Part-Time Quant (PTQ)

What it is

PTQ is a hosted workbench for one job: write a trading rule without code, backtest it, and find out whether the result was luck. You set entries, exits, position size, stop and risk per trade in a form-based builder, start from a template, type the idea in plain English, or — since 26 October 2026 — say it aloud and let the browser's speech recognition fill the fields. The backtest runs on PTQ's servers, charts the trades on real candlesticks, and then a validation gate scores it.

The gate is the product. Every backtest goes through five checks: walk-forward, the probability of backtest overfitting (Bailey, Borwein, López de Prado and Zhu), the deflated Sharpe ratio (Bailey and López de Prado), transaction costs charged inside the run, and maximum drawdown with open positions marked at every bar's close. Part of the history is held back while you build and scored last. An optional library adds purged cross-validation, Sharpe error bars, a stationarity test, Monte Carlo drawdown with a ruin bound, and a Kelly sizer that refuses to give a number when the data shows no edge. The vendor's FAQ says most strategies fail, and that this is the point. Nothing is sorted by return.

A strategy that passes can be traded with simulated money inside PTQ, against per-trade, daily loss, drawdown, exposure and single-instrument limits you set once. Everything is also reachable from Claude through an MCP connector, which can build, edit, backtest, validate and place practice orders on your account.

It is sold by Lewis Jackson Ventures Ltd, a company registered in England and Wales, to UK retail traders first; availability elsewhere is decided at checkout by jurisdiction. The firm states it is not FCA-authorised. The site says PTQ is in early access, its changelog starts on 23 September 2026, and it carries no testimonials because, in its own words, it has no real ones yet.

Pricing

One plan: $997 a year, billed once through Stripe in US dollars and renewing automatically. There is no monthly option, no cheaper tier and no trial; the pricing page states that nothing — no licence key, no credits, no access — is issued before payment.

The plan includes a credit allowance that backtests, validation and support chat draw on. The changelog shows the unit — a backtest and validation together cost 11 credits, a validation 10 — but no public page says how many credits a year buys. Credits come back when a run fails on the vendor's side, and re-validating the same backtest within 24 hours is free.

The refund policy offers the UK statutory 14-day cancellation, charged pro rata for the service already supplied, and states that beyond the law there are no refunds.

Data & coverage

The equities are American. The October 2026 competition rules list what the builder offers as "US stocks and ETFs, FX, futures and crypto spot", the changelog adds US pre-market and after-hours bars from 04:00 to 20:00 New York time, and no page names a non-US stock exchange. Coverage "varies by market, interval and date", and a missing range stops the run rather than being filled. History is fetched from configured providers and cached; there is no preloaded universal dataset. The two built-in sources on the integrations page are Yahoo Finance (stocks, forex and futures history) and Massive (stocks, forex and crypto); the competition names Yahoo for stocks, forex and futures and Bitget's public data for crypto. The product page says practice orders fill "at real-time prices" without naming the feed, and no public page states how far back a market goes or at which intervals.

You can bring your own prices as CSV, which the changelog has extended to 16 MB and 100,000 rows through Claude or Codex. A signal column of +1, 0 and −1 can be tested as a position held until it changes, with PTQ's stop, target and costs still applied — useful for re-checking a TradingView strategy under someone else's cost model. In that mode the overfitting check reports "not applicable", because PTQ chose nothing to overfit.

Rules can read a longer timeframe than the entries, using only the last fully closed bar; the composer has had 22 further indicators since 19 October 2026, including ADX, Stochastic RSI, Keltner channels and session VWAP.

Integrations

The Claude connector works from Claude Desktop, claude.ai and Claude Code; Codex and Cursor are listed as MCP clients too. Requests go to PTQ's servers and the connector does not install the engine locally. Strategies export to TradingView as a Pine v6 study that plots signals and never places an order.

The integrations page lists 101 brokers, exchanges and data sources, and says itself that a logo means only that PTQ can connect, nothing more. Nine are "built in" (Alpaca, Interactive Brokers and seven crypto spot exchanges) and are reached from a separate PTQ desktop app on macOS or Linux, never from the hosted product. Most of the rest are "your endpoint": PTQ sends signed orders to a bridge you run yourself, with TradersPost, n8n or Make as the suggested hosts. Data vendors such as Norgate, Databento and Tiingo are "bring data in": you test on prices you supply from them.

The same vendor publishes two local tools that can send rules to PTQ — a charting terminal and a Bloomberg-style terminal — and runs the Part-Time Quant Academy, a $29-a-month community on Skool. They are separate products. A monthly strategy competition on simulated money, with a $500 first prize paid by the vendor, is open to Academy members and PTQ subscribers.

Limitations

• No live trading in what you pay for. Hosted PTQ, the browser and the Claude connector place simulated orders only. The desktop app's live path is described by the vendor as experimental, tested on Alpaca's paper account, and — in the FAQ's words — no order has yet been sent to a live venue. Treat it as absent.
• No way to look before paying $997. No trial, no free tier, and no demo video; the site offers screenshots it states are from the real app.
• Early access, and young. The terms are version 1.0 from 27 August 2026, the changelog begins a month later, and the legal documents are, by the vendor's own account, still with a solicitor for review.
• Data depth is undocumented, and stocks are US only. No history depth or interval list is published; you find out per test. No page says whether the built-in stock history keeps delisted companies; the integrations page sends you to Norgate Data for that.
• Credits are metered against an allowance nobody publishes, so the cost of heavy use cannot be planned from the site.
• Rigour does not stop at the gate. The deflated Sharpe ratio counts versions PTQ saw; any searching you did elsewhere before uploading a signal is not in that number, which the vendor points out itself.
• The engine is proprietary and hosted. A strategy lives in PTQ and its JSON; the only portable output is a Pine study that plots signals.

Alternatives

Build Alpha, on Windows, and StrategyQuant X, on Windows, macOS or Linux, run their own robustness tests on your machine and export the result as platform code, for a one-off licence rather than a subscription — but they search for rules rather than test the one you wrote. Composer is the hosted builder that actually trades the rule, through its own US brokerage, and is free to backtest; it tests on daily closes and lets you mark an out-of-sample date, which is one of PTQ's five checks rather than all of them. QuantConnect if you can write Python and want the backtest and the live deployment in one place.