[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_210":211,"_212":213},"routes/_app.x.$handle.$name",{"_99":98,"_31":44,"_108":89,"_26":109,"_13":14,"_110":-5,"_11":111,"_112":113,"_114":115,"_116":117,"_118":119,"_20":-5,"_120":-5,"_121":-5,"_122":24,"_123":124,"_125":24,"_126":-5,"_127":-5,"_128":129},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_42":43,"_31":44,"_37":45,"_46":-5,"_47":24,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":70,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":82,"_83":84,"_85":86,"_87":24},"version",6,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":-7,"_23":24,"_25":24,"_26":27,"_28":27},"file",{"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":14,"_40":41,"_23":24},"previous","deleted",false,"deletedHere","path","lib/generate.ts","latestPath","fileId","b4be5fe6-7ff6-11f1-833e-1607ee4eb77e","name","generate.ts","fileType","script","content","// Stage 4 — Write the obituary via the Claude API.\n// The ONLY stage that costs money (~$0.05–0.10/issue on claude-opus-4-8).\n//\n// Fabrication guards (the failure that killed the old vals, 2026-07-12):\n// 1. The prompt contains ONLY evidence this pipeline fetched, and forbids\n// facts from anywhere else.\n// 2. Escape hatch: the model must set insufficient_evidence=true rather\n// than pad thin material — the orchestrator then moves to a fallback\n// candidate instead of publishing filler.\n// 3. Structured output (json_schema) — a malformed or un-rendered result\n// is impossible at the schema level.\n// 4. Post-checks: domain appears in copy; sources only cite hosts the\n// evidence actually came from; no pipeline vocabulary leaks into prose.\n//\n// Model note: `thinking` is deliberately omitted (= off on Opus 4.8) to keep\n// the call inside Val Town's per-run wall clock and the per-issue cost low.\n// If generation ever feels flat, set RIPURL_MODEL=claude-sonnet-5 (faster)\n// or raise maxTokens — both are env/config-level changes, not code changes.\n\nimport Anthropic from \"npm:@anthropic-ai/sdk\";\nimport type { Evidence, Obituary } from \"./types.ts\";\n\nconst OBITUARY_SCHEMA = {\n type: \"object\",\n additionalProperties: false,\n required: [\n \"insufficient_evidence\",\n \"domain\",\n \"subject_line\",\n \"prime_years\",\n \"category\",\n \"intro\",\n \"the_rise\",\n \"the_fall\",\n \"manner_of_death\",\n \"disposition\",\n \"legacy\",\n \"lineage\",\n \"sources\",\n \"screenshot_note\",\n \"x_post\",\n \"tumblr_post\",\n ],\n properties: {\n insufficient_evidence: {\n type: \"boolean\",\n description:\n \"Set true ONLY if the evidence cannot support an honest obituary; leave every other field as an empty string/array in that case.\",\n },\n domain: { type: \"string\" },\n subject_line: {\n type: \"string\",\n description: 'Format: \"In Loving Memory of : \"',\n },\n prime_years: { type: \"string\", description: 'e.g. \"2007–2009\"' },\n category: { type: \"string\", description: 'e.g. \"Social Music Streaming Network\"' },\n intro: {\n type: \"string\",\n description:\n \"Opening paragraph. May use / sparingly. Sets the scene before the site existed.\",\n },\n the_rise: { type: \"string\", description: \"Narrative of the ascent — scenes, arc, specifics. HTML em/strong allowed.\" },\n the_fall: {\n type: \"string\",\n description:\n \"Narrative of the death. Must diagnose the real cause, not just narrate events in order.\",\n },\n manner_of_death: { type: \"string\", description: 'Short, e.g. \"Acquired & shut down (X, 2009)\"' },\n disposition: {\n type: \"string\",\n description: \"What the domain/corpse is doing now, from the registration evidence.\",\n },\n legacy: {\n type: \"string\",\n description:\n \"The editorial standfirst: why this death matters beyond one company. Original copy, no quote marks.\",\n },\n lineage: {\n type: \"array\",\n items: {\n type: \"object\",\n additionalProperties: false,\n required: [\"year\", \"label\"],\n properties: { year: { type: \"string\" }, label: { type: \"string\" } },\n },\n description: \"3–6 dated beats: founded → pivot(s) → peak → death. Dry chronology lives here, not in prose.\",\n },\n sources: {\n type: \"array\",\n items: {\n type: \"object\",\n additionalProperties: false,\n required: [\"cite\", \"url\"],\n properties: { cite: { type: \"string\" }, url: { type: \"string\" } },\n },\n description:\n \"Academic-style citations. ONLY the Wikipedia article and the archived capture supplied in the evidence.