About Upside Down Research
Upside Down Research
We build software that is reliable, correct, and powerful. No simplifications. No hand-waving. The hard problems are where the value is, and we do the work to solve them right.
Upside Down Research LLC is based in Spokane, Washington.
Why this exists
Human-sourced information is messy. LLMs make it worse by projecting confidence where none exists. Everyone nods along because the output sounds plausible. That's not good enough when real decisions are on the line.
The question was whether software could reason in more than true and false, return answers such as "likely", "disputed" or "not enough evidence", and leave a record an analyst could audit. It can. The work began more than 15 years ago, when the company's principal did graduate research across hundreds of source articles. That research required tracking how information changes, conflicts and ages from one source to the next. In early 2026 those lines of work became a single method. Arbiter is the result: software for questions without a certain answer, which records how it reached each verdict.
What we measure
We measure the sources you give us against what they say. We do not adjudicate the world, and we will not pretend to. Arbiter does not go looking for evidence on its own — it processes only the actors, claims, and evidence submitted to it. Every verdict it returns is a statement about that source corpus — who said what, how reliable they are, how well their claims hold together — not a ruling on what actually happened. That is a narrower promise than "we fact-check the world," and it is the one we can keep.
Who
Paul Nathan
Principal — M.S. Computer Science
Designed Arbiter and founded Upside Down Research. Has spent 15 years studying how information changes, conflicts and decays, and what can be proved about it.