Partner Briefing · September 2026
Systematic, AI-augmented trading across global markets.
One platform, every asset class, improving every second. Research produces a validated portfolio; the terminal runs it under hard risk controls and reports every day; the reports and the agents’ journals feed the next research cycle.
01 · Markets, asset classes and venues
Venue- and asset-agnostic from day one.
Every centre below has strategies, contract specifications and a working feed and execution path today. Pin weight shows deployment status, not capital.
- Ten feeds, ten execution gateways, one interface. A strategy never knows where it is routed.
- Multi-currency by construction. Every leg is converted to the book’s base through a live rate service.
- A second venue and a second data source run beside the primary ones, each fault-isolated, validated bar for bar — so every instrument has an independent second opinion and a hot cut-over.
- Smart order routing (designed). A cost model, not a price hunt: fees plus our own measured slippage plus route health decide the venue, and a failover tier flattens risk when one goes down.
- Rolls are automated, inside per-exchange windows, with a calendar in the terminal.
02 · Executive summary
The industrialised loop around strategies.
Seven years building one platform that researches, validates, deploys, executes and continuously rebuilds portfolios of systematic strategies — across every major asset class, in every market above.
What makes it different is not any single strategy. It is everything around the strategies.
| Capability | What we have | Status |
|---|---|---|
| Markets | 40+ futures roots across nine exchanges in Hong Kong, the US, Japan, Singapore, Korea and mainland China; US and China A-share equities; crypto spot and perpetuals; FX | Live · live-paper · broker-paper |
| Strategy engine | Graph / state-machine strategies assembled from a library of reusable property modules, thousands of typed pattern nodes, seven timeframes from one minute to weekly | Live |
| Research factory | Sequential and Bayesian property optimiser with thirteen scoring objectives; portfolio constructor; four independent validation gates plus purged cross-validation | Live, weekly on a dedicated research server |
| Research IDE | Browser IDE where the desk edits properties, runs optimisers, reviews candidates, promotes a portfolio and deploys it through one audited, reversible pipeline | Live — the book is re-validated and redeployed whenever the evidence says so |
| Risk & controls | Broker-resting stop and target on every position, a portfolio circuit breaker that flattens the book, position reconciliation every fifteen minutes, backtest-vs-live divergence checks, per-user permissions with 2FA and device-signed mobile actions | Live |
| AI layer | Multi-agent LLM debate as a macro filter; Agent Trader — an autonomous LLM portfolio agent with its own paper book, a mandatory stop on every order, an auditable memory store, a weekly portfolio review, a replay-gated improvement loop and gated intervention rights over the systematic books | Live (paper and associate) |
| Operations | Two clusters, self-healing watchdogs, an ops channel, an automated daily trading report, web and mobile trading terminals | Live |
| Engineering | ~462,000 lines of Python, 5,400+ automated tests, 92 design specs, 110 implementation plans; spec → plan → test-driven build discipline | Ongoing |
Three books run today: sixteen strategies live over Hong Kong, US, Japan and Singapore index futures; five in paper on China futures; and a third, fully autonomous, run by an agent across nineteen China roots.
We are seeking partners who want exposure to a repeatable process rather than a single strategy: the same pipeline that continuously re-validates and redeploys the live portfolio can absorb new markets, new capital and new signal sources without changing shape.
03 · Why this matters
Complexity, turned into routine.
Regimes shift, correlations break, venues fail and yesterday’s edge decays. Our answer is not a cleverer prediction. It is one process, run the same way every time, in every market.
Most systematic shops fail in one of three ways. We engineered against each.
Overfitting
A backtest that looks like 6.0 Sharpe and lives like 0.6.
Every candidate has to pass four statistically independent gates and a purged cross-validation at the portfolio level. On the latest research corpus only nineteen of 147 candidate cells passed. The pipeline is designed to say no.
Operational fragility
The strategy is fine, but a missed roll, a stale feed, a stop that never reached the broker or a manual deploy at 2 a.m. costs more than the alpha.
Production is a supervised fleet: every process has a watchdog, every position a broker-resting bracket, every book a circuit breaker; the broker’s view of positions is reconciled with ours every fifteen minutes and a full trading report is generated and posted every night.
