The incentive in paid backtesting runs one way: a flattering equity curve is what earns a five-star review, and a negative verdict is the harder thing to sell. Your rule gets coded exactly as you describe it, tested properly, and written up as what it actually does — including when the answer is that it does nothing.
Behind this is two years in the markets — discretionary trading first, then systematic — and a futures research programme currently running: 84 strategy specifications frozen and recording live, continuous tick capture across 15 instruments, and automated checks that have caught real errors in its own results. Most of what gets tested fails. Reporting that plainly is the entire point — a backtest that always finds an edge has told you nothing.
What this is, precisely. Technical work: coding, testing, data infrastructure and written analysis, on a strategy you specify.
What it is not. No recommendations on what to trade, no signals, no managed money, no investment advice. No claim or promise is made about future returns, and nothing I deliver should be read as a personal recommendation. If you are making decisions about real money, speak to a qualified, regulated financial adviser.
Not a support service. Setup work is handed over complete and documented, and any alerting reports to you rather than to me. I do not monitor client systems, and there is no service-level agreement or guaranteed response time. If something needs fixing later, ask and I'll quote for it.
Send a backtest over. It gets checked against eight known failure modes — lookahead, costs, benchmark, trial count, holdout integrity, overlapping windows, fill assumptions and data seams — and comes back as a one-page verdict on each. Fixed scope, fixed price, no conversation needed first.
Before you build on data, find out what is wrong with it: rollover seams recorded as real price moves, missing sessions, timezone and session-boundary errors, duplicate or out-of-order bars, survivorship. You get the findings and the script, so it can be re-run any time.
Your rule, coded exactly as you describe it, tested on real data with costs and realistic fills, split out-of-sample before anything runs. You get the code, the full trade list, a benchmark comparison and a written verdict — including when the verdict is that it does not work.
Your own tick or minute data, recorded reliably rather than scraped. Automatic contract rollover, an audit that checks no rollover gap was recorded as a real move, stall detection, and a runbook so you know what each alert means. Alerts go to you. The optional monthly part is off-site backup: your data copied to managed storage, with an automated check each month confirming it arrived and is readable — because almost nobody has a verified second copy until the day they need one.
A rule you trade by hand turned into code that runs: Python for a standalone runner or your broker's API, or C# natively in Quantower and NinjaTrader. Comes with a historical test and documentation of exactly what it does and does not do. Or a focused web tool that does one thing well — this site is the evidence of what that looks like.
I also take on historical data acquisition and cleaning, alerting set up on a system you already run (£200–400, one-off) and research write-ups from results you already have. If what you need isn't here, describe it and I'll tell you whether I'm the right person — including when I'm not. On platforms specifically: Python and C# are covered, Pine Script and MetaTrader are not, and anything else is worth asking about.
These are the actual documents, not samples written for a brochure. They are the fastest way to judge whether this is worth your money.
Scope in writing first, including what is explicitly out of scope. For anything involving a trading rule, the first deliverable is a written specification of that rule — every condition stated unambiguously, with the edge cases decided — for you to approve before any code is written. It is the cheapest hour in the project and it is where misunderstandings get prevented rather than argued about.
You get the code, not just a report, with instructions to reproduce the result yourself. Payment is half up front, half on delivery, and the work is yours outright once the invoice is paid.
You can change your mind. If you are buying as an individual rather than a business, you have 14 days to cancel and get your money back — and if you have asked me to start in the meantime, you pay only for what has actually been done. The terms set out how that works.
On data. Most platforms export their own history, so the first question is simply what you already have — that costs nothing and is usually enough. Where it is not, the data is bought for the job and passed on at cost, agreed before anything is spent; typically £10–40 per market. Nothing is marked up, and the purchase is yours to keep.
AI tooling is used heavily here, as it is in most development now. The checking is what you are paying for: AI writes working code quickly and produces plausible wrong answers just as quickly, and knowing where it will be confidently wrong is most of the skill. Nothing goes out unverified.
Describe the rule in plain English — one paragraph is enough for me to quote from. Tell me the instrument and the timeframe and I'll tell you what testing it properly involves.
Richard Hodgson, trading as Point of Control — a sole trader based in Cleckheaton, West Yorkshire, United Kingdom. Prices are in pounds sterling and exclude VAT; I am not VAT registered. Full terms apply to commissioned work.
Everything else on this site stays free, with no accounts, no paywall and nothing collected about you. That is not changing.