Re-Underwriting 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 comparison is pinned: the frozen decision snapshot, never a reconstructed memory of it.
2. The variable set is fixed (spec Section 56): capacity, COD, permits, interconnection, CAPEX, contracts, sponsor, dependencies, financing.
3. Severity aggregation uses fixed weights: HIGH = 1, MEDIUM = 0.5, LOW = 0.15.
4. Background change detection that fires alerts is pipeline rule R1, not this model.

Derivation

The model diffs the frozen snapshot against the current record field by field over the Section 56 set, then re-runs the affected models against both states so the comparison reads in model terms - 'physical delivery confidence fell from 78% to 61%' - with the field-level reasons attached.

The material change score aggregates the severities the Materiality Screen (R2) assigned to the underlying changes:

Identification strategy

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

• Pinning the snapshot is the identification: change is measured against what was actually recorded at decision time, so the answer cannot be moved by hindsight or by edits to the current record's history.
• Re-running the models on both states makes the diff interventional rather than descriptive: the delta in delivery confidence is the recomputed difference, not a narrative about the fields that moved.

Inputs and outputs

Calibration and gate evidence

Validation design

• Replay: pick historical decision dates on resolved projects, re-underwrite, and check the flagged changes against what actually went wrong.

Evidence on record

• Replay 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.

• No decision snapshot on the case: there is nothing pinned to compare against and the model says so rather than reconstructing one.
• Fields absent from the snapshot are reported as unrecorded, not as unchanged.

Sensitivity

• The severity weights set how the score trades one big change against several small ones: HIGH=1 vs LOW=0.15 means one material COD slip outweighs six minor field edits.