Strategy testing & validation
Sound familiar? The problem isn't your strategy. It's that nobody checked where it breaks. We do — with real costs, real fills and data it has never seen — before you put money behind it.
01 — The problem
None of these show up as an error. They show up as a curve that goes up and to the right, and then doesn't. Every engagement we run is, in the end, a hunt for these four.
A signal computed on the bar it trades, an indicator that quietly peeks at the close, a universe built from today's index membership. Look-ahead and survivorship don't exaggerate an edge — they invent one.
Same-bar entry at the exact price, no commission, no slippage, unlimited size at the touch. Thin edges live or die here: a strategy at 0.1R expectancy is one tick of slippage away from being a losing system.
Sweep enough parameters and something always looks brilliant. The winner looks good because it won, not because it works. Without deflating for how many configurations were tried, a backtest is a lottery result.
An equity curve is a single ordering of your trades. Reshuffle them and the drawdown you'd actually have lived through may be twice as deep. Sizing off the curve that happened is how accounts get closed in month four.
02 — Who this is for
From one trader with an idea to a desk that needs overflow research capacity. The process is the same; the scope and the reporting depth change.
You've got a setup that works on the chart and a suspicion it won't survive costs. You want an answer before you fund it, not after.
You already run automated strategies and need more research throughput than one person can produce — sweeps, walk-forwards, and a second set of eyes on the method.
You have the infrastructure and the people; what you don't have is spare cycles. We run scoped research and validation work alongside your team and hand back reproducible output.
Not for you if you're looking for signals, tips, managed money, or someone to tell you what to trade. We don't do any of that — see the disclaimer at the foot of this page.
03 — What we do
No dashboard to learn, no template to fill in. A person reads your idea, writes the spec, runs the work and answers your questions.
Your rules coded exactly as specified and run over clean historical data — with commission, slippage and realistic fills modelled, not waved away.
Parameter sweeps, walk-forward analysis and sensitivity surfaces — tuned toward robustness rather than the prettiest possible curve.
Every entry and exit drawn on the chart, so you can see when your logic fires, when it stays flat, and where it hurts.
Take a strategy you've validated and put it live on your platform, with monitoring, logging and a documented kill-switch.
A rough idea turned into a precise, testable specification — filters, sizing, exits and the edge cases you hadn't decided yet.
You've trained a model — we stress-test whether the edge is real or an artefact of how it was validated. Leakage, decay and honest out-of-sample scoring.
We don't originate strategies, sell signals, manage money, or tell you what to trade. You specify the strategy; we test it and report what we find.
That boundary is deliberate. It's also what keeps our results honest.
04 — The deliverable
This is the depth every engagement ships with — headline metrics, risk and tail statistics, Monte Carlo, regime analysis and overfit diagnostics. Scroll it: it's the whole thing, the unflattering parts included.
Headline metrics tell you what a strategy earned. These tell you what you'd have had to sit through to collect it — which is what actually decides whether a system gets switched off halfway down a drawdown.
Out-of-sample performance below in-sample is normal and expected. A strategy that performs better out-of-sample usually means the split leaked — we say so when it happens.
Position sizing should be set against the worst 5% row, not the median and not the realised curve. Most accounts are not closed by the drawdown that was backtested; they're closed by the one that wasn't.
A single Sharpe hides the question that matters at deployment: does the edge hold everywhere, or is one regime carrying it? Sessions are clustered on the strategy's own realised volatility — that describes the conditions this system actually experienced, and is not a macro regime call.
Any sweep produces a winner. These four numbers are what separates a winner that means something from one that doesn't — and they are reported whether or not they flatter the result.
Backtested and simulated results are hypothetical. They do not represent actual trading, carry no guarantee of future performance, and are subject to the modelling assumptions listed above. This report describes the behaviour of a client-specified strategy; it is not advice and not a recommendation to trade.
05 — How it works
You don't have to hand over your rules to start. A sentence about the market, the idea and what you want answered is enough — we'll come back on whether clean data exists and what testing it would involve.
Your idea becomes an unambiguous written specification — entries, exits, filters, sizing, session rules, edge cases. Nothing gets coded until you approve it. You can start this yourself with the spec template.
Coded to spec and run over clean data with costs and slippage. Out-of-sample held back. Parameter surface swept, Monte Carlo run, overfit probability measured — so you can see whether the edge is real or fragile.
