Live Decision Ledger: Analytics vs. Coaching Divergence

## Entry 001: Nick Sirianni, Eagles vs. Falcons — Sept. 16, 2024 - **Sport:** NFL - **Teams:** Philadelphia Eagles at Atlanta Falcons - **Situation:** 4th-and-3 from Falcons' 10-yard line, Eagles led 18-15, ~1:40 remaining - **Model recommendation:** Go for it (strong consensus: ESPN, NFL NextGen Stats, NFL4th.com; Wharton leaned go with wide CI) - **Coach's choice:** Kick 28-yard field goal - **Stated reasoning:** "Go up six points" - **Outcome:** Atlanta drove for TD, won 22-21 - **Coach aggressiveness vs. model:** Less aggressive - **Sources:** CBS Sports, Philadelphia Inquirer, Wharton Sports Analytics - **Notes:** Classic case of playing not to lose. Model consensus was unusually strong; Wharton's wide confidence interval is a useful reminder that even strong recommendations carry uncertainty. --- *This ledger is updated weekly. Entries are tagged by sport, coach, decision type, and whether the coach was more or less aggressive than the model.* ## Entry 002: Kevin Cash, Rays vs. Brewers — World Series Game 5, Oct. 2026 - **Sport:** MLB - **Teams:** Tampa Bay Rays vs. Milwaukee Brewers - **Situation:** Drew Rasmussen pulled after 4 innings, 48 pitches, 1 run allowed, Rays trailing 0-1 - **Model recommendation:** No cited analytics model recommended keeping him in; one independent blogger noted efficiency metrics favored staying, but this is not an industry model - **Coach's choice:** Pulled Rasmussen per pre-set plan ("60 pitches or four innings") - **Stated reasoning:** "We had gone into the game saying it was 60 pitches or four innings. Stuck to that. That was the easiest call in the world to take him out of the ball game." - **Outcome:** Bullpen allowed 4 runs, Brewers won 4-1 - **Coach aggressiveness vs. model:** N/A — not a clear analytics divergence; more a pre-set plan vs. in-game adaptability question - **Sources:** AOL, Yardbarker - **Notes:** Criticism focused on pulling an effective pitcher, but no analytics model was cited as recommending the opposite. Raises question: when should a pre-game plan yield to in-game evidence? ## Pending: Red Sox Wild Card — Sept. 30, 2026 - **Sport:** MLB - **Teams:** Boston Red Sox vs. New York Yankees - **Situation:** Manager reportedly pulled Sonny Gray with 2 outs in 5th inning - **Claimed divergence:** Newsweek said data favored keeping him in (xwOBA .231), manager said "you feel like you've got to be aggressive and go" - **Caveat:** Manager name "Chad Tracy" is unverified; no independent source found for the data claim - **Status:** NOT ADDED to ledger pending verification ## Search Note: Counter-Examples (Nov. 2, 2026) Searched NFL decisions (2024–2025) where a coach overrode analytics and was demonstrably correct—not lucky. **Result: none found.** - **Dan Quinn** (Commanders vs. Eagles, Nov. 2024): Went for it on 4th-and-2 when model said kick. Failed. - **Mike Macdonald** (Seahawks vs. 49ers, Sept. 2025): Kicked when model said go. Lost; outcome turned on a single play (Purdy's 3rd-down TD). - **Nick Sirianni** (Eagles vs. Falcons, Sept. 2024): Claimed analytics supported kick; not a clear override. **Implication:** Either models are genuinely better at these decisions, media coverage is biased toward cataloging conservative mistakes, or successful overrides are under-reported. Flagged for follow-up in other sports or broader historical search. ## Search Note: NBA Counter-Examples (Nov. 2, 2026) Searched NBA decisions (2024–2025) where a coach overrode analytics and was demonstrably correct. **Result: none found.** - **Rick Carlisle** (Pacers, 2025 NBA Finals Game 1): Held timeout, Haliburton hit game-winner. But no source says a model recommended a timeout; shot-clock logic may have been analytics-consistent. - **Joe Mazzulla** (Celtics, 2025 East Semis Game 2): Didn't call timeout, lost. - **Rick Carlisle** (Pacers, 2025 East Finals Games 1 & 3): Skipped timeout, lost. - **David Adelman** (Nuggets, 2025 Playoffs Game 2): Held timeout, lost. - **Tom Thibodeau** (Knicks, 2025 Playoffs Game 5 vs. Pistons): Delayed timeout, lost. **Pattern:** Media documents conservative mistakes and failed overrides, but not successful overrides. Same as NFL. **Implication:** Either models are genuinely better at these decisions, successful overrides are under-reported, or the bar for "demonstrably correct" is too high for single-game decisions. Need to search MLB, NHL, soccer, or historical decisions. ## Search Note: Soccer/Football Counter-Examples (Nov. 2, 2026) Searched soccer/football decisions (2024–2025) where a manager overrode analytics and was demonstrably correct. **Result: none found meeting full criteria.** - **Fabian Hürzeler (Brighton vs. Man City, Aug. 2025):** Quadruple sub while trailing, won 2-1. Described as "a feeling that comes from inside." But **no documented analytics recommendation against the changes**. Guardian frames it as instinct winning, but this is opinion, not measured outcome. Cannot separate sound reasoning from luck. - **Manolo González (Espanyol, La Liga 2024-25):** Beat relegation forecasts (Opta: 56.8% in Dec.). But this is a **season-long forecast being beaten, not a specific decision override**. Survival compatible with both sound coaching and luck. **Pattern continues:** Media documents conservative mistakes and failed overrides, but not successful overrides. Clubs rarely publish internal model disagreements. **Next:** Search MLB (beyond World Series) and NHL. ## MLB Findings (Nov. 2, 2026) ### Carlos Mendoza, Mets vs. Cardinals — April 28, 2024 - **Situation:** Tie game, 8th inning. Matchup data favored removing Quintana for a right-handed reliever. - **Analytics recommendation:** "The book, the binder, the computer — they all suggested removing Quintana in favor of a right-handed pitcher." (MLB.com) - **Override:** Kept Quintana in based on pitcher's conviction and body language — information a matchup model does not contain. - **Outcome:** Quintana retired the next batter, game stayed tied, Mets won 4-2 in extra innings. - **Assessment:** Closest match found across all sports searched. Override is documented, reasoning points to information a model lacks. But single plate appearance cannot separate skill from luck. No pre-play record shows how often this type of read is right. - **Source:** https://www.mlb.com/news/carlos-mendoza-trusting-gut-feel-as-mets-manager ### Key Pattern After searching NFL, NBA, soccer, MLB, and limited NHL (2024–2025): **no fully verified case exists** where a coach/manager overrode a documented analytics recommendation and was demonstrably correct (not lucky). The Mendoza case is the closest match. Possible explanations: 1. Models are genuinely better at these decisions than human intuition 2. Media coverage is biased toward reporting conservative mistakes over successful overrides 3. Successful overrides are under-reported because clubs rarely publish internal model disagreements 4. The bar for "demonstrably correct" is too high for single-game decisions