Research
Long-form investigations from our lab — including, especially, the ones that didn't work. Negative results are the most expensive content in this industry, because nobody selling picks can afford to publish them. We can.
The A/B/C Model: Full Daily Archive, and Why We Retired It
Retired August 7, 2026
Four and a half months of daily sharp-money model results, every slate it ever graded, and the expected-value gate reading that shut it off. The Grind used to be this; it now follows the handicapper experiment.
Crossing Two Experiments: We Pointed the AI Panel at the Touts. So Far It Says Every One of Their Sides Is Overpriced.
Thursday, August 6, 2026
We already audit X's handicappers with their own timestamps, and we already run a three-model AI panel over the bets our rules fire. This is what happens when you run one through the other: cappers propose, the panel scores, and a third arm — every other game on the board — keeps both of them honest. Paper only, and the early disagreement is unanimous.
The Consensus: Three AIs Argue Over Every Bet We Fire. Can Their Score Predict Our P&L?
Sunday, August 2, 2026
Every play our mechanical rules fire now gets scored 1-10 by an AI panel before first pitch: a research pass, an adversarial debate — one model argues the dog, another argues the favorite — and a third model ranks the confidence. Paper-only, leak-guarded, and the test isn't win rate: it's whether the ranking correlates with profit over 100 plays.
We're Auditing X's Handicappers With Their Own Timestamps. Records Are Easy to Fake — Closing Line Value Isn't.
Sunday, August 2, 2026
A tracker that pulls handicappers' timestamped picks from X and YouTube, re-grades every one independently (never trusting a ✅), and measures them against our own closing-odds database. The goal isn't to find someone to tail — it's to find out which betting angles, if any, carry real closing line value. Early returns: the crowd of touts is a fade, not a follow.
Two Models, One Ledger: How an Audit and a Bucket-by-Bucket Drill-Down Narrowed Our Whole System to a Single Niche.
Saturday, July 11, 2026
We handed three months of our own betting ledger to GPT-5.6 for a forensic audit, then had Claude (Fable 5) drill into every bucket — market, rule, price band, home vs. away — to find where the profit actually lives. Most of it didn't survive. What did is now the only thing we bet live, preregistered, with a public tripwire that reverts the whole change if the paper trail beats the money.
Does the Sharp-Money Playbook Cross Sports? A Live WNBA Experiment.
Monday, June 29, 2026
We extended the baseball machinery to the WNBA — loaded a season of betting splits, ran two independent signal hunts (ours and an outside model's), and found a market that's mostly efficient with one stubborn, contrarian candidate. Now we're forward-testing it at $5 a game, in public, win or lose.
Does the Path Predict the Outcome? We Asked the Wrong Question and Found a Better Answer.
Monday, June 15, 2026
A market sitting at 60¢ that climbed from 20¢ and one that fell from 80¢ feel different — does the trajectory predict resolution beyond the price? We tested it on ~4,800 settled Polymarket markets. The momentum hunch was wrong. But the calibration table showed something that survived every check we could throw at it — and one honest wall still stands.
We Went Looking for Sharp Money on Polymarket. Here's What $80M of On-Chain Tape Told Us.
Thursday, June 11, 2026
Sportsbook betting splits live behind paywalls. Polymarket is a transparent exchange where every trade is public, forever. So we asked: can the blockchain replace — or beat — the sharp-money signal? One day, three experiments, ~500,000 settled fills. The answers surprised us twice.