\",\n },\n screenshot_note: {\n type: \"string\",\n description: \"One clause describing what the reader is seeing in the archived capture, given its year.\",\n },\n x_post: { type: \"string\", description: \"Cross-post teaser for X, under 280 characters, link goes at the end as {URL}.\" },\n tumblr_post: {\n type: \"string\",\n description: \"Cross-post for Tumblr: 2-3 sentences + a line of #tags. Link placeholder {URL}.\",\n },\n },\n} as const;\n\nconst VOICE = `You are the obituary writer for RIP · URL, a weekly publication by Unlikely Archive that\nperforms autopsies on dead domain names. Voice: darkly funny, elegiac, archival — an obituary\nwriter with taste, not a content marketer.\n\nHard rules:\n- Story, not fact-dump. Push dry chronology into the lineage timeline; use the_rise/the_fall for\n narrative — scenes, arc, specifics.\n- Analyze the why and why-it-matters. the_fall and legacy must diagnose the real cause of death\n and say what the story means beyond one company.\n- No pipeline language. Never describe how anything was found, matched, scored, or verified.\n Never mention Wikidata, RDAP, CDX, SPARQL, databases, or automation. Citing Wikipedia or the\n Internet Archive by name in a source citation is fine.\n- EVIDENCE ONLY. Every factual claim must come from the evidence block in the user message.\n If the evidence doesn't say it, you don't write it. No dates, numbers, names, or events from\n memory. If the evidence is too thin for an honest obituary, set insufficient_evidence=true.\n- Write like the exemplar below (Issue No. 1, imeem.com — study its register):\n intro: \"Before a single green Spotify icon existed, there was a website where you could type in\n almost any song and hear it, in full, for free... It was called imeem, and for a few bright\n years it was the sound of the social web.\"\n legacy: \"imeem was right about everything except how to pay for it... imeem is the ghost that\n walked so Spotify could run.\"`;\n\nclass InsufficientEvidenceError extends Error {}\nexport { InsufficientEvidenceError };\n\nconst BANNED_IN_PROSE = /\\b(Wikidata|RDAP|CDX|SPARQL|pipeline|automated|auto-generated)\\b/i;\n\nexport async function writeObituary(evidence: Evidence): Promise<{ obituary: Obituary; model: string }> {\n const apiKey = Deno.env.get(\"ANTHROPIC_API_KEY\");\n if (!apiKey) {\n throw new Error(\n \"ANTHROPIC_API_KEY is not set. Add it in the Val Town environment-variables UI — the pipeline cannot write copy without it.\",\n );\n }\n const model = Deno.env.get(\"RIPURL_MODEL\") ?? \"claude-opus-4-8\";\n const client = new Anthropic({ apiKey });\n\n const c = evidence.candidate;\n const evidenceBlock = {\n company: c.company,\n domain: c.domain,\n founded: evidence.founded ?? \"not recorded\",\n dissolved: c.dissolvedDate,\n wikipedia_description: evidence.wikipedia.description,\n wikipedia_lead_text: evidence.wikipedia.extract,\n wikipedia_article_url: evidence.wikipedia.url,\n domain_registration: evidence.rdap.reachable\n ? {\n registered: evidence.rdap.registered,\n paid_up_until: evidence.rdap.expires,\n registrar: evidence.rdap.registrar ?? \"unknown\",\n status_flags: evidence.rdap.status,\n }\n : \"domain no longer registered (RDAP lookup found nothing)\",\n archive_record: {\n first_capture: evidence.wayback.firstCapture,\n last_capture: evidence.wayback.lastCapture,\n months_with_captures: evidence.wayback.monthsArchived,\n screenshot_capture_timestamp: evidence.wayback.screenshotTimestamp,\n screenshot_archived_page: evidence.wayback.screenshotSourceUrl,\n },\n };\n\n const response = await client.messages.create({\n model,\n max_tokens: 8000,\n system: VOICE,\n output_config: { format: { type: \"json_schema\", schema: OBITUARY_SCHEMA } },\n messages: [\n {\n role: \"user\",\n content:\n `Write this week's obituary. Evidence file (the ONLY permitted source of facts):\\n\\n` +\n JSON.stringify(evidenceBlock, null, 2),\n },\n ],\n });\n\n if (response.stop_reason === \"refusal\") {\n throw new InsufficientEvidenceError(`model refused for ${c.domain}`);\n }\n const text = response.content.find((b: any) => b.type === \"text\")?.text;\n if (!text) throw new Error(\"no text block in model response\");\n const ob = JSON.parse(text) as Obituary;\n\n if (ob.insufficient_evidence) {\n throw new InsufficientEvidenceError(`model judged evidence too thin for ${c.domain}`);\n }\n\n // Post-checks — fail loudly rather than publish a defective issue.