Equities
Index and single-name context for the macro impulse.
Signals for what the market is pricing. Permissions for what may run.
Equities · Rates · Foreign exchange · Commodities
01 · Problem
A headline is not a signal. A narrative is not risk control.
- 01
Headline Noise
A loud headline is not necessarily what the market is pricing. What happened is not what is being traded.
- 02
Hidden Regime Change
A real regime change can show up first in price, volatility, and liquidity, not in the headline.
- 03
Unbounded AI
An AI system without a power boundary will treat an explanation as an action. AMOS does not allow that path.
Markets
Global macro, read as one book.
A signal that cannot see across markets is a headline with a ticker.
Rates
Policy, the curve, and the price of time.
Foreign exchange
Which regime the world is actually paying for.
Commodities
Whether the real economy confirms the story.
03 · System flow
From evidence to a permission.
The model may explain. Only a signed, versioned, risk-cleared instruction can leave the control plane.
Global events and economic data
Events
Governed data and event layer
Intake
What the market is pricing
Pricing
Regime recognition
Regime
Strategy ENABLE / THROTTLE / DISABLE
Permission
Portfolio and risk control
Risk
Research, backtest, and shadow
Research
Controlled execution
Isolated
02 · Five questions
Understand the regime before enabling the strategy.
Five questions. One permission at the end.
What happened, and what evidence was knowable at the time?
The input is an authorized event, producer identity, and observation, receipt, and revision times under a versioned schema. The output is a replayable event record, not a free-text conclusion.
Event record · Replayable
The signal
Three permissions. No narrative.
AMOS does not publish a view. It publishes a permission: run, run smaller, or stand down.
ENABLE
Evidence, regime, and risk budget agree. The strategy may run.
THROTTLE
The regime is unstable, or the evidence is thin. Size comes down.
DISABLE
The failure rule has fired. The strategy stands down.
Each permission names its evidence, its regime, its risk budget, and the condition that kills it.
05 · Safety
AI may form a judgment. It cannot move capital.
LLM can explain. It cannot move capital.
| Role | May do | May never do |
|---|---|---|
| LLM / statistical model | Extract, summarize, classify, propose hypotheses, seek counterevidence, explain evidence, and state confidence | Place orders, change risk limits, bypass a deterministic strategy, or turn natural language into a capital action |
| Deterministic strategy engine | Compute versioned strategy permissions, risk constraints, and failure conditions | Invent a permission from free text or silently change a threshold |
| Independent risk system | Veto a strategy, limit exposure, and trigger the Kill Switch | Be bypassed by an LLM or by a single strategy |
| Human owner | Approve releases, models, risk budgets, data licenses, and any execution eligibility | Present engineering progress as unproven investment capability |
| Optional execution plane | Execute only signed, versioned, risk-cleared structured instructions | Accept LLM output, arbitrary text, or an unsigned decision |
09 · Evidence
Claims no broader than evidence.
A lower grade of evidence cannot prove a higher-grade conclusion. Engineering contracts and local verification are not investment capability, and they are not the safety of capital.
- L1
Engineering evidence
Proves software contracts, migrations, idempotency, and supply-chain discipline. It does not prove investment capability.
- L2
Research evidence
Requires point-in-time correctness, out-of-sample tests, and cost and liquidity assumptions. No research performance is shown.
- L3
Shadow evidence
Unfunded and continuous. A backtest is not a substitute.
- L4
Execution evidence
Isolated from the model. Only a signed, versioned, risk-cleared instruction is accepted.
Standard
What a signal has to survive.
Conviction is the last step, not the first.
Point in time
A fact is usable only as it was known then.
No lookahead
A later revision does not improve yesterday's decision.
Out of sample
A result that lives only inside the sample is not a signal.
Deterministic gate
Permissions are rules. They are not model prose.
Human authority
A person, and a kill switch, sit above the model.
Replay
Every permission can be reconstructed.
08 · Roadmap
Research proceeds in one direction.
01
Events
Authorized sources, identity, and a record that can be replayed.
02
Pricing
Whether the market has already moved, and in which window.
03
Regime
Probability, confidence, and a condition that would falsify the state.
04
Permission
ENABLE, THROTTLE, or DISABLE, bound to a strategy version and a risk budget.
05
Shadow
Unfunded, continuous, and kept apart from the story that produced it.
06
Execution
Isolated from the model. Only a cleared instruction is accepted.
10 · FAQ
The boundary questions, answered directly.
Is AMOS an automated trading bot?
No. AMOS is macro decision and risk-control infrastructure. The LLM has no order path.
Does AMOS predict whether prices rise or fall?
The aim is a regime assessment with evidence, probability, and failure conditions, not a deterministic forecast.
What does the LLM do?
It extracts, summarizes, classifies, proposes hypotheses, and looks for counterevidence. It cannot change a risk limit or create an order.
Is a broker connected?
Execution sits outside the signal. It accepts only a cleared instruction.
Can live data be connected now?
A source enters with a license, a timestamp, a revision, and a lineage. Nothing else is treated as evidence.
Can AMOS guarantee returns?
No, and it will not claim to. Investment capability has to be established by point-in-time research, out-of-sample tests, cost, liquidity, and long-running shadow evidence.
Why a modular monolith?
The domain boundaries are still moving. A modular monolith reduces distributed failure and operating cost, and it keeps a path to split through ports and adapters.
Contact
Evidence for every macro judgment. A boundary for every strategy permission.
Briefings are for institutions. Name the desk, the markets, and the question. The note is prepared on this device and is not uploaded.
United States+1 (626) 989-7273