Named-vs-Pile Model

Assumptions

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

1. The staircase is deterministic bucket accounting: every ask is classified by fixed inclusion rules - printed pile, tracked, named (utility / queue / docket confirmed), bound, evidenced, P50 deliverable, P90 deliverable.
2. No estimation happens inside the buckets. The only modeled step is the Delivery Model applied to the evidenced bucket.
3. The pile is never presented as deliverable. Buckets past P90 are shown, not averaged away.

Derivation

The staircase is a monotone filter chain over the load books. Each rung keeps only the asks that clear its inclusion rule:

Rung membership is accounting, not judgment: an ask is 'named' when a utility, queue or docket record confirms it; 'bound' when a binding instrument attaches; 'evidenced' when the evidence graph carries it. Only then does any probability touch the numbers: the Delivery Model runs on the evidenced bucket and yields the P50 and P90 deliverable rungs. The waterfall is reported bucket by bucket, with the tail past P90 always shown.

Identification strategy

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

• The model's identification is its refusal to estimate: every drop between rungs is a documented inclusion rule, so the gap between the printed pile and the named number is attributable bucket by bucket - not a haircut whose size depends on an analyst's mood.
• The ERCOT pile is the canonical case: the gap between the pile's headline capacity and its named capacity is not 'risk'; it is asks that never cleared a naming rule, and the model keeps those two ideas separate by construction.

Inputs and outputs

Calibration and gate evidence

Validation design

• Bucket rules are tested by forward audit: asks classified 'evidenced' must resolve into real projects at a rate consistent with the bucket's label.

Evidence on record

• Forward-audit results are not yet published in the sealed source.

Failure modes

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

• An ask that fails a naming rule drops a rung; it is never carried in a higher bucket on the strength of a headline.
• Asks with unresolved identity stay at their last evidenced rung - deduplication failures would double-count the pile, so ambiguous asks do not advance.

Sensitivity

• The named rung is the sharp cliff: most of the pile's GW fails there, which is the model's central empirical finding, not a tuning artifact.
• P50 vs P90 spread inherits the Delivery Model's width on the evidenced bucket.