Hedge funds
Research infrastructure that assumes an allocator will ask how you know. The evidence is the output.
What's actually behind the buttons
The archive behind every fill, settle and verdict.
Back through 2020 and 2022 — regimes a calm-year backtest never meets.
FX majors and crosses, indices, metals, oil, crypto.
One of the published studies — this one came back negative.
Idea to verdict to live — one shape the whole way
Every strategy on the platform travels the same road, and the shape never changes between stations: what you backtest is what papers and what exports.
One pipeline, one strategy shape end to end. What you backtest is byte-for-byte what papers and what exports — there is no re-implementation step where drift can hide.
Why you run a book and a research process
Out-of-sample by time, significance, and honest degradation — not a curve that goes up.
Enterprise workspaces can inherit the accumulated knowledge instead of training from nothing.
Workspaces, roles, and dedicated data isolation where you need it contractually.
The honest, out-of-sample test rejects most strategies — and it will reject the ones that don't hold up. That's the feature.
138 million one-minute bars behind the fills. When a bar held both the stop and the target, the order was measured, not assumed — and where the data was missing it booked the loss. That is a question you get asked in a diligence meeting.
Every strategy carries the number of forward trades its edge needs before significance, computed from the size of that edge. "Promising" has a finish line rather than an opinion attached.
Describe it. Test it. Know.
Research throughput matters, but an allocator's first question is epistemic: how do you know? The wizard's whole design is that the evidence a strategy generates is the same evidence you would show in diligence.
Describe the setup the way you'd tell a colleague. The AI turns it into executable rules and — this is the part that matters — tells you what it ASSUMED and what it COULD NOT express, instead of quietly approximating. A strategy silently rounded to something else is the worst thing a builder can hand you.
Pine Script, MQL, Python — pasted in, translated, and shown back to you for approval before anything counts. Export goes the other way too: Pine, MQL5, Python and LEAN, and the export refuses rather than shipping a file that silently behaves differently. More
Screenshot a setup and the reader turns it into English you can correct — deliberately not straight into a strategy, because a chart shows what happened, never what the rule was. You confirm the rule; then it builds.
Sessions, trend regime, news windows, multi-timeframe confluence, seasonality, relative strength, fair-value gaps, liquidity sweeps, level flips, ranges, freshness, VWAP, currency strength, the dollar index, price channels. Each card carries what our own research found about it — including when the finding was 'no edge'.
Every backtest is graded on a held-out, recent stretch of history the strategy never saw. 'Profitable' is only said when the out-of-sample evidence supports it, and a promising strategy carries the number of forward trades its edge still needs — counted down, not implied. More
Long runs show a live progress bar driven by bars the engine has actually simulated — not a timer animating toward a guess. Small thing. Same principle as everything else here.
The journal shows each trade on real candles with the exact conditions that fired it — entry, stop, target, and the predicate values at the signal bar. "Why did it take this?" has a literal answer.
Trade the past bar by bar, by hand, filled at the backtester's own cost rules. The future is not hidden in your browser — it is never sent, so there is nothing to peek at. Every hand decision is then graded by the same grader that scores strategies, and rule inference will tell you what you were actually trading and whether it survives a decade.
Crypto does not pay EUR/USD's spread here. Per-instrument cost profiles are floors — they can make a result worse, never better — and every run reports exactly what it charged. An invisible cost model is indistinguishable from a flattering one.
Position sizing plus a Monte-Carlo risk-of-ruin read over your strategy's real trade distribution — what a losing streak actually looks like at your risk per trade, before a live losing streak teaches you.
Deploy a strategy and pull live signals through a small client that holds only the strategy's ID — the rules never leave the platform, and entitlement is re-checked on every call.
Every run's trades are recorded as a labelled, exportable dataset — market, side, timing, outcome, strategy fingerprint. Twenty iterations become training data instead of twenty screenshots.
The two rules that keep every number honest
Anyone can show you a curve that goes up. These are the mechanics that decide whether a result here is evidence — drawn, because they are the part most platforms keep vague.
A strategy is judged on a recent stretch of history it was never shown. An in-sample curve — the number most backtesters lead with — is a rehearsal, and the platform labels it as one. The engine also re-proves on every run that no rule could see past the wall.
Most backtesters silently assume the answer that makes the result look better. This platform measures it on minute data — and when the evidence isn't there, it charges you the stop. Your paper trades and your backtests are settled by the same rule.
The full grading discipline — walk-forward, significance, Monte-Carlo over the real trade distribution — is on the methodology page. The grading rule itself is not published; the boundary is stated, not blurred.
Train our learning system on your strategies
The Conscious is the desk's continuous-research layer: condition-level attribution, evolution under out-of-sample discipline, reconciliation of failures with proposed-and-proven fixes, and a decision layer constrained server-side to proven edges. Enterprise workspaces can inherit the accumulated research instead of cold-starting.
The same learning system that runs our own research desk, pointed at your strategies. You choose which folders it may learn from. Improved copies must beat the original on forward paper trades before promotion is even offered — and a promotion that stops holding up is demoted back to testing by the evidence, never quietly kept.
One click sends any strategy you've built or imported to the desk. From that moment it is worked like one of ours: backtested across condition lenses, stress-checked, and paper-traded around the clock.
