Most recently scraped:

If the computer turns off or restarts mid-scrape: no data is lost — every builder already scraped is saved to disk immediately, and the scraper always resumes from the next un-scraped builder, never from zero.

What it scrapes: real properties only, one at a time, in this order — the 43 verified seed addresses first, then this app's own real REA sold-listing pool (3,594 addresses across 38 QLD suburbs), then each page's own "Similar Properties" neighbours. Nothing is invented; an address with no Land Use formula estimate on file is marked N/A — not computed rather than guessed. Per property it captures the RP Data valuation band, bedrooms, bathrooms, car spaces, land size, and full sale/listing history, and compares the valuation to this app's own formula estimate for the same address.

The rules it runs under: nightly at 1am, once, unattended, using the count set above (capped at 200); also hard-capped at 60 minutes of actual scraping regardless of count, so it never runs for hours; 4–9s between properties with a longer pause every 20 (not a rapid-fire loop); stops itself outright if RP Data's own bot-detection has flagged it, with a real cooldown before trying again (shown above); logging in is never automated — clicking "Log me in" opens a real, visible Chrome window and waits for Brad to type his own password.

What happens with what it finds: once any one council reaches 30 real comparisons, its Land Use formula starts auto-adjusting from the real gap between formula and RP Data for that council — nothing further to build, it's just waiting on volume.