\n const prose = [ob.intro, ob.the_rise, ob.the_fall, ob.legacy].join(\" \");\n if (!prose.toLowerCase().includes(c.domain.split(\".\")[0].toLowerCase())) {\n throw new Error(`obituary never mentions ${c.domain}`);\n }\n if (BANNED_IN_PROSE.test(prose)) {\n throw new Error(\"pipeline language leaked into prose\");\n }\n for (const s of ob.sources) {\n const host = new URL(s.url).hostname;\n if (![\"en.wikipedia.org\", \"web.archive.org\", \"www.wikipedia.org\"].includes(host)) {\n throw new Error(`source cites unapproved host: ${host}`);\n }\n }\n const required = [ob.subject_line, ob.intro, ob.the_rise, ob.the_fall, ob.legacy, ob.manner_of_death, ob.disposition];\n if (required.some((f) => !f || f.trim().length < 20)) {\n throw new Error(\"a required obituary field is missing or too short\");\n }\n ob.domain = c.domain; // never trust the echo\n return { obituary: ob, model };\n}\n","createdAt",["D",1784083299603],"branchVersion","branchId","5e98dbb4-3939-4a4a-ba98-35721eff0730","id","1a4814cf-ad06-40ea-a498-5913ef00d415","ripurl",["D",1784083111215],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","RIP·URL — automated obituaries for dead domains","imageUrl","051b2cc5-9320-44ef-88e7-826bfa8daa60","pinnedFileId","ffe21efe-7ff6-11f1-8d93-1607ee4eb77e","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","019ee5b5-1072-70b4-8a0f-c844ad002ea1","tursoName","1a4814cfad0640eaa4985913ef00d415","tursoHostname","1a4814cfad0640eaa4985913ef00d415-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789438866375],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_99":98,"_107":103},"org",{"_99":98,"_100":101,"_102":103,"_18":54,"_104":105,"_106":24},"branch",{"_13":88,"_31":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"isOwned",21,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","unlikelyarchivist","handle","displayName","Eryn Montgomery","avatar","https://lh3.googleusercontent.com/a/ACg8ocJEkHIy674nn2OrWifP7cPPXWj2A5ftzagptlIz_AXo3OPt_A=s96-c","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_42":43,"_31":44,"_37":205,"_46":-5,"_47":24,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":206,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":207,"_83":208,"_85":209,"_87":24},"tree",[152],"files",[152,153,154,155,156,157,158],"counts",{"_141":62,"_142":135,"_143":62,"_144":145,"_146":147},"contents",[130],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_42":131,"_31":132,"_29":133,"_134":-5,"_40":41,"_39":135,"_136":58,"_37":137,"_13":62,"_33":138,"_35":-5,"_23":24,"_139":-5,"_26":140},"02028c83-19b6-4a86-ac51-e2f29188dbc2","lib","226233b4-9852-41f8-9f1e-a633ceb6013a","parentId",1,"latest",["D",1784083130616],"directory","label",[132],"pulls","branches","forks","envVars",{"_148":150,"_151":62},"connections",{"_148":62,"_149":62},"count","unfilledCount",4,"emptyCount",{"_42":131,"_31":132,"_29":133,"_134":-5,"_40":41,"_39":135,"_136":58,"_37":202,"_13":62,"_33":138,"_35":-5,"_23":24,"_139":-5,"_164":203,"_26":204},{"_42":194,"_31":195,"_29":196,"_134":133,"_40":41,"_39":197,"_136":58,"_37":198,"_13":62,"_33":34,"_35":199,"_23":24,"_139":-5,"_164":200,"_26":201},{"_42":187,"_31":188,"_29":189,"_134":133,"_40":41,"_39":150,"_136":58,"_37":190,"_13":62,"_33":34,"_35":191,"_23":24,"_139":-5,"_164":192,"_26":193},{"_42":183,"_31":32,"_29":30,"_134":133,"_40":41,"_39":14,"_136":58,"_37":184,"_13":62,"_33":34,"_35":36,"_23":24,"_139":-5,"_164":185,"_26":186},{"_42":175,"_31":176,"_29":177,"_134":133,"_40":41,"_39":178,"_136":58,"_37":179,"_13":62,"_33":34,"_35":180,"_23":24,"_139":-5,"_164":181,"_26":182},{"_42":167,"_31":168,"_29":169,"_134":133,"_40":41,"_39":170,"_136":58,"_37":171,"_13":62,"_33":34,"_35":172,"_23":24,"_139":-5,"_164":173,"_26":174},{"_42":159,"_31":160,"_29":161,"_134":133,"_40":41,"_39":135,"_136":58,"_37":162,"_13":62,"_33":34,"_35":163,"_23":24,"_139":-5,"_164":165,"_26":166},"71aac12e-2665-436c-b94c-805a44ce9f09","types.ts","5005076c-7ff6-11f1-b33c-1607ee4eb77e",["D",1784083130616],"// Shared shapes passed between pipeline stages.