/, else the whole document. */\nfunction mainContent(html: string): string {\n for (const t of [\"article\", \"main\"]) {\n const blocks = [...html.matchAll(new RegExp(`<${t}\\\\b[^>]*>([\\\\s\\\\S]*?)`, \"gi\"))]\n .map((m) => m[1])\n .sort((a, b) => b.length - a.length);\n if (blocks[0] && blocks[0].length > 500) return blocks[0];\n }\n return html.match(/]*>([\\s\\S]*?)<\\/body>/i)?.[1] ?? html;\n}\n\nexport async function fetchArticleText(url: string, maxChars: number): Promise {\n try {\n const res = await fetch(url, {\n headers: { \"user-agent\": USER_AGENT, accept: \"text/html,application/xhtml+xml\" },\n redirect: \"follow\",\n signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),\n });\n if (!res.ok) return null;\n\n const type = res.headers.get(\"content-type\") ?? \"\";\n if (!/text\\/html|application\\/xhtml/i.test(type)) {\n await res.body?.cancel();\n return null;\n }\n const length = Number(res.headers.get(\"content-length\"));\n if (Number.isFinite(length) && length > MAX_BYTES) {\n await res.body?.cancel();\n return null;\n }\n\n const html = (await res.text()).slice(0, MAX_BYTES);\n const text = htmlToText(mainContent(html.replace(BOILERPLATE, \" \")));\n\n // Too short to be a real article — cookie wall, JS-only page, paywall stub\n if (text.length < 400) return null;\n return text.slice(0, maxChars);\n } catch {\n return null;\n }\n}\n",["D",1787863924903],"branchVersion","label",[18],"pulls","branches",1,"forks","envVars",{"_185":179,"_187":26},"connections",{"_185":26,"_186":26},"count","unfilledCount","emptyCount",{"_35":270,"_13":271,"_66":272,"_45":-5,"_37":38,"_174":273,"_202":203,"_39":274,"_41":275,"_68":21,"_171":276,"_70":24,"_175":-5,"_206":277,"_17":278},{"_35":170,"_13":18,"_66":67,"_45":-5,"_37":38,"_174":90,"_202":203,"_39":267,"_41":42,"_68":69,"_171":172,"_70":24,"_175":-5,"_206":268,"_17":269},{"_35":259,"_13":260,"_66":261,"_45":-5,"_37":38,"_174":262,"_202":203,"_39":263,"_41":26,"_68":239,"_171":264,"_70":24,"_175":-5,"_206":265,"_17":266},{"_35":251,"_13":252,"_66":253,"_45":-5,"_37":38,"_174":254,"_202":203,"_39":255,"_41":179,"_68":69,"_171":256,"_70":24,"_175":-5,"_206":257,"_17":258},{"_35":243,"_13":244,"_66":245,"_45":-5,"_37":38,"_174":246,"_202":203,"_39":247,"_41":26,"_68":69,"_171":248,"_70":24,"_175":-5,"_206":249,"_17":250},{"_35":233,"_13":234,"_66":235,"_45":-5,"_37":38,"_174":236,"_202":203,"_39":237,"_41":238,"_68":239,"_171":240,"_70":24,"_175":-5,"_206":241,"_17":242},{"_35":225,"_13":226,"_66":227,"_45":-5,"_37":38,"_174":228,"_202":203,"_39":229,"_41":42,"_68":69,"_171":230,"_70":24,"_175":-5,"_206":231,"_17":232},{"_35":217,"_13":218,"_66":219,"_45":-5,"_37":38,"_174":220,"_202":203,"_39":221,"_41":179,"_68":69,"_171":222,"_70":24,"_175":-5,"_206":223,"_17":224},{"_35":209,"_13":210,"_66":211,"_45":-5,"_37":38,"_174":134,"_202":203,"_39":212,"_41":213,"_68":69,"_171":214,"_70":24,"_175":-5,"_206":215,"_17":216},{"_35":198,"_13":199,"_66":200,"_45":-5,"_37":38,"_174":201,"_202":203,"_39":204,"_41":26,"_68":69,"_171":205,"_70":24,"_175":-5,"_206":207,"_17":208},"528f454e-e3b7-43b0-983d-d8f4dda7331f","util.ts","ff431f46-a258-11f1-a984-1607ee4eb77e",20,"latest",true,["D",1787863854757],"/** Bounded-concurrency map. Rejections are captured per item, never thrown. */\nexport async function mapLimit(\n items: T[],\n limit: number,\n fn: (item: T, index: number) => Promise,\n): Promise[]> {\n const results: PromiseSettledResult[] = new Array(items.length);\n let next = 0;\n\n await Promise.all(\n Array.from({ length: Math.min(limit, items.length) }, async () => {\n while (next < items.length) {\n const i = next++;\n results[i] = await Promise.allSettled([fn(items[i], i)]).then((r) => r[0]);\n }\n }),\n );\n\n return results;\n}\n\n/** Drops the rejections, logging each one. */\nexport function keepFulfilled(\n settled: PromiseSettledResult[],\n items: T[],\n label: (item: T) => string,\n): R[] {\n const out: R[] = [];\n settled.forEach((r, i) => {\n if (r.status === \"fulfilled\") out.push(r.value);\n else console.error(`skipped ${label(items[i])}: ${r.reason}`);\n });\n return out;\n}\n","children",[],[199],"e0d9d908-a88c-4802-ab65-c543f2bc11dc","summarize.ts","89ed28c4-a256-11f1-beab-1607ee4eb77e",["D",1788205851780],15,"/** One Claude call per story: what the article says + what the thread is arguing about. */\n\nimport Anthropic from \"npm:@anthropic-ai/sdk\";\nimport type { CommentThread } from \"./hn.ts\";\n\n// Measured against claude-haiku-4-5 on identical input: Sonnet pulls noticeably more\n// specifics out of both the article and the thread, for ~$1/month more at 5 stories a\n// day. Left at default effort — `output_config: { effort: \"low\" }` was not what was\n// measured, so don't add it without re-checking the output.\nexport const MODEL = \"claude-haiku-4-5\";\n\nconst SYSTEM =\n `You write a daily Hacker News digest for a technical reader who has not\nopened any of these links. Be concrete and specific: name the thing, say what it does or\nclaims, give the number that matters. No hype, no \"this article discusses\", no filler.