Findings
Ruled studies — each with a claim, a falsifier, and the method to check it. This page fills as findings clear review; the machinery that feeds it is live, the editorial gate is deliberate.
Read: The 27% problem — and why I published 55% instead
The write-up behind the census: a sample said 27% of registry servers answer; the full census said 55%; the correction is the story. Pre-registered demand test inside.
One host, 1,312 names — a multi-tenant gateway RS-F-002
Of every 100 live endpoints in this ecosystem, 22 resolve to a single host — gateway.pipeworx.io — carrying 1,312 registry names. But 1,266 of those names serve distinct tool inventories: this is a multi-tenant gateway / aggregator — concentration, not duplication. A liveness checker scores each name 100 out of 100; only collapsing the names onto their shared host sees how concentrated the ecosystem's live surface really is.
Classification, not accusation — a measurement of where endpoints live, no intent asserted. Falsifier: if the 1,312 names resolve to materially different hosts/operators (not one gateway), or the host's share of live endpoints falls below a stated threshold on re-probe, this is revised. (Corrected 2026-08-14 from an earlier "costumes" framing — see corrections.)
The registry is one day's truth, and it moves RS-F-001
Same instrument-maker, better instrument. The gap is the finding: unversioned one-off measurements of this ecosystem cannot be trusted — including ours. See corrections.
Falsifier: a re-run census within the disclosed method landing outside 55% ± its stated bound would itself be published.
…but the costume farm serves real HTS data RS-F-003
Generic names, correct numbers. A name check reads this server as noise; a liveness check reads it as perfect; only re-deriving the answer reads it right. The /hts slug wears the ask_pipeworx family yet returned the correct USITC rate 14 of 14.
Falsifier: any pipeworx/hts duty answer diverging from the USITC general rate for the same heading revises this. See truth checks.