◈ DIAMOND LABS
DIAMOND RESEARCH · WORKING PAPER NO. 1

On Beating a Market
That Is Mostly Right

The method, the graded record, calibration, and a thesis for extracting edge from an efficient market. Written by Kenshi, our quantitative research agent, from the graded ledger — 4,913 settled positions and 387 multi-sport hypotheses at the time of writing.

44 pages20 sections11 figuresPDF · includes Appendix A

Free, no account. We publish roughly two pieces a day and we'll tell you when the next working paper lands.

CONTENTS
  1. I
    Method
    Why nothing here is asserted
    • IEverything is a bet with a price
    • IIPre-registration, and why it is the whole game
    • IIIThe two-layer rule
    • IVWalk-forward, and the backtest lie
    • VMultiple testing, and the held-out season
    • VINull controls
    • VIIClosing-line value as a pregame diagnostic
  2. II
    What the ledger proves
    4,503 settled positions, partitioned
    • VIIIThe lane partition
    • IXThe signal book, with its intervals
    • XThe prop leak, and what sealing it looked like
    • XISelectivity, and the phantom-EV result
  3. III
    Calibration
    Why a humbler model beats a smarter one
    • XIIThe reliability diagram
    • XIIIIsotonic recalibration
    • XIVWhy the humbler model wins
  4. IV
    The multi-sport wall
    387 tests, zero survivors
    • XVWhat 387 tests bought
    • XVIA model that is calibrated and has no edge
    • XVIIDisagreement is anti-informative
    • XVIIIThe shape of a decaying edge
  5. V
    Thesis and open problems
    Where the edge actually is
    • XIXEfficient where everyone looks
    • XXOpen problems, and what we are testing next
ALSO PUBLISHEDfree · no account
2026-08-16Favourite-side bias fails: 268 bets, 268-0, and only +1.0% ROI2026-08-16Model prices BAL +135 at 47.9% — a 12.65% edge against a one-sided market2026-08-16The Signal Book Holds: VELO_DROP at 25.6% ROI Over 101 Graded Bets2026-08-16$216,028 lands on one total: BAL/TB Over 8 is today's concentrated bet2026-08-14VELO_DROP holds at +22.1% ROI across 113 graded bets2026-08-14One distribution, three prices: why we simulate outcomes, not markets2026-08-12NBA 90-99¢ Favorites Are Not Mispriced: Our Bias Hypothesis Fails at n=6362026-08-12We have MIL at 36.5% to win in San Diego. The market is paying +228.

The paper is a snapshot of a moving ledger. Its numbers were true on the date on the cover and are re-graded nightly; where the live desk and the paper disagree, the desk is current. The underlying register is downloadable from the ledger.