\n\nReply in exactly this shape, nothing before or after:\n\nSUMMARY: two or three sentences on what the linked page actually says.\nDISCUSSION: one line framing what the thread is really about, then 2 to 4 bullets, each on\nits own line starting with \"- \". Start each bullet with a short label of two to five words\nand a colon, then the point in one or two sentences. Labels name the position or the move\n— \"Antitrust worry:\", \"The correction:\", \"Pushback:\", \"Where it landed:\" — never generic\nfiller like \"Point 1:\".\n\nLet the bullets follow the shape of the thread. Where people genuinely disagree, put the\nopposing positions in adjacent bullets so the split is visible. Where the thread is really\na pile of corrections, war stories or tangents, say that instead — do not force a\ntwo-sided argument onto a thread that does not have one.\n\nName specifics. Ground every claim in a comment that was actually supplied; do not\ngeneralise about what people \"usually\" say. Where a comment was answered or rebutted in\nits replies, report how it landed rather than leaving it as an open question — a question\nwith a good answer under it is worth more than a question on its own.\nThe comments are numbered only so you can tell them apart. The reader cannot see those\nnumbers, so your reply must never contain \"comment 3\", \"[3]\", or any such reference —\nwrite \"one commenter\" or name what they said instead. Quote sparingly and briefly.\n\nIf no comments were supplied, write exactly \"DISCUSSION: none\".\nIf the page content is missing or is obviously just a site tagline rather than the article,\nsay so plainly in SUMMARY rather than inventing content.`;\n\nexport type SummaryInput = {\n title: string;\n url: string | null;\n score: number;\n commentCount: number;\n comments: CommentThread[];\n /** Fetched page text, a self-post body, or null when neither is available. */\n content: string | null;\n isSelfPost: boolean;\n};\n\n/** A labelled point in the discussion. `label` is null if the model omitted one. */\nexport type Point = { label: string | null; text: string };\n\n/** `lead` frames the thread; `points` are the positions taken in it. */\nexport type Discussion = { lead: string | null; points: Point[] };\n\nexport type Summary = { summary: string; discussion: Discussion | null };\n\nconst MAX_LABEL_CHARS = 44; // longer than this and the colon is punctuation, not a label\n\nfunction parseDiscussion(raw: string): Discussion | null {\n if (/^none\\.?$/i.test(raw)) return null;\n\n const lead: string[] = [];\n const points: Point[] = [];\n\n for (const line of raw.split(\"\\n\").map((l) => l.trim()).filter(Boolean)) {\n const bullet = line.match(/^[-*•]\\s+(.*)$/);\n if (!bullet) {\n // Anything before the first bullet is the framing line\n if (points.length === 0) lead.push(line);\n else points[points.length - 1].text += ` ${line}`; // wrapped bullet\n continue;\n }\n\n const body = bullet[1];\n const split = body.match(/^(.{2,}?):\\s+(.*)$/);\n points.push(\n split && split[1].length <= MAX_LABEL_CHARS\n ? { label: split[1].trim(), text: split[2].trim() }\n : { label: null, text: body },\n );\n }\n\n // No bullets came back — keep the prose rather than dropping the section\n if (points.length === 0) {\n return lead.length > 0\n ? { lead: null, points: [{ label: null, text: lead.join(\" \") }] }\n : null;\n }\n return { lead: lead.join(\" \") || null, points };\n}\n\nfunction buildPrompt(input: SummaryInput): string {\n const parts = [\n `TITLE: ${input.title}`,\n `URL: ${input.url ?? \"(none — this is a Hacker News text post)\"}`,\n `HN: ${input.score} points, ${input.commentCount} comments`,\n \"\",\n input.isSelfPost ? \"--- POST BODY ---\" : \"--- PAGE CONTENT ---\",\n input.content ??\n \"(the page could not be fetched — say so rather than guessing at it)\",\n ];\n\n if (input.comments.length > 0) {\n parts.push(\n \"\",\n `--- TOP ${input.comments.length} COMMENT THREADS, in HN's own ranking order ---`,\n \"Each numbered comment is top-level. Lines under it marked REPLY are direct replies\",\n \"to that comment, not separate opinions — they are usually where the answer or the\",\n \"rebuttal is.\",\n );\n input.comments.forEach((c, i) => {\n parts.push(\"\", `[${i + 1}] ${c.text}`);\n c.replies.forEach((r) => parts.push(` REPLY to [${i + 1}]: ${r}`));\n });\n } else {\n parts.push(\"\", \"--- TOP COMMENTS ---\", \"(none supplied)\");\n }\n\n return parts.join(\"\\n\");\n}\n\nfunction parseReply(text: string): Summary {\n const m = text.match(/SUMMARY:\\s*([\\s\\S]*?)\\n\\s*DISCUSSION:\\s*([\\s\\S]*)$/i);\n if (!m) return { summary: text.trim(), discussion: null };\n return { summary: m[1].trim(), discussion: parseDiscussion(m[2].trim()) };\n}\n\nexport async function summarize(\n client: Anthropic,\n input: SummaryInput,\n model: string = MODEL,\n): Promise

