This is the tenth article in our /learn series, and the synthesis of everything we've covered: the structural math from article #6, the variance reality from article #7, the bankroll discipline from article #9, the prop-pricing mechanics from article #3, and the operator-level distinctions from earlier in the series. The question this article answers: how do you actually evaluate whether a specific pick is worth playing tonight?
The answer is a framework, not a checklist. Good pick evaluation is a sequence of comparisons — what the market is pricing, what you think the true probability is, how big the gap is, and whether your bankroll can absorb the variance behind the decision. None of the individual steps are complicated; the discipline is in doing them honestly. Most of this article is about that discipline. The math is already settled in earlier articles in the series — we'll point back where the derivations live.
Start with the market price
Different DFS and sportsbook markets have different efficiency. The most-traded prop markets — sportsbook standard sides and totals, top-star MLB props at major books, PrizePicks standard lines on most-played players — price close to true probability because they absorb sharp money quickly and adjust fast. Thinly-traded markets — backup hitter props, secondary stat types, demon/goblin tiers on PrizePicks — leave more room for edge, but they also have more idiosyncratic noise and harder evaluation.
For most casual players, the actionable rule is: the market is your starting point, not your enemy. The operators and books have already processed lineups, weather, recent form, ballpark factors, and matchup history before they posted the line. Your edge isn't usually in finding a pick the market missed — it's in finding spots where your read on a player differs from the market's by enough margin to matter.
The first practical step on every pick is to write down what the market is pricing it at. For a sportsbook prop, that's the American odds (Aaron Judge over 1.5 hits at -120). For a pick'em slip, it's the standard line and tier (Judge over 1.5 standard, with demon at 2.5 and goblin at 0.5). Until you know what the market is saying, you don't have anything to compare your read against. Comparison is the whole game.
This is the same point our sports betting vs. DFS guide made structurally: in both products, you're estimating probability against a market-set price. The product wrapping (slip multiplier vs. American odds) differs; the underlying decision is the same.
Compare your estimate to the market's
Once you know the market price, the comparison work begins. This is where most of the actual evaluation happens — and it has three layers that get conflated even by experienced players.
Take Aaron Judge over 1.5 hits, priced at -120 over / +100 under tonight. Three different probabilities are at play:
- Implied probability (over): 54.5%. Calculated directly from the price (120/220). This is what you'd need to win as a long-run hit rate just to break even at this price.
- No-vig probability (over): around 52%. The implied number with the bookmaker's margin stripped out. This is the market's actual belief about the true probability.
- Your estimate. What you actually think the true probability is, based on your read of the matchup, recent form, lineup spot, and the other factors you'd consider.
To play the over profitably, your estimate has to beat 54.5% (the implied / break-even) — not just 52% (the no-vig). The vig is the structural cost the market charges you for participating, and you have to clear it on top of beating the market's actual belief. Our sports betting vs. DFS guide walked through the implied-probability math; we won't re-derive it here.
The hardest part of pick evaluation for most casual players isn't the math — it's the honest self-assessment. Poker research has documented for decades that recreational players overestimate their edge; DFS is no different. Most picks you feel confident about don't actually have meaningful edges — they have small edges hidden under tilt-amplifying self-talk. The structural math of our math behind DFS losses piece demands this: in any rake market, most players are net losers, but most players think they're winners. Two of those statements have to be wrong; the math fixes which two. Acting on overestimated edge compounds losses faster than any specific pick error.
The honest version of pick evaluation: most picks you consider don't have edges. The picks worth playing are the ones where your estimate beats the implied probability by a meaningful margin — typically several percentage points, not fractions of one.
Account for variance
Suppose you've done the comparison and decided a pick has a real edge. You still have to account for variance: even real edges produce losing picks more than half the time on most prop bets, and produce losing slips most of the time on pick'em.
A 4-pick power slip with a real 12% true cash rate — slightly above break-even at 10x — loses 88 out of every 100 entries on average. The fact that one specific slip "feels" great doesn't change that. Every individual pick is a probability draw, and probability draws come in clumps. Our variance guide walked through what normal cold stretches look like; the takeaway here is that no single pick's outcome tells you whether the pick was good. Only patterns over hundreds of picks do.
Two practical consequences:
- Don't validate picks by single-game outcomes. A losing pick can have been a great pick; a winning pick can have been a terrible one. Process beats outcomes — evaluate your decision against what you knew before the game, not against what happened in it.
- Don't tilt-amplify after a win. A hot stretch doesn't make tonight's marginal pick more attractive. The market hasn't changed; your variance has.
