01 · The start
Down five with 2:04 left at their own 20: the model gives Tampa Bay 14.6 percent.
Week 2 of the 2025 season, fourth quarter in Houston. The Texans have just gone up 19–14, and Tampa Bay takes over at its own 20 with 2:04 left and three timeouts on each side. The site's logistic model turns that game state into one number from 12 inputs: score, clock, field position, down, distance, timeouts, venue and the pregame spread, here −2.5 for the road team. Its answer is 14.6 percent.
02 · First down
A 12-yard catch lifts it to 16.5 percent; a catch for no gain drops it to 14.8.
Mayfield hits Irving for 12 yards and a first down at the Tampa Bay 32, and the number rises to 16.5 percent: better field position and a fresh set of downs. The next snap is another short pass to Irving, caught at the same 32 for no gain. The two plays read almost alike in the play-by-play, but they are separate snaps, 3762 and 3787. Second and 10 with 1:37 left: 14.8 percent.
03 · Fourth and 10
Two incompletions leave fourth and 10 at their own 32: the drive's low, 11.1 percent.
Two deep throws fall incomplete, to Evans down the left side and to Egbuka down the right. Each lost down costs the offense in the model: third and 10 is 13.2 percent, and fourth and 10 at the Tampa Bay 32 with 1:24 left is 11.1 percent, the lowest point of the drive. A failed fourth down would hand Houston the ball with the lead and almost no clock left to defend.
04 · The scramble
Mayfield scrambles 15 yards on fourth down, and the number jumps to 16.4 percent.
On fourth and 10 Mayfield runs up the middle for 15 yards to the Tampa Bay 47. The conversion resets the downs and gains field position, and the number jumps from 11.1 to 16.4 percent, even though the clock now reads 0:59. For a team that is behind late, the clock term works against the offense on every snap that does not end in points.
05 · Into range
Two completions reach the Houston 23 with 42 seconds left: still only 21.3 percent.
Otton catches a short pass and gets out of bounds at the Houston 45 for 8 yards, 18.6 percent, and Houston calls its first timeout. Then Irving takes a short pass 22 yards to the Houston 23. First and 10 there with 0:42 left is 21.3 percent. Field position now counts for more than anything else, but a touchdown is still required, and the model treats it as a long shot.
06 · Goal to go
Second and goal from the 2, nine seconds left: the model says 23.0 percent.
Irving gains 5 to the Houston 18: 21.0 percent. Evans catches an 11-yarder at the 7, first and goal with 0:13 left: 21.5 percent. Otton gets out of bounds at the 2, and Houston uses its second timeout. Second and goal with 0:09 left: 23.0 percent. Each yard is paid for with clock, and in this model the two almost cancel out, so the number barely moves.
07 · The touchdown
White scores from the 2 with 0:09 left; the scoreboard, not the model, settles it.
White runs 2 yards behind the right guard for the touchdown. Tampa Bay leads 20–19; the two-point try fails, and Houston's kickoff return ends the game. The final score makes the probability 100 percent. The model never priced this drive above 23.0 percent, which on its own is not an error: roughly one chance in four or five still comes in, and this was one of them.
08 · Against the record · the check
Since 2016, offenses in this spot won 13 of 21; the model averaged 27 percent.
A lookup in nflverse play-by-play, 2016–2025: 21 drives had a first-to-third-down snap at the opponent's 2 or closer in the final 60 seconds, trailing by 4 to 6, this drive included. Sixteen scored a touchdown and 13 won, 62 percent; the model averaged 27 percent. If it were right, 13 or more wins would happen about 1 time in 1,500. nflfastR's tree-based model, which can bend its field-position effect, put this snap at 49.9 percent.
09 · The whole drive
Eleven snaps, one line: 14.6 at the start, 11.1 at the low, 23.0 before the touchdown.
Put together, the drive is a line of 11 segments, one per snap, each computed on this page from clock, field position, down, distance, score, timeouts, venue and the pregame spread. The published values are rounded to four decimals, and every value here matches them within 0.00005. The same coefficients run the win probability calculator, where any of these game states can be entered by hand and pushed a yard or a second either way.
The data · 11 snaps
Every snap of the drive and the probability before it
Each row is the game state at the snap, as recorded in nflverse play-by-play (game 2025_02_TB_HOU), and the play that followed. The first two rows describe two different plays that both ended at the Tampa Bay 32; they carry their own play IDs, 3762 and 3787, and both stay in. Win probability is Tampa Bay's, before the snap.
