors", "type": "main", "index": 0 } ] ] }, "Read Articles from Sheets": { "main": [ [ { "node": "Combine Sheet and Live Articles", "type": "main", "index": 0 } ] ] }, "Select Embedding Provider": { "main": [ [ { "node": "Post to Hugging Face API", "type": "main", "index": 0 } ], [ { "node": "Post to OpenAI Embeddings", "type": "main", "index": 0 } ] ] }, "Assemble Approved Articles": { "main": [ [ { "node": "Append Approved to Sheets", "type": "main", "index": 0 }, { "node": "Compile Article Outcomes", "type": "main", "index": 0 } ] ] }, "Scheduled Midnight Trigger": { "main": [ [ { "node": "Set Workflow Context", "type": "main", "index": 0 } ] ] }, "Set Pending Article Status": { "main": [ [ { "node": "Log to Seen Ledger", "type": "main", "index": 0 } ] ] }, "Clear Seen Ledger in Sheets": { "main": [ [ { "node": "Resume After Clear", "type": "main", "index": 0 } ] ] }, "Clear Package Data in Sheets": { "main": [ [ { "node": "Clear Seen Ledger in Sheets", "type": "main", "index": 0 } ] ] }, "Read Seen Ledger from Sheets": { "main": [ [ { "node": "Remove Duplicate URLs", "type": "main", "index": 1 } ] ] }, "Combine Sheet and Live Articles": { "main": [ [ { "node": "Remove Duplicate URLs", "type": "main", "index": 0 } ] ] } } }