\n// Every stage reads/writes these — nothing hides in loose objects.\n\nexport interface Candidate {\n company: string;\n qid: string; // Wikidata entity id, e.g. Q65073105\n domain: string; // derived from the entity's own P856 official-website claim\n website: string;\n dissolvedDate: string; // YYYY-MM-DD\n dissolvedYear: number;\n wikipediaTitle: string;\n wikipediaUrl: string;\n // filled during enrichment:\n extract?: string; // Wikipedia lead text (plain)\n description?: string; // Wikipedia short description\n monthsArchived?: number; // distinct months with Wayback captures\n firstCapture?: string; // YYYYMMDDhhmmss\n lastCapture?: string;\n score?: number;\n}\n\nexport interface Evidence {\n candidate: Candidate;\n wikipedia: {\n title: string;\n url: string;\n description: string;\n extract: string; // full lead section, plain text\n };\n rdap: {\n reachable: boolean;\n registered?: string; // ISO date\n expires?: string;\n status?: string[];\n registrar?: string;\n };\n wayback: {\n firstCapture: string;\n lastCapture: string;\n monthsArchived: number;\n screenshotTimestamp: string; // capture chosen for the hero image\n screenshotSourceUrl: string; // the archived page the render points at\n };\n founded?: string; // from Wikidata P571, if present\n gatheredAt: string; // ISO timestamp — every issue carries its evidence trail\n}\n\nexport interface Obituary {\n insufficient_evidence: boolean;\n domain: string;\n subject_line: string;\n prime_years: string;\n category: string;\n intro: string;\n the_rise: string;\n the_fall: string;\n manner_of_death: string;\n disposition: string;\n legacy: string;\n lineage: { year: string; label: string }[];\n sources: { cite: string; url: string }[];\n screenshot_note: string;\n x_post: string; // <=280 chars, for manual cross-post\n tumblr_post: string; // short + tags, for manual cross-post\n}\n\n// The cron advances one Job through these stages, one tick at a time,\n// so no single run risks the platform wall-clock limit.\nexport type Stage =\n | \"discover\"\n | \"evidence\"\n | \"screenshot\"\n | \"generate\"\n | \"assemble_send\"\n | \"done\"\n | \"failed\";\n\nexport interface Job {\n id: number;\n stage: Stage;\n attempts: number; // per-stage attempt counter\n candidates: Candidate[]; // ranked; [0] is current, rest are fallbacks\n evidence?: Evidence;\n obituary?: Obituary;\n error?: string;\n createdAt: string;\n updatedAt: string;\n}\n","children",[],[132,160],"428c8295-a5fa-48d1-8bc3-0ee279632cf2","state.ts","6069690e-7ff6-11f1-a76e-1607ee4eb77e",2,["D",1784083158129],"// Persistence: sqlite (val-scoped) for jobs, issues, and the used-domain\n// exclusion list. Blob storage holds the big artifacts (screenshot JPEG,\n// final HTML) keyed by issue number / domain.\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport type { Job, Stage } from \"./types.ts\";\n\nexport async function initTables() {\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS used_domains (\n domain TEXT PRIMARY KEY,\n used_at TEXT NOT NULL\n )`);\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS issues (\n n INTEGER PRIMARY KEY,\n domain TEXT NOT NULL,\n subject TEXT NOT NULL,\n model TEXT NOT NULL,\n evidence_json TEXT NOT NULL,\n created_at TEXT NOT NULL\n )`);\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS jobs (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n stage TEXT NOT NULL,\n attempts INTEGER NOT NULL DEFAULT 0,\n data_json TEXT NOT NULL,\n error TEXT,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )`);\n}\n\nexport async function usedDomains(): Promise> {\n const r = await sqlite.execute(\"SELECT domain FROM used_domains\");\n return new Set(r.rows.map((row: any) => String(row.domain)));\n}\n\nexport async function markUsed(domain: string) {\n await sqlite.execute({\n sql: \"INSERT OR IGNORE INTO used_domains (domain, used_at) VALUES (?, ?)\",\n args: [domain.toLowerCase(), new Date().toISOString()],\n });\n}\n\nexport async function seedUsedDomains(domains: string[]) {\n for (const d of domains) await markUsed(d);\n}\n\n// --- jobs ---------------------------------------------------------------\n\nexport async function activeJob(): Promise {\n const r = await sqlite.execute(\n \"SELECT * FROM jobs WHERE stage NOT IN ('done','failed') ORDER BY id DESC LIMIT 1\",\n );\n if (r.rows.length === 0) return null;\n return rowToJob(r.rows[0]);\n}\n\nexport async function createJob(): Promise {\n const now = new Date().toISOString();\n const data = JSON.stringify({ candidates: [] });\n await sqlite.execute({\n sql: \"INSERT INTO jobs (stage, attempts, data_json, created_at, updated_at) VALUES ('discover', 0, ?, ?, ?)