Drive significant performance improvements autonomously.

Foundation models read the state. RL policies decide. The decision writes back to the physical controller — safely, continuously, inside SIS-bounded envelopes. Continuous-control for chem / steel / refining / grid / mining mills; discrete-combinatorial for A&D and contract manufacturing. Roughly 70–80% of unit-economics value across the seven verticals is materially leveraged by CLO. Pricing is outcome-aligned — the Phaidra pattern, 5–15% of value liberated.

Dashboards don’t close the loop. Policies do.

Every plant has dashboards. Almost none has policies. Operators translate signals into setpoints by hand. Decisions wait for shifts. Variability compounds. The cost shows up as yield, energy, scrap, unplanned downtime — and at sub-second timescales (BESS regulation, EAF furnace control, autonomous cells), human reaction is structurally incapable of capturing the value.

Closed-Loop Optimization writes back to the asset. Three pretraining sources, combined: physics simulators (Aspen, OLGA, Eclipse, PSS®E, JKSimMet, METSIM, AnyLogic — domain anchors per vertical); decades of accumulated historian / operational data (the substrate the simulator misses); and MFM-encoded state representations from the per-vertical foundation models. Online improvement is safety-bounded — SIS hardware authority (IEC 61511 / ISA-84) preserved, exploration shaped by safety margins, write-back gated at the cell controller / DCS interface, auto-revert on alarm. Not ‘sim-to-real’ — that’s robotics jargon imported awkwardly. This is historian-grounded representation learning plus simulator-aided counterfactual exploration plus safety-bounded online improvement.

6 policies online
Live
BESS revenue stackPower
97%
EAF energySteel
96%
Refining FCC + hydrocrackerO&G
95%
Flotation + comminutionMining
94%
Distillation + batch S88Chem
92%
Shop schedulingContract mfg
93%

Two RL shapes, one safety contract

Continuous or combinatorial — never unsafe.

Industrial decisions fall into two shapes. Continuous-control for regulated processes; discrete-combinatorial for scheduling, routing, and assignment. Both run inside an SIS hardware envelope the policy cannot override. Pricing is outcome-aligned — the Phaidra pattern, 5–15% of value liberated, demonstrated at 16% (Pfizer) and 40% (Google) in adjacent settings.

1

Continuous control

Setpoint optimization for regulated processes. Mining: comminution + flotation + leach. O&G: drilling ROP + refining FCC + hydrocracker + blending. Power: BESS revenue stack — energy arbitrage + FRC + FRR + RegD + RegA + capacity + emergency, multi-market simultaneous in sub-50ms response windows that humans cannot make. Chem: distillation + batch S88 + polymerization + reactive distillation. Steel: EAF energy (electricity is 30%+ of mini-mill cost) + scrap-mix + caster + heat-treatment.

2

Discrete-combinatorial

Routing, sequencing, scheduling. A&D: MES routing, MRO depot allocation, EVM forecasting, supplier-delivery prediction. Contract mfg: shop scheduling, CAM toolpath generation (the manual-programming bottleneck), tool-life policies, cell coordination across pick-and-place + welding + machining + finishing.

3

Safety-bounded online RL

Three pretraining sources (physics simulator + historian + MFM state). Shadow-mode rollout before write-back. Customer-approval workflows for regulated industries. Auto-revert on alarm. SIS retains IEC 61511 hardware authority — the policy operates inside the envelope, never around it.

The loop in action

Perceive, decide, write back — in under a second.

A signal from the floor flows through MFM state, RL policy, and SIS-bounded write-back faster than any human could close the loop. Every decision is shadow-tested before it ships; every execution is audit-logged. When the policy effectively runs the plant, the platform earns operator-of-record positions — operator-streamer in mining, integrated grid IPP, molecule-to-market in chem.

From signal to setpoint

240ms from drift detected to setpoint applied

Perceive the state via MFM, decide via RL policy, write back inside the safety envelope — continuously, autonomously, auditably.

[14:23:07] SIGNAL Cell 4 — spindle load drift detected
PerceiveMFM state
120ms
DecideRL policy
80ms
Write backSIS-bounded
40ms
3 stages
safety-bounded
240ms total
Yield protected
How a policy ships

Three-source pretrain. Shadow. SIS. Outcome-aligned.

Policies are pretrained from three sources, MFM-conditioned, shadow-mode-validated, SIS-gated at every step — and priced on the value they liberate.

1
Three-source pretrain

Domain simulators per vertical (Aspen / HYSYS / OLGA / Eclipse / PSS®E / JKSimMet / METSIM / MAGMA / DEFORM) seed valid behavior; decades of historian / fleet / EMS / BMS / MES capture what the simulator misses; MFM-encoded state gives compressed transferable representations the policy doesn’t have to relearn.

2
Shadow before write-back

Every policy runs against operator decisions before it ever moves a setpoint. Customer-approval workflows for regulated industries. OTA staged-canary rollout via Edge AI; auto-revert on performance regression.

3
SIS keeps the keys

IEC 61511 / ISA-84 hardware safety authority preserved. Policy actions constrained by SIS envelopes; out-of-bound commands blocked at the cell controller / DCS interface. Every action logged in Standard Engines’ provenance ledger.

4
Outcome-aligned economics

Phaidra pattern: 5–15% of attributable margin / energy / throughput / quality uplift. Demonstrated at 16% (Pfizer) and 40% (Google) in adjacent settings. Cross-customer policy transfer via shared MFM substrate sharpens the policy for every later customer. When CLO runs the plant, the platform earns operator-of-record positions.

From dashboards to decisions.

Close the loop on one constraint — yield, energy, throughput, scrap. Outcome-aligned pricing, audit trail included, SIS in charge of safety.