# Site Selection Atlas: decision validity, v14

This release fixes cases where an attractive preference score concealed a failed project requirement, inadequate annual price evidence, or a different land-development pathway. The tool now separates **project requirements**, **comparable financial evidence**, **development pathway** and **weighted regional preferences**.

The result is a shortlist for investigation. It does not establish the best purchasable parcel, a profitable investment, available utility service or development approval. Unknown commitments remain unknown; a known failed requirement is no longer treated as merely a missing score component.

## Visible before-and-after behavior

| User action or finding | Previous behavior | v14 behavior |
| --- | --- | --- |
| Select a solar location whose closest node has only 360 December hours | A short seasonal sample could be annualized; its low coverage then reduced the influence of poor economics | Candidate price references are screened for annual and monthly sufficiency. An adequate nearby reference is preferred by evidence quality and distance, not favorable price. If none qualifies, annual valuation and the headline site score are withheld |
| Enter service below facility peak, an energization date after the target, or insufficient cooling water | The physical deficits appeared in the report while a favorable score could remain | The headline says **Project brief not met**, shows the failed requirements, and has no numerical preference score |
| Increase requested MW beyond entered usable acreage or turbine layout | Output could shrink while the location retained a favorable site score | Requested capacity and the actual modeled layout are checked separately. An undersized layout fails the brief; unknown parcel area remains unverified |
| Investigate the reviewed battery winner at 34.7, -120.6 | Regional indicators did not distinguish the military land pathway | Bundled BLM geometry identifies U.S. Air Force management. The ordinary greenfield score is withheld and the report identifies the military-host pathway |
| Place a candidate on recorded solar arrays or wind turbines | Existing infrastructure could appear as an ordinary new-build opportunity | The report identifies documented occupied infrastructure. Ordinary greenfield rankings require a compatible existing-site, co-location or repowering pathway |
| Encounter military/tribal/conservation and occupied infrastructure together | A single headline category could conceal a second required pathway | Every intersecting special pathway is checked. A single matching selection cannot bypass other requirements; **Mixed land / existing-asset partnerships** explicitly acknowledges multiple paths without claiming consent |
| Model hydrogen that cannot meet stated customer demand | Production potential and a high regional score could coexist with inadequate delivery | Actual customer withdrawals are compared with the stated service target. Failed delivery requirements suppress the site score. A production-only brief remains explicitly exploratory |
| Compare adjacent high-scoring grid samples | Many near-identical nearby points could fill the shortlist | Regional finalists are separated by at least **25 km**. The heatmap retains every sampled score; the shortlist provides geographically distinct alternatives |
| Read a high score with negative NPV | Poor economics could be buried beneath resource and map context | **Negative scenario NPV** appears beside the headline score. A preference score is explicitly not a profitability finding |

## 1. Annual price admissibility

Annual historical valuation requires a complete calendar reference, **at least 90% of annual price hours and at least 75% of every calendar month**. These are transparent screening thresholds, not a claim that historical prices predict future revenue. They prevent a short seasonal sample from masquerading as an annual project scenario. Missing hours are not filled with invented observations.

Prepared monthly hour counts in the bundled index support deterministic reference selection before loading a price vector. The actual timestamped vector supersedes metadata once loaded. Selection preserves the strongest supported electrical association, then prefers annual evidence sufficiency, distance and a stable node identifier within that tier. An inadequate directly associated reference does not silently yield to a weaker electrical association. Geographic fallback references are limited to 50 km. Price magnitude does not determine the selected node.

Battery and generation eligibility require sufficient annual price evidence and a finite technology-specific lifetime valuation. When that evidence is missing, changing preference weights cannot restore a headline score. Supporting observed prices and physical output can still be inspected. A generation model can therefore retain known weather-based production while withholding unsupported annual revenue and NPV.

For large loads, wholesale LMP is not substituted for an absent delivered bill. Comparable cost evidence requires one uniquely mapped serving utility and one supported conditional tariff scenario. Overlapping utilities and conditional tariff alternatives remain visible, but the model does not choose the cheapest utility automatically or imply a retail service agreement.

Hydrogen's current co-located renewable-supply design uses its process production cost for exploratory eligibility. It does not require an unrelated LMP history to describe production potential. A customer-supply brief also has to pass its delivery requirements.

## 2. Mandatory requirements precede preferences

`project-fit.js` assesses pass, fail, unknown and not-applicable conditions. `decision-eligibility.js` supplies independent gates to the shared score summary. A failed gate cannot be compensated by changing user weights. Missing commitments remain explicitly unconfirmed instead of being converted into zero physical capacity.

