When Goliath Falls: The Biggest ELO Upsets in NFL History

Methodology

ELO ratings condense a team's strength into a single number that updates after every game. We took each team's pregame rating (elo_home_pre, elo_away_pre), found games where the heavier-rated side lost outright (home_win / away_win), and ranked them by the size of the pregame ELO gap — the bigger the gap, the more improbable the upset.

import pandas as pd
df = pd.read_csv("spreadspoke_enhanced.csv", low_memory=False)
d = df.dropna(subset=['elo_home_pre','elo_away_pre']).copy()
d['gap'] = d.elo_home_pre - d.elo_away_pre
home = d[(d.gap>0) & (d.home_win==False)]      # home favorite lost
away = d[(d.gap<0) & (d.away_win==False)]       # away favorite lost
# rank both by absolute ELO gap and take the top 12
print(home.nlargest(6,'gap')[['schedule_date','team_home','score_home','score_away','team_away','gap']])

The Analysis

ELO ratings make a previously fuzzy question — “how big an upset was that, really?” — precise. By boiling each team down to one number that rises and falls with results, we can rank every game in history by how mismatched the two sides were on paper, then surface the times the favorite still lost. The result reads like a highlight reel of the impossible.

At the top sits a genuinely famous one: on 1973-12-09, the Miami Dolphins entered rated about 454 ELO points above the Baltimore Colts and lost 3–16 anyway. The rest of the registry is a parade of similar shocks across all 60 seasons — the kind of games that make fans groan and make the dataset fun to dig through.

Beyond the trivia, the list is a reminder of why single-game upsets are so hard to predict and so lucrative when they hit: even a 400-point ELO edge — roughly the distance between a great team and a bad one — is not a guarantee. The full table is below, sorted by pregame ELO gap.