NukeWorker Outage Schedule Scorecard

The Challenge

Nuclear utilities publish outage schedules months in advance, but the reality rarely matches the plan. Refueling outages routinely extend days or weeks beyond the published end date. And when an unplanned shutdown occurs, there's no published estimate at all — operators and stakeholders are left guessing.

We set out to answer two questions: Can we predict how long a scheduled outage will actually take? And when a forced outage happens, how long will it last?

Our Approach

Rather than relying on a single prediction method, we developed and rigorously tested multiple competing models against our historical database. Each model was backtested against thousands of real outage outcomes to measure its accuracy before being put into production.

Rigorous Backtesting

Every model earns its place. Before deploying a prediction approach, we backtest it against historical outcomes — running the model as if we were predicting in real time, then comparing the prediction to what actually happened.

We tested over a dozen competing prediction strategies — from simple historical averages to adaptive models that adjust based on how long an outage has already lasted. The winning models were selected based on mean absolute error (how many days off from actual), coverage (what percentage of outages get a prediction), and consistency across both BWR and PWR reactor types.

Key Insight: BWRs and PWRs Are Different

One of our most significant findings was that boiling water reactors (BWRs) and pressurized water reactors (PWRs) exhibit meaningfully different forced outage behavior. BWR forced outages are more predictable, while PWR forced outages have higher variance and require more sophisticated modeling.

Rather than forcing a one-size-fits-all approach, our production models use different prediction strategies for each reactor type, each independently validated against that type's historical data.

Continuous Improvement

Our models aren't static. As new outage data comes in, the system recalculates predictions and accuracy scores automatically. The leaderboards on this page reflect live, up-to-date accuracy measurements — not cherry-picked results.

Every unit's accuracy score is computed by comparing what we predicted against what actually happened, weighted toward recent performance. If a unit's behavior changes, our model adapts.