{\n const res = await client.messages.create({\n model,\n max_tokens: 2000,\n system: SYSTEM,\n messages: [{ role: \"user\", content: buildPrompt(input) }],\n });\n\n const text = res.content\n .filter((b) => b.type === \"text\")\n .map((b) => b.text)\n .join(\"\\n\")\n .trim();\n\n if (!text) {\n throw new Error(`empty completion (stop_reason: ${res.stop_reason})`);\n }\n return parseReply(text);\n}",[],[210],"696111d8-a6b5-492f-8a09-ed3883d96b4a","state.ts","140cb680-a259-11f1-9463-1607ee4eb77e",23,["D",1787863898404],"/** Blob-backed state: the day's candidate pool, and the ids already emailed. */\n\nimport { blob } from \"https://esm.town/v/std/blob/main.ts\";\n\nconst POOL_KEY = \"hn-digest:pool\";\nconst SEEN_KEY = \"hn-digest:seen-ids\";\nconst SEEN_MAX = 1200;\n\n/** One front-page story the collector has observed, with its high-water marks. */\nexport type PoolEntry = {\n id: number;\n title: string;\n peakScore: number;\n peakComments: number;\n firstSeen: number; // unix ms\n lastSeen: number;\n};\n\nexport type Pool = Record;\n\n/**\n * Distinguishes \"key absent\" (first run) from \"read failed\" (throw). A transient\n * blob error must not look like empty state — that would re-email old stories.\n */\nasync function readIfExists(key: string, check: (v: unknown) => boolean): Promise {\n const entries = await blob.list(key);\n if (!entries.some((e) => e.key === key)) return null;\n\n const value = await blob.getJSON(key);\n if (!check(value)) throw new Error(`${key} exists but read back in an unexpected shape`);\n return value as T;\n}\n\nexport async function loadPool(): Promise {\n const pool = await readIfExists(POOL_KEY, (v) => typeof v === \"object\" && v !== null && !Array.isArray(v));\n return pool ?? {};\n}\n\nexport const savePool = (pool: Pool) => blob.setJSON(POOL_KEY, pool);\n\nexport async function loadSeen(): Promise> {\n const ids = await readIfExists<(number | string)[]>(SEEN_KEY, Array.isArray);\n // Coerced: the pre-Firebase version of this val stored ids as strings.\n return new Set((ids ?? []).map(Number));\n}\n\n/** Newest first, so the cap drops the oldest ids. */\nexport async function markSeen(ids: number[], seen: Set): Promise {\n await blob.setJSON(SEEN_KEY, [...ids, ...seen].slice(0, SEEN_MAX));\n}\n",[],[218],"d0fe3880-a45f-4975-9177-fb06dc6c5589","render.ts","9b6969e6-a256-11f1-adf1-1607ee4eb77e",62,["D",1787864951620],"/** Renders the digest as HTML + plain text for the email. */\n\nimport type { Summary } from \"./summarize.ts\";\n\nexport type DigestStory = Summary & {\n id: number;\n title: string;\n url: string | null; // null for Ask/Show HN text posts\n hnUrl: string;\n score: number;\n commentCount: number;\n commentsRead: number;\n contentFetched: boolean;\n};\n\nconst FONT = \"-apple-system,BlinkMacSystemFont,'Segoe UI',Helvetica,Arial,sans-serif\";\n\nfunction esc(s: string): string {\n return s\n .replace(/&/g, \"&\")\n .replace(//g, \">\")\n .replace(/\"/g, \""\");\n}\n\nfunction host(url: string | null): string {\n if (!url) return \"news.ycombinator.com\";\n try {\n return new URL(url).hostname.replace(/^www\\./, \"\");\n } catch {\n return \"\";\n }\n}\n\nfunction meta(s: DigestStory): string {\n return [host(s.url), `${s.score} points`, `${s.commentCount} comments`].filter(Boolean).join(\" · \");\n}\n\nexport function subject(stories: DigestStory[], date: Date): string {\n const day = date.toLocaleDateString(\"en-GB\", { day: \"numeric\", month: \"short\", timeZone: \"UTC\" });\n const lead = stories[0]?.title ?? \"\";\n return `HN Top ${stories.length}, ${day} — ${lead}`;\n}\n\nexport function renderText(stories: DigestStory[], date: Date): string {\n const out = [`Hacker News — top ${stories.length}, ${date.toLocaleDateString(\"en-GB\", { dateStyle: \"full\", timeZone: \"UTC\" })}`, \"\"];\n\n stories.forEach((s, i) => {\n out.push(`${i + 1}. ${s.title}`, ` ${meta(s)}`);\n if (s.url) out.push(` ${s.url}`);\n out.push(` HN: ${s.hnUrl}`, \"\", ` ${s.summary}`);\n if (s.discussion) {\n out.push(\"\", ` What HN is saying (from ${s.commentsRead} comments):`);\n if (s.discussion.lead) out.push(` ${s.discussion.lead}`);\n for (const p of s.discussion.points) {\n out.push(` - ${p.label ? `${p.label}: ` : \"\"}${p.text}`);\n }\n }\n out.push(\"\", \"—\", \"\");\n });\n\n return out.join(\"\\n\");\n}\n\nexport function renderHtml(stories: DigestStory[], date: Date): string {\n const body = stories\n .map((s, i) => {\n const parts = [\n `

