Settled model recap · 2026-09-23

MLB Game & Prop Picks Recap for September 23, 2026: 3-6, -3.23 units

Review 1 game Pick and 8 pitcher-prop Picks from September 23, 2026 with the exact saved sportsbook prices, model probabilities, lineup or player values, final results, and plain-English postgame review.

Published after the saved Picks settled. The original prices and inputs stay attached.

Record3-6-0

Net units-3.23u

Game Picks1

Player Props8

One-minute recap

The card missed: 1 game Pick and 8 pitcher-prop Picks finished 3-6 for -3.23 units. The useful question is which assumptions failed, not how to explain the result after the fact.

Decision coverage note

Decision coverage was incomplete in the pitcher strikeout lane (11 opportunities expired despite a prior decision opportunity) and the additional pitcher props lane (34 opportunities expired without a decision). This recap is complete for the 9 settled Picks that were actually recorded, but the slate does not meet full-lifecycle acceptance. No missing decision was reconstructed, and these incidents do not count as Picks or change the record above.

Saved Picks and reviews

Open any Pick for the why

MLB Game Picks

1 game Pick, reviewed from the saved lineup, player, team, and moneyline snapshot.

Game Pick

TOR moneyline

TOR at BAL · TOR 2, BAL 4

BetRivers +120lossSee the review ↓

Postgame review

Lost: TOR +120

TOR did not get there in a TOR 2, BAL 4 final. The saved BetRivers +120 price cost 1.00 unit on the game-pick record.

The pregame case started with BetRivers TOR +120. We made the side 45.8%; the price needed 45.5%. That left 0.3 percentage points of room over the price's break-even mark. The strongest public push toward TOR was Long-run team strength, worth 0.08 probability points in the saved calculation. BAL carried the higher starter rating: Max Scherzer rated 48.3, compared with Chris Bassitt at 50.9. The two confirmed lineups graded about even: TOR 48.6, BAL 49.2. The top displayed hitters were Kazuma Okamoto (53.5) and Pete Alonso (59.5). BAL had the higher bullpen rating: TOR 51.0, BAL 53.0.

The price case was there; the result was not. We will compare this starter, lineup, and bullpen profile with future frozen reads before deciding whether the miss was noise or a repeatable blind spot.

Saved pregame inputs

Model probability45.8%

Price break-even45.5%

Calculated difference+0.3 pts

Selected side

TOR

Team form50.5

Lineup form48.6

Starting pitcher48.3

Bullpen available51.0

Opponent

BAL

Team form49.7

Lineup form49.2

Starting pitcher50.9

Bullpen available53.0

Values use the saved 0–100 rolling scale. They are inputs to the selection screen, not direct probability-point weights.

MLB Player Props

8 pitcher-prop Picks, covering strikeouts and reviewed from the saved workload, opponent, and price snapshots.

Pitcher prop

Logan Henderson Under 5.5 strikeouts

MIL at PHI · 3 strikeouts · under 5.5

DraftKings -121winSee the review ↓

Postgame review

Won: Logan Henderson Under 5.5 Ks

Logan Henderson finished with 3 strikeouts, cashing the saved under 5.5 and adding +0.83 units to the prop record.

At DraftKings -121, the model gave the under 57.0% against a 54.8% break-even mark, a +2.3-point gap. The saved projection was 5.0 strikeouts across 21.2 batters faced. The pitcher graded 54.8/100 and the opponent strikeout matchup graded 55.4; the confirmed lineup carried a 20.1% strikeout rate.

The actual result finished 2.0 strikeouts below the point projection. The ticket and the point projection told the same story. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability57.0%

Price break-even54.8%

Calculated difference+2.3 pts

Frozen pitcher-prop inputs

Logan Henderson vs. PHI

Projected Ks: 5.0
Actual Ks: 3

Pitcher value54.8

Opponent K matchup55.4

Lineup K rate20.1%

Lineup coverage100%

Projected batters faced21.2

Evidence quality89%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.40 Ks. Robustness: 75%.

Pitcher prop

Foster Griffin Over 4.5 strikeouts

CLE at BOS · 7 strikeouts · over 4.5

FanDuel +102winSee the review ↓

Postgame review

Won: Foster Griffin Over 4.5 Ks

Foster Griffin finished with 7 strikeouts, cashing the saved over 4.5 and adding +1.02 units to the prop record.

At FanDuel +102, the model gave the over 50.2% against a 49.5% break-even mark, a +0.7-point gap. The saved projection was 5.0 strikeouts across 23.1 batters faced. The pitcher graded 52.4/100 and the opponent strikeout matchup graded 45.1; the confirmed lineup carried a 21.3% strikeout rate.

