Repository navigation
Look before you send, and a dated answer to "does this broker's form still work?". Read Limits before you rely on either: both are narrower than the roadmap hoped.
New
aidr scan(F2): searches each broker that can be searched, for each person, and records found, not found or could not tell, with the listing URL and a private screenshot. It submits nothing and sends no email. It honours--onlyand--skip, and exits 2 when every search came back "could not tell".aidr run --found-only(F2): submits only where the latest scan found you.- A scan result counts only if it was made with the same search the run would do now, and is no older than
scan.maxAgeDays(default 30). - Only a broker whose search step has been validated can be skipped on "not found". Everything else (could not tell, stale, never scanned, cannot be searched, and the email and generic passes) follows
scan.unknown:submit(the default) orskip. - The run prints, per person, how many brokers it submits or skips and why.
- A scan result counts only if it was made with the same search the run would do now, and is no older than
- "Never seen" is no longer "removed" (F2):
aidr verifyreports a broker where no listing was ever seen apart from one where a listing was seen and is now gone. Only the second counts as a removal in the score and the report. A scan that finds you counts as listed, and a later sighting overrides an earlier "absent". - A weekly broker health check (F3): a scheduled job loads each explicit broker's opt-out page with a plain browser and checks that the page answers and that the form fields the tool relies on are still there. It never types, clicks or submits, and uses no personal data.
- Each broker gets
ok,drifted,blocked,deadorerrorwith a date.STATUS.mdis generated from that, and the README shows the count. - One issue is opened per drifted broker after two checks in a row, and closed when the broker is
okagain. aidr doctor --brokersruns the same check from your machine.
- Each broker gets
- USPhonebook can be searched, validated against the live site in both directions.
Changed
- A page must look like a results page before "not listed" is believed. It has to be on the same site as the search and name the person searched for. Otherwise the answer is "could not tell".
aidr verifymay now say "could not tell" where it used to say clear. - Radaris and PeekYou are removed. Both domains were transferred to another company by court order, and every address on radaris.com now shows the same notice, which the tool had been reading as "not listed" for everyone.
brokers.jslists 48 explicit brokers. STATUS.mdis generated byscripts/generate-status.js. Every broker used to be markeduntested.aidr verifyandaidr scanrefuse flags that mean nothing to them (--pollute,--resume,--solve-captchas,--snapshot,--fields) where they used to ignore them.- The dashboard returns only the host of a scan's listing URL, since the path carries your name.
Limits
- Few brokers can be searched. The roadmap aimed for about 20. There are six (Spokeo, WhitePages, FastPeopleSearch, MyLife, Nuwber, USPhonebook), and only USPhonebook's search has been validated, because the other sites refuse a plain browser and the tool does not get around bot checks. For most brokers a scan says "could not tell" or cannot search at all, so with the default
scan.unknown: submit,--found-onlychanges little today. - The first health check could not confirm most forms. From a home connection on 2026-11-17, of 44 opt-out pages: 12
ok(4 automated forms; the other 8 are manual brokers, where only "the page answers" is checked), 13drifted, 18blockedand 1error.driftedmeans the form the tool expects is not visible on the page; it is a prompt to look, not proof the opt-out fails. - The health check looks at the shape of the form. It cannot tell you whether a submission deletes anything;
aidr verifyis the check for that. - The scheduled job has not yet run on GitHub. Its first runs are the real test of the workflow file.
How it was checked
Written test-first. An independent review of the release branch found 1 P1, 9 P2 and 7 P3 issues before release; all are fixed, each with a test that failed first. A mutation check planted 24 bugs in the new rules: 21 were caught, 1 could not change behaviour, and the 2 that slipped through now have tests.