What's new in PTQ

This week was mostly about keeping things working, plus a few additions to Quant Terminal.

Bot X-Ray runs on Polymarket's new data API

Polymarket switched off its old data API on 2 November. Bot X-Ray read wallets, trades and the "Who's winning now" list from it, so it could no longer look any of them up.

Bot X-Ray now reads Polymarket's new data API for wallet lookups, trade history, P&L history and the monthly leaderboard. It finds usernames through Polymarket's profile search.

To get it, download it again from /polymarket, or paste the prompt from /polymarket/prompt again. The prompt updates your copy in place and keeps your settings and paper trades.

Quant Terminal asks you to approve scripts one at a time

The Quant Terminal install prompt used to end with a tip: add a permanent allow rule for its scripts so a safety check wouldn't stop you next time. That rule lets those scripts run without asking.

The tip is gone. If a safety check asks before a script runs, your Claude now tells you in one line what the script does, so you can read it and approve it once.

Before you run a download, you can check it against the checksums at /kit/quant-terminal-sha256.txt. If you added the old rule, remove it with /permissions in the Quant Terminal folder.

Your locked strategy on the competition page

In a running round, the competition page only showed your return line. The builder below it holds your draft for the next round, so the strategy actually being measured wasn't anywhere on the page.

Standings now has a "Your locked strategy" line with its name, market, timeframe and direction. Open it to see the risk per trade and the full strategy. It's at app.part-timequant.com/competition.

Quant Terminal

• RSI Candle Colours. A new indicator colours each candle by the RSI zone it closed in. There are six zones, cut at 20, 30, 50, 70 and 80, and you can move every cut in its settings. That takes Quant Terminal to 88 indicators.
• Your own indicators can colour candles too, the way barcolor does in Pine Script.
• Paste Pine Script into the indicator editor and it now tells you it's Pine Script and how to get Claude or Codex to turn it into a Quant Terminal indicator. Before, you got a short error about the file format.

Get it at /tradingview.

Fixes

• Quant Terminal: custom indicators written with module.exports now save as they are.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

Most of this week's changes are fixes to things members ran into, plus updates to Quant Terminal.

Daily VWAP and a signal candle stop in the Academy tools

The Academy tools only had a rolling VWAP and a swing stop of two bars or more. A rule like "buy the cross above VWAP, stop under the signal candle" got saved as BLOCKED on Day 3.

Days 3 to 6 now have a VWAP that starts again each day at the time you choose, the same as the VWAP on a chart. They also have a stop at the signal candle's own low or high.

To use them, send the Day 3 prompt dated 27 Oct in the same chat, or ask your assistant to update the Academy tools. Then redo Day 3 with daily VWAP and a signal candle stop.

Bot X-Ray waits out order book blocks

Polymarket's order book server sometimes blocks an internet address for a few minutes. Bot X-Ray gave up on the read straight away, so each block cost you a scan.

Now it waits. It retries after 20 seconds, then 40, 80, 160 and 240, which is about 9 minutes in all. It tells you on screen and in the log that it's waiting, and carries on when the order book answers. If a block lasts longer than that, it stops with a clear message.

Download it again at /polymarket and leave it running.

Competition builder shows the limits before you submit

You could type a stop wider than 20 ATR or a risk above 10% into the competition builder. The entry was only refused when you pressed submit, with an engine error that didn't say what to change.

The builder now sets the same limits the engine runs with and tells you before you submit: a stop above 0 and up to 20 ATR, and risk up to 10%.

The round notice link shows the current round

The "Read the round notice" link on the competition page was still opening October's notice in the run-up to the November round, so it looked like there was nowhere to enter.

The link now opens the December notice, and the older notices are linked from the bottom of it. December entries are open at app.part-timequant.com/competition. They close at midnight Chicago time as 30 November turns into 1 December: 6am UK, 1am New York and 5pm Sydney, all on 1 December. Entries are measured on simulated money, and first place gets $500. The full terms are at /competition-round.

Quant Terminal

• Crypto perpetual futures. Type BTCUSDT.P, ETH.P or any coin with .P to chart its perpetual, live from Binance futures, or from Bybit where Binance futures is blocked. Search a coin and its perpetual sits in the list next to spot.
• Your own OANDA token. Add it to data/feeds.json and forex and metals chart from OANDA first. Anything else your account lists charts as OANDA:XAU_USD or OANDA:NAS100_USD, refreshing every 10 seconds. The prices are OANDA's own rather than CME data.

Get it at /tradingview.