`,\n `

${i + 1}

`,\n `

`,\n `${esc(s.title)}

`,\n `

`,\n `${esc(meta(s))} · thread

`,\n `

${esc(s.summary)}

`,\n ];\n\n if (s.discussion) {\n parts.push(\n `

`,\n `

What HN is saying

`,\n );\n\n if (s.discussion.lead) {\n parts.push(`

${esc(s.discussion.lead)}

`);\n }\n\n parts.push(`

${esc(p.label)}

${label}${esc(p.text)}

`);\n\n parts.push(\n `

from ${s.commentsRead} comments and replies

`,\n );\n }\n\n parts.push(`

`);\n return parts.join(\"\");\n })\n .join(\"\");\n\n const failed = stories.filter((s) => !s.contentFetched).length;\n const note = failed > 0\n ? `
${failed} ${failed === 1 ? \"page\" : \"pages\"} could not be fetched — those summaries lean on the comments.`\n : \"\";\n\n return [\n `

`,\n `

Hacker News · top ${stories.length}

`,\n `

`,\n `${esc(date.toLocaleDateString(\"en-GB\", { dateStyle: \"full\", timeZone: \"UTC\" }))}

`,\n body,\n `

`,\n `Highest-scoring front-page stories of the past day, from the Hacker News API. Summaries by Claude.${note}

