National Bank Open 2026: 131 Flagged Picks, +17.2% ROI, the Full Graded Ledger

The National Bank Open is in the books. Ben Shelton defended his Montreal title, the first man to do that in Canada since Rafael Nadal in 2019, beating Brandon Nakashima 6-3, 7-6(4) without dropping a set all tournament.

Our ledger closed with the trophy. Across the ATP draw from August 1 to 14, the model flagged 131 bets. They went 73-57 with one void, good for +22.40 units at flat 1-unit stakes, a +17.2% ROI over 130 settled picks. Every one was recorded before its match and graded after, including the losers. The full 131-row ledger is published with the announcement on @PredictiumAI.

A few conventions so the numbers can't hide anything. Every pick is staked at a flat 1 unit. A win pays the decimal price minus one, a loss costs one, and a void (a retirement or walkover before the bet has action) returns the stake and comes out of the denominator. ROI is units divided by settled picks. When both sides of a market crossed our threshold at different times as prices moved, each entry was recorded the first time it crossed. Nothing is retro-picked, and nothing is graded at a better price than the one we logged.

The headline belongs to the least glamorous row. 110 settled moneylines at +11.6% is what carried this tournament: a barely-above-water 57-53 record that still cleared eleven percent, because the model's edges tend to sit on prices the market undervalues rather than on coin-flips it happens to like. A 51.8% hit rate that pays +11.6% is what edge-hunting looks like when the sizing and prices are honest. For contrast, a 70% hit rate on heavy favorites can easily lose money.

Spreads and totals went a combined 16-4 for +9.59 units, a 48% ROI. We're not leading with that, for two reasons.

First, twenty settled picks proves nothing. Anything can happen in twenty bets. Those rows ride along with the ledger; they don't headline it.

Second, our own longer record disagrees with the hot streak. On the live ATP scoreboard, which tracks every flagged pick across events, spreads are actually negative all-time: down 0.7 units over 51 settled. Totals sit at +8.7% over 43. Montreal's 16-4 fortnight is real and it's graded, but the longer ledger says the jury is still out on those markets. We'd rather tell you that next to the good news than let a two-week sample impersonate a track record.

Tournament recaps are snapshots. The permanent record is the closing-line-value scoreboard on the tennis backtest page, which logs every flagged pick with its entry price, the closing line, and the result. A tour's headline record stays hidden until it reaches 200 settled picks, so no tier can debut on a lucky streak.

Montreal pushed the ATP tier past that gate. It now stands public at 215 settled: 112-103, +25.0 units, +11.6% ROI on moneylines all-time. Same market that carried this tournament, at nearly twice the sample. WTA and Challenger picks are accumulating toward their own gates and stay marked beta until they get there.

That scoreboard also answers the only fair question about a post like this one: how do you know we didn't just write up our good tournament? You don't have to trust us. The scoreboard runs continuously, includes every event between the ones we write about, and its worst rows publish exactly as prominently as its best.

The tour moved straight to Cincinnati, and the match board is already flagging picks there under the same rules, timestamped before each match and graded after. The US Open outright boards have been live since late July, priced before the draw even exists. That includes the runners we're against, and the one row where we've said plainly that the market knows more than we do.

Nothing here is betting advice. It's a model, a fortnight of its claims, and the standing public record they rolled into.

Live tennis board · CLV scoreboard · US Open futures · How the tennis model works