Finder
What this doesOnce a month we ask Brave Search to find Common Data Set pages for schools that still have no listing. If this goes quiet, we stop discovering new files.
Every dispatch station on collegedata.fyi, when it last ran on schedule, and what it found. If a step goes quiet, the lamp turns red — including the monthly search that sat idle from April to August 2026. Locked doors (Tableau, SharePoint IRM, SSO) ship in the next slice; this page is the clocks.
What this doesOnce a month we ask Brave Search to find Common Data Set pages for schools that still have no listing. If this goes quiet, we stop discovering new files.
What this doesMany seeds point at a single old PDF. This step re-checks those schools so we can find the office's actual CDS listing instead of one frozen year.
What this doesWhen a PDF seed is really a landing page — the office's CDS index — we rewrite it so weekly archive can collect every year they publish.
What this doesEvery night we decide which schools are due for a fresh look and put them on the archive queue. Schools we already checked recently are skipped.
What this doesAbout every 30 seconds we take one school off the queue, download what we can, and archive it. This is the machine's pulse.
What this doesOnce a day a real browser opens the listings that block ordinary downloads, finds new Common Data Set files, and archives them. This is how WAF-gated schools stay current without a person clicking Save.
What this doesAfter a file is archived, we pull the numbers out of it. The daily job only drains a handful, so a backlog here is late — not necessarily a crash.
What this doesHourly we rebuild the public coverage ledger: which schools have a current CDS, an older one, or none we could find.
What this doesHourly we refresh the homepage and API caches so the public site is not reading live tables on every visit.
What this doesOnce a month we check whether NCES has posted a new IPEDS Access database. A quiet "no new release" is still a healthy tick.
What this doesOnce a year an operator loads the new CDS year schema. Gray until that happens; red only if it has been more than 18 months.
What this doesOnce a year an operator loads College Scorecard. Gray until that load is recorded; red if it is more than 18 months old.
Manual sources (not on the strip): Directory enqueue — no heartbeat · Mirror ingest — no heartbeat
Each station writes a public heartbeat when it starts and when it finishes. Lamps are computed at read time from the scheduled clock. A manual GitHub run cannot green a missing cron. If the scheduled heartbeat is older than the SLA, the lamp is red — including when a dispatch succeeded yesterday. Missing data is red, not gray.
This page is not coverage. Coverage asks whether we have a CDS. This board asks whether the machine is running. JSON snapshot: /pipeline-observation.json.