Your rules, running and watched — while you're at work.
Finding a rule that works is the part most traders can already do. The part that defeats them is keeping it running for months, and knowing the moment it stops working. That second part is what I sell.
This is a pilot, not a running business. No paying clients yet. The first validations are free — I'm testing the pipeline on real rules, and I'd rather say that plainly than let a price list imply otherwise.
The prices further down are what this would cost if it becomes a service. Nothing is charged today, and there is no checkout link.
Decebal Flos — four decades in IT (Assembler to C++ to databases), former IT firm owner, based in Graz, Austria. I build and run validated trading automation for myself; this service applies the same discipline to your rules.
The problem
You have rules you trust. You're at your desk 8 to 5, and your setups don't wait for you. So you automate — and the first 80% goes fast. Then comes the hard 20%: reliable data, execution edge cases, the connection dropping at 3am. A job that dies quietly at 02:00 and reports nothing looks exactly like a job that had nothing to do.
Buying a black box means trusting a stranger's curve instead. And a rule that worked for two years can stop working without announcing it — the equity curve flattens, and by the time you notice, you've given back a year.
What supervision actually means
Not a dashboard you have to remember to look at. Specific, boring machinery:
- Every run leaves a heartbeat. A job that succeeds says so. A job that stops saying so raises an alert before you'd have noticed anything was missing.
- A failed dependency never produces a normal-looking result. When a price feed dies, the run says it died. It does not quietly use yesterday's number and hand you a clean report.
- Retries with growing waits, then a message. Not one attempt and silence, not an infinite loop — a bounded number of tries, then somebody gets told.
- Logs you can actually read, at a path you know, kept long enough to investigate and cleared before they fill the disk.
- Decay watch. A monthly read on whether the rule is still earning its place, against the same thresholds it passed on — not against how you feel about it.
I built this for my own trading first, and it caught real failures: a node that ran for nineteen days on data nobody else agreed with, ten jobs that died overnight and stayed dead, a watchdog that reported everything green while it was unable to check anything at all. Those are the failures that cost money, and none of them announce themselves.
The gate before any of that
I don't supervise an unvalidated rule. Automating one just executes a losing process with better discipline, and I'd rather lose the sale than sell you that.
So every rule goes through the gauntlet first — walk-forward, out-of-sample folds, thresholds fixed before the data is seen. If it fails, we stop there. You pay for the truth and you don't pay to automate a loser.
The method is published and free — see the bottom of this page. If you'd rather run it yourself, run it yourself. What I charge for is what happens in the eleven months after the verdict.
How you get there
- Validation gauntlet. Your rules, exactly as you trade them, through walk-forward testing: out-of-sample yearly folds, no lookahead, parameter perturbation, 3× cost stress. Thresholds fixed before the data is seen. Verdict PASS / MARGINAL / FAIL in a written report, same day. If your rules fail, that's where we stop — you'll know why, and you won't pay to automate a losing process.
- Rule-to-config translation. Rules that pass get translated into a configuration that runs on your machine and your brokerage account. Your keys never leave your hands. Running within a week of a passed validation.
- Supervision. While you're at work, the system watches your automation: every setup handled per your rules, logged and auditable. You get honest reports — including a decay watch that tells you when a rule stops earning its place.
Honest verdicts are the gate, not the product
Read the full report → — all 15 configurations, both gates, the thresholds fixed before the runs. 12 of the 15 failed.
My own ideas go through the same gate and my failures are published in the method guide — the elegant exit rule I dropped the same day it lost, the short-side thesis the data killed in two evenings.
Pricing
Supervision — monthly
€120/month
Starts after your onboarding.
Daily supervision and auditable logs, honest monthly reports, strategy decay watch. Cancel any month — the system earns its keep or it doesn't.
Ask about supervisionOnboarding — one time
€300
Validation gauntlet on your rules + translation of a passed rule into your running configuration. If validation fails, we stop at the gauntlet and you pay only for the truth.
Start — send your rulesYour rules stay yours
- I am not collecting strategies. I need rules to test the pipeline against — the pipeline is the thing I'm building, not a rule library.
- Send the minimum that makes the test runnable: the parameters of one rule. Not your whole system, not your position sizing, not your edge.
- I don't publish your rule, don't pass it on, and don't add it to my own trading. My slots run on my own rules, and they fail the same gauntlet.
- Inputs are deleted after the report unless you ask me to keep them. The report is yours; I keep only the aggregate — how many passed, how many failed — with nothing that identifies a rule.
- If that still isn't comfortable: don't send anything. The full method below is free and you can run it yourself. That option is deliberate, not a fallback.
What this is not — read before you pay
- I never hold, receive, or operate your money. Your account, your broker, your software, your keys — always.
- I never trade for you. I deliver configuration, validation, and monitoring of software you run.
- No signals, no tips, no investment advice. Verdicts are statistical tests of your own rules — they predict nothing.
- Most rules fail validation. That result costs you the onboarding gauntlet and saves you everything after it.
Questions people actually ask
The method is published. Why not just give it to an AI myself?
You should, if you want to — that's why the guide is free and why the thresholds are printed on this site. PBO and the deflated Sharpe ratio are published academic work, not my secret.
What an AI won't do is refuse you. Ask it whether your strategy is good and it will find a way to be encouraging. It also can't give you nine years of fee-modelled data, and it won't stop you from seeing the result and then adjusting the threshold — at which point you no longer have a threshold, you have a preference.
And none of it addresses the eleven months after the verdict, which is the part I actually charge for.
What if my rules fail?
Then we stop at the gauntlet and you don't pay to automate a losing process. Most rules fail — twelve of my own fifteen did, and I dropped every one.
The report tells you which gate failed and what would change the verdict. That's the deliverable, not a consolation prize.
What do you actually need from me?
The parameters of one rule: market, timeframe, entry, exit. Not your whole system, not your sizing, not your edge. If a rule can't be described without handing over everything, it's outside v1 scope and I'll say so plainly.
How long until something is actually running?
Validation is same-day. If a rule passes, translation into a running configuration takes about a week — on your machine, your brokerage account, your keys.
I run about two of these a week. That limit is real, not scarcity marketing.
Do you ever touch my money?
No. Your account, your broker, your keys — always. I deliver configuration, validation and monitoring of software that you run. No signals, no tips, no investment advice.
Start with the free part
The full validation method — including my published failures — is free, and you can run it yourself with any competent AI assistant. If you'd rather I run it: use the form below. v1 scope: moving-average cross, breakout/channel, RSI-pullback families; daily bars, long side. Rules outside that get a plain "no" — no force-fit.