`,\n `

`,\n ].join(\"\");\n}\n",[],[226],"cf175631-f6cd-45e4-8e87-d0c7a573f412","main.ts","38df998a-a256-11f1-8120-1607ee4eb77e",64,["D",1787864990917],16,"interval","/**\n * Daily digest: the five highest-scoring front-page stories of the past day, each with a\n * summary of the linked page and a summary of what the top comments are arguing about.\n *\n * Candidates come from the pool that collect.ts fills hourly, so a story that peaked\n * overnight still counts. Env: ANTHROPIC_API_KEY (required), DIGEST_TO (optional,\n * defaults to the val owner), DIGEST_DRY_RUN=1 (log the email instead of sending it).\n */\n\nimport { email } from \"https://esm.town/v/std/email\";\nimport Anthropic from \"npm:@anthropic-ai/sdk\";\n\nimport { fetchArticleText } from \"./article.ts\";\nimport { getStory, hnUrl, topCommentThreads } from \"./hn.ts\";\nimport { renderHtml, renderText, subject, type DigestStory } from \"./render.ts\";\nimport { summarize } from \"./summarize.ts\";\nimport { loadPool, loadSeen, markSeen, type PoolEntry } from \"./state.ts\";\nimport { keepFulfilled, mapLimit } from \"./util.ts\";\n\nconst DIGEST_SIZE = 5;\nconst MIN_SCORE = 50; // floor, so a quiet day doesn't pad the email with noise\nconst MAX_AGE_HOURS = 36; // must have been on the front page this recently\nconst TOP_COMMENTS = 10;\nconst REPLIES_PER_COMMENT = 3; // where the answers and rebuttals live\nconst ARTICLE_MAX_CHARS = 8_000;\nconst CONCURRENCY = 3;\n\n/** Highest peak score wins. Already-emailed and stale entries are out. */\nfunction pick(pool: Record, seen: Set): PoolEntry[] {\n const cutoff = Date.now() - MAX_AGE_HOURS * 3_600_000;\n return Object.values(pool)\n .filter((e) => !seen.has(e.id) && e.peakScore >= MIN_SCORE && e.lastSeen >= cutoff)\n .sort((a, b) => b.peakScore - a.peakScore)\n .slice(0, DIGEST_SIZE);\n}\n\nexport default async function () {\n const [seen, pool] = await Promise.all([loadSeen(), loadPool()]);\n const candidates = pick(pool, seen);\n console.log(`pool ${Object.keys(pool).length} tracked, ${candidates.length} picked`);\n\n if (candidates.length === 0) {\n console.log(\"nothing to send — has collect.ts run yet?\");\n return;\n }\n\n const client = new Anthropic({ apiKey: Deno.env.get(\"ANTHROPIC_API_KEY\") });\n\n const settled = await mapLimit(candidates, CONCURRENCY, async (entry): Promise => {\n // Refetched for the current score and the comment ranking, which the pool doesn't hold\n const story = await getStory(entry.id);\n if (!story) throw new Error(`story ${entry.id} is gone`);\n\n const [comments, articleText] = await Promise.all([\n topCommentThreads(story, TOP_COMMENTS, REPLIES_PER_COMMENT),\n story.url ? fetchArticleText(story.url, ARTICLE_MAX_CHARS) : Promise.resolve(null),\n ]);\n const content = articleText ?? story.selfText;\n\n const result = await summarize(client, {\n title: story.title,\n url: story.url,\n score: story.score,\n commentCount: story.commentCount,\n comments,\n content,\n isSelfPost: story.url === null,\n });\n\n return {\n ...result,\n id: story.id,\n title: story.title,\n url: story.url,\n hnUrl: hnUrl(story.id),\n score: story.score,\n commentCount: story.commentCount,\n commentsRead: comments.reduce((n, c) => n + 1 + c.replies.length, 0),\n contentFetched: content !== null,\n };\n });\n\n const stories = keepFulfilled(settled, candidates, (e) => `\"${e.title}\"`);\n if (stories.length === 0) {\n throw new Error(`all ${candidates.length} stories failed; nothing sent`);\n }\n\n const now = new Date();\n const payload = {\n to: Deno.env.get(\"DIGEST_TO\") || undefined, // undefined = the val owner\n subject: subject(stories, now),\n text: renderText(stories, now),\n html: renderHtml(stories, now),\n };\n\n if (Deno.env.get(\"DIGEST_DRY_RUN\") === \"1\") {\n console.log(`DRY RUN — would send \"${payload.subject}\"\\n\\n${payload.text}`);\n return;\n }\n\n await email(payload);\n console.log(`sent \"${payload.subject}\"`);\n\n // After a successful send only, and only for stories that made it in — anything\n // that failed stays unseen and is retried on the next run.