The actual result finished 2.0 strikeouts above the point projection. The ticket and the point projection told the same story. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability50.2%

Price break-even49.5%

Calculated difference+0.7 pts

Frozen pitcher-prop inputs

Foster Griffin vs. BOS

Projected Ks: 5.0
Actual Ks: 7

Pitcher value52.4

Opponent K matchup45.1

Lineup K rate21.3%

Lineup coverage100%

Projected batters faced23.1

Evidence quality92%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.39 Ks. Robustness: 79%.

Pitcher prop

Gerrit Cole Under 5.5 strikeouts

TB at NYY · 8 strikeouts · under 5.5

DraftKings -116lossSee the review ↓

Postgame review

Lost: Gerrit Cole Under 5.5 Ks

Gerrit Cole finished with 8 strikeouts, leaving the saved under 5.5 on the wrong side of the line for a 1.00-unit loss.

At DraftKings -116, the model gave the under 55.3% against a 53.7% break-even mark, a +1.6-point gap. The saved projection was 5.2 strikeouts across 23.4 batters faced. The pitcher graded 50.3/100 and the opponent strikeout matchup graded 48.4; the confirmed lineup carried a 20.4% strikeout rate.

The actual result finished 2.8 strikeouts above the point projection. The projection supported the ticket, but the actual result crossed to the other side of the sportsbook line. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability55.3%

Price break-even53.7%

Calculated difference+1.6 pts

Frozen pitcher-prop inputs

Gerrit Cole vs. TB

Projected Ks: 5.2
Actual Ks: 8

Pitcher value50.3

Opponent K matchup48.4

Lineup K rate20.4%

Lineup coverage100%

Projected batters faced23.4

Evidence quality91%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.45 Ks. Robustness: 78%.

Pitcher prop

Chris Sale Under 7.5 strikeouts

CIN at ATL · 10 strikeouts · under 7.5

Betonlineag -132lossSee the review ↓

Postgame review

Lost: Chris Sale Under 7.5 Ks

Chris Sale finished with 10 strikeouts, leaving the saved under 7.5 on the wrong side of the line for a 1.00-unit loss.

At Betonlineag -132, the model gave the under 58.5% against a 56.9% break-even mark, a +1.6-point gap. The saved projection was 6.5 strikeouts across 23.8 batters faced. The pitcher graded 65.5/100 and the opponent strikeout matchup graded 73.6; the confirmed lineup carried a 25.9% strikeout rate.

The actual result finished 3.5 strikeouts above the point projection. The projection supported the ticket, but the actual result crossed to the other side of the sportsbook line. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability58.5%

Price break-even56.9%

Calculated difference+1.6 pts

Frozen pitcher-prop inputs

Chris Sale vs. CIN

Projected Ks: 6.5
Actual Ks: 10

Pitcher value65.5

Opponent K matchup73.6

Lineup K rate25.9%

Lineup coverage100%

Projected batters faced23.8

Evidence quality91%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.78 Ks. Robustness: 78%.

Pitcher prop

Kevin Gausman Under 5.5 strikeouts

MIA at CHC · 7 strikeouts · under 5.5

FanDuel +118lossSee the review ↓

Postgame review

Lost: Kevin Gausman Under 5.5 Ks

Kevin Gausman finished with 7 strikeouts, leaving the saved under 5.5 on the wrong side of the line for a 1.00-unit loss.

At FanDuel +118, the model gave the under 48.2% against a 45.9% break-even mark, a +2.3-point gap. The saved projection was 5.4 strikeouts across 23.5 batters faced. The pitcher graded 51.2/100 and the opponent strikeout matchup graded 52.1; the confirmed lineup carried a 22.1% strikeout rate.

The actual result finished 1.6 strikeouts above the point projection. The projection supported the ticket, but the actual result crossed to the other side of the sportsbook line. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability48.2%

Price break-even45.9%

Calculated difference+2.3 pts

Frozen pitcher-prop inputs

Kevin Gausman vs. MIA

Projected Ks: 5.4
Actual Ks: 7

Pitcher value51.2

Opponent K matchup52.1

Lineup K rate22.1%

Lineup coverage100%

Projected batters faced23.5

Evidence quality92%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.50 Ks. Robustness: 80%.

Pitcher prop

Seth Lugo Over 4.5 strikeouts

CWS at KC · 3 strikeouts · over 4.5

DraftKings +104lossSee the review ↓

Postgame review

Lost: Seth Lugo Over 4.5 Ks

Seth Lugo finished with 3 strikeouts, leaving the saved over 4.5 on the wrong side of the line for a 1.00-unit loss.