\",\n args: [data, now, now],\n });\n return (await activeJob())!;\n}\n\nexport async function saveJob(job: Job) {\n const data = JSON.stringify({\n candidates: job.candidates,\n evidence: job.evidence,\n obituary: job.obituary,\n });\n await sqlite.execute({\n sql: \"UPDATE jobs SET stage = ?, attempts = ?, data_json = ?, error = ?, updated_at = ? WHERE id = ?\",\n args: [job.stage, job.attempts, data, job.error ?? null, new Date().toISOString(), job.id],\n });\n}\n\nfunction rowToJob(row: any): Job {\n const data = JSON.parse(String(row.data_json));\n return {\n id: Number(row.id),\n stage: String(row.stage) as Stage,\n attempts: Number(row.attempts),\n candidates: data.candidates ?? [],\n evidence: data.evidence,\n obituary: data.obituary,\n error: row.error ? String(row.error) : undefined,\n createdAt: String(row.created_at),\n updatedAt: String(row.updated_at),\n };\n}\n\n// --- issues ---------------------------------------------------------------\n\nexport async function nextIssueNumber(): Promise {\n const r = await sqlite.execute(\"SELECT MAX(n) AS m FROM issues\");\n const m = r.rows[0]?.m;\n // Issue No. 1 (imeem.com) was produced by the local pipeline; cloud issues\n // continue the numbering, so the first automated issue is No. 2.\n return m == null ? 2 : Number(m) + 1;\n}\n\nexport async function recordIssue(\n n: number,\n domain: string,\n subject: string,\n model: string,\n evidenceJson: string,\n) {\n await sqlite.execute({\n sql: \"INSERT INTO issues (n, domain, subject, model, evidence_json, created_at) VALUES (?, ?, ?, ?, ?, ?)\",\n args: [n, domain, subject, model, evidenceJson, new Date().toISOString()],\n });\n}\n\nexport async function listIssues() {\n const r = await sqlite.execute(\n \"SELECT n, domain, subject, created_at FROM issues ORDER BY n DESC\",\n );\n return r.rows.map((row: any) => ({\n n: Number(row.n),\n domain: String(row.domain),\n subject: String(row.subject),\n createdAt: String(row.created_at),\n }));\n}\n\nexport async function getIssue(n: number) {\n const r = await sqlite.execute({ sql: \"SELECT * FROM issues WHERE n = ?\", args: [n] });\n if (r.rows.length === 0) return null;\n const row: any = r.rows[0];\n return {\n n: Number(row.n),\n domain: String(row.domain),\n subject: String(row.subject),\n model: String(row.model),\n evidenceJson: String(row.evidence_json),\n createdAt: String(row.created_at),\n };\n}\n\nexport async function daysSinceLastIssue(): Promise {\n const r = await sqlite.execute(\"SELECT MAX(created_at) AS m FROM issues\");\n const m = r.rows[0]?.m;\n if (m == null) return Infinity;\n return (Date.now() - new Date(String(m)).getTime()) / 86_400_000;\n}\n",[],[132,168],"948382bf-aa54-4b06-8bd2-5c20ece9c5c1","screenshot.ts","917293f4-7ff6-11f1-8754-1607ee4eb77e",5,["D",1784083240398],"// Stage 3 — Capture the hero screenshot without a browser.\n//\n// Verified 2026-07-14: WordPress mShots (s0.wp.com/mshots) renders Wayback\n// pages server-side and returns a real JPEG, but only if you (a) send a\n// browser User-Agent (403 otherwise), (b) use the `if_` flag in the Wayback\n// URL to strip the archive toolbar, and (c) poll — the first responses are a\n// small placeholder GIF until the render is cached (~10-40s).\n// (thum.io returns its loading spinner for Wayback pages; microlink captures\n// a blank page. Neither is usable — do not switch to them.)\n//\n// We fetch the JPEG bytes and store them in blob storage, so the published\n// issue never depends on a third-party image URL staying alive.\n\nimport { blob } from \"https://esm.town/v/std/blob/main.ts\";\n\nconst BROWSER_UA =\n \"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0 Safari/537.36\";\n\nexport function toolbarFreeWaybackUrl(timestamp: string, domain: string): string {\n return `http://web.archive.org/web/${timestamp}if_/http://${domain}/`;\n}\n\nexport function mshotsUrl(target: string, width = 1000): string {\n return `https://s0.wp.com/mshots/v1/${encodeURIComponent(target)}?w=${width}`;\n}\n\n// One polling attempt. Returns true once a real JPEG is stored.