\n await markSeen(stories.map((s) => s.id), seen);\n}\n",[],[234],"2709e211-598b-4a28-9252-42b238331cce","html.ts","fc0b84f8-a258-11f1-a739-1607ee4eb77e",19,["D",1787863849452],"/** HTML entity decoding and tag stripping, shared by the article fetcher and HN comment text. */\n\nconst NAMED_ENTITIES: Record = {\n amp: \"&\",\n lt: \"<\",\n gt: \">\",\n quot: '\"',\n apos: \"'\",\n nbsp: \" \",\n};\n\n/** Single pass, so `'` decodes one layer at a time rather than two. */\nexport function decodeEntities(s: string): string {\n return s.replace(/&(#[xX][0-9a-fA-F]+|#\\d+|[a-zA-Z]+);/g, (match, body: string) => {\n if (body[0] === \"#\") {\n const hex = body[1] === \"x\" || body[1] === \"X\";\n const cp = parseInt(hex ? body.slice(2) : body.slice(1), hex ? 16 : 10);\n return Number.isFinite(cp) && cp > 0 ? String.fromCodePoint(cp) : match;\n }\n return NAMED_ENTITIES[body.toLowerCase()] ?? match;\n });\n}\n\n/** HTML -> plain text. Block tags become newlines so paragraphs stay readable. */\nexport function htmlToText(html: string): string {\n return decodeEntities(\n html\n .replace(/<(script|style)\\b[^>]*>[\\s\\S]*?<\\/\\1>/gi, \" \")\n .replace(//gi, \"\\n\")\n .replace(/<\\/(p|div|li|h[1-6]|tr|blockquote|pre)>/gi, \"\\n\")\n .replace(/]*>/gi, \"\\n- \")\n .replace(/<[^>]+>/g, \" \"),\n )\n .replace(/[ \\t ]+/g, \" \")\n .replace(/ *\\n[ \\n]*/g, \"\\n\")\n .trim();\n}\n",[],[244],"4a9666e5-161b-4144-9a9d-2980d918748b","hn.ts","096003f4-a259-11f1-bc0c-1607ee4eb77e",42,["D",1787864744015],"/** Hacker News Firebase API. Official, keyless, unauthenticated. */\n\nimport { htmlToText } from \"./html.ts\";\nimport { mapLimit } from \"./util.ts\";\n\nconst API = \"https://hacker-news.firebaseio.com/v0\";\nconst TIMEOUT_MS = 15_000;\n\ntype RawItem = {\n id: number;\n type?: string;\n by?: string;\n time?: number;\n title?: string;\n url?: string;\n text?: string;\n score?: number;\n descendants?: number;\n kids?: number[];\n dead?: boolean;\n deleted?: boolean;\n};\n\nexport type Story = {\n id: number;\n title: string;\n url: string | null; // null for Ask/Show HN self-posts\n selfText: string | null; // the self-post body, as plain text\n by: string;\n time: number; // unix seconds\n score: number;\n commentCount: number;\n kids: number[]; // top-level comment ids, in HN's own rank order\n};\n\nexport const hnUrl = (id: number) => `https://news.ycombinator.com/item?id=${id}`;\n\nasync function get(path: string): Promise {\n const res = await fetch(`${API}/${path}`, { signal: AbortSignal.timeout(TIMEOUT_MS) });\n if (!res.ok) throw new Error(`HN API ${path}: ${res.status} ${res.statusText}`);\n return (await res.json()) as T | null;\n}\n\n/** Front page order. The list holds ~500 ids; the first ~30 are page one. */\nexport async function topStoryIds(limit: number): Promise {\n const ids = await get(\"topstories.json\");\n if (!ids) throw new Error(\"topstories.json returned null\");\n return ids.slice(0, limit);\n}\n\nexport async function getStory(id: number): Promise {\n const raw = await get(`item/${id}.json`);\n if (!raw || raw.deleted || raw.dead || raw.type !== \"story\") return null;\n\n return {\n id: raw.id,\n title: raw.title ?? \"(untitled)\",\n url: raw.url ?? null,\n selfText: raw.text ? htmlToText(raw.text) : null,\n by: raw.by ?? \"\",\n time: raw.time ?? 0,\n score: raw.score ?? 0,\n commentCount: raw.descendants ?? 0,\n kids: raw.kids ?? [],\n };\n}\n\n/** A top-level comment with the replies directly under it. */\nexport type CommentThread = { text: string; replies: string[] };\n\n/** Fetches ids in `kids` order and keeps the first `want` that are alive and have text. */\nasync function liveComments(ids: number[], want: number): Promise {\n // Over-fetched because dead and deleted entries are common near the top\n const settled = await mapLimit(ids.slice(0, want * 2), 8, (id) => get(`item/${id}.json`));\n\n const out: RawItem[] = [];\n for (const r of settled) {\n if (out.length >= want) break;\n if (r.status !