For 21 seasons, a Premier League side leading by one goal when the clock reached 90:00 went on to win 93.3% of the time: 2,870 of 3,077 leads. From 2023-24, referees stopped adding a nominal allowance for stoppages and began adding the time actually lost. In the three seasons since, the same lead has been held 367 times in 430, 85.3%. About one in 15 used to slip away after 90:00; now one in 7 does, 2.2 times as often. Each of the last three seasons sits below all 21 before it. The change happens after 90:00. In the last five minutes before it, a one-goal lead is as safe as it ever was.

Sourcing. Every completed Premier League match on ESPN's public scoreboard from 2002-03 to the 50 played so far this season: 9,170 matches, each with the minute ESPN records for every goal. For 2016-17 onward, ESPN's match summaries add the minute each half ended. Two checks came first. ESPN's minutes reproduce the league's own published counts exactly: 102 added-time goals in 2016-17, the record before the change, and 138 in 2023-24 when the league counted them in April 2024, 12.8% of all goals at that point. And the feed lists 2009-10 twice: 357 extra rows, filed under another season's label, repeat fixtures that are already listed, and 337 of them carry a different score. We dropped them, which leaves 380 matches in each of the 24 complete seasons. Twelve matches, all before 2017-18, whose goal lists do not match the final score stay out of the timing counts, which therefore run on 9,158 matches. The data is pulled by our pull scripts, and every figure below is recomputed by our chart script, which exits non-zero if any of them stops reproducing.

What changed, and when

Before 2023-24, the time added at the end of each half was an allowance: a nominal amount for each goal, substitution or injury. From the first weekend of 2023-24, Premier League referees added the exact time lost to goals and their celebrations, substitutions, injuries and other stoppages, as the league and Opta's analysts both described at the time. A year later the league trimmed the rule: the first 30 seconds after a goal are no longer added. Both announcements are listed in the sources below.

ESPN's match summaries show what that did to the clock. From 2016-17 to 2022-23 the second half ended, on average, at 90+5.3 on ESPN's clock, which shows the minute of the final whistle (90+5 is the fifth added minute). Since 2023-24 the average is 90+7.6: 90+8.1 in the first season, 90+7.1 after the 30-second trim and 90+7.5 last season. First halves went from 45+3.1 to 45+4.3. Before the change, 61.3% of second halves were over by 90+5 and 3.9% ran to 90+9 or later. Since, the figures are 17.8% and 28.8%. ESPN records the whistle in two places, the match's key events and its commentary, and the two agree on 3,825 of 3,847 matches. All 708 second-half added-time goals since 2016-17 fall by the recorded whistle.

Goals followed the minutes. Second-half added time produced 0.155 goals a match from 2016-17 to 2022-23 and 0.252 since. Last season it produced 104, the most of the 24 seasons and one more than the 103 of 2023-24, while the 90 minutes before it produced 2.36 goals a match, the fewest since 2008-09. Counting both halves, 147 of 1,045 goals in 2025-26 came in added time, 14.1%. No season before 2023-24 was above 9.6%.

When the halves ended, and what a one-goal lead at 90:00 was worth, by season. "Ended" is the average of ESPN's displayed minute for the end of each half. Source: ESPN public scoreboard and match summaries, retrieved 2026-10-05.
SeasonSecond half endedFirst half endedGoals in second-half added timeOne-goal leads at 90:00 held
2016-1790+5.145+2.968127 of 137 (92.7%)
2017-1890+5.045+2.849126 of 137 (92.0%)
2018-1990+5.245+2.972127 of 136 (93.4%)
2019-2090+5.645+3.451134 of 142 (94.4%)
2020-2190+5.145+2.750141 of 151 (93.4%)
2021-2290+5.445+3.372125 of 136 (91.9%)
2022-2390+6.145+3.649136 of 143 (95.1%)
2023-2490+8.145+4.7103116 of 137 (84.7%)
2024-2590+7.145+3.980122 of 142 (85.9%)
2025-2690+7.545+4.2104129 of 151 (85.4%)
2026-27 (50 matches)90+6.245+3.21017 of 20

Where the leads go

The chart's left panel is the headline: 21 seasons between 90.5% (2003-04) and 96.8% (2005-06), then 84.7%, 85.9% and 85.4%. The right panel is the test. Take every side leading by one at 85:00 and ask what happened to the ones that did not win. Before the change, 6.3% had lost the lead by 90:00 and not won it back, and 5.8% were still ahead at 90:00 and then failed to win. Since, 6.6% lost it before 90:00, about the same, and 12.5% lost it after. In those last minutes, the extra losses arrive after 90:00 and nowhere else.