| Technology | Required conditions evaluated |
| --- | --- |
| Battery | Entered project acreage; positive import/export paths; explicit required import/export MW when supplied; feasible reserved contract power and contracted cycling |
| Solar/wind | Requested capacity fits usable area, setbacks and actual whole-machine layout; entered developable area meets the layout envelope; positive or explicitly required export capability; useful output |
| Geothermal | Entered acreage and export requirements; stated sustainable reservoir capacity meets requested gross MW; useful generation after the chosen pumping/design assumptions |
| Large load/data center | Electrical service meets complete facility peak; energization meets target year; cooling-water supply meets peak consecutive 24-hour demand; cooling equipment meets hourly thermal demand; stated backup and independent-fiber requirements |
| Hydrogen | Positive production and, when specified, positive customer demand; actual delivered mass-service target; optional fully served-hour and maximum-interruption requirements; explicitly required storage buffer; separate renewable land requirements; constrained process-water scenario |
| Hydro/bioenergy | Positive useful output under the geographically matched or explicitly entered process resource and project-area assumptions |

Default wind capacity now budgets enough whole turbines to meet the requested project rather than silently rounding the project down. In the reviewed default 150-MW cases, the selected reference machine installs **151.65 MW**. Capital and fixed O&M budget those machines; the requested export design remains 150 MW. A user-entered smaller layout remains a failed brief.

The checks are intentionally architecture-specific. An entered generation export limit below nameplate can represent deliberate clipping unless the brief specifies a higher required export MW. Likewise a battery connection limit can deliberately derate operation; separate required import/export fields express non-negotiable capacity. Entered capacities, dates and water volumes are scenario inputs, not verified reservations.

## 3. Public land and occupied infrastructure

The new September 23, 2026 bundle contains:

- **53,342** BLM California surface-management polygons.
- **893** USGS California large-scale solar array geometries.
- **4,935** USGS California wind turbine locations, including rotor diameter where reported.

The same local data are used for scanning and individual reports. Pin interaction does not trigger a live land lookup. Source URLs, query filters, feature counts, source edit dates when available, decoded agency codes and hashes are saved in `data/developer-land-metadata.json`.

All three bundled layers must be nonempty and match their declared feature counts. A missing or truncated bundle withholds the headline score; it cannot silently become a clean land finding. A complete bundle with no local intersection remains usable evidence, with parcel availability and permission still unverified.

The assessment checks a drawn footprint, an equivalent area from entered acreage, or an MW-dependent required-area envelope. Management and solar overlap presence includes polygon boundaries, holes, narrow crossing strips and enclosed small features. Management percentages remain deterministic area samples; they are not cadastral clipping. Wind screening checks recorded turbine points and reported rotor envelopes, not an invented wind-farm boundary or legal setback.

Military, tribal, conservation and existing-asset requirements are collected independently. A compound intersection cannot be cleared merely by selecting its most prominent category. Ordinary BLM/state/local management retains an agency-authorization warning without being categorically prohibited. Selecting any development pathway acknowledges an intended approach; it does not prove agency approval, tribal consent, land control, compatibility or owner agreement.

Mapped private land or no polygon match is never labeled available or buildable. The BLM geometry is generalized by approximately 4–6 m and is not a parcel survey; the age and completeness of documented generation infrastructure also require verification. See [the land-method note](developer-land-v14.md).

## 4. Hydrogen architecture, supply land and water

The current hydrogen architecture is co-located renewable supply. Its grid-proximity weight is **zero**, including when an older saved brief supplies a nonzero grid weight. Remaining priorities are normalized. Nearby substations remain context rather than a prerequisite for isolated renewable production. Grid-electrolysis sensitivity remains a separate calculation.

The process parcel does not establish the buildability of its renewable supply. Solar array acreage and the much larger wind-spacing envelope are reported separately. Insufficient explicitly entered supply acreage fails the brief. Absent renewable parcels remain unverified instead of being assumed to fit inside the electrolyzer site.

Hydrogen now reuses the public water-supplier planning and recycled-water leads previously available for large loads. Modeled average and peak raw process-water demand accompany those leads. Supplier boundaries, planning balances and nearby recycled-water assets do not establish a water allocation, treatment suitability or connection. Cooling and domestic water remain additional to this raw process-water balance.

Production-only exploration has no established customer requirement. Once demand is supplied, actual hourly withdrawals must satisfy the entered reference-year mass-service target; ending inventory is not delivery. Optional service-hour and maximum-shortage requirements express a stricter delivery brief. Additional weather-year stress calculations do not silently change scan/report eligibility.

## 5. Shared ranking, reports and prepared results

The scan, selected report, mapped-node comparisons and reusable criteria use the same summary and eligibility logic. Invalid designs, failed briefs, incompatible pathways and insufficient critical evidence do not receive a second favorable numerical score elsewhere. Evidence coverage is still displayed to show what is known.

The headline report now puts failed or unconfirmed project requirements, development pathway, missing financial evidence and negative NPV next to the score. Detailed calculations and source context remain available below. Negative NPV is prominently identified rather than hidden or converted into a universal legal/technical exclusion: a developer may investigate a different contract, cost assumption or design, but the current scenario is not represented as profitable.

The regional shortlist selects up to 12 distinct areas separated by at least 25 km; matched hydro assets retain their individual asset basis. This is geographic diversity, not a claim of statistical independence between weather cells or market references. The complete heatmap still uses every evaluated grid sample.

Prepared criteria require the exact physical-input scenario, model fingerprint and data fingerprint. The v14 schema stores the new independent eligibility gates. Weight changes reuse those gates and unweighted criteria; obsolete v13 bundles cannot satisfy the new identity. Changes to physical assumptions trigger a matching prepared scenario or recalculation.

