Rachet
Systematic · Fully automated

A trading system
built to not
overshoot.

Systematic, fully automated, and bounded by construction. An AI layer that can veto a position but never enlarge one — behind hard limits it cannot reach.

ONE DIRECTION ONLY

The edge is refusing to overshoot.

01 //Design principles

Restraint, expressed as architecture.

01.1

Bounded by construction

The worst case is set by the architecture, not by judgment in the moment.

01.2

No discretion

The system does not ask permission and cannot be talked into a trade. Human involvement is limited to halting it.

01.3

The AI can only subtract

The analyst layer may veto or reduce. It has no path to increase exposure, by design rather than by policy.

01.4

Fail closed

When a component is unavailable or state cannot be verified, the system does nothing rather than something.

01.5

Validated before trusted

Nothing runs that has not cleared a written acceptance standard fixed in advance.

02 // The guardrailsThe edge is refusing to overshoot.One direction only

Most of this system is machinery for not losing money. That is deliberate.

  • G-01A fixed fraction of equity at risk on any single position.
  • G-02A hard cap on concurrent open positions.
  • G-03A ceiling on total open risk across the portfolio.
  • G-04A daily loss limit that halts trading automatically.
  • G-05A losing-streak cooldown that pauses new entries.
  • G-06Per-position size caps.
  • G-07A price sanity band that rejects implausible market data.
  • G-08A volatility emergency stop.
  • G-09A kill switch with three independent triggers — operator confirmation, a file on disk, or automatic on detected state mismatch. It cancels all orders, flattens all positions, and halts until manually cleared.
  • G-10Startup reconciliation against the exchange. Unexplained state trips the kill switch rather than proceeding.
  • G-11Protective orders held at the venue, so they survive process failure.
  • G-12An independent monitoring witness that alarms on a missed heartbeat, running separately from the trading process itself.
03 //Validation

Measured against a standard fixed in advance.

Profit factor
2.01
Sharpe
1.17
Max drawdown
17.4%
CAGR
18.9%
Out-of-sample trades
242
Profit factor, stress test
1.97

Out-of-sample walk-forward results, 2018–2026, net of modeled fees, slippage and funding. Not live trading performance.

04 //What we rejected

Published results are worthless without the discarded ones.

These were tested against the same standard and rejected.

  1. R-01

    Configurations that raised returns while breaching the drawdown ceiling.

  2. R-02

    A broader asset universe that improved headline performance and failed the risk standard.

  3. R-03

    Candidates with attractive headline profitability whose risk-adjusted results revealed small-sample luck rather than durable edge.

  4. R-04

    An entire category of position that was permanently disabled once its financing cost was shown to consume the whole expected return — a result the cost model predicted before the test was run.

The full validation method, acceptance standard and limitations.

Read the approach