Fitted to one battery. Challenged before deployment.
A second ERCOT battery in commercial operation. We used its own instruction history to build the physical twin, forecast context, and decision program around the asset — then put that candidate through a promotion gate before deployment.
121 days
of operating history replayed at 15-minute decision cadence with a 24-hour receding horizon
Asset-fitted replay
A public physical twin, forecast fan, and receding-horizon policy were fitted to one battery's operating record.
What was already operating.
Same benchmark, same limits: the disclosed Base Point path, integrated at actual event spacing and valued against the same 15-minute price and throughput-cost boundary as the candidate. It is the instruction record, not the strategy that produced it — we can see what the asset was told to do, not what it was trying to do.
01
121 days
Operating history
02
Twin + fan
Asset and market state
03
15 min
Receding decision
- operating history reconstructed
- 121 days
- decision cadence
- 15 min
- receding planning horizon
- 24 hr
How the replay candidate was built.
- 01
Reconstruct
Turn the public Base Point instruction path, 15-minute prices, and public asset facts into one auditable proxy.
- 02
Fit
Use a public physical twin, the median path from a Chronos-2 forecast fan, registered intraday re-fans, and a receding-horizon 24-hour optimizer.
- 03
Advance one interval
Optimize the full state path, advance only the first 15-minute action, and re-solve each quarter using the latest registered forecast fan.
Built to be challenged before it trades.
The reconstructed-price replay produced a strong candidate. A stricter price reconciliation became the release gate, so the policy stayed in research rather than advancing on an unconfirmed result.
Process study built from public Base Points and reconstructed price data. No settled-P&L claim is made.
- The asset's Base Point path became the operating benchmark.
- The twin and forecast fan were re-solved every 15 minutes rather than applied as one static rule.
- Promotion depended on the operator-agreed data and accounting boundary, not the headline replay alone.
What we would do on your asset.
Rebuild the same program on your private operating record, agree the release boundary, and promote only what survives it.