Forecasting infrastructure for power-market desks.
One decision core reads the live grid and market state you already watch, then returns the next trade, bid, or dispatch, scored against your objective and risk limits. It plugs into the desk tools you run.
Forecasts miss on the days that cost the most.
The published ISO forecast and the desk spreadsheet are fine on a normal day and wrong on the few that move P&L. Metis is built for those days, and grades itself on the same settled numbers you do.
- Heat-onset evening ramps
- Scarcity pricing events
- Congestion and binding constraints
- Outage-driven price separation
- Steep net-load ramps
Here's how Metis works, and why it's more accurate.
One loop, every desk: Metis reads the live world, forecasts the futures that matter, scores your candidate moves, and hands back the call. Your own feeds tell it your system; the weather, outage, and cross-market data we bring tell it what's about to hit. Mixed, the forecast is sharper on the few days that move P&L, and we prove the models in the open at tsfm.ai.
Pick a desk for the exact data it reads, the call, and the packet it returns.
Set the day-ahead position and hedge before the market clears.
For tomorrow's peak block at this hub, do I go long or short day-ahead versus real-time, how big, and where do I set the hedge?
Scenariosee the setup
ERCOT North hub, tomorrow's HE15 to HE20 peak. A heat-onset day, 8 to 10F above last week; the desk is flat and the DAM curve is pricing a normal afternoon. Metis sees load above MTLF into the evening, net load staying high as solar falls off faster than wind picks up, and two gas derates the curve has not absorbed.
Data in
Yours
- LMP by hub and node (energy, congestion, loss)
- System load and net load, evening ramp shape
- Generation stack, wind/solar, reserve margin
Metis adds
- Weather across load and renewable zones
- Outage and derate notices, scarcity alerts
- Settled DA/RT prices, congestion history
Plus cross-ISO history and pretrained time-series models.
Metis decision core
- 1Read
- 2Forecast
- 3Score
- 4Recommend
Same core, every desk.
Decision packet
WhyGo long day-ahead at North across the HE18 to HE20 ramp (expecting RT to settle above DA), sized to a moderate fraction of the block limit, and hedge the wings with a short on the soft midday HE13 to HE14 block. A shape trade, not an outright long.
Medium-high on direction, lower on magnitude (scarcity tail)
A repeatable read on the days that move P&L: better positioning into the evening ramp and the DA-versus-RT spread, with calibrated confidence and worst-case framing inside risk limits, every call traced so the edge compounds.
Plugs into the systems you already run.
Decision support, not a replacement for your systems.
Into the tools you run
Decision packets land in your blotter, ETRM, risk system, or controller. Nothing to rip out.
API and MCP native
Every decision is callable, so your own models, copilots, and automated strategies can pull the same recommendation.
Leakage-safe and graded
We validate out-of-sample against the baseline you already grade on, with the worst-day behavior shown, not hidden.
FAQs
How is this different from a vendor price or load forecast?
Vendors stop at a number. Metis attaches the forecast to the decision, scores your candidate trades, bids, or dispatch against your objective and risk limits, and returns a recommended move with a confidence band and a trace. The forecast is the input; the decision is the product.
Is it a black box?
No. Every decision packet carries its drivers, a calibrated confidence band, and a trace from inputs to settled outcome, so you can audit why it made the call and grade it against the benchmark you already use.
Do we have to replace our models, EMS, or market systems?
No. Metis is decision support that delivers into the tools you already run: blotter, ETRM, risk system, controller, API, or MCP. It plugs in alongside your stack rather than replacing it.
How do you prove it beats what we have?
We run a paid discovery against the baseline you already grade on, the ISO forecast or your own model, leakage-safe and out-of-sample, and show the lift before you commit to anything.
What happens to our data?
You set the data boundary and what Metis can use. Your data stays yours, and every recommendation is traceable back to the inputs that produced it.
Prove it on your hardest call.
Bring one baseline you already trust, and we'll show the lift on a decision you make today.