Inside the seven-layer control loop

AUMERA replaces one opaque agent chain with seven bounded layers. Each has a distinct authority, and none can skip ahead to the next.

Five stages carry the decision

• L01 · Market evidence normalizes public feeds into one reference state, stamped with source and age, before anything downstream reasons about it.
• L02 · Research intelligence is where the specialist agents build and challenge a thesis.
• L03 · Capital authority is the independent veto: the only layer that can say no on financial grounds.
• L04 · Execution control carries out only what L03 cleared. Today, every fill is on paper.
• L05 · Financial truth reconciles what filled against what was intended, so the system’s record of itself cannot drift from reality.

"Splitting the path into layers with narrow authority means each one can be inspected, tested, and replaced without touching the invariants the others depend on." AUMERA Core Engineering Desk

Two supervisors watch them

• L06 · Calibration studies outcomes after the fact and feeds them back into research and routing, without relaxing any limit.
• L07 · SafeState sits above everything, ready to pull the whole system to reduce-only or a full halt, whatever any single layer wants.

Why layers, not one model

A single end-to-end model that goes straight from market data to an order is hard to audit and easy to trust too much. Splitting the path into layers with narrow authority means each one can be inspected, tested, and replaced without touching the invariants the others depend on. A failure in one layer degrades safely instead of quietly producing a bad order.

Every layer, its inputs, its output contract, and its fail-closed behavior are documented and inspectable in the architecture guide.