At DraftKings +104, the model gave the over 50.6% against a 49.0% break-even mark, a +1.6-point gap. The saved projection was 5.1 strikeouts across 24.0 batters faced. The pitcher graded 44.6/100 and the opponent strikeout matchup graded 43.8; the confirmed lineup carried a 24.5% strikeout rate.

The actual result finished 2.1 strikeouts below the point projection. The projection supported the ticket, but the actual result crossed to the other side of the sportsbook line. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability50.6%

Price break-even49.0%

Calculated difference+1.6 pts

Frozen pitcher-prop inputs

Seth Lugo vs. CWS

Projected Ks: 5.1
Actual Ks: 3

Pitcher value44.6

Opponent K matchup43.8

Lineup K rate24.5%

Lineup coverage100%

Projected batters faced24.0

Evidence quality93%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.43 Ks. Robustness: 80%.

Pitcher prop

Merrill Kelly Over 3.5 strikeouts

ARI at COL · 2 strikeouts · over 3.5

Bovada -135lossSee the review ↓

Postgame review

Lost: Merrill Kelly Over 3.5 Ks

Merrill Kelly finished with 2 strikeouts, leaving the saved over 3.5 on the wrong side of the line for a 1.00-unit loss.

At Bovada -135, the model gave the over 60.3% against a 57.4% break-even mark, a +2.9-point gap. The saved projection was 4.9 strikeouts across 24.6 batters faced. The pitcher graded 44.8/100 and the opponent strikeout matchup graded 37.0; the confirmed lineup carried a 22.9% strikeout rate.

The actual result finished 2.9 strikeouts below the point projection. The projection supported the ticket, but the actual result crossed to the other side of the sportsbook line. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability60.3%

Price break-even57.4%

Calculated difference+2.9 pts

Frozen pitcher-prop inputs

Merrill Kelly vs. COL

Projected Ks: 4.9
Actual Ks: 2

Pitcher value44.8

Opponent K matchup37.0

Lineup K rate22.9%

Lineup coverage100%

Projected batters faced24.6

Evidence quality91%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.37 Ks. Robustness: 79%.

Pitcher prop

Yoshinobu Yamamoto Under 6.5 strikeouts

SD at LAD · 4 strikeouts · under 6.5

Betonlineag -108winSee the review ↓

Postgame review

Won: Yoshinobu Yamamoto Under 6.5 Ks

Yoshinobu Yamamoto finished with 4 strikeouts, cashing the saved under 6.5 and adding +0.93 units to the prop record.

At Betonlineag -108, the model gave the under 55.1% against a 51.9% break-even mark, a +3.2-point gap. The saved projection was 5.8 strikeouts across 25.1 batters faced. The pitcher graded 57.8/100 and the opponent strikeout matchup graded 53.5; the confirmed lineup carried a 21.4% strikeout rate.

The actual result finished 1.8 strikeouts below the point projection. The ticket and the point projection told the same story. We will compare the same workload, opponent, line, and price profile before changing production weights.

Saved pregame inputs

Model probability55.1%

Price break-even51.9%

Calculated difference+3.2 pts

Frozen pitcher-prop inputs

Yoshinobu Yamamoto vs. SD

Projected Ks: 5.8
Actual Ks: 4

Pitcher value57.8

Opponent K matchup53.5

Lineup K rate21.4%

Lineup coverage100%

Projected batters faced25.1

Evidence quality92%

Pitcher and opponent values use the frozen 0–100 rolling scale. Projection standard deviation: 2.62 Ks. Robustness: 80%.

Logan Henderson: What had to hold

Strikeout count: the saved expectation was 2.59881 to 7.39113 strikeouts. The pregame model centered on 5.0 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Logan Henderson: What actually happened

The saved decision settled as a win, with +0.83 units on 1 units of reference risk.

Strikeout count finished at 3 strikeouts, inside the saved 2.59881 to 7.39113 range.

Logan Henderson: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Foster Griffin: What had to hold

Strikeout count: the saved expectation was 2.58282 to 7.36994 strikeouts. The pregame model centered on 5.0 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Foster Griffin: What actually happened

The saved decision settled as a win, with +1.02 units on 1 units of reference risk.

Strikeout count finished at 7 strikeouts, inside the saved 2.58282 to 7.36994 range.

Foster Griffin: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Gerrit Cole: What had to hold

Strikeout count: the saved expectation was 2.73498 to 7.63913 strikeouts. The pregame model centered on 5.2 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Gerrit Cole: What actually happened

The saved decision settled as a loss, with -1.00 units on 1 units of reference risk.