(Multi-pick slips add a correlation layer — picks on the same lineup share game-level variance — which complicates the math further. A future article will go deeper on stacking and correlation; the glossary entry is the short reference for now.)
Check the quality of your underlying data
Your probability estimate is only as good as the inputs that produced it. Several things should make you trust your estimate less:
Small sample on the player. A rookie with 30 MLB plate appearances doesn't have enough history to support a confident projection. Recent call-ups and players returning from extended injuries fall into this bucket.
Low expected plate appearances. A bench player you're projecting as a starter is a projection error waiting to happen. Even if your stat read is right, the player has to actually get the at-bats for the prop to clear.
Recent role change. A player traded last week, called up from AAA two days ago, or rolled into a new lineup spot is operating in conditions that historical data doesn't reflect well. Manager-tendency context becomes incomplete in these cases.
Unusual matchup context. A pitcher returning from injury, a hitter facing a knuckleballer they haven't seen in two years, an interleague game where the lineup includes the DH for the first time this week — anything that pushes the matchup outside the training data your projection is built on.
RunsLeft surfaces quality flags for several of these conditions on /dfs/edges — low expected PA, small sample, recent role change. The flags are transparency labels that tell you to trust the projection less when they fire, not filters that hide projections.
Line-shop if you can
The same pick prices differently across operators. Aaron Judge over 1.5 hits might be -120 at FanDuel Sportsbook and -115 at DraftKings; the same stat line on PrizePicks vs. Underdog can have slightly different tiers and payouts. The mechanics differ — sportsbook American odds vs. pick'em multipliers — but the principle is identical: line shopping across operators captures expected value that compounds across hundreds of bets.
Practical version: if you're betting at sportsbooks, have accounts at two or three and check the line on every prop before betting. If you're playing pick'em, comparing standard lines across PrizePicks, Underdog, and DraftKings Pick6 catches the cases where one operator's number is meaningfully softer than the others. The work is small per bet; the impact compounds.
Size honestly
None of the evaluation work matters if your bankroll can't absorb the variance behind it. The cleanest pick at the strongest edge still loses 88% of the time as a 4-pick slip; a 15-slip cold streak isn't unusual on a real 12% edge; bankroll growth on real edges is measured in baseball seasons, not weeks.
The full sizing framework is in our bankroll guide. The short version that applies to every pick evaluation:
- Pick a percentage of bankroll (1-2% conservative, 3-5% aggressive). Stick to it.
- Don't tilt-resize after wins or losses. Same percentage tonight as last night.
- Size against your real edge, not your hot-streak feel. Most casual players overestimate their edge — defer to the conservative end.
- Treat entry fees as entertainment expenses you'd be happy to spend even on a losing night.
The sizing decision is independent of the pick decision. A great pick at reckless sizing is still a bankroll-blower. A marginal pick at conservative sizing is survivable. The two decisions multiply — both have to be right for the math to work over time.
Where RunsLeft fits
The framework above — start with the market, compare to your estimate, account for variance, check data quality, line-shop, size honestly — is what tonight's DFS edges page automates each night. Our model generates probability estimates for player props across the slate, compares them to the market lines on offer, surfaces the spots where the gap is large enough to clear the operator's rake and vig, and attaches quality flags and context blocks (pull pattern, recent form, lineup spot) for the conditions that should make you trust the projection more or less.
The framework is yours to use whether you use RunsLeft or not. We don't make the bankroll, line-shopping, or variance decisions for you. What we do is the comparison work systematically — every night, across PrizePicks, Underdog, DraftKings Pick6, salary-cap DFS, and sportsbook props. The signal is operator-agnostic; the math is the same across products.
What we surface isn't "tonight's winners." It's tonight's spots where the math is friendlier than the operator's default — with the reasoning shown so you can decide whether to act on each one.
Where to go from here
The /learn series is just getting started. Coming in future articles: pitcher props (strikeouts, outs recorded, hits allowed), operator-specific strategy (PrizePicks vs. Underdog vs. DraftKings Pick6 deep dives), correlation and stacking math for multi-pick slips, and seasonal/playoff-specific approaches as the calendar moves through MLB, NFL, NBA, and NHL.
The full /learn index is at /learn. If you landed here from search and haven't read the underlying articles, the natural reading order for the foundations is: what DFS is → how props work → the math behind losses → variance → bankroll. The glossary is the reference layer underneath.
For nightly application of the framework above, the live page is /dfs/edges.