Stagnation
The first portfolio ships, then research and production drift apart and the book is never rebuilt.
Research and production share one code base, one data pipeline and one IDE. A candidate found on Saturday’s research run can be validated, promoted and deployed by Monday, with backups and a rollback path generated automatically.
Why partner with Etherius
- A process, not a bet. Every portfolio is validated by four independent gates and deployed with a rollback path — the same way, every time.
- Breadth on day one. Every major asset class across nine exchanges and eleven venue gateways; two brokers live, more built. New markets are configuration.
- Risk engineered in. Broker-resting stops, a circuit breaker that actually flattens, reconciliation, divergence gates and a daily report nobody has to remember to write.
- AI where it helps, deterministic where it must be. Models modulate and manage; tested code executes.
- Transparent. Every decision, deploy and trade is journaled; partners get the same daily report the desk gets.
04 · Platform at a glance
One loop, four surfaces, one core.
Four surfaces, one core. Everything beneath the surfaces executes, risk-manages and reports the same way for every market and asset class.
Research
Optimisers sweep the property library and propose candidates.
Validate
Four independent gates and a purged cross-validation reject most of them.
Deploy
One audited, reversible pipeline puts the survivors into production.
Monitor
Brackets, circuit breaker, reconciliation, sync checks, a daily report.
Improve
Those reports and the agents’ journals open the next cycle.
Deployment topology
Two clusters, three boxes each, one terminal in front of both.
Design principles
- Config-driven, no hard-coded instrument logic. Adding a market is a config file, a contract specification and a feed connector — not a code fork.
- One code base for backtest, paper and live. The same strategy class runs in all three modes; only the header config changes.
- Deterministic core, swappable intelligence. Execution, risk and journaling are deterministic and unit-tested; models sit behind narrow contracts and can be replaced or switched off without touching the pipeline.
- Everything observable. Heartbeats, pattern-state exports, NAV logs, alerts, daily reports, and a browser terminal on desktop and phone.
05 · Strategy engine
Properties as state machines.
Every position the platform takes traces back through six layers, each a reusable, independently tested unit with a single job. Because the layers are separable, an improvement anywhere — a better indicator, a sharper property, one more validated strategy — lifts everything above it without a rewrite, and a weakness stays isolated to the layer that caused it.
- 6
Agent Trader
Manages the resulting book: protective and closing actions unattended, exposure changes only with approval.
- 5
Portfolio
A validated composition of strategies, weighted by the constructor and clamped by a circuit breaker.
- 4
Strategies
State machines over typed nodes; a trace through the graph is a pattern, and a completed pattern is an order.
- 3
Properties
Reusable, independently scored signal primitives, each evaluated on one timeframe and typed — trend, mean-reversion, reversal, momentum, breakout, close, filter.
- 2
Indicators
The indicator engine computes a wide bank of engineered features per instrument across every timeframe.
- 1
Raw market data
Ticks, bars and quotes from brokers, exchanges and data vendors, on every timeframe from one minute to weekly.
06 · The research factory
A pipeline designed to say no.
Candidates enter on the left. Each gate is independent of the others, and a candidate is rejected by the first one it fails.
The most recent candidate matrix
147 cells, partitioned by outcome. A cell is rejected by the first gate it fails.
Latest research corpus, Global cluster. The deployed portfolio carries a backtest overfitting probability of 0.33 and passed four of four walk-forward windows.
Portfolio constructor
Seven weighting methods, a correlation cap between legs, and a search for out-of-sample performance subject to a full-sample floor. The same pass sets each candidate’s daily loss clamp.
Data plumbing that keeps research honest
- Two cache sets per market, refreshed weekly — one for testing, one for training.
- Sources are never mixed. Every dataset carries its origin.
- Hourly re-runs keep every candidate comparable to the live book.
07 · Research IDE
From idea to production, with an audit trail.
Where the desk lives: run an optimisation, work the candidate queue, build a portfolio, test it against the live book, study the agents.
