Caffè Cloud is a private research-to-execution stack: regime detection, walk-forward validation, and risk controls as code — brewed on one small server.
Private by design. Nothing trades until it survives the validation gate.
A hidden Markov model classifies the tape into regimes, and each regime gets the playbook that historically deserved it.
Every change re-runs rolling train/test windows. In-sample glory means nothing here; only the held-out record counts.
Position caps, drawdown stops, and symbol lists live in one YAML file. Changes arrive as pull requests and are gated by CI.
Crash, chop, and squeeze scenarios replay against the current configuration before it ever touches an order.
Deterministic crypto and stock screens run nightly on fixed rules, publish their candidates, and feed hypotheses to the gate — never orders.
Version-stamped deploys, nightly backups, and a main branch that only moves when both test suites are green.
Behind the login
Every deploy stamps its git SHA into the page footer and /version.json — what's running is never a mystery.
The dashboard sits behind authentication. This page is the only public face.
A cron job archives state every night and keeps two weeks of history, fetchable with one command.
Nothing reaches the server that didn't pass both test suites. The deploy key can run exactly one command.
The validation gate
Performance claims come from held-out windows the optimizer never saw.
Same data, same config, same result — every run is reproducible down to the trade.
The record keeps rejections too. When the regime-gate hypothesis didn't hold up, the record said so — and that's the point.
Touch the risk config and walk-forward validation re-runs automatically on deploy.
How it compares
| Discipline | Caffè Cloud | Spreadsheet + gut feel | Off-the-shelf bot |
|---|---|---|---|
| Regime-aware strategy routing | ✓ | — | rare |
| Out-of-sample validation gate | ✓ | — | — |
| Risk limits enforced in code | ✓ | manual | varies |
| Scenario stress testing | ✓ | — | — |
| Audit trail of every deploy | ✓ | — | — |
| Runs on infrastructure you own | ✓ | ✓ | — |
This comparison describes categories of tooling in general terms, not any specific product. Caffè Cloud is private research infrastructure, not a commercial service.
Looking ahead
An assistant that lives inside the dashboard and already knows the data: ask about regimes, validation runs, risk posture, or why a strategy sat out a session, without clicking through pages to piece it together.
A server-side agent that watches the pipeline and proposes improvements on its own branches. It can suggest anything; it can merge nothing. Main stays human-approved and CI-gated.
The final stretch of the pipeline: strategies that cleared walk-forward and stress gates route real orders per regime — paper first, tiny size after, and only while the validation record stays green.
Nothing on this page is a schedule or a promise. Scope and order change as the validation record dictates.
The stack
The daily view: what's running, what changed, what needs attention.
Where ideas queue, gate-test, and earn paper approval — evidence first.
The part that says no: gates, limits, and reproducible records.
Questions?
A private, self-hosted trading research platform: a regime-detection engine, a walk-forward backtesting pipeline, risk configuration managed as code, and a dashboard to watch it all — running on a single small server.
Not currently. Caffè Cloud is personal research infrastructure, and this page is its public face. The dashboard is private.
A strategy must survive rolling walk-forward windows it was never fitted on, plus crash, chop, and squeeze stress scenarios, before it's eligible to trade. Failed hypotheses stay in the record.
On one modest VPS. Code lives on GitHub; pushing to main runs both test suites and, only on green, deploys over a single-purpose SSH key. Nightly backups keep two weeks of history.
No. Nothing on this site is a recommendation to buy or sell anything. It documents private research tooling.
Regimes detected, records kept, risk capped — before a single order leaves the house.