A research workbench for US equities and crypto. The AI proposes and explains; a deterministic engine validates, simulates, and enforces the risk boundaries — and every figure you see is that engine's own output.
Simulation only · Not investment advice · For research & education
The only performance figure on this page is our own fixed reference strategy, run in the open over a disclosed universe — net of costs, per run, hypothetical. Then run a simulation on your own idea and see what it actually did.
Run a historical simulationA real, reproducible illustration
One un-optimized Reference SMA 10/30 crossover run through the exact engine your backtests use, over 2 of 5 instruments and 98 delayed historical bars. Equal-weight, per-run average — not a portfolio, not compounded.
Read the universe before you read the number: it is a small, hand-picked list of currently-listed large caps and major crypto pairs, selected with hindsight, and a symbol only counts once enough history has been backfilled for it. These averages are not survivorship-free and are not a broad-market study.
Net of modeled costs — commission 2 bps/side + slippage 1 bps. Aggregated across 0 hypothetical trades. These are per-run figures over a limited sample and are NOT annualized (annualized metrics like CAGR/Sharpe/Sortino are omitted — they mislead over short windows). Per-run average, not indicative of future results and not an achievable return. Engine 0.1.0 · spec 40d6776a99.
| Instrument | Return / run | Max DD | Trades | Bars |
|---|---|---|---|---|
| BTC/USD | 0.00% | 0.00% | 0 | 49 |
| ETH/USD | 0.00% | 0.00% | 0 | 49 |
Runs over the DELAYED historical bars held for each instrument (polled market data, not a real-time SIP feed). Only symbols with enough history to warm up the slow SMA are included; coverage grows as more data is backfilled.
Six things the platform actually does today — each one visible in the engine, none of them a roadmap item written in the present tense.
Bars are replayed one at a time and orders fill at the next bar's open — never the signal bar's close. The engine cannot act on a price it would not have had.
Commission (2 bps per side) and adverse slippage (1 bps) hit every fill, and the API exposes no way to switch them off. Reported figures are net, with total commission shown beside them.
Every simulation records a spec hash, an engine version, and a data fingerprint — so a figure you saw last month traces back to the exact inputs that produced it.
Describe an idea and the copilot drafts one canonical JSON spec. It has to pass schema validation before it can run, and nothing is auto-executed — the AI proposes, the engine decides.
Position caps, a max-daily-loss halt, and consecutive-loss limits run inside the engine, not as a warning label. Every time one fires it is recorded on the result.
Paper sessions run the same engine against delayed bars on a virtual ledger. The real-money seam refuses by construction: the controls it requires are not built, so it answers 403.
Indicators, entry and exit rules, sizing, and risk limits live in a single versioned JSON spec. Write it by hand in the editor or have the copilot draft it — the same validator and the same engine run either way. There is no separate drag-and-drop builder: one spec, one execution path, nothing hidden between them.
Candles with SMA/EMA/RSI overlays and oscillator panes, computed causally by the same indicator code the engine uses — warm-up windows left blank rather than back-filled. Prices are polled, not streamed, and are labelled DELAYED with the time they were last fetched, wherever they appear.
A historical simulation under assumptions you can read — not a forecast and not a promise. Results are hypothetical; past performance says nothing about the future.
Create your account