Most strategies aren't good or bad — they're good somewhere. The Conscious sweeps your rules across sessions, trend regimes, volatility states, currency-strength and dollar-index conditions, and price structure, and reports the edge per condition instead of one averaged-away number. More
When a filter combination genuinely improves your strategy out-of-sample, the improved copy is deployed to paper as a challenger — your original untouched. It must beat the original on real forward trades before promotion is even offered.
Every promotion is your click. And a promotion is not forever: the evidence is re-read continuously, and a strategy whose post-promotion record stops holding up is demoted back to testing — never quietly kept, never deleted. Retiring a strategy is a decision only you can make.
Losing trades are read as evidence. The reconciliation loop finds what they had in common, proposes a change to the rule, and proves the change against the same grading as a new strategy — then asks you. More
The orchestrator turns every module's opinion into a single decision with the evidence attached — and it is limited, server-side, to edges that have actually proved out. It advises. It cannot arm anything. More
Brain scope is per folder: point the Conscious at the strategies you want worked and exclude the rest. Your data trains your workspace's understanding — it isn't pooled into someone else's.
The desk's paper book renders inside the platform — open positions, closed trades, the equity line — so 'is it actually trading my strategy?' has a visible answer, not a support ticket.
Practice on the past — without the past leaking
Every serious platform has bar replay. Ours is the only one where the future physically cannot be peeked at, and where your hand-trading is graded by the same engine that grades strategies.
Practice on real history bar by bar, place trades by hand, and get filled by the same cost rules as the backtester. Client-side replays ship every future bar to your browser and merely hide them; here the cursor lives on the server, so honesty is structural. Every hand-placed decision is then graded by the same grader that scores every strategy — and the platform will tell you, in plain terms, when your sample is still too small to mean anything.
Answers to the questions a diligence meeting actually asks
Not 'what did it return' — 'what would survive an adversarial read of your process'. These are the process answers, stated plainly.
Split by time, never shuffled; the verdict comes only from the held-out period; and the engine re-proves on every run that no rule could see across the wall. In-sample output is labelled a rehearsal.
Every strategy carries the number of forward trades its edge size requires before the evidence is statistically meaningful — computed, displayed, counted down. A strategy with no positive edge gets no countdown, because there is nothing to prove.
138 million one-minute bars behind the settles. Ambiguous bars are measured, not assumed; missing data books the loss; costs are per-instrument floors that can only hurt results; and every run reports the exact costs charged.
A decade across 30+ instruments, with every incoming bar reconciled against held history — a feed that disagrees is refused, not blended. Sources and reconciliation mechanics are disclosed under NDA, not on a marketing page.
Workspaces with roles and folder-level scope over what the learning layer may touch; API access to everything the UI does; export of strategies and of the labelled research corpus. Dedicated isolation where the mandate requires it.
It will not show performance marketing, it will not let a backtest silently become 'live results', and it will not arm live execution by itself. The refusals are load-bearing: they are why the numbers it does show survive scrutiny.
We'll go as deep as an NDA allows.
We publish what didn't work, too
Credibility is not a claim; it's a track record of saying no. These findings ship inside the product, next to the filters they inform.
≈ 0R over 87,000 sweeps across a decade. Unselected, the famous setup pays nothing.
Flat across 530,000 samples. We do not gate or size by volume anywhere — and we say so.
No edge over 1,401 graded trades on a decade of data.
A real out-of-sample edge, consistent across every yearly window we held out — fragile to tight stops, which the platform tells you rather than hiding.
Materially better than holding, measured across 286,000 fade situations.
These are our own studies, run on our own archive, and the negative results ship inside the product next to the positive ones. A platform that only ever finds edges is selling you something. Most ideas don't work; the value is knowing which — before money does the experiment for you.
Hedge funds — asked and answered
Is this a fund, a signal service, or software?
Software. We build research infrastructure and tooling; we do not manage money, sell trade advice, or publish performance. The output of the platform is evidence about your strategies, generated by a process you can put in front of an allocator.
Can our researchers use their own tooling on top?
Yes — the platform is an API. Authoring, backtests, grading, exports (Pine, MQL5, Python, LEAN) and the labelled trade corpus are all reachable programmatically, with workspace-scoped keys.
What does 'inheriting the accumulated knowledge' mean?
Enterprise workspaces can start from the platform's validated research base — the condition-level findings our own desk has accumulated and proven — rather than training the learning layer from zero. Your strategies and results remain yours and are not pooled back.
How is the learning layer kept honest?
Everything it proposes must survive the same out-of-sample grading as a human-authored strategy; improvements deploy as paper challengers that must beat the original on forward trades; promotions are re-audited continuously and demoted when the post-promotion evidence fails. And it is structurally advisory — the constraint to proven edges is enforced in the backend, not by model behaviour.
What happens in an NDA conversation that doesn't happen here?
Methods. Grading rules, split mechanics, detection internals, data sourcing and reconciliation. This page tells you what the machine does and what it refuses to do; the demo goes as deep as the NDA allows.
Put your process in front of the machine.
A demo walks your desk through the evidence chain end to end — as deep as an NDA allows.
No card required · sales is for desks and teams