\n// The cron calls this each tick while the job sits in the \"screenshot\"\n// stage, so mShots gets rendering time between ticks for free.\nexport async function tryCaptureScreenshot(\n timestamp: string,\n domain: string,\n): Promise {\n const target = toolbarFreeWaybackUrl(timestamp, domain);\n const url = mshotsUrl(target);\n // Poll a few times within this tick; if not ready, the next tick retries.\n for (let i = 0; i < 3; i++) {\n const resp = await fetch(url, { headers: { \"User-Agent\": BROWSER_UA } });\n const type = resp.headers.get(\"content-type\") ?? \"\";\n const bytes = await resp.arrayBuffer();\n // Guard: a real render is a JPEG well above placeholder size. Small or\n // non-JPEG responses are the loading GIF / an error page — reject them.\n if (resp.ok && type.includes(\"image/jpeg\") && bytes.byteLength > 25_000) {\n await blob.set(`img:${domain}`, bytes);\n return true;\n }\n await new Promise((r) => setTimeout(r, 8000));\n }\n return false;\n}\n\nexport async function screenshotStored(domain: string): Promise {\n try {\n const r = await blob.get(`img:${domain}`);\n const b = await r.arrayBuffer();\n return b.byteLength > 25_000;\n } catch {\n return false;\n }\n}\n",[],[132,176],"c720cff2-b4ac-4d29-9f49-a8ea0928511f",["D",1784083299603],[],[132,32],"b60019b9-8275-4c1e-b0d0-ca5bf382c03d","evidence.ts","85ea2e48-7ff6-11f1-ab0b-1607ee4eb77e",["D",1784083221049],"// Stage 2 — Gather the evidence file for the chosen candidate.\n// Everything the obituary is allowed to say comes from what this stage fetched.\n//\n// Sources (all free, all verified live 2026-07-14):\n// • Wikipedia action API — full lead section, plain text\n// • RDAP (rdap.org) — who still pays to hold the domain, since when\n// • Wayback CDX — first/last capture + screenshot pick\n// • Wikidata entity JSON — founding date (P571) when recorded\n\nimport type { Candidate, Evidence } from \"./types.ts\";\n\nconst UA = \"RIPURL-newsletter/2.0 (montgomery.k.eryn@gmail.com; val.town cron)\";\n\nasync function fullExtract(title: string): Promise {\n const params = new URLSearchParams({\n action: \"query\",\n prop: \"extracts\",\n explaintext: \"1\",\n exintro: \"0\",\n exchars: \"6000\",\n titles: title,\n format: \"json\",\n formatversion: \"2\",\n redirects: \"1\",\n });\n const resp = await fetch(`https://en.wikipedia.org/w/api.php?${params}`, {\n headers: { \"User-Agent\": UA },\n });\n if (!resp.ok) return \"\";\n const j = await resp.json();\n return j.query?.pages?.[0]?.extract ?? \"\";\n}\n\nasync function rdapLookup(domain: string) {\n try {\n const resp = await fetch(`https://rdap.org/domain/${domain}`, {\n headers: { \"User-Agent\": UA },\n redirect: \"follow\",\n });\n if (!resp.ok) return { reachable: false };\n const j = await resp.json();\n const events: Record = {};\n for (const e of j.events ?? []) events[e.eventAction] = e.eventDate;\n let registrar: string | undefined;\n for (const ent of j.entities ?? []) {\n if ((ent.roles ?? []).includes(\"registrar\")) {\n registrar = ent.vcardArray?.[1]?.find((f: any[]) => f[0] === \"fn\")?.[3];\n }\n }\n return {\n reachable: true,\n registered: events[\"registration\"],\n expires: events[\"expiration\"],\n status: j.status ?? [],\n registrar,\n };\n } catch {\n return { reachable: false };\n }\n}\n\nasync function foundedDate(qid: string): Promise {\n try {\n const resp = await fetch(\n `https://www.wikidata.org/wiki/Special:EntityData/${qid}.json`,\n { headers: { \"User-Agent\": UA } },\n );\n if (!resp.ok) return undefined;\n const j = await resp.json();\n const claims = j.entities?.[qid]?.claims?.P571;\n const t = claims?.[0]?.mainsnak?.datavalue?.value?.time; // \"+2003-01-01T00:00:00Z\"\n return t ? t.replace(/^\\+/, \"\").slice(0, 10) : undefined;\n } catch {\n return undefined;\n }\n}\n\n// Pick the screenshot capture: the archived homepage closest to the midpoint\n// between first capture and dissolution — \"the site in its prime\". This is\n// the deterministic stand-in for the local pipeline's human contact-sheet\n// pick, and the known quality ceiling of the automated path.