Read the full writeup →
World-record half-lives46 events · 1,251 reigns · October 2026

Which world records fall before LA 2028

I treated every ratified athletics world record in the 46 Olympic events as a survival time: it starts the day it is set and ends the day it is broken. The records come from World Athletics' own progression lists, checked against its records page on 28 October 2026. A model fitted on what was known at the end of 2010 beat the historical base rate on what happened next, and here is what it says about the 21 months to the LA 2028 Olympics.

Most likely to fallBefore 30 July 2028
1Pole vaultDuplantis, men6.31202648%
25000 mChebet, women13:58.06202535%
310,000 mChebet, women28:54.14202429%
4MarathonSawe, men1:59:30202629%
5High jumpMahuchikh, women2.10202428%
6Pole vaultIsinbayeva, women5.06200928%
7DiscusAlekna, men75.56202527%
81500 mKipyegon, women3:48.68202525%
9MarathonChepngetich, women2:09:56202424%
10400 m hurdlesMcLaughlin-Levrone, women50.37202424%

From tables/watchlist.csv (run of 28 October 2026). All 46 records, with intervals and a five-year horizon, are in the sortable watchlist. The backtest says to trust the order more than the levels.

The backtest

I cut the data at the end of 2010, fitted the model on what was known then, and asked for each of the 46 records standing that day: what is the chance this falls by 2025? 26 of them fell. The model's Brier score was 0.235 against 0.329 for the historical base rate and 0.250 for the Kaplan-Meier method the July version used. A rolling version, refitting every year from 1985 to 2020, shows the weak spot: records the model gave about a coin flip fell only about a quarter of the time.

Reliability plot of the rolling backtest. Below 40 percent predicted, the share of records that fell is close to the prediction; above that it falls well short, for example 24 percent observed where 50 percent was predicted.
Rolling backtest, 1,535 five-year predictions. Points below the dotted line are overconfident.

What makes a record last

  1. The period it is at risk in. Since 2000, records fall at about half the rate of 1980 to 1999 (hazard ratio 0.53).
  2. Margin. Each percentage point by which a record beat the one before it cuts the hazard by about a fifth (0.79).
  3. A busy event. Events that have just seen records keep seeing them (1.28 per unit of log count of records in the last three years).
  4. Event group. Throws and jumps records fall about twice as fast as sprint records. Half are gone inside a year.
Kaplan-Meier survival curves for sprints, middle distance, jumps and throws. Jumps and throws drop below one half before one year; sprints last longest with a median of about three years.
Kaplan-Meier survival for four of the ten event groups. The write-up has all ten, with the model's curve.

Rebuilt from World Athletics' lists

The July 2026 version scraped Wikipedia and kept 9 of 21 events. It missed rows in the women's 800 m and still listed Kiptum's 2:00:35 as the marathon record, three months after Sawe ran 1:59:30. This version reads World Athletics' statistics handbook and progression pages, keeps hand and automatic timing apart, and fails its own run if any current record differs from World Athletics' records page. The rules are in docs/data_decisions.md.