Solonic
Level of explanation, for the entire site

Verify

Send us something to check.Send us a paper.
We'll check every claim.Submit an artifact for gated verification.Artifact in. Typed verdicts out.

We read the important sentences, look them up, and stamp each one — so you can see what's solid before anyone else does. Your writing stays private. Know where it stands before the world does. Every claim your argument rests on gets a verdict, the evidence behind it, and a trail you can re-run yourself. We don't publish your work or its verdicts. Load-bearing claims identified and tested against primary sources and internal logic by independent model families; verdicts typed, evidence attached, audit trail re-runnable. Artifacts and verdicts are never published. Intake below. Verdict schema PROVED · VERIFIED · OPEN · CORRECTED · UNSUPPORTED; nothing exits unlabeled. Artifact confidentiality: not published, not entered in the Ledger.

Send a paper See prices

What it costsFour gates. Pick one.GatesGates

A small check, a full check, a hard check — or a re-check after you fix things. If we're late, it's free. Each of the first three includes everything in the one before it. Miss the promised time and you get your money back. Cumulative depth across the first three gates; turnaround is an SLA with a full refund on breach. Times below are the single source for every mention on this site. SLA per gate, refund on breach. Canonical values mirrored in /s-claims.json.

  • Individuals

    Check the citations

    Jürge
    $49
    per paper · report in 2 hours

    Every reference resolved to a primary source and checked to say what you cite it as saying.

    • Every citation resolved or flagged
    • Quotation accuracy checked
    • Fabricated and hallucinated references caught
    • Report in 2 hours, or it's free
    Choose Jürge
  • Individuals · labs

    Check every claim

    Caliber
    $99
    per paper · report in 24 hours

    A verdict on every load-bearing claim, with the ones holding up your conclusion named.

    • Everything in Jürge
    • Claim-by-claim verdicts, tagged
    • Load-bearing claims identified
    • Report in 24 hours, or it's free
    Choose Caliber
  • Labs · funds · firms

    Attack the argument

    Caliber Deep
    $149
    per paper · report in 48 hours

    We attack the argument end to end and show exactly where it holds, where it needs work, and what would break it.

    • Everything in Caliber
    • Adversarial review by independent model families
    • A falsifier named for every surviving claim
    • Report in 48 hours, or it's free
    Choose Caliber Deep
  • Post-review revision

    Re-check after you revise

    Refit
    $79
    per revision cycle · scheduled at intake

    Revised after our review? Refit re-runs the gate diff-aware — cheaper than a fresh review because the diff bounds the work.

    • Diff against your prior report
    • New verdicts on changed claims
    • Regression check on unchanged claims
    • Updated audit trail
    Choose Refit
Standing engagements
Labs, funds, and firms who need this continuously should hire us rather than buy per paper.
Books and corpora
Quoted by scope. Tell us what it is and how long, and we'll quote it.
Privacy
We don't publish your work or its verdicts. The Ledger is our own work, not yours.

How the check worksHow it worksMethodPipeline

  1. You send it.

    A paper, a preprint, a report, or a whole book. PDF, Word, LaTeX, or Markdown. Pick how deep you want us to go.

  2. We check every claim.

    First the citations: does each source exist, and does it say what you say it says? Then the claims themselves, and which ones your conclusion depends on. At the deepest level we attack the argument and name what would break it.

  3. You get a report.

    Every load-bearing claim tagged with one of the five words, the evidence beside it, and a trail so anyone can re-run the check.

The checking is done by a panel of independent AI model families with different failure modes — correlated checkers don't add up — and every report names the families that actually sat. How the panel is built is in the methods, which are public.

Send us the paper

PDF, Word, LaTeX, or Markdown. We'll confirm by email within a day, and the clock starts when we confirm.

Where the confirmation and the report go. Never shared, never sold.

Up to one file. For a book or a corpus, send a zip or a link in the notes and we'll quote it.

How deep
We'll confirm scope and payment by email before work starts. Nothing is charged on this page.

Prefer email? Send the file to [email protected] with the depth you want in the subject line. Same queue, same clock, same report.

What a report looks like

A real excerpt. We ran our own biggest project through our own toughest check, and published what it found. An excerpt from a real Caliber Deep review — of our own flagship, the Superintelligence Problem map. We published what it found. Caliber Deep, self-applied to the Superintelligence Problem map, v2.0→v2.1. For a full multi-family review with per-family scores and spread, see the sample report on a philosophy paper. Excerpt: self-applied Caliber Deep. Full multi-family sample with per-family scores: /sample-review.html.

Caliber Deep · excerptArtifact: The Superintelligence Problem, v2.0 → v2.1
Unsupported

Each number links to its source. No hyperlinks existed in 120+ estimates; the citation layer was evidence-in-prose. Corrected in v2.1.

Open

Generation growth gai = 0.67. Rests on a single one-year extrapolation. The result is scale-invariant; the date is not. Shown as a band in v2.1.

Corrected

§1.5 branches declared conditional on ASI. Contained a plateau-before-ASI branch — incoherent. Relabeled as the full 50-year distribution.

Verified

Versioned estimates with visible v1→v2 deltas, two-method reconciliation, and pre-registered revision triggers. Rare, and correct.

Read a full sample report Read the map itself, errata and all →

Questions people ask

How long does it take?

Jürge: 2 hours. Caliber: 24 hours. Caliber Deep: 48 hours — each from the moment we confirm your submission, and each free if we miss it. Refit, books, and standing engagements are scheduled at intake.

What does "load-bearing" mean?

A claim your conclusion depends on. If it fell, the argument would fall with it. We name these explicitly, so you know which verdicts matter most.

Who does the checking?

A panel of independent AI model families with different failure modes, run through a published method. Correlated checkers don't add up, so the panel is built for diversity, and every report names the families that actually sat. The methods are public.

Do you certify that a paper is true?

No. There is no mark of truth to issue. We return verdicts with evidence and a trail you can re-run. What we guarantee is that nothing leaves unlabeled.

Can you check AI-written work?

Yes. Much of what arrives is partly or wholly machine-written. The checks are the same: do the sources exist, do the claims hold, and what would break the argument.

Is my work kept private?

Yes. We don't publish your paper or its verdicts, and we don't add them to the Ledger, which records only our own claims. Your report is delivered to you and to no one else.

What if I revise and want it checked again?

That's Refit: $79 per revision cycle — cheaper than a fresh review because the diff bounds the work — with a regression check on what you didn't change. Send the new version and say it's a revision.