Promote → deploy → rollback
Promotion runs fourteen audited steps by itself — configuration, database rows, a pre-flight snapshot, file backups with a manifest, a sync check, a gated start, circuit-breaker arming from the live NAV, and an archive of every deployed file. One command undoes all of it.
Latest compositions
| Portfolio | Strategies | Backtest Sharpe | Out-of-sample Sharpe | Max drawdown | Note |
|---|---|---|---|---|---|
| Global — live | 16 | 4.52 | 3.12 | −0.1% | Selected on out-of-sample Sharpe subject to a full-sample floor; walk-forward four of four; overfitting probability 0.33 |
| China — paper | 5 | 4.98 | 3.36 | −0.8% | Hole-free build on repaired data; gated starts, circuit breaker armed |
Backtest and validation statistics, not live results. Drawdown is stated with the circuit breaker applied.
The live book is re-tested against every new candidate on a rolling basis. It gets rebuilt when the evidence says so — not on a calendar.
08 · AI layer
From macro filter to autonomous trader.
8.1 LLM Debate — the book argues with itself
A macro filter, not a signal generator. Every four hours a cast of model agents argues over each instrument, and the verdict becomes a small machine-readable bias the strategies read on every bar.
- The cast. Regional analysts, five advocates over two rounds, a red team paid to attack the consensus, a risk agent, a cross-asset checker and a synthesiser — each able to run on a different model.
- What they see. 152 engineered indicators, the calendar with its event-risk windows, the regime state and the news flow.
- What they produce. Per instrument: weights by signal type, a vote with conviction, and the agreement a signal must clear before it may act.
- Fail-safe by construction. No bias file means no change, and the layer can be switched off per instrument without touching a strategy.
- It has to earn its keep. A cycle is skipped rather than half-applied when calls fail, and every cycle is scored against what the market actually did.
8.2 Agent Trader — a model that runs a book, safely
The model writes a decision. Everything that turns it into an order is ordinary, tested code.
- A stop on every open, enforced by both the executor and the monitor, inside per-instrument, cluster and margin caps it cannot argue past.
- Paper to live is a file edit by a person. No button, API call or model output can promote an agent.
- It is told in advance what it may not do — and how often it proposes it anyway is measured and shown beside P&L.
- Memory is an auditable note store, not a growing prompt: the model proposes, a nightly pass writes lessons under evidence floors, operators pin instructions.
- Every Saturday it reviews its portfolio’s week, not just its own trades.
- Bound to a systematic book, each management action sits on its own shadow, approve or automatic rung. Anything that changes position size waits for a human; anything that protects or closes one does not.
- New rules ship switched off, evaluating and journaling what they would have done until the measurement justifies arming them.
The improvement loop — propose, replay, gate, promote
An agent improves the same way a strategy does: by beating the incumbent on evidence it could not have seen.
- One change at a time. A challenger differs from the champion in its prompt or in its memory setting — never both.
- Replayed on history, month by month, with incomplete windows marked unusable rather than scored.
- Only windows after the model’s own knowledge cut-off may decide. Everything earlier is directional evidence.
- To be promoted it must win most of those windows, earn a positive expectancy, not deepen drawdown, clear zero on a bootstrap interval, and not get sloppier. Promotion carries a cooldown and a one-action rollback.
09 · Live books today
Three books, one runbook.
Global book
Sixteen strategies over five instruments — mini HSCEI, mini Hang Seng, Micro Nasdaq-100, Nikkei 225 mini and FTSE China A50. Backtest Sharpe 4.52, in-sample 5.15, out-of-sample 3.12, drawdown −0.1% with the circuit breaker, walk-forward four of four, overfitting probability 0.33.
The circuit breaker is armed: a daily loss limit and a daily profit lock, set in absolute currency at arming and re-based on the live NAV. On breach the portfolio manager sends live market orders to flatten every position and blocks new opens until the next daily reset.
China book
Five legs across index and commodity futures — CSI 500, CSI 300, corn, pulp and alumina. Backtest Sharpe 4.98, out-of-sample 3.36, drawdown −0.8%; circuit breaker armed.
Real market data, a hidden stop and target path inside the gateway, a daily order-manager reset and contract-roll awareness. China has exercised exactly the same runbook as Global from its first paper trade.
