Eligibility comes before ranking
The standard picks engine considers future, unstarted regular-season games and excludes cancelled or postponed games. The matchup reports begin with up to four skaters per team ranked by season shots per game. This is a limited candidate pool, not a scan of every NHL player.
A player needs verified recent game logs, at least three usable games before the slate date, and sufficiently fresh relevant source coverage. The engine uses up to ten recent games. Reported out, inactive, injured-reserve and suspended players are excluded. Roster and published lineup checks apply when those feeds are available; unresolved participation is not proof that a player will play.
Shots: an example of the actual ranking rule
For the 2+ shots market, the score is (qualifying games + 2) ÷ (sample games + 4). If player A reached two shots in 8 of 10 games, the score is 10 ÷ 14, approximately 0.714. Player B with 5 qualifying games in 6 has a score of 7 ÷ 10, or 0.700.
A ranks above B under this rule even though B’s raw hit rate is higher. The smoothing reduces the influence of very small samples. It does not establish that A has a 71.4% chance tonight. “Lock of the Night” names the top eligible shot selection; it does not mean a guaranteed result.
Goals are different markets
Anytime-goal ranking uses (games with a goal + 1) ÷ (sample games + 4). Multiple goals in one game still count as one qualifying game for that market. First-goal ranking instead uses (entire-game first goals + 1) ÷ (covered team games + 10), with at least three covered games and at least one observed first goal.
Scoring a team’s first goal after the opponent has already scored does not win an entire-game first-goalscorer pick. Coverage and settlement must match the exact market.
A likely outcome is not automatically a good price
A sportsbook quote and a model score answer different questions. For illustration, decimal odds of 1.50 imply a break-even probability of about 66.7% before other considerations. A positive expected-value claim would require a defensible probability above that threshold. Our historical ranking alone does not supply that probability.
Compare the exact player, market, threshold, sportsbook and quote time. A stale quote is not a current offer. Where an illustrative payout uses model-estimated odds, BenchWise labels it separately; it is not a sportsbook return you could necessarily have obtained.
Parlays need every leg
The standard board constructs different-player shots and goal packages, while certain markets use separate games. Some players can share a game, creating correlation. Multiplying marginal estimates assumes independence and can misprice a same-game ticket.
A winning parlay enters winning history only when the whole tracked ticket wins. A successful leg from a losing parlay does not become a separate winning straight. The winning gallery is a showcase of verified winners, not a full record of all losses, turnover or profitability.
What Premium adds today
Premium reports compare the baseline with supported context and preserve the original pregame assessment for later grading. They do not currently use an outside language model to invent selections. Missing current line assignments, power-play roles or other evidence are marked unavailable. Premium is a research workflow, not proof of a stronger betting edge.
Inspect tonight’s selections · Read the source and grading method