Standard KPIs plus any custom metric you ask for, the full assumption list, and a plain-English read on what held up and what didn't — including when the answer is "this doesn't work".
Scope, timeline and deliverables are agreed in writing before we start. Every engagement has a deadline.
06 — Scale & method
A person owns your engagement end to end. What sits behind that person is a compute pipeline that can evaluate far more of the problem space than a laptop and a spreadsheet ever will.
Six parameters at twenty values each is sixty-four million configurations. We evaluate the surface — plateaus, cliffs, dead zones — rather than reporting whichever single setting happened to win.
The same rule set run across every instrument with clean history, so you can see whether the edge lives in the logic or in one symbol's particular past. Cross-sectional robustness is the cheapest lie-detector there is.
Block-bootstrap resampling of the trade sequence, so you get a distribution of drawdowns and outcomes instead of the one path that happened to occur. Sizing decisions come from the tail, not the median.
Deflated Sharpe, CSCV overfit probability, purged and embargoed cross-validation, feature-stability and regime clustering. The machine-learning toolkit pointed at whether your edge is real — never at inventing one for you.
07 — Coverage
We're not limited to a fixed list of supported markets. Data availability is the only real constraint — and we'll tell you whether clean history exists before you commit to anything.
08 — Why trust the numbers
Not a variant we liked better. If we think a rule is a problem, we say so in writing and let you decide — then we test what you chose.
Commission and slippage are in every result by default. A backtest without costs isn't a backtest; it's a chart.
A held-back window no parameter ever touched. You see both numbers, side by side, even when the second one is worse.
We map the parameter surface and show you the whole thing. Edges that exist at one exact setting aren't edges.
Data source, fill logic, sizing, survivorship handling. If you can't reproduce our reasoning, we haven't finished the job.
Plenty of ideas don't survive contact with costs and out-of-sample data. Finding that out on paper is the cheapest outcome available to you.
09 — Questions
Scope and cost are agreed in writing before any work starts, and nothing begins until you've approved both. What it comes to depends on the complexity of the rules, how many instruments are involved and how much data preparation is needed — which is why we quote against a specific brief rather than publish a price list. Send the idea and you'll get back a scope, a deadline and a fixed quote, with no obligation to proceed.
Both. Most enquiries come from individual traders with a rule set they want tested properly. We also work with desks, prop firms and fund teams — usually on larger sweeps, on validating a model before capital is committed, or as overflow capacity for research queued behind a small team. The process is identical; scope and reporting depth change.
No. We are a testing and reporting service. You specify the strategy; we test it and report what the data shows. We don't recommend instruments, don't sell signals, don't manage money, and nothing we produce is investment advice.
Yes. Your idea is yours. We don't trade it, don't resell it, don't reuse it for another client, and we'll sign an NDA before you send anything if you want one on file. You're also welcome to describe the shape of the problem first and share the actual rules only once you're comfortable.
That's normal and it's fine. Most ideas arrive as a paragraph and a screenshot. Turning that into a precise, testable specification is part of the work — we'll ask the questions you haven't answered yet and write it up for your sign-off.
Whatever you have: rules in plain English, a Pine/Python script, chart screenshots, or a description of what you watch for. If you'd rather arrive prepared, the strategy spec template is the same document we'd build with you — filling it in gets you a faster, tighter quote. No email required to read it.
Anything tradable with available data — US equity options, index futures, NIFTY F&O, equities, ETFs, FX, commodities, crypto. If clean history for your instrument doesn't exist or is unreliable, we'll tell you that up front rather than quietly producing a result you can't trust.
It depends on the complexity of the rules and the data involved. You get a scope and a firm deadline in writing before any work starts, so you're never waiting on an open-ended promise. Enquiries are answered within 24 hours.
Yes — custom KPIs are part of the service. If you care about a specific metric, exposure breakdown, regime split or risk measure, tell us and it goes in the report.
You get the report anyway, with the reasons. We'd rather hand you an honest negative result than a flattering one — and we'll point at which assumptions, if any, were doing the damage.
10 — Start
Tell us roughly what you trade and what you want answered. We'll come back with what we'd need, what we'd do, and how long it takes — before you share anything proprietary.
Prefer email? taurus.consultancyservice@gmail.com