Imports priced at EIA retail rate despite 'wholesale' label

Problem

electricity.imports.costs: wholesale (the default) prices import links using EIA's electricity/retail-sales endpoint (eia.py, _ElectricityCosts) — the monthly average retail price per state, all sectors blended. For CA 2019 that is roughly $165–170/MWh, several times a typical wholesale clearing price (~$30–50/MWh at SP15/NP15 in 2019).

Retail rates include T&D, fixed-cost recovery, and utility margins — components unrelated to the avoided cost of importing a marginal MWh. Priced this way, imports behave as a near-last-resort resource, biasing capacity-expansion results toward in-state build. Combined with a volume cap (e.g. 25% of load), imports are doubly penalized.

Additional wrinkles in the same code path (add_extra_components.py, load_import_export_costs / format_import_export_costs):

• The price year is the weather year, not the planning horizon (year = weather_year[0]).
• If the year is >= 2024, the code routes to the AEO national projection with state='USA', which format_import_export_costs fails to match against state codes and silently drops.
• All import interfaces receive the same statewide monthly series, forward-filled hourly (12 steps/year).

Suggested direction

• Rename or re-source the wholesale mode to an actual wholesale proxy (e.g. EIA wholesale/hub data, or an hourly LMP series such as CAISO SP15 DA for the weather year).
• In the interim, document that costs: <float> or costs: <carrier> (e.g. CCGT) are better proxies for marginal import cost.

Found while building the California (CPUC SERVM) configuration, where imports are enabled with a 25% annual volume limit.

Start in eia.py at _ElectricityCosts, then trace load_import_export_costs and format_import_export_costs in add_extra_components.py. Compare how the weather year, USA state, and statewide monthly series are handled for the California configuration. Done means the wholesale mode uses an appropriate import-cost source or is clearly documented with its limitations, without silently dropping projected data.