Research
What we tested and threw out.
Most betting ideas don't survive a fair test. Here are the ones we tried, including some members asked for and some we liked, and why none of them became a call.
Precision Sports Mathematics · Nov 14, 2026
The short version
- Our original halftime rules faded on recent seasons, so we retired them in October 2026.
- Basketball halftime has no edge at any trigger we tried. Those spots are shown as labeled leans, never bets.
- NHL failed twice: at the opening line and at the intermissions. No hockey calls.
- A college football Over looked great recently and failed on older seasons. Killed.
- Buying points isn't an edge by itself. On a weak pick it just costs more.
Every test below was written down before results were run, and scored on seasons the idea was never built on. Backtest figures are hypothetical, not results of actual bets.
Why publish the failures
A pick service that only shows what worked is showing you the lucky survivors. When you test dozens of ideas, a few will look great by chance. The only defense is to decide the test in advance, score it on seasons it never saw, and throw out what fails, even ideas we liked. This page is that list.
Our original halftime rules
HalfSharp launched on rules backtested on 2013–2023. In October 2026 we re-tested them on the seasons since. They didn't hold up:
- College football Under when the book moved 13+: 47.3% on 237 recent bets. Over every season on file, 52.4% on 2,265, which is break-even. Retired.
- College football blowout Under (lead 22+): 52.7% over 2,299 bets, all seasons. Break-even at best. Retired.
- NFL Under when the book moved 7+: 50.3% on 153 recent bets, 53.5% over 1,018 bets since 2007. Kept on probation only: about zero return at the −115 books charge at the half.
- The NFL "strong" tier (14+ moves): 18–30 recently. The label was dropped.
- Pregame college football Under: 45.6% on 125 bets at the closing line, where the app bet it. It still won 55.6% against the opener, which led to our opening-line research.
We replaced the college football rules with a selective model that passed on unseen seasons. Calls made under the old rules stay public on the Results page under Retired rules.
Basketball at the half
NBA and college basketball halftime totals showed no edge in either direction at any trigger, old or new. For example, the NBA Under at a 20-point move won 48.6% on 1,068 bets, and the Over 49.1% on 983. A full model topped out around 53–54% on its best spots, at or just under break-even at −115. So the strongest basketball halftime spots are shown in the app as LEAN, with their real rate and their own record, and never called bets. Basketball's real edge turned out to be at the opening line.
An NFL halftime model
The same selective approach that worked in college football found nothing in the NFL. It predicted worse than a plain average, and its most confident spots went 45–48% on recent seasons. The best NFL option, the Under after a 7+ move when the favorite leads at the half, went 55.6% on 333 unseen bets: suggestive, but not proven.
A high-total NFL Over
The idea: if a game expected to be high-scoring is low at the half, bet the 2nd-half Over. Over every season on file it went 70–66 (51.5%), below break-even. Low-total games in the same spot did almost as well (49.6% on 967). Books already set the 2nd-half line about 10 points above the 1st-half pace, so the "better number" isn't there.
A college football Over that looked too good
College football 2nd-half Overs after the book dropped its number 17+ went 63.5% on 96 recent bets. Tempting. But on the older unseen seasons (2007–2014) the same rule went 48.0% on 440. Combined, 50.7% on 536. After testing about 40 trigger and side combinations, one result this good is what chance produces. Killed.
NHL, twice
- At the opening line: two hockey models over about 19,000 games. Neither beat the opener on moneylines or totals by our standard. Hockey totals barely move (the opening total equals the closing one in about 80–87% of games), so there's little stale number to catch.
- At the intermissions (the hockey version of halftime), which members asked for: no rule beat break-even on unseen seasons. The best one won in 2024-25 and lost in 2025-26, which is what noise looks like. DraftKings prices the rest of the game about the way past seasons played out.
So there are no NHL halftime or opening-line calls. The app shows NHL scores, and the daily deep dive can add up to two NHL picks at 1 unit, made on research alone and counted in the Deep Dive record.
Buying points on its own
Buying 3 points on any NFL or college football 2nd-half total wins about 61–64%. Books know that and price it in: a fair price for 3 NFL 2nd-half points is about −172. Buying 3 points on the NFL Under rule won 58.6%, worse than buying on a random NFL Under (63.3%). On a weak pick, buying points just turns a coin flip into a more expensive coin flip. It only helps when the pick is already good, and at a price below what the extra wins are worth. The app's buy points calculator checks that price for you.
Related research
- The college football halftime model
- The opening-line edge
- Month one: the live record so far
- All research
See every live call on the Results page, or how the whole system works on How it works.
Backtest figures are hypothetical. They are not results of actual bets, and past results don't predict future results. No call is guaranteed. HalfSharp is an analytics and information service for adults 21+; we don't accept wagers. Gambling problem? Call 1-800-GAMBLER.