A measuring instrument for trading

Know,
don't hope.

Rexagone promises no gains. It executes your trading plans and tells you the truth about your results — uncertainty included.

Create an accountSign in
  • No projected gains
  • Paper · Demo · Live
  • Every figure with its interval
Fig. 1a plan card, as you receive it
Rejecting weighs as much as approving. Stop and target come from the strategy, not from the mood of the moment.

Six sides, six steps

Every trade goes through six steps. The last one is what everything else exists for.

  1. 1Candlethe market as it is
  2. 2Signalwhat the strategy reads in it
  3. 3Planthe trade proposed, sized, bounded
  4. 4Decisionapproved or rejected, by whom, when
  5. 5Orderwhat really leaves for the broker
  6. 6Outcomethe result, measured in R

What the measure says

A statistic has three lines: the value, its interval, its sample.

Below the threshold, the tile turns grey and says so. It does not disappear: it tells you that you don't know yet.

Cumulative R

+4,3 R

64 closed plans

ExpectancyInconclusive

+0,07 R

[−0,18 ; +0,24] · n = 64

Costs / gross result

7,2 %

spread 0.05 R · carry 0.01 R

Fig. 2expectancy per plan · 95% interval
−0,4 R0 R+0,4 R+0,8 RModèle 2022n = 201Silver Bulletn = 144Donchian 4 hn = 204Donchian journaliern = 696
Four strategies, measured. 124 declared trials, at the broker's real costs. None passes validation: every interval touches or contains zero. This is what Rexagone tells you about your own plans.

Write, test, delegate

Write your strategy. The bench measures it under a protocol. An agent can write it for you; it never grades itself.

Everything that runs here is a TypeScript module on the Rexagone SDK. Yours, or the official ones you read and copy. Nothing goes live without a validation.

The workshop

Your source, checked in a sandbox: no network, no clock, no reading of the future. A version is a fingerprint; the bench only knows versions.

The bench, under protocol

Hypothesis, decision rule, disjoint windows, declared trials: written before the first run. The validation is read once, on its whole window, and seals the protocol.

@Builder and @Judge

@Builder writes and corrects. @Judge reads the result and returns PASSED, FAILED or INCONCLUSIVE, with its reasons. Two separate calls: no agent grades its own work.

One cycle
  1. 1Write@Builder, on your thesis
  2. 2Checkthe sandbox compiles, or refuses and says why
  3. 3Benchone calibration trial of the protocol
  4. 4Judge@Judge, on the interval and the costs
Each cycle uses one trial. The loop stops on PASSED, at the cycle cap, or after two cycles without improving the lower bound of the expectancy. The validation is never run by an agent. Every version written by an agent is your draft: you choose the one you keep.

What you read

  1. 1

    Every plan says why

    Entry, stop, target, risk in R and in currency, and the rules that triggered it. You approve or you reject.

  2. 2

    No rate without its interval

    8 winners out of 13 prove nothing. Every statistic carries its confidence interval and its sample size.

  3. 3

    Strategies are validated — or not

    Those that fail validation stay visible, with their numbers, and only run in paper mode.

  4. 4

    You decide, the machine executes

    Suspend is one gesture. Close all asks for a confirmation with the figure. No liquidation is ever decided by the machine.

Three modes, one order

PaperDemoLive

Start on paper, with no broker account. Move to demo. Live is earned: a strategy that failed its validation never goes there.

Measure first.

An account, a paper account, your first plans. No card, no promise.

Create an accountContact

CFD trading carries a high risk of loss. Past performance does not predict future results. Rexagone never displays projected gains.

Rexagone

© 2026 Rexagone

A measuring instrument, not a casino.

ContactSign in