About
Who made this, and why.One person, a lot of machines, and one rule: show your work.A verification company, from first principles.Operator and provenance.
A professor who spent twenty-five years studying broken ideas built a company that finds broken claims. Solonic is a verification company. It checks claims — in the papers people send us, and in the research we publish ourselves — and keeps a public record of when it was right and when it was wrong. Founded by Kevin Scharp — Professor of Philosophy, UIUC; author of Replacing Truth (OUP, 2013). Conceptual engineering, industrialized. Human principal: Kevin Scharp, UIUC. Generation and verification separated by design; model families named per report. Public record: /ledger.html, /errata.html.
What Solonic is
People send us papers and books. We check every claim in them and show our work. We also publish our own research — maps, clocks, censuses — checked the same way, for free. And we keep a list, in public, of what we got right and what we got wrong. Two lines of work. Verification as a service: per-paper reviews and standing engagements, with typed verdicts, evidence, and a re-runnable audit trail. And open instruments: the methods, the Atlas, the censuses, the packages — public because verification that requires secrecy is not verification. Two lines of work. Verification as a service: per-artifact gates and standing engagements — typed verdicts on load-bearing claims, evidence attached, trails re-runnable by third parties. And open instruments: methods, corpora, and packages published under the same gate. Formal proof handles the claims that can be formalized; almost everything that matters — evidence quality, citation rigor, reasoning chains — cannot, and that is where we work. Our founding bet, still tagged OPEN in our own Ledger: verification, not generation, is the binding constraint on knowledge. Service: per-artifact gates (/verifier.html) and engagements (/hire.html). Instruments: index at /research.html. Founding bet tagged OPEN in the Ledger: verification, not generation, is the binding constraint.
Kevin Scharp
Founder · Professor of Philosophy, University of Illinois Urbana-Champaign · kevinscharp.com
Kevin has spent twenty-five years finding ideas that don't quite work and building better ones. His book Replacing Truth argues that even the idea of "truth" is broken, and shows how to fix it. That sounds abstract, but it's exactly what checking is: finding where an argument breaks, and saying so precisely. Solonic does that at scale. Kevin's work is on defective concepts — concepts that are internally inconsistent and must be replaced rather than repaired. Replacing Truth (Oxford University Press, 2013) argues that truth itself is one of them, and constructs two replacement concepts that do the work without the paradoxes. The same method that diagnoses a defective concept diagnoses a broken claim in any domain; Solonic is that method, industrialized. PhD in Philosophy, University of Pittsburgh (under Robert Brandom); undergraduate degree in mathematics. Replacing Truth (OUP, 2013) argues truth is a defective concept and constructs ascending and descending truth as replacements; Semantics for Reasons forthcoming. Twenty-five years on conceptual engineering — the systematic design of better concepts — and, recently, AI conceptual disruption, agent moral status, and the philosophy of verification. When a concept is defective, every argument built on it inherits the defect; identifying the defect is verification. Principal's method: conceptual engineering — inconsistency identification, maximal consistent subsets, replacement construction — applied as claim verification. Bibliography at kevinscharp.com.
Why "Solonic"
Named for Solon, who posted the laws of Athens where any citizen could walk up and read them — the first act of making power legible. Solonic checks claims and shows its work, not because transparency is a nice value, but because verification that requires secrecy is not verification.
Four rules
- No mark of truth
- There is only evidence, and it points both ways. Our gate never certifies. It tags, tracks, and corrects.
- Every claim carries its status
- PROVED, VERIFIED, OPEN, CORRECTED, UNSUPPORTED — and, on our own Ledger, REFUTED for claims the evidence turned against. Untagged output is a bug, not a result.
- Methods are free
- Judgment is the product. The methods are published because verification that requires secrecy is not verification.
- The moat is the record
- Not secrecy — a public track record of being right, and of saying so when we were not. See the Ledger and the errata.
Who writes the research
Most of what we publish is written by AI systems and checked by other AI systems, with one human in charge who signs off. We say which is which, and every item shows its status. Most of what we publish is drafted by AI systems and checked by other, independent AI systems, with one human directing the pipeline and signing off. Every report and instrument names the model families that sat. Generation and verification are separated by design: one model family drafts; independent families with uncorrelated failure modes check; the output carries the verdict of that check, including when the verdict is failure. Human oversight: Kevin Scharp, UIUC. Pipeline: generator family ≠ verifier families; verdict travels with output. Oversight: human principal signs off. Family roster named per report.
Contact
- Anything at all
- [email protected]
- Verify something
- solonic.ai/verifier.html
- Hire us
- solonic.ai/hire.html
- Where
- Champaign–Urbana, Illinois