## Validation evidence

Focused land testing passed **16 tests**, including the actual reviewed USAF candidate, holes, thin boundary conflicts, arrays, rotor overlap, acreage scaling, hydrogen supply land, empty/truncated bundle rejection and a complete-source local no-match. **Six additional integration tests** cover compound land pathways, public-land exceptions and the actual greenfield gate.

An independent **600-envelope** Shapely 2.1.2/GEOS comparison matched all **1,200** management and solar intersection-set checks. The JavaScript land assessment, including initial indexing, completed that particular run in about 1.56 seconds; this is not a full-model or browser timing claim. See [independent land validation](developer-land-validation-v14.json).

An independent local-data integration review exercised **12 default technology/location cases plus two deliberately inadequate briefs**. The reviewed USAF and occupied candidates left ordinary greenfield rankings; default wind retained the requested capacity; insufficient load service/date/water and hydrogen delivery failed the brief. Annual-price examples used 8,712 supported observed hours. See [integration results](developer-land-integration-v14.json).

The complete release suite passed **615 tests**, with zero failures or skipped tests. Nine actual-model counterfactuals also passed **45 adversarial weighting and cached/fresh comparisons**. A separate nine-test offline partition suite verified exact checkpoint preservation and rejected incomplete or corrupted shard sets.

The locally installed prepared release contains **20 bundles and 129,064 records**: six technologies at three scan resolutions for the default 2030 scenarios, plus two native hydrogen-region scenarios. Every grid, content hash and source/model identity passed publication checks. **360 sampled fresh-calculation and weighting comparisons** matched the prepared results, including independent eligibility gates. See [cache validation](precomputed-validation-v14.json) and [the release validation record](decision-validation-v14.json).

An independent audit verified the complete bundle matrix and all **99,948 detailed-grid records**. The former leading battery, wind and geothermal samples now require a different development pathway; the former load leader lacks comparable critical evidence. The reviewed solar sample changes from 90 to 79.9, with adequate annual prices and prominent negative economics. This corrects misleading eligibility without retuning weights to imitate the observed development map. See [complete ranking diagnostics and former-leader outcomes](decision-ranking-validation-v14.json).

Under the default merchant/cost scenarios, almost all scorable battery, solar, wind and geothermal cells still have negative NPV: 13,580 of 13,582 battery cells, 13,165 of 13,169 solar cells, 11,252 of 11,253 wind cells, and all 6,433 geothermal cells. These are relative regional prospects under unfavorable financial assumptions, not profitable recommendations. An actual revenue contract, different design or defensible cost scenario must be evaluated explicitly; altering preference weights does not establish profitability.

All **19 statewide Chromium checks passed**, with no page or worker errors. Every scan used matching prepared criteria. Opening the top finalist preserved the exact unrounded score, evidence coverage, eligibility status and coordinates. All fine scans contained 16,658 unique samples; each shortlist contained no more than 12 alternatives separated by at least 25 km. Four focused report scenarios, five saved/evidence interaction checks and the numeric/formula-safe CSV regression also passed. See [browser validation](browser-validation-v14.json).

Prepared criteria eliminate recalculation of the full grid, while evidence loading and complete finalist reports still take time. In this local run, fine scans took 55.6 seconds for battery, 34.8 seconds for load and about 9–10 seconds for the other technologies. Most coarser scans took 0.5–2.2 seconds; battery took 19–21 seconds, and native hydrogen took 3.9 seconds. These are observations from this machine, not guaranteed timings. Calculation and cache agreement do not establish predictive siting or yield accuracy.

## What this release does not resolve

The EIA comparison was regenerated against v14 and current input hashes on September 23, 2026. All 70 plant physical outputs and aggregate cohort statistics remain unchanged: modeled energy exceeds observed 2025 generation by 9.46% for solar, 64.73% for wind and 84.06% for geothermal. The decision fixes are not a yield calibration.

The previous held-out operating-plant comparison found material wind and geothermal overprediction. This release does not tune resource scores or apply a blanket correction to hide those discrepancies. Geothermal regional capacity/requested-MW scores can still saturate; they are modeled regional potential, not proved reservoir productivity. Equipment, historical operations, resource resolution and reservoir-specific evidence remain important limits on exact yield and profitability claims.

Solar and wind still need measured or engineered site yield, layout/access and lifetime operating assumptions. Geothermal needs tested temperature, flow, permeability, sustainable extraction and reinjection evidence. Battery forecasts do not establish multi-market awards or contracted revenues. Hydro and bioenergy need owner/permit, water-release or current feedstock agreements; arbitrary bioenergy regional scanning remains disabled.

Across technologies, public information does not establish purchasable or leased parcels, contiguous developable acreage, a utility connection offer, upgrade responsibility, energization commitments, firm water allocations, independent carrier routes or a hydrogen buyer/delivery contract. These remain explicit next-diligence items. The v14 fixes make unsupported and failed decisions visible and non-compensable; they do not turn regional screening into a development-ready project selection.