Strikeout count finished at 8 strikeouts, outside the saved 2.73498 to 7.63913 range.

Gerrit Cole: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Chris Sale: What had to hold

Strikeout count: the saved expectation was 3.68743 to 9.25286 strikeouts. The pregame model centered on 6.5 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Chris Sale: What actually happened

The saved decision settled as a loss, with -1.00 units on 1 units of reference risk.

Strikeout count finished at 10 strikeouts, outside the saved 3.68743 to 9.25286 range.

Chris Sale: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Kevin Gausman: What had to hold

Strikeout count: the saved expectation was 2.87644 to 7.87902 strikeouts. The pregame model centered on 5.4 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Kevin Gausman: What actually happened

The saved decision settled as a loss, with -1.00 units on 1 units of reference risk.

Strikeout count finished at 7 strikeouts, inside the saved 2.87644 to 7.87902 range.

Kevin Gausman: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Seth Lugo: What had to hold

Strikeout count: the saved expectation was 2.6718 to 7.53086 strikeouts. The pregame model centered on 5.1 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Seth Lugo: What actually happened

The saved decision settled as a loss, with -1.00 units on 1 units of reference risk.

Strikeout count finished at 3 strikeouts, inside the saved 2.6718 to 7.53086 range.

Seth Lugo: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Merrill Kelly: What had to hold

Strikeout count: the saved expectation was 2.51606 to 7.2651 strikeouts. The pregame model centered on 4.9 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Merrill Kelly: What actually happened

The saved decision settled as a loss, with -1.00 units on 1 units of reference risk.

Strikeout count finished at 2 strikeouts, outside the saved 2.51606 to 7.2651 range.

Merrill Kelly: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

Yoshinobu Yamamoto: What had to hold

Strikeout count: the saved expectation was 3.19471 to 8.43541 strikeouts. The pregame model centered on 5.8 strikeouts, with uncertainty around that estimate. The case becomes weaker if: the actual count falls outside the saved descriptive range; this alone does not diagnose the cause.

This range was frozen before the game as a descriptive check, not a calibrated confidence interval.

Yoshinobu Yamamoto: What actually happened

The saved decision settled as a win, with +0.93 units on 1 units of reference risk.

Strikeout count finished at 4 strikeouts, inside the saved 3.19471 to 8.43541 range.

Yoshinobu Yamamoto: What enters model review

These are checks of the original assumptions, not proof of what caused the result. A win does not validate every assumption, and a loss does not invalidate every forecast.

The measured observations can enter a source-verified review dataset. This recap changes no model weights; any proposed change requires separate chronological and prospective evaluation.

What comes next

What we take forward

Each result can raise a useful question, but no single game rewrites the model. Any adjustment has to repeat across comparable frozen reads and survive a prospective test first.

Technical receipt

Fact pack: 3b0d1622a591e8fd419bfe600bc6cfffeb8ed753c70803e1c0fa4524832b5986

Decision receipts: mlb-current-decision:53d9796af91d9b4b258fcbf026e46a79, mlb-k-prop-b48045e85656849a27360b5c, mlb-k-prop-d94dc9d429bc1ee5f0dc70f9, mlb-k-prop-a39016af83eabd03f743cfb4, mlb-k-prop-56ce502e0f13d05ed3671cf6, mlb-k-prop-1f7d1d18619df6be44844ed0, mlb-k-prop-379a3f45ba5f30639d547bd3, mlb-k-prop-d583f17b13c123fbe066c9ed, mlb-k-prop-7d7ed9f99aa98f1914ddd7c1

Settlement receipts: mlb_live_settlement_b7b6f56d17eeaf26b3068468, 9a2b51107d374bac123740b534a53a53b563648fcfe8f26799dfff11ebbfcf2a, 7f538c47c2c1dc8f1a0dafe016a548b9e8b1946bb3813d36b504d5bedc18e9f7, 7f585e1bc6eb6ac76320d9a79f70814fb7533d8f263ffd48d212f41e17e55fcf, cdc6a416788440eb6c0767883295753e88563fee22fabdd22f25a565008bbcff, 2cd2f8500a27b4323c1934ed058114657318838e9f3b9baf9e200416181d3ebb, e85866a3fbfc48fd9473dc93ee59fe9a4e28a07c9263fc941fb8c84eeb98ce56, 33f69addf3d0ace3927ec1f8e94d317b538f0bdbced215becff87ed36d19752b, 67a67a9966d2c36bceebf07f07b8bbdfa55fa2e1db40f1f997675a39f13aaa21