Documentation · v0.2

How the Atlas scores.

Every state and site gets six rankings — now expanded with compute-hub and major charging-network context. Below is what goes into each ranking, how confidence is measured, and how we keep the model current.

On this page

  1. Principles
  2. The six modes
  3. Six scoring components
  4. The scoring formula
  5. Map layers
  6. Confidence & caveats
  7. Roadmap

1. Principles

Every number is citable. If a metric can't trace back to a named public source with a version and pull date, it isn't in the scoring. See the source registry.

Modes over monoscores. A single "energy score" is dishonest — Texas looks perfect for merchant solar and terrible for regulated PPAs. So we publish six lenses, and let you pick the one that matches your model.

Recency is explicit. Atlas v0.2 applies freshness weighting by source cadence. Signals from annual filings have lower confidence than weekly or monthly operational feeds unless corroborated by multiple datasets.

2. The six modes

Each mode is a different weighted blend of the six components in §3. Pick the mode that matches how you monetize a site.

3. Six scoring components

Every mode's total is a weighted average of these six 0–100 subscores.

3.1 Grid Access

Interconnection queue time (median months from application to COD), transmission headroom (regional import/export capacity), and curtailment risk (recent-year negative-LMP hours). Sourced from FERC Form 715, EIA 861/930, and each ISO's public queue reports.

3.2 Market Opportunity

Wholesale LMP hub prices (24-mo average, off-peak + peak), capacity market clearing (where applicable), and REC / SREC prices. In non-RTO states, we substitute avoided-cost and utility IRP forecasts.

3.3 Renewable Potential

Solar GHI and DNI from NSRDB, wind class distribution from NREL WIND Toolkit, and geothermal / small-hydro potential from open NREL reV runs.

3.4 Policy Score

RPS stringency and CES target, DSIRE-tracked incentives, interconnection cost socialization, PURPA activity, and whether the state has a functioning PUC stakeholder process. Points added for IRA §48E / §45Y adders eligibility.

3.5 Demand Growth + Compute

Forecast load growth from EIA STEO + AEO regional splits, hyperscale datacenter announced MW, and derived compute-hub clustering (regional MW concentration, active-build share, operational share).

3.6 Land + Water + Mobility Access

Land/water constraints remain foundational, with a mobility-access adjustment from major-network EV charging density and DC-fast port presence in target corridors for transport and logistics load scenarios.

4. The scoring formula

For each state s and mode m:

// weighted total with confidence decay, floored at 0, ceilinged at 100 score(s, m) = clamp( Σᵢ wᵢ,m · componentᵢ(s) · confidenceᵢ(s), 0, 100 ) // where Σ wᵢ,m = 1 for every mode

Weights per mode are published in lettuce-energy-atlas.js as the MODES constant. Confidence terms are derived from source age, source reliability class, and cross-source agreement.

Why not ML? We deliberately use interpretable, hand-weighted composites in v0.1 — a developer needs to defend every knob when the site scores well or poorly. A learned model earns its keep in v0.3+ once we have enough site-outcome labels to avoid overfitting to headline noise.

5. Map layers

The map recolors under any of six layers, decoupled from the scoring mode:

6. Confidence & caveats

7. Roadmap

v0.3 (Q4 2026) — tile-backed heavy layers by default, county-level queue risk, and compute-aware parcel ranking.

v0.4 (H1 2027) — scenario planner with explicit build-vs-buy energy strategies for AI campuses, fleets, and industrial loads.

Contribute — the source registry lives in registry JSON. Open a PR or email atlas@lettuce.energy to nominate a source.

Cite this doc Lettuce Energy Atlas v0.2 — Methodology. lettuce.energy. Retrieved .