Method
One rule set. Ten seasons. Nothing re-fit.
QuantSight's engine evaluates every Division I men's game daily using predictive ratings data and applies fixed statistical criteria to identify qualifying games. The criteria concern predicted scoring margins. They were derived once from historical data and have never been revised: no parameter was re-fit after that, in any season, including the live one. The same criteria that produced the ten-season historical record produce the daily publication, without modification.
The ratings inputs are commercially licensed and are not proprietary to us. The criteria applied to them are the licensed asset.
Fixed rules are not a parameter sweep
A common and justified criticism of published backtests is that they are the survivors of a search: test enough parameter combinations against history and some combination will look brilliant by chance. The shape of a record cannot prove how its rules were arrived at, and this page does not claim that it can. What the record shows is narrower and checkable: one rule set, applied unchanged to all ten seasons, with every season, every drawdown, and every losing stretch that rule set produced. Every figure is reproduced by our engine from raw game data. Nothing was re-fit between seasons. Every historical evaluation uses the ratings as they stood before tip-off on that date; no post-game or end-of-season information enters a historical evaluation.
How this record is kept
- Coverage
- Every Division I men's college basketball game with a model prediction is evaluated, every day of the season, against the same fixed statistical criteria. Games that do not meet the criteria are passed. A day when no game qualifies is recorded as a no-play day, not a gap. From 2026-27, a game scheduled to tip before the morning publication is outside that day's slate, and so is a game without a known tip time, so each published position goes out before its game's scheduled tip. The ten-season record evaluates every game, including games that tipped before a morning publication would have gone out.
- Selection
- Qualification is mechanical. No game is added or removed by judgment, including ours. If the criteria produce twelve positions, twelve are published; if they produce none, none are.
- Stake rule
- Every published position is sized identically at one unit. A parlay card is one unit. No position is ever sized up or down.
- Pricing basis
- Historical and 2025-26 units are computed at one fixed assumed price, identical for every position. From 2026-27, each published position carries the market price captured at publication, and its units are graded at that price.
- Grading
- Positions are graded from final scores. A published game that is not played on its scheduled date grades as void, zero units, and stays on the record.
- Line availability
- Whether the published line is still offered by the time a license holder acts is not yet measured. From 2026-27 each position records the market price at publication, and how often the line had moved before it could be taken is measured and published from that season.
- Publication
- From 2026-27, every published position carries a publication timestamp taken before tip. The record is append-only: rows are never edited and never deleted. Daily rows publish the date, the day's units, and the running total. Season tables publish complete counts and unit totals.
- Fingerprint
- Each published slate is fingerprinted in the same database step that publishes it: a SHA-256 hash of the day's positions and the publication time, chained to the fingerprint before it. The fingerprints are public. A license holder can hash the exact text of their part of each day, from the Picks page, and check its positions against the pick email received that morning. The fingerprints and their times are kept in our own database, on our own clock. The fingerprints →
- What we withhold, and why
- The selection criteria are the licensed asset. The record publishes what the criteria produced, not the criteria themselves. Results are the part a reader can check.
The risk, stated plainly
Worst season drawdown across all ten seasons: -28.2 units. Worst single day: -9.5 units.
Past performance does not guarantee future results.
Losing days are normal operation. The 2025-26 season finished at +195.2 units and still contained a -11.2 unit drawdown and a -4.9 unit day. A subscriber who cannot sit through that stretch without abandoning the rule will not experience the system's record; they will experience the stretch.
Source: the QuantSight results ledger. Data through Apr 6, 2026.
Units, not win percentage
The record's headline measure is units, not win percentage, for a reason. The system's positions are alternate lines priced far from even money: each win pays a fraction of a unit while each loss costs a full unit, so the win rate runs far higher than records stated at standard pricing and the two are not comparable. On that basis: 12,467-2,712 (82.1%) on qualifying singles across nine seasons of historical data, and 1,387-317 (81.4%) in the 2025-26 season. Historical and 2025-26 units are computed at one fixed assumed price, identical for every position. The units, the drawdowns, and the day-by-day ledger are the honest measure of what that record was worth.
Source: the QuantSight results ledger. Data through Apr 6, 2026.
The economic premise
The market. Division I men's basketball plays thousands of games a season across several hundred teams. Pricing effort follows betting volume, and volume follows the names.
The information. The inputs are a commercially licensed ratings model that anyone can subscribe to. The premise is not private information; it is a fixed rule applied without exception, every day, to a market that does not apply one.
Why alternate lines. A book's main number on a game is its most contested price. The premise is that the alternate ladder around it is set more by convention than by model. What the criteria select, and on what basis, is the licensed asset and is not described.
The constraints. A market that prices loosely also limits the size it will take and the accounts that keep winning in it. That is why the research is licensed rather than kept, and why the capital and discipline requirements are stated before the price.
What would undermine it. If books repriced the alternate ladder to the model, or withdrew it, the record would show it first and the ledger would publish it. From 2026-27 every position carries the market price it was published at, which is the measurement that would show a thesis weakening before a season's total does.
Why sell access
The honest version of "why sell it if it works" is capacity. The record was built in a constrained corner of the market: exposure across every eligible game, at lines a book will not take in unlimited size, from accounts that books limit. Licensing the research is the part of this that scales. Betting it is the part that does not.
What we refuse to publish, and why
The selection criteria themselves: thresholds, margins, and qualifying bands never appear on this site, in the app, or in support conversations. They are the licensed asset.
Daily pick counts and splits in the historical ledger: at one uniform price, per-day counts would let a reader reverse toward the criteria against public data. The daily ledger therefore publishes date, units, and running total; season tables carry complete counts.
Closing line value: the record measures realized units at published lines. We state that plainly rather than claim a benchmark we do not publish.
The question a careful reader asks
What happens on a day the criteria qualify twelve games and I do not have the stomach for it?
The system does not adjust. Coverage is all-or-nothing by design: the record is the product of taking every qualifying game, and a follower who skips games is no longer following the system, they are running their own judgment on top of it. This is the single most important thing to know before buying a license, and it is why the capital requirement is stated here rather than pointed at.
The system has required as many as 66 units of exposure in a single day. Sizing assumes a unit is a small fixed fraction of capital dedicated to this purpose, on the order of 0.25 to 0.5 percent. If a 66-unit day is more than you can hold to without deviating, this license is not for you.
Worst season drawdown across all ten seasons: -28.2 units. Worst single day: -9.5 units.
Every claim on this page points at the ledger. See the complete record →
Licenses run one full year from the day of purchase. Pricing →
The objections this page raises are answered head-on. FAQ →
Disclosures
- 1QuantSight Analytics is a statistical research service. It is not a sportsbook, does not accept or place wagers, and does not provide betting advice.
- 2Past performance does not guarantee future results. All figures are units, not currency, on the price basis stated above.