Fixes

• Quant Terminal: an ad blocker or browser shield can stop the chart library loading, which left the chart area empty with no explanation. The app now tells you what's happened and what to do.
• Clicking the unlocked prompt on /polymarket, /tradingview and /bloomberg now copies it.
• Screenshots on /polymarket and /tradingview open full size.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

One new page in PTQ this week, and updates to both terminals.

Say your strategy

There's a new page in the menu called Say your strategy. Press Start and talk it through the way you'd explain it to a friend. For example: "Buy the S&P on the daily chart when it breaks above the 20 day high. Risk half a percent per trade, stop at 2 ATR, no more than 1 trade a day."

As you talk, PTQ reads the words into builder fields: the market, long or short, the chart, the rule, its numbers, risk per trade, the stop and trades per day. The rule can be a breakout, a moving average trend or mean reversion. Next to each field it shows the words it took it from.

Anything you didn't mention gets listed separately with the starting value it was given, so nothing is set without you seeing it. If a number is outside what the builder allows, it tells you what you said and what it set instead. If you say a percent stop, it points out that the builder sets stops in ATR, so you can check it.

Press Open it in Build and it opens as a new strategy for you to check. Nothing is tested and no credits are spent until you say so.

If your strategy has more to it than Build can hold, the page gives you a prompt to paste into Claude with PTQ connected. Claude asks about anything you left out before it saves.

Speech to text uses your browser's own speech recognition, in Chrome, Edge or Safari. The audio goes to the browser's speech service, not to PTQ. In other browsers you can type or paste instead. Your last 20 are kept on your device so you can use one again.

Quant Terminal

• One alert can watch a whole watchlist. When you create an alert, set Check to "Every symbol on" a watchlist and the same conditions run on every symbol in it. Each pop-up and webhook message names the symbol that fired.
• Alerts from your own indicator. If you port a Pine indicator that uses alertcondition, each signal becomes an output that is 1 when it's true, and an alert can watch it. docs/alert-example.js is a working example to copy or show Claude.
• The only official download is part-timequant.com/kit/quant-terminal.zip. To check yours is the real one, run shasum -a 256 quant-terminal.zip and compare it with part-timequant.com/kit/quant-terminal-sha256.txt.

Get it at /tradingview.

Amber Terminal

• GEX. Type SPY GEX (or SPX, QQQ, AAPL) to see dealer gamma exposure at each strike from CBOE's public delayed option data, with net GEX per 1% move, the call wall, the put wall and the gamma flip. GEX 1 shows the nearest expiry only.
• Delayed prices now show the time they're from.

Get it at /bloomberg.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

A big week for the strategy composer. Most of what's here started with a report from you.

Rules can read a longer timeframe

Every rule in the composer used to run on the same timeframe as your entries. It could only compare one indicator with another, and the indicator list was short. So a filter like an hourly EMA 200 on 15-minute entries couldn't be written.

Each indicator in a rule can now have its own longer timeframe. It only reads the last bar of that timeframe that had fully closed, so a backtest never sees a price that hadn't happened yet. Open the longer-timeframe note on a backtest to see which closed bar each value came from.

Rules can also take a transform: change, slope, regression line, rising or falling, rolling high, low, average or spread, z-score, percentile rank, absolute value or clip. And there are 22 more indicators, including ADX and DI, Stochastic and Stochastic RSI, Keltner channels, Chandelier exits, an ATR trailing stop, OBV, relative volume and session VWAP with bands.

Rules can do arithmetic on indicators too (add, subtract, multiply, divide, min, max or % difference), so "close more than 2% above its average" works. For now Claude writes those, or you add them in the strategy's JSON. Ask Claude for something like "ADX above 15 with a 3-bar slope above 1".

Enter once, then wait

If your entry rules stayed true, a composed strategy could enter again on the same bar its stop or target closed the trade. There was no way to say "one trade per setup".

The composer now has a switch: "Enter once, then wait for the entry condition to turn false". With it on, the strategy enters when your rules turn true, then waits for them to be false on a finished bar before it can enter again. It works with your trades-per-day limit, which only counts trades that actually opened.

More choices when your signal is your position

Last week you got "My position" mode for a +1, 0, -1 signal. It always sized trades from your risk per trade with an ATR stop, and closed them at the next bar's open, so a trade that went flat on your last candle stayed open.

There are now three settings in the builder, all off unless you pick them. You can also ask Claude to set them.

• Size each trade: a fixed amount, in whole units (rounded down) or fractional.
• Stop: a percent from entry, for longs and shorts.
• Exits fill at: the next open (as before), the close of the bar where your signal changed, or a fill price column in your upload.