| # | Clock | Situation and play | WP |
|---|---|---|---|
| 1 | 2:04 | 1st & 10 · TB 20(2:04) (Shotgun) 6-B.Mayfield pass short left to 7-B.Irving to TB 32 for 12 yards (0-A.Al-Shaair; 5-J.Pitre). | 14.6% |
| 2 | 1:56 | 1st & 10 · TB 32(1:56) (Shotgun) 6-B.Mayfield pass short left to 7-B.Irving to TB 32 for no gain (39-H.To'oTo'o; 98-T.Settle). | 16.5% |
| 3 | 1:37 | 2nd & 10 · TB 32(1:37) (No Huddle, Shotgun) 6-B.Mayfield pass incomplete deep left to 13-M.Evans. | 14.8% |
| 4 | 1:32 | 3rd & 10 · TB 32(1:32) (Shotgun) 6-B.Mayfield pass incomplete deep right to 2-E.Egbuka. | 13.2% |
| 5 | 1:24 | 4th & 10 · TB 32(1:24) (Shotgun) 6-B.Mayfield scrambles up the middle to TB 47 for 15 yards (8-C.Gardner-Johnson). | 11.1% |
| 6 | 0:59 | 1st & 10 · TB 47(:59) (No Huddle, Shotgun) 6-B.Mayfield pass short right to 88-C.Otton pushed ob at HOU 45 for 8 yards (5-J.Pitre). | 16.4% |
| 7 | 0:52 | 2nd & 2 · HOU 45(:52) (Shotgun) 6-B.Mayfield pass short middle to 7-B.Irving to HOU 23 for 22 yards (0-A.Al-Shaair) [95-D.Barnett]. | 18.6% |
| 8 | 0:42 | 1st & 10 · HOU 23(:42) (Shotgun) 6-B.Mayfield pass short left to 7-B.Irving pushed ob at HOU 18 for 5 yards (39-H.To'oTo'o). | 21.3% |
| 9 | 0:34 | 2nd & 5 · HOU 18(:34) (Shotgun) 6-B.Mayfield pass short middle to 13-M.Evans to HOU 7 for 11 yards (2-C.Bullock). | 21.0% |
| 10 | 0:13 | 1st & goal · HOU 7(:13) (No Huddle, Shotgun) 6-B.Mayfield pass short left to 88-C.Otton pushed ob at HOU 2 for 5 yards (8-C.Gardner-Johnson). | 21.5% |
| 11 | 0:09 | 2nd & goal · HOU 2(:09) 1-R.White right guard for 2 yards, TOUCHDOWN. | 23.0% |
The model · 12 inputs
How one number is made: second and goal at the 2
The model is a logistic regression. With t the share of the game left (seconds left ÷ 3600), field position (50 − yards to the goal line) ÷ 50 and the pregame spread from the offense's side (−2.5 for Tampa Bay, the road underdog), each input is multiplied by its coefficient, the products are added to the intercept, and the sum z becomes a probability through 1 ÷ (1 + e−z). These are the terms for the last snap before the touchdown.
| Input | Value | Coef. | Term |
|---|---|---|---|
| Intercept | +0.5403 | ||
| Score difference | -5.0000 | 0.0685 | -0.3426 |
| Score difference × (1 − t)² | -4.9750 | 0.0997 | -0.4962 |
| Score difference ÷ √(minutes left + 1) | -4.6625 | 0.3316 | -1.5460 |
| Pregame spread × t | -0.0063 | 0.1986 | -0.0012 |
| Field position | 0.9600 | 0.2407 | +0.2311 |
| Field position × (1 − t) | 0.9576 | 0.4650 | +0.4453 |
| Second down | 1.0000 | -0.0681 | -0.0681 |
| Third down | 0.0000 | -0.1828 | +0.0000 |
| Fourth down | 0.0000 | -0.3539 | +0.0000 |
| ln(yards to go) | 0.6931 | -0.0518 | -0.0359 |
| Timeout difference × (1 − t) | 0.9975 | 0.0645 | +0.0643 |
| Home team | 0.0000 | -0.1289 | +0.0000 |
| z, then probability | -1.2090 → 23.0% |
Field position enters twice, both times as a straight line: a yard gained at midfield counts exactly as much as the last yard before the goal line. That is why the model underrates the last snaps of a drive like this one; the check above measures by how much.
Frequently asked questions
What is win probability in football?
Win probability estimates how often a team in a given game state goes on to win. The model on this page is a logistic regression with 12 inputs covering score, clock, field position, down, distance, timeouts, venue and the pregame spread. On Tampa Bay's final drive at Houston in week 2 of the 2025 season, it started at 14.6 percent with 2:04 left, the offense down five at its own 20.
How does the logistic model produce a probability?
Each input is multiplied by its coefficient, the products are added to an intercept, and the sum z becomes a probability through 1 ÷ (1 + e to the power of −z). Time enters as t, the share of the game left (seconds left ÷ 3600), and field position as (50 − yards to the goal line) ÷ 50. For second and goal from the Houston 2 with 0:09 left, the terms add up to 23.0 percent.
Why did win probability drop after plays that gained no yards?
For a team behind late, every lost down and every second off the clock count against it. After a 12-yard catch lifted Tampa Bay to 16.5 percent, a catch for no gain left second and 10 with 1:37 left at 14.8 percent. Two incompletions then pushed it to 13.2 and 11.1 percent, the low of the drive at fourth and 10. A 15-yard scramble on fourth down reset the downs and lifted it to 16.4.
Why did the model rate the final snap so low?
Field position enters the model as a straight line, so the last yard before the goal line counts no more than a yard at midfield. In nflverse play-by-play from 2016 to 2025, 21 drives had a comparable snap at the opponent's 2 or closer in the final 60 seconds, trailing by 4 to 6. Of those, 13 won, 62 percent, while the model averaged 27 percent. nflfastR's tree-based model put this snap at 49.9 percent.
Does a touchdown after a 23.0 percent estimate mean the model was wrong?
Not on its own. An estimate of 23.0 percent means roughly one chance in four or five, and outcomes like that are expected to happen regularly. The stronger evidence comes from many similar drives: if the model's average of 27 percent were right, 13 or more wins out of 21 would occur about 1 time in 1,500. That pattern, not one touchdown, points to the straight-line field position term.