Agent books
One agent runs its own paper book across all nineteen China futures roots, with an hourly full evaluation and a faster intra-session pass; every open carries a stop.
Two associate agents are bound to the systematic books with a per-operation ladder of rights. All three run the same machinery: their own memory store, the nightly consolidation, the Saturday portfolio review, hygiene measurement and replay on the research boxes.
Gated go-live. A new leg starts only when the model is flat, holds no pending signal, and its sync check is green. A starter waits for that window and promotes it.
10 · Data Terminal
The macro cockpit.
It reads the world so that research, strategies and agents do not have to.
- Brief and regime call — the drivers that matter and a regime label, by region and globally.
- Investment committee — the same debate, twice each weekday, across eleven markets, with every role’s report kept.
- Calendar and news — impact-ranked, with actual against forecast, and a merged feed in English and Chinese.
- Macro and commodity matrices — index, FX, yields, policy rate, inflation and growth per market; the four commodity complexes side by side.
- Alerts on regime changes and data surprises, and a box that answers questions from the terminal’s own data.
11 · Risk management and controls
Controls at every layer.
12 · Operations and infrastructure
A supervised fleet.
Every process has a starter, a watchdog and somewhere to report.
- Two clusters, three boxes each — market data, trading, and research — laid out identically in both regions.
- Supervision. Per-minute watchdogs revive strategies, the circuit breaker, the agents and the feeds; the desk starts and stops any of it from the ops channel.
- Deploys to production are surgical, with backups, checksum parity and a check that the running process is the one we shipped.
- One terminal, desktop and phone — positions and their stops, orders, watchlist, the risk panel, health and attribution, the roll calendar, the debate and the agents, across every portfolio.
- A read-only research app for studying candidates and agents from a phone.
13 · Engineering discipline
How a small team ships this.
- Scale. ~462,000 lines of Python; more than 5,400 automated tests.
- Process. Every feature is a written spec, then a plan, then a test, then the code — 92 specs and 110 plans so far.
- AI-augmented engineering. An AI coding agent works as a full partner, with a persistent memory that carries operational lessons between sessions. It is how a small team ships this much.
- Safety rules are codified, not tribal. No manual intervention in a live book, no restart in live hours without approval, and no production change without a rollback path.
Appendix · Facts and figures
The platform in numbers.
| Item | Value |
|---|---|
| Code base | ~2,050 Python files · ~462,000 lines · 6,690 commits since February 2019 |
| Tests | 323 test modules · 5,485 test functions |
| Strategy library | 621 property modules · 158 tracemaps · 3,559 typed nodes · 205 strategy configurations |
| Design archive | 92 specs · 110 implementation plans · runbooks |
| Timeframes | 1m · 5m · 15m · 60m · 4h · daily · weekly |
| Feeds and gateways | 10 feed connectors · 10 execution gateways |
| Optimiser | 13 objectives · grid and Bayesian · dedicated research server |
| Validation | Hold-out · walk-forward · deflated Sharpe · permutation test · purged cross-validation |
| Latest gate matrix | 147 cells: 19 pass · 79 deflated-Sharpe reject · 35 cross-validation reject · 9 permutation reject · 5 empty |
| Global live portfolio | 16 strategies · Sharpe 4.52, in-sample 5.15, out-of-sample 3.12 · drawdown −0.1% · walk-forward 4/4 · overfitting probability 0.33 · circuit breaker armed |
| China paper portfolio | 5 legs · Sharpe 4.98 · out-of-sample 3.36 · drawdown −0.8% · circuit breaker armed |
| Agent Trader | 3 agents · 54-module deterministic package · 1,900+ dedicated tests · memory store, hygiene metrics and weekly portfolio review live on all three |
| LLM Debate | 16 model calls per four-hour cycle · 152 indicators · multi-provider |
| Smart Order Router | Designed: cost-model routing across two brokers, failover, synthetic close; build pending |
All Sharpe and drawdown figures are backtest or validation statistics computed on FX-converted NAV in the book’s base currency, with the circuit breaker applied where stated. Live performance records are available to qualified investors on request.