Two-panel chart. Left: for each Premier League season from 2002-03 to 2025-26, the share of one-goal leads at 90:00 that ended in a win. The 21 seasons to 2022-23 sit between 90.5% and 96.8% around a pooled 93.3% (2,870 of 3,077). The three seasons since 2023-24, when referees began adding the exact time lost, sit at 84.7%, 85.9% and 85.4%, pooled 85.3% (367 of 430). An open marker shows 2026-27 so far at 17 of 20. Right: of the one-goal leads at 85:00, the share lost before 90:00 was 6.3% before the change and 6.6% since; the share lost after 90:00 was 5.8% before and 12.5% since.
Left: one-goal leads at 90:00 that ended in a win, by season. Right: leads at 85:00 that were not won, split by whether the lead was gone before or after 90:00. Data: ESPN public scoreboard, every completed Premier League match 2002-03 to 2026-27 to date, goal minutes as ESPN records them.

The same split from earlier checkpoints tells the same story. From every one of them, the share of leads lost after 90:00 is 1.9 to 2.3 times the old rate. From 75:00 the share lost before 90:00 also rose, from 15.3% to 18.3%, but most of that is one season. 2023-24 scored 2.89 goals a match in regulation time, the most of the 24, and lost 21.6% of its 75th-minute leads before 90:00. The next two seasons lost 16.3% and 17.5%, within about two points of the old 15.3%.

One-goal leads at four points in the match and what became of them. "Lost before 90:00": the lead was gone by 90:00 and the side did not win. "Lost after 90:00": still ahead at 90:00, did not win. Won includes the few sides that lost the lead before 90:00 and won after it. Source: ESPN public scoreboard, 9,158 matches with goal times.
Lead atSeasonsLeadsWonLost before 90:00Lost after 90:00
half-time2002-03 to 2022-233,36069.9%27.5%2.5%
half-time2023-24 to 2025-2650062.2%33.0%4.8%
75:002002-03 to 2022-233,25380.1%15.3%4.5%
75:002023-24 to 2025-2645971.5%18.3%10.2%
85:002002-03 to 2022-233,10587.9%6.3%5.8%
85:002023-24 to 2025-2644181.0%6.6%12.5%
90:002002-03 to 2022-233,07793.3%0.0%6.7%
90:002023-24 to 2025-2643085.3%0.0%14.7%

It is not about who was leading

A lead held by a stronger side is safer, and home sides hold leads better than visitors, so a change in who tends to lead could move the rate on its own. It did not. Home sides held 56.6% of the one-goal leads at 90:00 before the change and 52.8% since. At the old mix, the new home and away rates would give 85.6%, nowhere near 93.3%. The side with the better points-per-game record that season was the leader 65.2% of the time before and 64.9% since. And the drop shows up inside every group: home leaders 94.8% before and 88.1% since, away leaders 91.2% and 82.3%, leaders with the better record 95.2% and 87.8%, the rest 89.6% and 80.8%.

Two goals are still enough

None of this touches a two-goal lead. Sides two up at 90:00 won 1,679 of 1,681 times before the change and 244 of 244 since. The longer added time has made a one-goal lead fail about one time in seven. In three seasons, no side two goals down at 90:00 has come back to take a point.

What a lead at 90:00 is worth now

Points make the change concrete. Since 2023-24 the 430 one-goal leaders at 90:00 won 367 matches, drew 56 and lost 7: 3 × 367 + 56 = 1,157 points, and 1,157 ÷ 430 = 2.6907 points a lead. Before the change, 2,870 wins and 204 draws from 3,077 leads made 3 × 2,870 + 204 = 8,814 points, and 8,814 ÷ 3,077 = 2.8645 a lead. The gap is 0.1738 points a lead. At the old rate the 430 leads would have earned 430 × 2.8645 = 1,231.7 points. They earned 1,157, so 74.7 points slipped away over three seasons, about 25 points a season.

