Site and Market Suitability Model

Assumptions

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

1. Not a universal score: hard gates apply first - PASS, FAIL, or UNRESOLVED - and unknown is never a fail and never a pass.
2. Dimensions are named and their basis printed; the weight policy is visible and versioned.
3. A weighted total exists only when all four dimensions are measurable - it is never computed over filled-in guesses.
4. The Pareto frontier is the decision object; sensitivity to the policy choice prints.
5. A sponsor stub is identity, not quality: sponsor record scores only with 3 or more resolved projects from the same sponsor, else UNKNOWN.
6. Coverage that is not comparable across candidates returns NO_DECISION - thin coverage is never normalized into comparability.

Derivation

Each candidate passes through the mandate's hard gates first. A gate checks only what the record shows - capacity against a minimum, state against an allowed list - and missing evidence resolves UNRESOLVED: distinct from FAIL, and never inferred into a pass.

Four named dimensions are then scored from the record: capacity fit (MW scaled to a 1000 MW reference), sponsor record (the operating share of the sponsor's resolved projects, requiring at least 3), queue position (a fixed status ladder), and schedule confidence (COD slip against a two-year reference). Every score carries its printed basis.

The weighted total is a plain weighted sum, defined only when every dimension is measured:

With any dimension UNKNOWN, no total exists - the candidate is carried on its measured dimensions, never on filled-in guesses. Candidates are compared on a Pareto frontier over the measured vectors, where one candidate dominates another only if it is at least as good on every dimension and better on one:

The answer is RANKED only when coverage is comparable - at least half the surviving candidates measurable on two or more dimensions; otherwise NO_DECISION with the reason printed. Policy sensitivity prints alongside: when the mandate weights and equal weights would crown different leaders, the page says the policy choice matters and the frontier, not a blended score, is the decision object.

Identification strategy

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

• Gates separate failed from unknown: UNRESOLVED is a third state, so missing data can never be laundered into a pass or counted as a fail.
• The all-or-nothing total is the anti-fabrication rule: a blended score exists only over fully measured candidates, so nothing is ranked on guesses.
• The Pareto frontier identifies what a weighted sum hides: candidates that trade dimensions against each other stay visibly incomparable instead of collapsing into one number.
• NO_DECISION is the comparability gate: when too little of the field is measurable, the model declines to rank rather than normalizing thin coverage.

Inputs and outputs

Calibration and gate evidence

Validation design

• Gate-verified discipline: missing data never infers a pass, unknowns stay separate from fails, no total is computed over filled-in guesses, policy sensitivity prints, and the decision is RANKED or NO_DECISION.

Evidence on record

• Smoke gates at seal (7 m19-specific): gates never infer a pass from missing data; unknown dimensions stay unknown, separate from fail; no total over filled-in guesses; policy sensitivity printed; decision is RANKED or NO_DECISION; determinism (identical input hash on re-run); audit row written for every run.

Gates at seal

• smoke: 7 m19 gates incl. missing-never-passes, no totals over guesses, RANKED-or-NO_DECISION, determinism + audit row (suite counts at seal per the build agent's seal report)
• sealed at 456d820 on atlas-nl-shock-redesign

Failure modes

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

• Capacity or state missing on a gated candidate: the gate resolves UNRESOLVED and says why - the candidate is not eliminated and not passed.
• Sponsor with fewer than 3 resolved projects: sponsor record is UNKNOWN - a stub is identity, not evidence of quality.
• Status off the queue ladder, or no announced/current COD pair: that dimension is UNKNOWN, and if the field thins past comparability the model returns NO_DECISION.

Sensitivity

• Policy choice can move the leader: the sensitivity line compares the mandate weights against equal weights and names the divergence when they disagree.
• Dimension scales are fixed references (1000 MW capacity, two-year slip): a change of reference would move every score together, and the chosen references are printed.