Performance
Track how the patterns performed, compare markets, and review the results behind each pick. Choose a week or view the season to date. Hit rates use settled results only.
Player status +
On the selected slate
| Market | Patterns | Hits | Misses | Open / ungraded | Settled hit rate |
|---|
View results by streak length
Build two-leg research combinations from qualifying Weekly picks. Each leg comes from a different player and game. Expand a combination to inspect its lines, history and price coverage.
Results by week
How combinations and prices work
Combinations use different players and games on the same scheduled date. Leg estimates must be within 10 percentage points. Rank by the weakest leg first, then combined estimate; each player appears only once in a week’s selected set. Return up to six combinations, with fewer when the rules cannot be met. These are product grouping rules, not an optimized or validated betting edge. The pool is capped at 18 eligible legs and three candidates per game. Different games reduce shared-game exposure, but do not establish independence. The combined estimate multiplies individual estimates under an unvalidated independence assumption; it is not a measured parlay win probability.
Pattern mode keeps the original weekly thresholds. Safer mode applies the same 5-percentage-point gain, 80–93% estimate range and exact-price ceiling as Safer lines. Sportsbook snapshots remain supporting context. A common-book price is shown only when every exact threshold was captured at that book at −800 or better. Market lines never replace the selected threshold.
Multiplying singles prices gives an illustrative payout, not a bookmaker parlay quote. Missing prices stay unverified. Historical results are reconstructed using today's rules; combinations can still share games and are not independent trials. Push, void/DNP or unconfirmed-stat legs are ungraded. No historical profit or verified pregame publication is established.
Season market insights
Market lines are captured sportsbook snapshots, not live consensus odds. A different market line is context, not the Pattern Bets selection. Missing lines are shown as unavailable.
How safer lines work
Start with a qualified weekly pick and move toward an easier threshold: lower for overs, higher for unders. Both thresholds use the same production curve and prior player history, with four model-weighted prior appearances. The safer estimate must be 80–93% and improve by at least 5 percentage points. This comparison is separate from the continuation score used to rank Weekly picks.
Exactly −800 is permitted; worse exact quoted prices are excluded. Unquoted safer thresholds remain price-unverified. The minimum gain is a product rule, not a proven optimum. Market lines are snapshots, not live offers; estimates and reconstructed hit rates do not establish profitability.
Explore what players have produced before this week. No weekly-pick probability or streak minimum applies. One row per player and statistic; averages describe history, not a forecast.
Regular-season appearances only. Byes and DNPs do not count as zero-stat appearances. Histories stop before the selected week; that week's result is shown separately.
Player histories
No player histories match these filters.
Try another player, team or statistic.
CHANGING ROLES · CHANGING PRODUCTION
Who is moving up or down?
A rising trend can still be an under at a high enough sportsbook line. These shifts describe production and workload; they do not automatically change bet probabilities or create picks. Signals require six appearances in the same season and team, with two recent games making meaningful contributions. Byes and DNPs do not become zeros.
Follow the NFL’s changing production. Compare positions, find teams gaining or losing momentum, and see where opponents have been giving up the most.
Team-position totals per recorded game · regular season only · games before the selected slate. Changes describe production, not sportsbook line movement or betting returns. Byes are skipped; unavailable games stay missing.
EVIDENCE BY MARKET
What the data is telling us
Compare streak cohorts within each statistic, not just one streak rule for every player.
Week-by-week streak audit / 2026
Open the experiment methods, calibration and full research tables
TEST THE ASSUMPTIONS
Is 3+ really the best streak length?
| Minimum streak | Checks | Repeat rate · 95% interval | Existing model expectation | Extra repetition · pp |
|---|
2024–2025, predictions made before each week. Intervals resample whole games. These cohorts overlap and use different research thresholds; a higher rate does not prove that adding streak length causes an edge. The expectation comes from a probability model trained through 2023. Extra repetition is a calibration diagnostic, not betting value.
See season and statistic breakdowns
| Minimum | 2024 checks / repeat | 2025 checks / repeat | Rushing | Receiving | Receptions | Passing |
|---|
Does explicitly adding streak length improve the model?
TREND RESEARCH
Which changes persist?
| Detector | Usage check | Development persistence | 2024–2025 checks | Later persistence · 95% interval | Forecast skill* |
|---|
Choose the detector on 2020–2023 only, using the lower confidence bound of improvement over a smoothed prior-history direction benchmark (500+ development signals). Then check 2024–2025 without retuning. All variants use the same outcome: does the next appearance remain above/below the first three of the prior six games? Ties are excluded and counted. *Forecast skill compares extrapolating the recent level with forecasting the old baseline, using market-scaled absolute error. Negative means worse; persistence is not a precise stat forecast.
These are retrospective tests of research signals, not sportsbook win rates. Role and injury changes, opponent strength and scheduling can affect results. Multiple variants were compared; this is exploratory evidence.
PROBABILITY BEFORE PRICE
How the shortlist is chosen
A streak is a starting point. The model also considers recent production relative to the line, comparable players against the defense, sample size and workload changes. It shrinks small-sample hit rates instead of treating 4/4 as a 100% chance.
| Later-season test | Observations | Games | Model Brier ↓ | History baseline ↓ |
|---|
A price needs an estimated edge of at least 3 percentage points after a 5-point probability haircut, 8+ prior appearances, 3+ consecutive results and stable workload. Imported quotes must be fresh and the game unstarted. A candidate is experimental; this is not evidence of sportsbook profitability. At most one shortlisted pattern per player; multiple props in one game are flagged as correlated.
DOES THE PATTERN REPEAT?
Next-appearance check
Consecutive streaks were found before each week, then checked on the player's next appearance. The table covers runs of at least 3 straight results. Reference thresholds are used because historical player-prop prices are absent from this dataset.
| Statistic | Checks | Hits / misses | Hit rate | Player-only baseline |
|---|
| Season | 3+ patterns | Hits / misses | Hit rate |
|---|
Six seasons test repeatability across more roles and environments; 2019 supplies warm-up history. More years do not create independent observations or prove an edge. Game-cluster bootstrap intervals are shown for the later-season tests. These are research thresholds, not historical sportsbook prices or a betting record.
How the pattern finder works +
Player form
Current streaks across up to 16 appearances, ending at the latest game. Misses and pushes break the streak; byes and inactive games are skipped. Last 3/5/8 counts are extra context, not streak lengths. This includes zero-stat games when the player took offensive snaps. Byes and inactive games are not fabricated as zeros. Changes of team reset the team's player history.
Comparable opponents
Only players with a meaningful role before that game enter the defensive comparison: WRs need 4 prior targets per game, TEs and receiving RBs need 3, rushing RBs need 8 carries, and passing QBs need 15 attempts. Their prior workload must also fall between 60% and 160% of the target player's recent workload, so backups are not treated like featured starters. Comparisons cover the defense's last 5 games.
Reference lines and support
The scanner keeps the longest current over and under run within a reference band around the prior eight-game median (20%, with minimum bands of 10 yards, 25 passing yards, or 1 reception). Ties favor the tighter line nearest the median. It checks only preset thresholds, not arbitrary extreme lines. Over support blends a smoothed player hit rate (80%) with a smoothed comparable-defense hit rate (20%) when at least 5 comparisons exist. Without them, only player history is used. No ties are called wins.
Data and availability
Built from cached nflverse play-by-play, snaps, schedules, player positions, and injury reports. Participation and team are inferred from recent appearances. An absent injury report does not confirm availability. Stat corrections and sportsbook settlement rules may differ.