Structural Break: Real-Time Edition
Following last year’s success of the competition, ADIA Lab launches the Structural Break: Real-Time Edition, in partnership with CrunchDAO.
Most predictive systems assume that the world they model tomorrow will behave like the world they modelled yesterday. When the data-generating process changes, that assumption breaks. This moment is a structural break.
This challenge turns that problem into a real-time detection task. Given a time series, participants determine whether the underlying behaviour has shifted or whether the variation is noise.
What Changes in 2026
The 2026 edition keeps the same scientific question and changes the mechanics.
Participants receive a historical reference segment at the start. The online segment, where a break may occur, is revealed one observation at a time. After each new observation, participants output a score between 0 and 1 representing their cumulative confidence that a break has already occurred.
There is no way to look ahead. A break may occur at any point in the online segment, or not at all.
Why the Format Matters
Structural breaks do not occur at known boundaries. They appear within ongoing data, and missing them means continuing to operate under outdated assumptions.
This applies across finance, healthcare, industrial systems, and climatology, where early detection is critical.
Competition Details
Prize pool:
$100K USDC
Duration:
May 4 to September 17
Dataset:
10,000+ univariate
time series per split
Platform:
CrunchDAO
Teams: up to 5 members, with only the team leader ranked
Winners announced at the ADIA Lab 2026 Symposium (26–28 October)
Who Should Participate
Data scientists, machine learning engineers, statisticians, and researchers working on time series analysis, sequential modelling, and real-time systems.