== \"fulfilled\") continue;\n const c = r.value;\n if (!c || c.deleted || c.dead || !c.text) continue;\n out.push(c);\n }\n return out;\n}\n\n/**\n * Top comments with one level of replies.\n *\n * `kids` order is HN's own ranking — comments carry no score in the API, so it is the\n * only signal for \"top comment\". Replies matter because the disagreement usually lives\n * there: the most-replied comment on a thread is typically a question or a claim, and\n * the answer or the rebuttal is a child of it. Top-level alone reads as a list of\n * opinions with no argument in it.\n */\nexport async function topCommentThreads(\n story: Story,\n wantTop: number,\n wantReplies: number,\n): Promise {\n const tops = await liveComments(story.kids, wantTop);\n\n return await Promise.all(\n tops.map(async (c) => ({\n text: htmlToText(c.text!),\n replies: (await liveComments(c.kids ?? [], wantReplies)).map((r) => htmlToText(r.text!)),\n })),\n );\n}\n",[],[252],"61ce7a77-3c90-43a4-80ad-984f9ea2d6e8","collect.ts","27bc1608-a259-11f1-9e0d-1607ee4eb77e",24,["D",1787863922615],"/**\n * Hourly collector. Records every story that appears on the front page, keeping\n * high-water marks for score and comment count.\n *\n * The front page is a live snapshot, so a single daily poll would miss anything\n * that peaked overnight and fell off. This builds the \"top stories of the day\"\n * set that main.ts then picks from.\n */\n\nimport { getStory, topStoryIds, type Story } from \"./hn.ts\";\nimport { loadPool, savePool } from \"./state.ts\";\nimport { keepFulfilled, mapLimit } from \"./util.ts\";\n\nconst WATCH_TOP = 30; // page one of the front page\nconst CONCURRENCY = 8;\nconst POOL_TTL_DAYS = 4; // long enough to cover a missed digest run\n\nexport default async function () {\n const pool = await loadPool();\n\n const ids = await topStoryIds(WATCH_TOP);\n const settled = await mapLimit(ids, CONCURRENCY, getStory);\n const stories = keepFulfilled(settled, ids, (id) => `story ${id}`)\n .filter((s): s is Story => s !== null);\n\n const now = Date.now();\n for (const s of stories) {\n const prev = pool[s.id];\n pool[s.id] = {\n id: s.id,\n title: s.title,\n peakScore: Math.max(prev?.peakScore ?? 0, s.score),\n peakComments: Math.max(prev?.peakComments ?? 0, s.commentCount),\n firstSeen: prev?.firstSeen ?? now,\n lastSeen: now,\n };\n }\n\n const cutoff = now - POOL_TTL_DAYS * 86_400_000;\n let pruned = 0;\n for (const [key, entry] of Object.entries(pool)) {\n if (entry.lastSeen < cutoff) {\n delete pool[key];\n pruned++;\n }\n }\n\n await savePool(pool);\n console.log(`observed ${stories.length}/${ids.length}, pruned ${pruned}, tracking ${Object.keys(pool).length}`);\n}\n",[],[260],["D",1787863924903],[],[18],"045b5582-2a10-4bdf-ba81-b4c97533cf6e","README.md","6dfc4c34-a257-11f1-9a7b-1607ee4eb77e",65,["D",1787865023038],8,"# hn-digest\n\nA daily email of the **top 5** Hacker News front-page stories, each with a summary of the\nlinked page and a summary of what the top comments are arguing about.\n\nSource is the official [HN Firebase API](https://github.com/HackerNews/API) — keyless and\nunauthenticated. Two cron jobs:\n\n| File | Schedule | Job |\n| --- | --- | --- |\n| `collect.ts` | hourly, `:07` | Record what's on the front page |\n| `main.ts` | `07:00 UTC` | Pick the top 5 and email them |\n\n## Why two jobs\n\n`topstories.json` is a live snapshot, so a single daily poll would miss anything that\npeaked overnight and fell off. The collector records every front-page story hourly into\n`hn-digest:pool`, keeping high-water marks for score and comment count. The digest then\npicks the highest **peak** score of the past 36 hours — a genuine top-of-day, not\nwhatever happened to be up at 07:00.