Capacity-at-Risk Model

Assumptions

Everything the model believes, stated before the math. If you do not buy one of these, do not buy the output.

1. Requested capacity is not awarded capacity, and planned is not delivered - the model reads only the promised MW on the project record and the Delivery Model's delivered-MW path.
2. MW at risk is a P50 concept: promised MW minus the P50 expected delivered MW at the end of the horizon. It is a median shortfall, not a worst case.
3. Attribution is computed per delivery path (grid vs behind-the-meter, M14); the paths are never blended.

Derivation

Expected delivered MW by date is the Delivery Model's P50 delivered-MW path, computed separately per delivery path. MW at risk is the terminal shortfall:

Attribution is leave-one-dependency-out. For each incomplete dependency d, the Delivery Model is re-run with d forced complete; the recovered MW is the increase in terminal expected delivery:

The dependency with the largest recovery is the binding one. If no dependency's removal recovers anything, the shortfall is base-rate slippage, not a single point of failure - and the model says exactly that.

Identification strategy

Why the estimate supports the claimed meaning - and where the model refuses to claim more than the data carries.

• MW at risk is identified off the P50 path, not a scenario: it is the shortfall the median future already carries, which is the figure a credit book can act on without agreeing to a stress case.
• The LODO attribution identifies the binding dependency by counterfactual recovery, not by correlation with delay. A dependency is binding only if forcing it complete moves the delivery path - the same interventional logic as the Scenario Engine, read one edge at a time.

Inputs and outputs

Calibration and gate evidence

Validation design

• Shares the Delivery Model's rolling-origin backtest; attribution is sanity-checked against resolved delays in outcomes - the dependency the model called binding should be the one the record shows slipping.

Evidence on record

• No separate calibration results are published in the sealed source beyond the shared Delivery backtest design.

Failure modes

The ways this model declines, degrades or goes UNKNOWN - and what it prints when it does.

• No promised MW on the project record: MW at risk and binding dependency are UNKNOWN.
• No incomplete dependencies: attribution is empty and the model reports that the shortfall is base-rate slippage, not a binding edge.
• A dependency whose forced completion recovers nothing is reported with zero recovery - it is not silently dropped from the table.

Sensitivity

• The answer inherits the Delivery Model's sensitivities wholesale (sponsor multiplier, gate profile, wide prior on thin classes).
• MW at risk is a P50 figure; on a wide delivery distribution the same project can carry a far larger P90 shortfall - read it with the Delivery page, not alone.