Filling at the signal bar's close closes a trade that went flat on the last bar. Those results are marked as a PTQ simulation. Results on your own fill prices are marked as yours, for reconciliation only. Your risk per trade still caps the size, and PTQ's costs and cost stress test apply in every mode.

See where cost drag comes from

Cost drag used to be one percentage with nothing saying what went into it. Open Cost drag on any backtest now, including runs on your own data, and you'll see the formula, the starting equity it's measured against, the combined rate per side, how it grows with trade size, and what you paid on entries and exits. It also shows that your live slippage limit doesn't change a backtest.

Bring a strategy in from Amber Terminal

The strategy builder has a new panel: Import a strategy from Amber Terminal. Ask Claude for your BTST strategy as JSON and paste it in. The entry and exit rules load as Quant rules. PTQ keeps its own stop, target, size and costs. Anything it can't work out the same way is refused, and the message names it.

If your licence lapses, your exits still run

If your licence lapses, say your membership ends or the PTQ app is offline for more than 3 days and can't renew, the app on your computer now stops opening new trades and lets every exit, stop and close through. A paused strategy can resume in exits-only mode so it can close what it holds. Everything else that needs a licence stays locked until you renew.

This is in version 0.5.2 of the PTQ app. Get it from Integrations, then Get the app. It needs macOS or Linux and Node 20 or newer. Live execution is experimental and has never been tested against a live venue.

Quant Terminal and Amber Terminal

Two tools that run on your own computer. You install each by pasting a prompt into Claude Code, and you update by pasting it again in the same folder.

Quant Terminal is a charting app with TradingView-style charts, 87 indicators, alerts, a screener, bar replay, paper trading and a strategy tester. Every tester result can go through the same checks PTQ uses. With the PTQ app running on the same computer, Check with PTQ sends the rules to your PTQ account, which tests them on its own engine and saves the strategy there. Quant Terminal never places an order or arms a strategy. Get it at /tradingview.

• The quant checks' Monte Carlo now scales each trade by your position size, and the cost stress always tests at least 0.05% slippage per side.
• Search XAUUSD or gold for spot gold, and XAGUSD for silver.
• If Claude Code's auto mode blocks the install, Claude hands you one command at a time and shows you the rule to add with /permissions.

Amber Terminal is a market terminal in the style of Bloomberg's, with BTST for backtests. Get it at /bloomberg.

• Every BTST run through Claude saves a dated copy of the bars it tested on, with a SHA-256 hash. Ask Claude to rerun with replay and the snapshot id to get the result on exactly the same bars.
• BTST can enter on a limit at a level you choose, like an older high or low or a 20-bar channel. It fills only if price gets there within the bars you allow.
• QQQ DES and QQQ CHART work for ETFs. PORT ADD adds to a holding (PORT SET replaces it), PORT holds US options, and OMON takes a date or a month, like OMON AAPL JAN27.
• Its PTQ connector checks it's really talking to PTQ before it sends anything.

Fixes

• A backtest on one market ignored your trades-per-day limit, so a limit of 1 could show several trades on one day. It now stops opening trades once the limit is hit and starts again the next day (UTC). Each run records the limit it used and how many days it was reached.
• Saving a composed strategy on your own uploaded data was refused with "data not attached", even when the data was fine. It now saves, as long as the strategy's timeframe matches the dataset's.
• Adding your own price history through Claude or Codex failed with "outcome may be unknown" once the CSV passed about 1 MB. Nothing was saved and no credits were taken. Calls now take up to 16 MB and 100,000 rows, and one over the limit says so. If you use the downloadable connector, run the setup command on the Connect page again.
• Get the app on Integrations led to a dead end, and Trade for real on Connect was refused. Both now download the PTQ app and check it against its SHA-256 before saving it.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

A smaller week. Every change here started with a report from you.

Your signal can be your position

If you uploaded a TradingView strategy's signal (+1 long, 0 flat, -1 short) and tested it as your own indicator, PTQ read it as a score and picked its own levels to act on. So a 0 didn't close the trade. Two longs with a flat gap between them came out as one trade, and the trade count didn't match TradingView.

Strategies on your own indicator now have a signal mode. In the builder, next to the rule type, it asks "Your signal column is" with two options:

• A score. This works exactly as before and stays the default. PTQ picks the levels and tests 7 sets of them.
• My position. PTQ trades the column as given: +1 long, 0 flat, -1 short, held until it changes. On a long-only strategy, -1 means flat.

