01 / EVIDENCE
Performance gate
Walk-forward lift against the baseline, including adverse windows, risk, wear, and data failures.
Shadow before switch.
Start with one asset and a trusted baseline. Authority grows only after the system earns it in replay, shadow, and operator-approved production.
Autonomy is a permission level.
Every stage has a visible output and hard promotion decision. Metis may watch without acting; every material refit must clear the gates again.
- 01Authority level 1 of 4.
Public-data replay
Correct the assumptions.
We reconstruct one asset and one event window from public market data. You tell us where the physics, contracts, or operating policy are wrong.
Output
Asset replay memo
- 02Authority level 2 of 4.
No-control shadow
Beat the baseline in daylight.
For 30–90 days, Metis produces timestamped recommendations without touching control. Every call is graded against the baseline you already trust.
Output
Daily decision trace
- 03Authority level 3 of 4.
Operator-approved
Put the call into workflow.
Recommendations move into the desk, QSE, ETRM, or EMS workflow with explicit approvals, limits, and exception handling.
Output
Advisory operation
- 04Authority level 4 of 4.
Automated execution
Earn the right to act.
Only after calibration, contracted permissions, and fail-safes are in place does Metis submit bids or dispatch within bounded authority.
Output
Optimizer + auto-trader
A recommendation you can cross-examine.
Shadow mode publishes the call before the outcome, with enough evidence for operators, asset management, and risk to challenge it.
- Action
- Bid, hold, charge, discharge, reserve, or abstain
- Evidence
- Decision-time forecasts, drivers, and rejected alternatives
- Constraints
- Active physical, commercial, and risk boundaries
- Outcome
- Settlement, delivery, wear, regret, and override
01
02
03
04
Meet the asset where it operates.
Keep the systems you trust. Metis plugs into the stack already operating the asset, whether it is listed below or not — from EMS and SCADA through market, QSE, ETRM, and O&M workflows. We adapt to the interfaces available; nothing needs to be ripped out.
EMS
EMS / site controller
constraints · setpoints · status
Wärtsilä GEMSFluence OSTesla PowerhubSCADA
SCADA / historian
SOC · power · alarms · derates
AVEVA PISiemens WinCCSchneider
MKT
ISO / RTO / market
prices · awards · rules · settlement
ERCOTCAISONYISO
QSE
QSE / route to market
bids · schedules · dispatch
QSE APIBid uploadDispatch feed
RISK
ETRM / risk
positions · limits · contracts
ION EndurION AllegroMolecule
O&M
O&M / asset management
availability · work orders · warranty
IBM MaximoSAP EAMInfor EAM
No promotion on performance alone.
Profit alone is not enough. More authority requires evidence, physical, commercial, and control clearance together.
02 / PHYSICAL
Constraint gate
Validated power, energy, SOC, thermal, availability, interconnection, and warranty limits.
03 / MARKET
Commercial gate
Market eligibility, contracts, positions, bid gates, settlement, and risk limits.
04 / CONTROL
Authority gate
Approval, fail-safes, overrides, escalation, audit retention, and contracted action rights.
Nothing is promoted on a backtest alone.
Every candidate program — the first one and every refit after it — is scored against your accepted baseline under the information that was actually available at decision time.
01
Reconstruct
Restore what was known before each decision.
02
Hide the future
Keep settled prices and realized conditions out of reach.
03
Run both policies
Score candidate and baseline inside the same constraints.
04
Attribute the result
Compare revenue, risk, wear, overrides, and failures.
Start with the replay you can challenge.
One asset. One hard window. One assumption review. If the reconstruction holds, we define shadow mode together.
Benchmark an asset →