\nasync function pickScreenshot(c: Candidate) {\n const first = c.firstCapture!;\n const dissolvedStamp = `${c.dissolvedYear}0101000000`;\n const firstYear = parseInt(first.slice(0, 4), 10);\n const midYear = Math.min(\n Math.floor((firstYear + c.dissolvedYear) / 2),\n c.dissolvedYear - 1,\n );\n const target = `${Math.max(midYear, firstYear)}0601`;\n const resp = await fetch(\n `https://archive.org/wayback/available?url=${c.domain}×tamp=${target}`,\n { headers: { \"User-Agent\": UA } },\n );\n if (resp.ok) {\n const j = await resp.json();\n const closest = j.archived_snapshots?.closest;\n if (closest?.available && closest.timestamp <= dissolvedStamp) {\n return {\n timestamp: closest.timestamp,\n sourceUrl: closest.url as string,\n };\n }\n }\n // Fallback: first capture (origin story beats nothing).\n return {\n timestamp: first,\n sourceUrl: `http://web.archive.org/web/${first}/http://${c.domain}/`,\n };\n}\n\nexport async function gatherEvidence(c: Candidate): Promise {\n const extract = await fullExtract(c.wikipediaTitle);\n if (extract.length < 400) {\n throw new Error(`Evidence too thin for ${c.domain}: extract ${extract.length} chars`);\n }\n const rdap = await rdapLookup(c.domain);\n const founded = await foundedDate(c.qid);\n const shot = await pickScreenshot(c);\n return {\n candidate: c,\n wikipedia: {\n title: c.wikipediaTitle,\n url: c.wikipediaUrl,\n description: c.description ?? \"\",\n extract,\n },\n rdap,\n wayback: {\n firstCapture: c.firstCapture!,\n lastCapture: c.lastCapture!,\n monthsArchived: c.monthsArchived!,\n screenshotTimestamp: shot.timestamp,\n screenshotSourceUrl: shot.sourceUrl,\n },\n founded,\n gatheredAt: new Date().toISOString(),\n };\n}\n",[],[132,188],"161b47c2-ce44-45d6-8f06-36cf296049de","discover.ts","76ab2bd0-7ff6-11f1-ac3e-1607ee4eb77e",3,["D",1784083195461],"// Stage 1 — Discover candidates.\n//\n// Source: Wikidata SPARQL. We ask for entities with a recorded dissolution\n// date (P576), an official website (P856), and an English Wikipedia article.\n// Deadness is ground truth (a recorded dissolution), never a heuristic, and\n// the domain comes from the dead entity's OWN website claim — which makes the\n// name-mismatch fabrication bug of the old pipeline impossible by construction.\n//\n// Scoring (deterministic, inspectable — tune the weights here):\n// archive depth 3.0 × min(monthsArchived, 120) / 120\n// story length 2.0 × min(extract chars, 2000) / 2000\n// recency 2.0 × max(0, 1 − yearsSinceDeath / 15)\n// web-nativeness 2.0 if the description/extract reads web-native\n// short domain 1.0 × (domains ≤ 12 chars read better in a headline)\n//\n// Hard gates (fail any → discarded): unused domain, extract ≥ 400 chars,\n// ≥ 12 distinct months in the Wayback Machine, .com/.net/.org/.tv/.fm/.io TLD.\n\nimport type { Candidate } from \"./types.ts\";\nimport { usedDomains } from \"./state.ts\";\n\nconst UA = \"RIPURL-newsletter/2.0 (montgomery.k.eryn@gmail.com; val.town cron)\";\nconst SPARQL = `SELECT DISTINCT ?company ?companyLabel ?website ?dissolved ?article WHERE {\n ?company wdt:P576 ?dissolved ;\n wdt:P856 ?website .\n ?article schema:about ?company ;\n schema:isPartOf .\n FILTER(YEAR(?dissolved) >= 2000)\n SERVICE wikibase:label { bd:serviceParam wikibase:language \"en\". }\n}\nLIMIT 400`;\n\nconst WEB_NATIVE = /\\b(website|web ?site|online|internet|dot-?com|e-?commerce|streaming|social network|social media|app|platform|search engine|portal|forum|blog|browser|hosting|startup|video game|software)\\b/i;\nconst OK_TLD = /\\.(com|net|org|tv|fm|io|co)$/;\n\nfunction domainFromUrl(url: string): string {\n try {\n let host = new URL(url).hostname.toLowerCase();\n if (host.startsWith(\"www.\")) host = host.slice(4);\n return host;\n } catch {\n return \"\";\n }\n}\n\nexport async function fetchRawCandidates(): Promise {\n const params = new URLSearchParams({ query: SPARQL, format: \"json\" });\n const resp = await fetch(`https://query.wikidata.org/sparql?