[home]

where the next battery goes

an eight-bus ERCOT dispatch model, and what a home-battery fleet does to it

Every one of ERCOT’s 18,000-odd electrical buses gets its own five-minute price, but fewer than 1,000 of them are settlement points, and residential load is settled at four Load Zone averages. So a home battery is paid a number that is an average over a region it sits somewhere inside. This page solves a security-constrained dispatch across a reduced eight-bus Texas, prices every bus off the shadow prices of the binding constraints, and then grows a fleet of home batteries against it.

The price at bus b is LMP_b = λ − Σ_k SF_b,k · μ_k: one system-wide energy component, minus this bus’s shift factor on each binding constraint times that constraint’s shadow price. Everything below is read off the duals of the dispatch, not fitted to anything.

solving dispatch…

what the model says

  1. Congestion lives in surplus hours, not peak hours. On the mild spring day the Panhandle sits pinned at −$8 for sixteen hours behind its export constraint while the South coast clears near $30 — a $38 basis that is entirely congestion. On the summer peak evening, nothing binds at all. West Texas is not export-constrained when the whole state wants the power.
  2. Cheap is not the same as valuable. The Panhandle is the worst place on the system to put a battery despite being the cheapest bus on it. A binding export constraint depresses the price level all day; it does not create a time spread, and storage is only ever paid for moving energy through time. Reading a congestion map as a siting map points you exactly the wrong way. Capturing a spatial basis is what transmission and congestion revenue rights are for.
  3. Firmness dominates arbitrage by two orders of magnitude. On the winter morning the fleet earns around $5.50 per kW per day against roughly $0.03 on a mild spring day, and takes unserved energy from 18.3 GWh down to 2.9 GWh. Whatever the economics of a home battery are, on the days that matter they are not about the spread.

how it works

Each hour is a linear program: meet load at least cost, subject to every unit’s capability, a DC representation of the network, and an aggregate reserve requirement bought against a price-responsive demand curve. Shift factors come from the network’s susceptance matrix with a load-weighted distributed reference, the way ERCOT uses a distributed slack rather than a single swing bus. Two Generic Transmission Constraints sit on top of the individual line ratings, because ERCOT’s largest limits are stability limits on the sum of flows across a set of monitored elements rather than thermal limits on any one of them.

Storage is a price taker. Each fleet solves its own best arbitrage against the price series it is actually settled at, by dynamic programming over state of charge with a cyclic daily constraint, 86% round trip and $4/MWh of throughput for degradation. Its schedule and the price are then iterated to a fixed point by the method of successive averages — a fleet that best-responds at full step swings in unison and the iteration never settles.

The one control that changes the physics is settled at. A retail electric provider is paid the Load Zone price; a registered resource is paid its own bus. Switching it does not change what the batteries are, only what they can see, and the gap between the two settings is the honest measure of what nodal participation is worth.

The solver is checked rather than trusted: shift factors reproduce an independently coded DC power flow, an unconstrained hour prices every bus at λ, and the LMP at all eight buses equals the derivative of total dispatch cost with respect to load at that bus, by finite difference, in congested and scarcity cases alike. See tests/ercot.test.js.

where the numbers come from

Fleet totals and market rules are ERCOT’s published figures: 40,737 MW of wind and 39,794 MW of utility-scale solar as of July 2026, 14 GW of grid batteries entering 2026, 4,970 MW of nuclear, and roughly 13 GW of coal, 41 GW of combined cycle and 17 GW of peaking capacity. The summer day is anchored to the 91.1 GW record peak of 22 July 2026, met 48% by gas and 32% by solar — which is why the solar shape here holds its output so late and the price is made in the 19:00–21:00 ramp. The offer cap and value of lost load are $5,000/MWh against a −$251/MWh floor.

Reserves are co-optimised with energy, because Real-Time Co-optimisation plus Batteries went live on 5 December 2025. That retired the separate ORDC price adders — RTORPA and its siblings no longer exist — and scarcity now reaches the energy price through the ancillary service demand curves inside SCED, with the LMP carrying the opportunity cost of holding reserves.

The eight-bus network is not ERCOT’s. ERCOT publishes prices for more than 18,000 electrical buses and does not publish a reduced equivalent. The bus split of the fleet, the reactances, the line ratings, the GTC limits and the cost stack are a stylised reduction chosen to reproduce ERCOT’s characteristic behaviour — West and Panhandle export congestion in surplus hours, an evening net-load ramp in summer — not to reproduce any particular hour. Treat levels as indicative and directions as the finding.

base power, for scale

The two left-hand marks on the slider are Base’s actual position, as best as public reporting pins it down. Figures are the company’s own or its utility partners’, and deployment is moving fast enough that anything here is stale within a quarter.

Put against the model: 20,000 homes is about 0.23 GW and half a GWh, which is a rounding error next to the 14 GW of grid-scale batteries already on ERCOT. The signed programmes roughly double it. The interesting part of the slider starts about two orders of magnitude to the right of where the company actually is.

what it leaves out

prices from a linear program are step functions of quantity, so a small move on the slider can leave the price flat and the next one can jump it a whole block. that is marginal-cost pricing, not an artefact.

← back home