It learns from its mistakes — with your consent and our gates.
Every day the mistakes report, approvals and rejections and live per-signal results feed the next model. Your data is used only if you opt in, pseudonymised, and a new model ships only after it beats the old one out of sample.
| Gate | Current | Candidate | Baseline | Result |
|---|---|---|---|---|
| Directional accuracy | 54.1% | 55.3% | 50.0% | Pass |
| Net return vs baseline | +1.2% | +1.6% | 0.0% | Pass |
| Max drawdown | −9.8% | −9.1% | −14.0% | Pass |
| Calibration error | 0.071 | 0.064 | — | Pass |
| Abstain when unsure | 12% | 14% | — | Pass |
Over-weighted momentum into an earnings date; the next training set adds event-aware examples.
Illustrative paper-trading sample. Not a record of real results.
Learning loop and report card, in detail.
Opt-in, off by default
Pseudonymised
A daily mistakes report
Losers switch off
Gated releases
The same consent text on every device.
Web, desktop and mobile show the engine's consent text with one switch. Withdrawing consent is recorded and honoured from the next export.
Use my pseudonymised activity (ideas, approvals, fills and outcomes, amounts as % of account) to train future models. Off by default; change it any time.
Illustrative paper-trading sample. Not a record of real results.
See how the AI is improving.
The daily mistakes report and gate results, published so users and investors can check the model's progress.
| Gate | Current | Candidate | Baseline | Result |
|---|---|---|---|---|
| Directional accuracy | 54.1% | 55.3% | 50.0% | Pass |
| Net return vs baseline | +1.2% | +1.6% | 0.0% | Pass |
| Max drawdown | −9.8% | −9.1% | −14.0% | Pass |
| Calibration error | 0.071 | 0.064 | — | Pass |
| Abstain when unsure | 12% | 14% | — | Pass |
Over-weighted momentum into an earnings date; the next training set adds event-aware examples.
Design preview of a feature in development. Figures are illustrative.
Web, mobile, desktop and API — together.
Every feature ships on every surface at once, and on every interface: REST, GraphQL, MCP and the TypeScript client.
- WebAvailable
- iOS and AndroidAvailable
- macOS, Windows, LinuxAvailable
- REST · GraphQL · MCPAvailable
Do you train on my trades by default?
No. Training consent is off by default. If you opt in, your data is pseudonymised and amounts are shown only as a percentage of your account.
Who decides when a new model ships?
Automated gates first, then two named approvers. Releases go to a small share first, with automatic rollback.