What the RemNavi corpus actually pulls from — a source-taxonomy note.

For most of H1 2026 RemNavi described its index as drawing from "seven major sources." That phrase was inherited from an early-2026 internal note and propagated, with the best of intentions, into our State of Remote Hiring DRAFT and a few homepage copy passes. A careful reader of the v2 DRAFT noted that the source-count claim didn't reconcile against what /health.php and /jobs_api.php actually return. They were right. The methodology page itself had been updated to "90+ sources" some time ago; the surrounding copy hadn't caught up.

Here's the live taxonomy, as of the 2026-05-17 snapshot:

There are 9 platform-level sources contributing listings to the index: Jobicy, Remote OK, We Work Remotely, Remotive, Himalayas, Working Nomads, Arbeitnow, 4 Day Week, and the ATS layer (Greenhouse and Ashby), plus Lever where present. Together these surface 95 active feeds, when each named-employer ATS endpoint is counted as its own feed. Greenhouse alone contributes 33 named-employer feeds; Ashby contributes 9. The remaining feeds are board-level — a board is a single feed that returns the board's full inventory.

This matters for three reasons. First, the source mix is mostly ATS-driven rather than board-driven by listing volume: 4,826 listings flow through Greenhouse feeds and 1,355 through Ashby, against 88 from Remote OK and 100 from Jobicy. ATS amplifies the corpus far more than boards do, even though boards are what readers tend to picture when they hear "remote job aggregator." Second, the source mix is what it is because of where transparency comes from: employers who've adopted pay transparency tend to use Greenhouse or Ashby and to publish bands directly; boards that aggregate from those employers carry less structured data. Third, the source mix changes over the year as new employers turn on transparency or migrate ATSes — the named list on /methodology/ is updated quarterly to track this.

The takeaway for anyone citing our numbers: ground the figure in the methodology version and the data snapshot date. "RemNavi corpus 2026-05-17, RRS v1" carries forward through time and won't go stale even if the source mix changes. The methodology version page at /methodology/versions/ lists every published version of the rubric so any cited number can be reproduced against the rules in force on the date the data was taken.

This note is also a small commitment: we'll surface source-mix changes in subsequent notes so the index never drifts away from its published methodology again.