Chevy's Picks

Three picks a day. A public record. Picks that clear the evidence bar are flagged verified. The rest are pressure picks, forced action, labeled on the board and tracked in their own record. Every published pick stays on the board, wins and losses.

0-0
Combined Record
$1,000.00
Bankroll
--
Win %

Demo money. The bankroll started at $1,000, every pick is $100, and every pick is posted before the game starts, graded in public, and never edited. The full record is below.

Today's Card

Day 1 · Saturday, October 17, 2026

Past Days

Every earlier day and its 3 picks, graded. Slide sideways to look back through the cards.

Full History

Every pick ever posted, newest first. Days 1-2 are the legacy card; their odds are labeled unverified (see the house rules).

Verified Record

Verified picks only, standing on their own. No pressure picks and no legacy picks mixed in.

0-0
Verified Record
$0.00
Demo Profit
--
Win %

The house rules

Every pick is posted before the game starts and it is never edited after the fact. Wins and losses both stay on the board, and every one of them is counted in the records on this page.

How it works: three picks go up every day, in two tiers. Verified picks clear the full evidence checklist; pressure picks are forced fills, labeled as such on the board. Every pick is $100 out of a demo bankroll that started at $1,000, graded at its exact price, and the full history stays public.

These picks are tracked content for entertainment. Chevy places no real-money bets. Tail or fade at your own risk.

Inside the Machine

This is not a guy with a feeling about a game. It is a pipeline: ingest everything, price it honestly, grade it in public. Here is what actually runs before a single pick is allowed on the card above.

01 · The Archive

Every market is captured on both sides, timestamped, across the major books: 74,370 prices and counting. When the machine quotes a number, it can point to the exact pull it came from, the book that posted it, and the minute it moved. Opinions expire. The archive does not.

02 · Fair Price, Derived

A book's price is not a probability; it is a probability with the house's cut baked in. The engine strips the cut out properly: both sides of the same market at the same book are de-vigged as a matched pair, incomplete or stale pairs are discarded, and fair probability is the median across books. What survives is what the market actually believes.

03 · A Play-by-Play Brain

The NFL model is built from 13,179 filtered plays of play-by-play data, expected points added on offense and defense, resolved to the matchup level. Its probabilities are calibrated the honest way: walk-forward, out of sample, with the spread of its own past errors (12.47 points across 75 graded forecasts) deciding how much confidence any single number deserves. No backtest ego. The error bars are measured, not imagined.

04 · Watchers That Never Sleep

Injuries come off the official NBA report. NHL starting goalies are tracked as confirmed or projected. Kickoff weather is pulled from the National Weather Service. Rest and travel differentials are computed from the schedule itself. Structured feeds, all of them rechecked minutes before anything posts, because a lineup card beats a hot take every time.

05 · A Control Group for Itself

A second, independent model runs in permanent shadow: frozen specification, graded on the same finals, scored with Brier score and log loss against the market's own probabilities, and forbidden by rule from influencing the card. Most handicappers grade their feelings. The machine grades its own math against a control, and publishes the comparison either way.

06 · Protocol Over Instinct

Settlement rules are written before the games, never after. Every pick is refreshed at its exact named book immediately before publication, rechecked against injuries, goalies, inactives, and weather, and a pick whose game has already started never posts. Grading is exact price math, return on investment is reported separately from bankroll growth, and nothing here is edited after the fact. Even the model's first version, retired for a bug, stays in the log labeled invalid.

A human expert can watch a few games a week and remember a few hundred prices. This machine ingests the entire board, prices it against itself, cross-examines its own forecasts, and shows its work down to the pull, the pair, and the play. Not a hotter hand. A bigger brain, with receipts.