Count every match, not only one-goal leads, and the result at 90:00 differed from the final result in 484 of 7,968 matches before the change (6.1%, 23 a season) and in 129 of 1,140 since (11.3%, 43 a season). Last season had 52, the most of the 24. A match level at 90:00 is now decided after it 22.8% of the time, against 13.5% before.

The first 50 matches of 2026-27

This season's sample is small, and it points the same way. Through 20 September, the last league round before the break, sides a goal up at 90:00 held on 17 of 20 times. Three did not. Newcastle, 2-1 up at home to Liverpool, conceded at 90'+9' and drew 2-2. Bournemouth, 1-0 up at home to Everton, conceded at 90'+1'. Manchester United, 2-1 up at Everton, conceded at 90'+6'. One level match went the other way, when Manchester City beat Bournemouth 2-1 with a goal at 90'+1'.

Second halves this season have ended at 90+6.2 on average, earlier than in any of the last three seasons. The league changed several laws for 2026-27: substituted players have ten seconds to leave the pitch, clearly delayed throw-ins and goal kicks face a five-second countdown, and a player treated for an injury must stay off for a minute. Fifty matches cannot say whether those changes, or chance, explain the shorter added time. The league resumes on 10 October.

What this does and does not say

It is a before-and-after comparison, not an experiment. The timekeeping change is the obvious cause. The drop starts the season the change did, it sits after 90:00 and not in the five minutes before, it survives a season of low scoring, and it shows up for home and away leaders, stronger and weaker. Other things changed in those seasons too, and a comparison like this cannot rule all of them out.

ESPN's minutes are whole minutes. A goal at 89:30 is shown as 90' and counts as before added time, and a whistle shown as 90+7 came somewhere in the seventh added minute. ESPN once printed 89'+2' for a goal its own clock puts at 88:22; we counted it in the 89th minute.

Added time cannot be read match by match. A late goal and its celebration lengthen the added time they happen in, so long added time and late goals feed each other inside a match. That is why this compares seasons rather than matches with long and short added time.

Three seasons is a small sample. The 95% interval around 85.3% runs from 82.0% to 88.7%. That still sits wholly below the worst of the 21 earlier seasons.

A win-probability model fitted to the seasons before 2023-24 would now put a one-goal lead at 90:00 about eight percentage points too high, which is why how those models handle the clock matters more than it did. The trailing side's all-out push in those minutes is the same game-state effect that bends every other number in a match, and the World Cup's own version of the late-goal pattern is in When World Cup Goals Are Scored.

How the numbers are made

Everything here rests on one step: reading the score at 90:00 from ESPN's goal minutes. Every goal counts except those shown as 90'+N'. In Python, with each match holding its goals, the home side's id and the final score, that step looks like this; run on the same data, it finds the 3,077 and 430 leads above.

import re

def minute(shown):
    """ESPN shows "67'" or "90'+3'": return (67, 0) or (90, 3)."""
    base, added = re.match(r"(\d+)'(?:\+(\d+)')?", shown).groups()
    return int(base), int(added or 0)

def score_at_90(match):
    """Count every goal except those shown as 90'+N' (second-half added time)."""
    home = away = 0
    for goal in match["goals"]:
        base, added = minute(goal["min"])
        if base == 90 and added > 0:
            continue
        if goal["team"] == match["home_id"]:
            home += 1
        else:
            away += 1
    return home, away

def held(match):
    """None unless a side led by exactly one at 90:00; then True if it won."""
    home, away = score_at_90(match)
    if abs(home - away) != 1:
        return None
    final = match["hs"] - match["as"]
    return final > 0 if home > away else final < 0

The rest is counting: for each lead, the final result, and for each earlier checkpoint, whether the lead was still there at 90:00. Seasons run from August to July, so the June and July 2020 matches count in 2019-20.

Sources & Further Reading

Published by SoccerAnalytics from public data and the sources named in this piece. How this site is made.