Next step
Let’s talk.
We are happy to walk through any part of this in a live session on the platform itself — the research IDE, the terminal, or an agent’s journal.
Important disclosures and disclaimer
No offer or solicitation. This page has been prepared by Etherius Capital for informational and discussion purposes only. It does not constitute, and should not be construed as, an offer to sell or a solicitation of an offer to buy any security, fund interest, managed account, financial instrument or investment advisory service in any jurisdiction. Any such offer, if made, would be made only by means of definitive documentation and only to persons eligible to receive it under applicable law.
Not investment advice. Nothing here constitutes investment, legal, tax or accounting advice. Recipients should make their own independent evaluation of the platform, strategies and risks described, and consult their own professional advisers.
Hypothetical, backtested and simulated performance. Unless expressly stated otherwise, all performance statistics on this page — including Sharpe ratios, drawdowns, out-of-sample and walk-forward figures, validation gate results and portfolio metrics — are derived from backtests, simulations or paper trading. Hypothetical performance results have many inherent limitations. No representation is made that any account will or is likely to achieve profits or losses similar to those shown. There are frequently sharp differences between hypothetical results and the actual results subsequently achieved by any particular trading program. One limitation of hypothetical results is that they are generally prepared with the benefit of hindsight; in addition, hypothetical trading does not involve financial risk, and no hypothetical record can completely account for the impact of financial risk in actual trading — for example, the ability to withstand losses or to adhere to a particular trading program in spite of trading losses. Numerous other factors related to the markets in general or to the implementation of any specific trading program, including liquidity, slippage, commissions, financing costs, exchange and broker outages and regulatory changes, cannot be fully accounted for and may adversely affect actual results.
Past performance. Past performance, whether actual or simulated, is not indicative of future results. The value of investments and the income from them can go down as well as up, and investors may lose some or all of their capital.
Risk of futures, derivatives and digital-asset trading. Trading futures, options, FX, crypto-assets and other derivatives involves substantial risk of loss and is not suitable for all investors. Leverage can work against you as well as for you. The risk controls described here — stop-loss orders, circuit breakers, reconciliation, monitoring and other processes — are designed to reduce, not eliminate, risk; they may fail to operate as intended, may be subject to gaps, latency, broker or exchange failures, and cannot guarantee any limit on losses.
Illustrative screenshots. Screenshots on this page are taken from Etherius’s live systems. Monetary values displayed — including net asset values, profit and loss, market values, margin and circuit-breaker amounts — have been scaled by a uniform factor for confidentiality and do not represent actual account balances, assets under management or results. In the trading-terminal screenshots the realised and unrealised P&L, ROI and Sharpe/drawdown tiles contain sample figures for illustration only; they are not actual results and must not be read as a performance record. Other values shown — prices, quantities, instrument-level percentages, order and position detail — are unmodified system output at the time of capture, with monetary amounts scaled as described.
Forward-looking statements. This page contains forward-looking statements, including planned capital ramps, planned venue integrations and expected system behaviour. Such statements reflect current intentions and assumptions, are subject to change without notice, and are not guarantees of future outcomes. Etherius undertakes no obligation to update them.
Artificial-intelligence components. Certain components described here rely on third-party large language models. Their outputs may be inaccurate, incomplete or inconsistent; Etherius constrains them through deterministic risk and execution layers but does not warrant their behaviour.
Accuracy and completeness. The information here is believed to be accurate as of the date shown but has not been independently verified, is subject to change, and may be incomplete or condensed. Etherius makes no representation or warranty, express or implied, as to its accuracy or completeness and accepts no liability for any loss arising from its use.
Confidentiality and distribution. This page is provided to the recipient for their own evaluation and may not be reproduced, redistributed or disclosed, in whole or in part, without Etherius’s prior written consent. Its distribution may be restricted by law in certain jurisdictions; recipients are responsible for informing themselves of and observing any such restrictions.
Regulatory status. Etherius Capital operates as a proprietary trading firm. Nothing here should be taken as a statement that Etherius or any of its affiliates is licensed, registered or authorised to provide regulated investment services in any jurisdiction unless expressly stated in definitive documentation.