${params}`, {\n headers: { \"User-Agent\": UA, Accept: \"application/sparql-results+json\" },\n });\n if (!resp.ok) throw new Error(`Wikidata SPARQL ${resp.status}`);\n const json = await resp.json();\n const seen = new Set();\n const out: Candidate[] = [];\n for (const b of json.results.bindings) {\n const domain = domainFromUrl(b.website.value);\n const qid = b.company.value.split(\"/\").pop();\n const label = b.companyLabel?.value ?? \"\";\n if (!domain || seen.has(domain)) continue;\n if (!label || label === qid) continue; // no English label\n if (!OK_TLD.test(domain)) continue;\n seen.add(domain);\n const dissolved = b.dissolved.value.slice(0, 10);\n const articleUrl = b.article.value;\n out.push({\n company: label,\n qid,\n domain,\n website: b.website.value,\n dissolvedDate: dissolved,\n dissolvedYear: parseInt(dissolved.slice(0, 4), 10),\n wikipediaTitle: decodeURIComponent(articleUrl.split(\"/wiki/\").pop() ?? \"\").replaceAll(\"_\", \" \"),\n wikipediaUrl: articleUrl,\n });\n }\n return out;\n}\n\nasync function wikipediaSummary(title: string) {\n const resp = await fetch(\n `https://en.wikipedia.org/api/rest_v1/page/summary/${encodeURIComponent(title.replaceAll(\" \", \"_\"))}`,\n { headers: { \"User-Agent\": UA } },\n );\n if (!resp.ok) return null;\n const j = await resp.json();\n return { extract: j.extract ?? \"\", description: j.description ?? \"\" };\n}\n\nasync function waybackDepth(domain: string) {\n // collapse=timestamp:6 → one row per distinct YYYYMM with a capture.\n const resp = await fetch(\n `https://web.archive.org/cdx/search/cdx?url=${domain}&output=json&fl=timestamp&filter=statuscode:200&collapse=timestamp:6&limit=500`,\n { headers: { \"User-Agent\": UA } },\n );\n if (!resp.ok) return null;\n const rows = await resp.json();\n if (!Array.isArray(rows) || rows.length < 2) return null;\n const stamps = rows.slice(1).map((r: string[]) => r[0]);\n return {\n monthsArchived: stamps.length,\n firstCapture: stamps[0],\n lastCapture: stamps[stamps.length - 1],\n };\n}\n\nexport function scoreCandidate(c: Candidate): number {\n const months = c.monthsArchived ?? 0;\n const chars = (c.extract ?? \"\").length;\n const yearsDead = new Date().getFullYear() - c.dissolvedYear;\n const webNative = WEB_NATIVE.test(`${c.description} ${c.extract}`) ? 2 : 0;\n const shortName = c.domain.length <= 12 ? 1 : 0;\n return (\n 3 * Math.min(months, 120) / 120 +\n 2 * Math.min(chars, 2000) / 2000 +\n 2 * Math.max(0, 1 - yearsDead / 15) +\n webNative +\n shortName\n );\n}\n\n// Full discovery pass: fetch pool, filter, enrich a recency-biased sample,\n// gate, score, and return the ranked shortlist (winner + fallbacks).\nexport async function discover(maxEnrich = 12): Promise {\n const used = await usedDomains();\n const raw = (await fetchRawCandidates()).filter((c) => !used.has(c.domain));\n if (raw.length === 0) throw new Error(\"Wikidata returned no unused candidates\");\n\n // Bias toward recent deaths but keep variety: newest first, then sample\n // deterministically by week so consecutive runs don't always retry the\n // exact same losers.\n raw.sort((a, b) => b.dissolvedDate.localeCompare(a.dissolvedDate));\n const pool = raw.slice(0, 80);\n const weekOffset = Math.floor(Date.now() / (7 * 86_400_000)) % 5;\n const sample = pool.filter((_, i) => i % 5 === weekOffset).concat(pool.filter((_, i) => i % 5 !== weekOffset))\n .slice(0, maxEnrich);\n\n const passed: Candidate[] = [];\n for (const c of sample) {\n const wiki = await wikipediaSummary(c.wikipediaTitle);\n if (!wiki || wiki.extract.length < 400) continue; // gate: enough recorded history\n c.extract = wiki.extract;\n c.description = wiki.description;\n const wb = await waybackDepth(c.domain);\n if (!wb || wb.monthsArchived < 12) continue; // gate: enough bones to examine\n c.monthsArchived = wb.monthsArchived;\n c.firstCapture = wb.firstCapture;\n c.lastCapture = wb.lastCapture;\n c.score = scoreCandidate(c);\n passed.push(c);\n if (passed.length >= 5) break; // enough — winner + 4 fallbacks\n }\n if (passed.length === 0) throw new Error(\"No candidate passed the hard gates this pass\");\n passed.sort((a, b) => (b.score ?? 0) - (a.score ?? 0));\n return passed;\n}\n",[],[132,195],["D",1784083130616],[153,154,155,156,157,158],[132],["D",1784083111215],["D",1789438866375],{"_99":98,"_107":103},{"_99":98,"_100":101,"_102":103,"_18":54,"_104":105,"_106":24},{"_13":88,"_31":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"ctx",{"_81":-5,"_127":243,"_244":245,"_246":247,"_248":24,"_249":24,"_250":24,"_251":58,"_252":-5,"_253":-7,"_254":-7,"_255":62,"_256":24,"_257":24,"_258":-5,"_259":24,"_260":24,"_261":24,"_262":-5},"ENV",{"_214":215,"_216":217,"_218":89,"_219":220,"_221":222,"_223":224,"_225":226,"_227":228,"_229":230,"_231":232,"_233":234,"_235":236,"_237":238,"_239":240,"_241":242},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v6