NRFI weather talk usually arrives as a rule. Cold adds 4 to 7 points. Wind blowing out kills the lean. Humidity matters. The sample is never in the sentence.
The nightly replay (record_cache) stores pk, nrfi, and the two
model probabilities. It does not store temperature or wind. So this page
does not come from the replay file. It is a join we can actually run:
the public MLB Stats API schedule, hydrated with linescore, venue,
and weather, for every 2026 regular-season game through September 3
that has a completed first inning. 2,109 games. Weather was present on
all 2,109. 401 of them were dome or closed roof.
The league NRFI rate on the board’s own replay for the same dates is 1,059/2,134 (49.6%). Use that as the season figure. Use the 2,109 below as the weather sample. They differ by 25 games; we are not going to pretend they are the same count.
What the board does with weather
The Full Count multiplies both half-inning lambdas by a weather factor
(_weather_factor in server.py, weatherFactor in static/index.html):
- Missing weather: 1.00
- Condition contains “roof” or “dome”: 1.00, ignore temp and wind
- Otherwise start at 1.00, add
0.004 * (temp − 70) - If the wind string parses a speed and says “Out To”, add
0.008 * mph - If it says “In From”, subtract
0.008 * mph - Clamp to 0.85 through 1.15
At 55°F, no wind, that is a 0.94 multiplier on expected first-inning runs. At 90°F, 1.08. A 15 mph wind out adds 0.12 before the clamp. Those are model knobs, not measured lifts. This page is the measurement.
Humidity is not in the feed we get (condition, temp, wind). We have
no humidity split to publish.
Temperature, open air only
Dome and closed-roof games are held out of this table so a retractable roof in July does not land in the 90s band. 1,708 open-air games with a parsed temperature. 855 were NRFI (50.1%).
| Temperature | Scoreless 1st | NRFI rate | vs open-air 50.1% |
|---|---|---|---|
| Under 50°F | 35/61 | 57.4% | +7.3 pts |
| 50 to 59°F | 73/136 | 53.7% | +3.6 pts |
| 60 to 69°F | 149/299 | 49.8% | −0.3 pts |
| 70 to 79°F | 312/603 | 51.7% | +1.6 pts |
| 80 to 89°F | 235/491 | 47.9% | −2.2 pts |
| 90°F and up | 51/118 | 43.2% | −6.9 pts |
The cold-week rule that gets quoted as “4 to 7% NRFI lift under 55°F” was tested as 63/114 (55.3%) on open-air games under 55°F. That is 5.2 points above the 1,708-game open-air rate, and 5.7 points above the replay’s 49.6% league rate. 114 games is a thin band. It is also April and early May, which means it is also Coors-or-not, and it is also whichever teams opened in the cold. We are not turning 63/114 into a weather lift you can take to a window.
The hot band is the same size and the other direction: 51/118 (43.2%) NRFI at 90°F and above. More first-inning scoring on hot slates in this sample. Same caveat. 118 games, mixed with park and month.
Wind
Parsed from the MLB wind string (leading mph, plus “Out To” / “In From” / cross). Dome and roof games are not in these rows.
| Direction | Scoreless 1st | NRFI rate |
|---|---|---|
| Out | 338/719 | 47.0% |
| In | 189/360 | 52.5% |
| Cross (L to R, R to L) | 284/538 | 52.8% |
| Other / none | 44/91 | 48.4% |
Wind out versus wind in is the cleanest weather split we have: 5.5 points, and the samples are hundreds of games, not a hundred. Wind blowing out scored more often in the first inning. That matches the direction of the live multiplier.
The 15 mph cutoffs do not. Out at 15+ mph: 33/64 (51.6%) NRFI. In at 15+: 6/13 (46.2%). Those are small cells and they do not support a “strong wind out kills NRFI” rule. We would rather print the null than borrow someone else’s 4 to 7%.
Roof, rain, condition
| Setting | Scoreless 1st | NRFI rate |
|---|---|---|
| Dome or roof closed | 192/401 | 47.9% |
| Cloudy (open air) | 529/1,044 | 50.7% |
| Clear / sunny | 282/562 | 50.2% |
| Rain in the condition string | 4/8 | 50.0% |
Enclosed games sat a couple of points below the open-air 50.1%. Rain is 8 games. That is not a finding.
A postponed or suspended game is not graded. The replay drops any game without a completed first inning. What a sportsbook does with a delay after first pitch is a house rule. It is not a model output.
What we will not say
We will not publish a single weather lift percentage. The temperature gradient in the table is real in this sample and it is not identified. Park and calendar sit inside every band. The park table is the other half of that sentence.
The board still applies the live multiplier, because a true probability has to react to 48°F and a 20 mph wind in. Reacting is not the same as having proven the size of the effect. Methodology is where that line is drawn for the models; this page is the same line for weather.
Statistics and research, not betting advice. 21+. Gambling problem? Call 1-800-GAMBLER.