MONITOR SPORTS PICKS / UPDATED OCTOBER 1, 2026

Methodology

By Monitor Sports Picks · AI-assisted explanation · Updated October 1, 2026

Our daily board is a shortlist of up to five estimated game winners, not a list of bets proven to beat the market. The current method is experimental and has not been independently validated or backtested. We do not advertise a historical win rate.

Inputs and sources

We read moneylines and available season records from public ESPN odds pages for MLB, NFL, NBA, NHL, WNBA, and college football. Each pick includes a source link and bookmaker label. We collect factual game information, not another publisher's written picks. Source references do not imply endorsement or affiliation. Page formats and access can change, and coverage is not guaranteed.

How the estimate works

For a negative American moneyline, implied probability is the absolute odds divided by the absolute odds plus 100. For positive odds, it is 100 divided by the odds plus 100. We divide each team's implied probability by the sum of both probabilities to normalize out the bookmaker margin.

If both teams have at least ten recorded games, we add five wins and five losses to each record to reduce small-sample extremes. Ties count as half a win. With adjusted win rates A and B, the record-only estimate is A × (1 − B) / [A × (1 − B) + B × (1 − A)]. The final estimate is 80% normalized market probability and 20% record estimate. Those weights are provisional design choices, not fitted or validated parameters. If records are missing or too short, the estimate uses the normalized market only.

A worked example: from records to an estimate

Consider two fictional teams in a two-way market. Suppose Team A's normalized market probability is 60%, Team A has a 15–5 record, and Team B is 10–10. These are illustrative inputs, not a current game or a tested prediction.

1. Adjust the records. Adding five wins and five losses gives A a rate of 20/30, or about 66.67%, and B a rate of 15/30, or 50%. This pulls A's raw 75% season win rate toward 50%; it does not account for schedule strength.
2. Compare the adjusted rates. With B at 50%, the record comparison is [(2/3) × (1/2)] ÷ [(2/3) × (1/2) + (1/2) × (1/3)] = 2/3, or about 66.67%.
3. Blend the two inputs. 0.80 × 60% + 0.20 × 66.67% is about 61.33%. The card rounds that estimate to one decimal place: 61.3%.

The record component raises this example's estimate by about 1.33 percentage points above the normalized market. It does not add 20 percentage points. If either team's record is missing or contains fewer than ten games, the record component is omitted and this example stays at 60%.

Why the percentage is not a confidence guarantee

The displayed estimate is a single output of the calculation, not a confidence interval or an accuracy score. Adding the hypothetical wins and losses moderates extreme records; it does not establish that the resulting percentage is calibrated. The 80/20 blend is a design choice, and combining records with market prices may reuse information already reflected in those prices.

We would need a retained record of pregame predictions and outcomes to assess whether selections near 60% actually win near 60% over a suitable sample. Our guide to prediction value, sample size and calibration explains that distinction. The example above explains the calculation only; it is not evidence that the method beats the market.

For related context, read how moneylines become implied probabilities and why the odds observation time matters. These public explanations do not require email signup.

Our output does not estimate winning margins. See why winning outright and covering a spread are different predictions for a worked comparison.

Same record, different opponents: what the adjustment cannot see

Two teams can both be 8–2 without having faced the same opponents. Our record component receives their wins, losses and ties; it does not receive an opponent-by-opponent history. Changing that history while keeping the record and market price fixed therefore cannot change this calculation.

Consider an original hypothetical comparison: A and B are each 8–2, and A has a normalized market probability of 60%. These are invented inputs, not current teams or a forecast.

1. Both adjusted record rates are (8 + 5) / (10 + 10) = 13/20, or 65%.
2. The record-only comparison is (0.65 × 0.35) / [(0.65 × 0.35) + (0.65 × 0.35)] = 50%.
3. The blend becomes 0.80 × 60% + 0.20 × 50% = 58%. It lowers A's estimate by two percentage points.

That reduction is a consequence of the formula, not evidence that the market overrated A. Imagine A faced stronger opponents than B, then imagine the reverse. With the same two 8–2 records and the same prices, this method returns 58% in both scenarios. It cannot decide which schedule was harder or how large an adjustment would be justified. Adding five wins and five losses moderates a record; it does not correct the identity of the opponents.

Read the definition behind a schedule-strength number

NFL.com's 2026 schedule discussion identifies a preseason measure based on opponents' previous-year winning percentage. That is a dated input, not an observation of how those opponents will perform in the new season.

Before comparing two schedule figures, check whether each covers games already played or games still to come, which season supplies the records, and whether the measure uses records or a separate rating model. Mixing those definitions does not create a like-for-like comparison. A schedule statistic alone also does not supply a calibrated probability for today's matchup.

Our current method adds no independent schedule-strength correction. The market input may contain information missing from the record component, so this example is a reason to evaluate the blend against retained outcomes, not to assume blending improves it. See the prediction-record audit and calibration example. Source and calculation checked October 1, 2026.

Daily selection rules

We use the Eastern Time calendar day. Games must not have started at the moment the day's board is first created, must have two complete moneylines, and must have a winner estimate of at least 55%. We deduplicate games, rank the higher-probability team in each game, and keep the first five. There is no required selection from each league. Heavy favorites can therefore dominate the list.

The first available daily edition remains fixed for that day. Source checks are cached for six hours. Displayed odds are the publication snapshot, not live odds. A started selection stays visible with a started label; it is not a recommendation to place an in-play bet. If too few games qualify, we show fewer than five. We never invent games to fill the board.

What this method misses

Season records do not fully represent opponent strength, home advantage, rest, injuries, roster changes, or starting pitchers. Markets may reflect those factors, but this method does not independently model them. A 70% estimate is not proof that comparable selections win 70% of the time. Ranking likely winners is different from identifying positive expected value at the offered price.

Access and transparency

The method is public. The daily selections require an email submission and are withheld by the server until access is granted. Email ownership is not verified. Odds availability can fail independently of the signup service, in which case the board shows an unavailable message instead of sample picks.