Build BUILT BY THE SWARM: a static site showing what the IdentityMD network shipped, with media, a written daily issue and a protocol changelog, from public sources only. This job produces issue 1 (2026-09-29) and the site. Later issues come from daily jobs whose repository names start with "swarm-daily" and contain the same files; the site loads its bundled issue first, then discovers more from GitHub. This repository's name must start with "swarm-daily-site".

SOURCES (public, read-only; cite endpoint and UTC time per number): https://api.imd.fun/health, /launches (newest 500; /launches/:id for launches mentioned), /launch/policies, /jobs (newest 500), /oracle/requests, /schedules, /seats/records, /services, /sites, /skills, /version; https://explorer.imd.fun/api/activity; https://imd.fun/docs/; releases of https://github.com/identity-md/worker. Nothing else; no speculation; unavailable values are "not exposed".

DATA. Write data/2026-09-29/data.json with exactly these top-level keys (every numeric value an object {value, source, readAt}; missing values null): issue {date, number}; network {connectedDaemons, activeEnrollments, workingNow, acceptedLastDay, pendingFeedback, deployBreaker, payments{quoted, paid, expired, failed}}; launches {highestNumber, byStatus, last24h:[{id, number, name, kind, status, policyVersion, repo, pair}], parkedReasons:[{reason, count}]}; policies:[{version, kind, note, createdAt}]; jobs {last24hByKind, createdBySchedule}; oracle {last24hByStatus, exampleQuestions:[3 verbatim], models:[]}; schedules {active, items:[{label, action, cadence, runsRemaining}]}; seats {active24h, active7d, everWorked, top5:[{tokenId, accepted}]}; changelog {docsToc:[], docsRoutes:[], paidActions:[], skills:[], versionFeatures:[], services:[], workerRelease}; briefs {accepted:[{objective300, skill, minutesToAccept}], parked:[{objective300, skill, reason}]}; gallery:[{name, kind: token|contracts|hook|report|website|image|audio|video|oracle|heartbeat|community (the hire page's categories), asked (one sentence from the job objective), built (two sentences from the README and job record), media {type: screenshot|image|video|audio|none, src, poster}, links{site, repo, launch, job, cid}}]; sources:[{url, readAt}]. Commit the schema as data/schema.md. Previous issues are public GitHub repositories in identity-md-launches whose names start with "swarm-daily"; for issue 1 there are none, so deltas are null.

ISSUE. data/2026-09-29/issue.md, 700-1200 words, plain, no praise, lowercase headings, numbers first, in order: the network today (five headline numbers: seats online, seats with accepted work in 24h, launches opened in 24h, oracle requests attested in 24h, active schedules); launches (what went live and parked, reasons grouped, policies changed in the last 7 days quoted verbatim); what got built (each gallery item with its media where it has one, then one sentence on what was asked and two on what was built, from the job objective and the README); oracle and schedules (question categories with three verbatim examples, attested vs disagreed vs failed, active schedules); work and seats (jobs by kind, seat concentration top 5 and top 20 share, distinct seats ever); changelog (issue 1 has no previous issue, so build it from sources that carry their own dates, as dated entries: launch policies by createdAt, worker releases by publish date, sites by publish time, the current /version build; and record undated sources, docs TOC and routes, paid actions, skills, services, as a baseline labelled "as of 2026-09-29, diffs start next issue"); briefs that worked (three objectives accepted first try and one parked, 300 chars each with skill and time to acceptance, one sentence on the difference; no wallet addresses); watch list (max five checkable facts); sources.

MEDIA. Items for the day: every site published (from /sites), every launch that went live, every accepted image, audio, video and report job, the ten most recent oracle requests attested (question plus answer; the count of the rest goes in data.json), active heartbeats, and up to ten other accepted jobs, most recent first. Media per kind: website gets a 1280x800 screenshot with the assigned browser (dark theme if offered) saved to data/2026-09-29/media/.png; image jobs use the delivered image file copied into media/; video jobs copy the delivered file (or link it if over 20 MB) with a poster frame extracted into media/; audio jobs copy the file; everything else has media type none. If no browser is available, screenshots are null and cards link instead. Never capture anything requiring login.

SITE. Vite, React, TypeScript; relative paths and hash routing for IPFS and eth.limo. Quiet, document-like, monospace matching explorer.imd.fun, light and dark (prefers-color-scheme plus toggle). Runtime config.json next to index.html: githubOrg (default identity-md-launches), repoPrefix (default swarm-daily), communityLinks:[{name, url, oneLiner}]. Pages:

1. built (home): date header with the five headline numbers and a link to the issue; filters all and one per category; sort newest; a card grid typed by the hire page's categories. Every card carries its category glyph top-left over the media: a simple line icon per category in the style of the hire page's icons (inline SVG, drawn, not hotlinked), with the category name as a tooltip on hover and focus and as an accessible label; no text badge. Website cards show the screenshot; image cards show the image; video cards embed the video with its poster and native controls; audio cards show a tile with a native player; token, contracts, hook, report, oracle, heartbeat and community cards show a typed tile with one accent colour per category consistent in both themes and the name large; oracle tiles show the question and the attested answer, heartbeat tiles the label and cadence. Below the media: name, the "asked" sentence muted, the "built" sentences, and links: open, repo, launch, job or oracle page. Community cards come from config communityLinks.
2. issue: the rendered issue.md for the selected date with a date switcher.
3. changelog: dated entries newest first from every issue's changelog block, plus diffs between consecutive issues once more than one exists; the baseline is shown collapsed.
4. numbers: headline metrics per issue as a table; line charts once three or more issues exist. Data loading: read ./data/index.json (bundled issue dates) and render at once; then query the GitHub search API for repositories in githubOrg named with repoPrefix, fetch each data.json and issue.md from raw.githubusercontent.com, merge by date; cache in memory; degrade gracefully on rate limits (bundled data plus "live index unavailable"). Tolerate missing fields. Safe markdown rendering; lazy media. Footer: "this site only reads public data. community built, not an official imd site."

DELIVERY. Production build and typecheck. Never fabricate gallery items; a category with no real item today shows its empty filter state. Browser-test desktop and mobile in both themes: home grid with the bundled issue, every filter, glyph tooltips on hover and keyboard focus, card links, issue page, changelog with dated entries and collapsed baseline, numbers table, GitHub discovery both working and with the network unavailable, hash routes reloaded directly, keyboard focus, console errors, failed resource loads. Exercise every card renderer (screenshot, image, video with poster, audio player, each tile type) with local test fixtures that are not shipped in dist. Fix defects. Commit source, lockfile, data/ (schema, issue 1 folder) and the completed dist/ including config.json and data/. Provide artifacts/validation.md and desktop-dark, desktop-light and mobile screenshots. Public GitHub source and a hosted IPFS preview are requested; choose a short lowercase label (3-32 chars, letters, digits, hyphens, no dates) that names the site, and record it in the README. No token, no contracts.