Watch QA Reef turn a real county records portal into a live API.
The button below runs the real thing, right now: QA Reef opens the public septic-records portal of Garfield County, Colorado, walks to a live septic case, reads the parcel number from the county's own metadata, asks the portal to export one real document, and captures the PDF off the wire, verified by its magic bytes, then discarded. No browser, no login, no CAPTCHA, no tricks: the whole run is plain HTTP against a portal any citizen can open.
This is our own workload, dogfooded. Parcel Perfect, the septic-records product built by the same team, harvests county septic archives at scale using exactly this engine. The demo below is one polite, single-document slice of that production flow.
A recorded run (15 seconds): one real Garfield County septic PDF, magic-byte verified. Or run it yourself below.
Run it against the live portal
Target: records.garfield-county.com/WebLink → Community Development / Building / Septic (5,800+ public OWTS case files). A real run takes 10 to 30 seconds; repeat clicks within a few minutes replay the cached run instead of re-hitting the county.
What actually happens when you press Run
The demo endpoint on app.qareef.com performs the same sequence QA Reef's harvest engine runs in production, over plain HTTP, because the portal's AJAX API needs no JavaScript at all:
- Warm a session on the portal's Septic folder, same as opening it in a browser.
- List the live folder and pick a real septic case (a permit-numbered folder like
SEPT-08-26-…). - Resolve the parcel from the case's own metadata, the county's
ParcelNumfield, never inferred or guessed. - Ask the portal to export one document as PDF, poll the export job, and capture the finished bytes off the wire.
- Verify the capture starts with
%PDF(magic bytes), hash it, measure it, then throw it away.
If any step fails, you see the real failure. Nothing on this page is simulated or pre-recorded.
The same run, as an API
This is the product point: a portal with no API becomes one. The response includes the exact API-mode shape, the single GET that stands in for the whole portal session:
GET /county/co-garfield/lookup?parcel=239328464001
→ { "parcel": "239328464001",
"documents": [ { "title": "…",
"content_type": "application/pdf",
"bytes": 3121844, "sha256": "…" } ] }In production this is a QA Reef Website → API flow: recorded once, exposed as an HTTP endpoint, with the browser lane kept as ground truth.
The honest fine print
- Public records only. Garfield County publishes these septic (OWTS) case files for anyone to browse, no login, no CAPTCHA. The demo touches nothing gated and circumvents nothing.
- One document per run, then it's gone. The demo captures a single PDF, proves it's real (magic bytes, size, SHA-256), and discards it. It never bulk-downloads and never stores county documents.
- Deliberately polite. Runs are rate-limited, requests inside a run are spaced out and capped, and repeat clicks are served from a short cache so a busy demo page never hammers a county server.
- Failures are shown, not hidden. If the portal is down or returns something that isn't a PDF, the demo says exactly that. An honest error is part of the product, QA Reef's verdicts distinguish "it failed" from "we couldn't measure".
- The parcel number is read, never invented. It comes from the county's own metadata field on the case file. No-fabrication is a hard rule across everything we ship.