Three studies sit behind this desk. Each one asks a single question, shows the measurement rather than the conclusion, and states what would have to be true for it to be wrong.
One of them failed, and it is still here. A strategy search that only publishes its wins is not a search — and a desk that hides its dead ends is asking you to trust the ones it kept.
Each page follows the same five steps: question, method, sample, verdict, caveats.
Live — feeds the desk
Can a fight model beat the Polymarket price?
The study behind the signal the desk publishes. A model scores each moneyline from fight history alone, and its own guards veto most of what it likes.
Closed — negative result
Is Polymarket cheaper than the sharp books?
No — and the interesting part is by how little. The disagreement with Pinnacle is real, and smaller than the spread and fee it costs to collect. A negative result, published on purpose.
Open — sample too small
Can a market maker earn the spread on a prediction market?
Unanswered. The stack works end to end on paper — recorder, simulator, quoting models — but the sample is far too small to separate the strategy from variance.
Every conclusion on these pages is computed from the underlying ledger when the page is built. If a later run overturns a finding, the page changes with it rather than keeping a sentence someone typed once.
All three studies are read-only and paper only. No wallet, no keys, no order has ever been placed. Costs are measured from real order books, but no capital was at risk.
Each study states its own evidence level, and two of the three are honest that they have not measured enough to conclude anything. Treat a thin sample as thin, whichever direction it points.
The desk itself is a separate thing.
These pages explain the reasoning. The signals the desk publishes, and their graded record against the closing line, live in History.