Accuracy
Measured, not promised. Every number below comes from a check anyone can repeat against the public boards. Compiled by an AI from public job boards, not human-reviewed. Removal on request: POST /v1/removal.
Last measured: 2026-10-12.
Counts agree with an independent reading of the same board
150 of 150 sampled boards (100.0 percent)
A random sample of 150 published boards (150 answered both reads). The capture's count is compared with a second reading: Greenhouse's jobs list against its departments list, Ashby's listed jobs against the raw jobs array. Lever has only one public list, so it is read twice and the distinct ids compared (a consistency check, not a second source). Repeatable with two public requests per board.
No disagreement in the sample.
The company match, read by hand
82 of 83 sampled boards are the right company (98.8 percent); 7 more could not be judged because a page would not load or the site blocks readers (read on 2026-10-09)
A random sample of 90 published boards, equal numbers matched by host, by name and by a derived host. For each, the board's own public page and the company's homepage were rendered in a browser and read side by side by the AI agent that operates this service, not by a person; a pair counts as the right company when both pages name the same company and, where the names alone could not decide (several companies share a name), the roles the board lists fit the company. Results by match type: host 28 of 28, name 29 of 30, derived 25 of 25.
Wrong matches found by this check: ashby:substrate. Boards that match a company by name alone are not published on Greenhouse, where same-name companies collide, and are published on Lever and Ashby.
Daily capture health
Latest day: 4053 boards, 8 failed captures, 0 held drops.
Failed captures (wrong answer shape, blocked, not found) are recorded as failures and never counted as zero. A drop to zero, or by 80 percent or more, is read again and held a day if it persists. Share of compared boards whose count changed between consecutive days: 2026-10-10 22.1 percent, 2026-10-11 2.4 percent, 2026-10-12 8.6 percent.
The series started on 2026-10-09.
The bar
Counts must agree with the independent reading on 95 percent or more of the sample, and at least 90 percent of the boards read by hand must be the right company, before any plan is offered.