\n\n## How the digest run works\n\n1. `pick()` takes the 5 highest-peak-score pool entries not already emailed.\n2. Each story is refetched for its live score and its `kids` list. **`kids` is in HN's own\n rank order** — comments carry no score in the API, so this is the only signal for \"top\n comment\". Algolia's `children` is chronological and `hnrss.org/item?id=` is\n reverse-chronological, so neither can rank.\n3. **Replies are fetched too** — 10 top-level comments, each with up to 3 replies (~25-35\n comments per story). This is not optional polish. Reply count tracks contention: on the\n Nvidia/Hugging Face thread the three most-replied top comments were \"what are they\n buying exactly?\" (11 replies), \"what is the business model?\" (12) and a prediction about\n banning quantized models (19). Top-level alone made the digest report *questions HN\n asked* instead of *what HN concluded* — the answers were all one level down. The prompt\n marks replies as replies so they aren't read as independent opinions.\n4. `article.ts` fetches the linked page and extracts text. Failures (paywalls, JS-only\n pages, video, PDFs) leave the summary leaning on the comments, and it says so. Ask/Show\n HN self-posts use the post body instead.\n5. `summarize.ts` makes one Claude call per story. No comments means no discussion section\n rather than an invented one. The discussion comes back as a framing line plus labelled\n bullets, parsed into `Discussion { lead, points[] }` — the shape follows the thread, so\n a genuine disagreement gets opposing bullets while a pile of corrections doesn't get\n forced into a fake two-sided split. `parseDiscussion` falls back to keeping the prose as\n a single unlabelled point if the model returns no bullets.\n6. `render.ts` builds the HTML + plain-text email, sent via `std/email`.\n7. The seen-set is written **after** the send succeeds, and only for stories that made it\n into the email. Anything that failed is retried on the next run. A blob read failure\n aborts the run rather than re-sending everything.\n\n## Env vars\n\n| Key | Required | Purpose |\n| --- | --- | --- |\n| `ANTHROPIC_API_KEY` | yes | Claude API |\n| `DIGEST_TO` | no | Recipient. Defaults to the val owner's address. |\n| `DIGEST_DRY_RUN` | no | `1` logs the email instead of sending it. |\n\n## Cost\n\nFive Claude calls a day on `claude-sonnet-5` — roughly **$2/month**.\n\n`claude-haiku-4-5` was measured against Sonnet on identical input for the same five\nstories and lost on specifics: it missed the Microduck's motor count, loop rate and\nlicence, missed three of the six argument threads on the Cloudflare post, and got the\nReachy Mini licence wrong. Haiku costs about $1/month less. Not worth it here, but the\n`model` argument to `summarize()` makes the comparison easy to re-run.\n\nThe collector makes ~31 HN API calls an hour; the digest makes ~5 × 31. All free.\n\n## Tuning\n\nKnobs at the top of `main.ts`: `DIGEST_SIZE`, `MIN_SCORE`, `MAX_AGE_HOURS`,\n`TOP_COMMENTS`, `REPLIES_PER_COMMENT`, `ARTICLE_MAX_CHARS`. `WATCH_TOP` and `POOL_TTL_DAYS` are in\n`collect.ts`. Prompt and tone live in `SYSTEM` in `summarize.ts`.\n",[],[271],["D",1787862662938],["D",1789545036283],{"_11":12,"_148":73},{"_11":12,"_143":144,"_72":73,"_30":71,"_145":146,"_147":24},{"_41":134,"_13":16,"_35":38,"_135":97,"_136":-5,"_137":-5,"_138":-5,"_139":-5,"_140":-5,"_141":14,"_142":12},"ctx",{"_56":-5,"_167":317,"_318":319,"_320":321,"_322":24,"_323":24,"_324":24,"_325":203,"_326":-5,"_327":-7,"_328":-7,"_329":26,"_330":24,"_331":24,"_332":-5,"_333":24,"_334":24,"_335":24,"_336":-5},"ENV",{"_288":289,"_290":291,"_292":16,"_293":294,"_295":296,"_297":298,"_299":300,"_301":302,"_303":304,"_305":306,"_307":308,"_309":310,"_311":312,"_313":314,"_315":316},"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"]

History · article.ts