Paper-only execution; the system never sends live orders.
Operator decides every trade. Engine learns from paper fills only.
Desk Commander
A private, VIP-only research workspace for systematic traders. Runs on a market-hours cron, drafts ideas with AI, records every signal whether it gets traded or not. We built and run it as a private operator tool — shown here as proof of capability.
- Started
- Early 2026
- Operator
- One systematic trader
- Environment
- Paper, validated in public
Before
What the trading day looked like before.
Mornings started with 60–90 minutes of manual scanning — same routine, every weekday, before the market open.
Signal ideas lived in scattered notes, browser tabs, and a spreadsheet that never quite stayed in sync with reality.
Strategy adjustments happened on instinct after losing trades, with no record of what changed or why — easy to over-fit yesterday's regret.
There was no honest record of "what I almost traded but didn't" — only what hit the broker. Half the learning was invisible.
Sunday review meant reconstructing the week from memory and screenshots — useful, but exhausting, and inconsistent week to week.
What we built
A research engine that runs itself, then hands you the brief.
Desk Commander is a research engine built around the rhythm of the trading day. It runs on a scheduled cron — twice a weekday, plus a Sunday knowledge job — and executes the same workflow every time without manual involvement.
Each run scans the market for setups that match the operator's filters, then uses an AI pass to enrich each candidate with structured context: sector, recent price action, gap behaviour, the plan if it were taken. Bracket orders are placed against a paper account so the system has real fills + slippage to learn from, without putting money at risk.
Every signal is recorded — taken or skipped, profitable or not — in a database the operator owns. A Sunday job rolls the week up into a knowledge brief: what triggered, what filled, what was avoided, what the regime felt like. The operator reviews. The operator decides. The system never sends orders to a live broker on its own.
After
What changed when the engine took over the morning.
Mornings now open with the scan already done. The first 60 minutes are spent reviewing AI-enriched candidates, not assembling them.
Every signal — taken, skipped, hypothetical — lives in one place with structured context. The dataset for retrospection is finally complete.
Strategy parameters are version-controlled. A change requires a commit; the system can't be quietly over-fit after a losing week.
The Sunday knowledge brief writes itself. Weekly review used to take 90 minutes; now it's a 20-minute read of what the engine already summarised.
The operator's attention moved from "finding signals" to "making decisions" — the part of the workflow only a human can do.
Audit stack
Every decision documented, every parameter versioned.
ADR-001Accepted ADR-002Accepted Twice-daily cron + a Sunday knowledge job.
Pre-open scan, mid-session check, weekend retrospection.
ADR-003Accepted AI critic is optional, not gating.
Every signal ships with structured context; AI commentary is additive, never required.
ADR-004Accepted Log every candidate — taken, skipped, hypothetical.
Half the learning lives in the trades we almost took. Now they're in the record.
ADR-005Accepted Strategy parameters live in version control.
No quiet over-fitting after a losing week. A change is a commit.
ADR-006Accepted Telegram is the primary delivery surface for alerts.
Mobile, latency-tolerant, decision context in the message body.
Proof
What it ships into your inbox each session.
Per-session signal alerts arrive via Telegram. Ticker, conviction band, entry / stop / target, and a one-line plan for the setup. No charts to interpret on a phone — the decision context is in the message itself.
Next
Want a research engine that runs without you?
If your trading workflow has a repeatable shape that you're executing by hand every day, we can build the engine around it. Book a free intake to scope it.