You can also ask Claude to set it. In "My position" the column must hold only -1, 0 or 1. Any other value is refused, never rounded, and a missing value holds the position you're in. PTQ's stop and target still apply, so a stop can close a trade before your signal does.

One thing to know before you validate. PTQ didn't choose anything in this mode, so the overfitting check has nothing to compare. It says "not applicable" with the reason and counts as not passed. The luck check counts 1 attempt, so any searching you did before uploading isn't in that number.

"Both" stays "Both", and you're not charged twice to validate

A strategy Claude made to trade long and short turned into long-only when you opened it in the web app. Pressing Test there saved that change and charged 11 credits to backtest and validate a strategy you'd already validated.

"Both" is now kept, and the builder has a Both button next to Long and Short.

Validating the same backtest again within 24 hours now returns the verdict you already have and charges nothing. A second click while a validation is still running is turned away without a charge. If you do want a fresh validation of the same backtest, ask Claude to re-run it and it charges as normal.

Credits back when a run fails on our side

If a backtest or validation fails because something broke at our end, the credits it took go straight back to your balance, and the error says so. If they couldn't be returned, the message tells you to contact support. A run we turn down because of its settings still costs what it cost.

SMT divergence in the Academy course

The course's tester used to read one market at a time, so a rule comparing two markets couldn't be written. SMT divergence between EURUSD and GBPUSD is the common one: one pair takes out its swing low or high and the other doesn't.

Day 3 can now name a second market. Day 4 loads it from the same price source, or from its own CSV, and lines the two up bar by bar. If the second market has no data or its bars don't line up, it stops and says why. There's a new SMT divergence building block, bullish on lows and bearish on highs, with a lookback and a number of bars it stays valid for. It works alongside the session window, fair value gaps, swing stops and R targets. Day 5 and the Day 6 paper loop use the same numbers.

Download the new "(5 Oct)" prompt and Academy tools from the Day 1, 3, 4 and 5 lessons. Re-send the Day 3 prompt in the same chat and add the second market to your spec.

Fixes

• Validating a backtest that ran on your own uploaded data crashed with "'str' object has no attribute 'get'", and the 10 credits were still taken. It now finishes and gives you a verdict. Re-run validation on any backtest built on your own data.
• In the October competition, daily strategies lost their first trading day. The 1 October bar was stamped at midnight, before the window opened at 05:00 UTC, so it became the starting point and 7 of 22 entries showed as not measured. Each bar now counts from its close, so daily entries are measured from day one.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

A big week in the app. Most of it started as a report or a question from you.

Describe your strategy and PTQ writes the rules

Until now you had to turn your idea into the builder's rules yourself. If you didn't know which indicator or comparison to pick, the strategy didn't go in.

Now there's a box at the top of the builder. Write the idea the way you'd say it and press Build it. PTQ asks about anything missing, with options you can tap, then proposes the rules. The engine checks them before you see them. If part of your idea can't be tested yet, PTQ tells you and leaves that part out. It never swaps in something close.

You can change anything it proposes before you save. Backtests and validation work exactly as before. It's in the competition builder too.

Search for ideas

PTQ used to test one idea at a time. The new Search for ideas page tests hundreds or thousands in one go.

You set the markets, the kinds of idea, the lookbacks, the stop sizes and the pass rules before anything runs. PTQ holds back the most recent years and scores every idea on the rest. It then marks the results down for how many ideas were tried, and gives only the finalists one test on the held-back years.

Most ideas fail. Most searches end with nothing at all, and the page says so. It also shows the ones that looked good and didn't hold up, so you can see what luck looks like. Anything that survives opens in the builder as a draft for you to test properly.

You see the size and the cost before you run it: 1 credit per 100 backtests. For now it searches daily bars only.

The builder does a lot more

Twelve more indicators: EMA, RSI, ATR, rate of change, standard deviation, volume and average volume, the Bollinger upper and lower bands, and the MACD line, signal and histogram. Each has an input for its settings.

• Read any value from a number of bars ago.
• Put a group of either/or conditions inside a rule. "RSI crosses above 30 and the close is above yesterday's high, or the MACD histogram turns positive" can now be written.
• Every indicator is worked out only from bars that had already closed. A setting the engine can't use is refused with the reason.

In Quant mode you can also limit entries to a time window in New York time, with an optional time to be out by. US stocks can be tested on pre-market and after-hours bars, 04:00 to 20:00.

And a rule can run on a watchlist of up to 40 tickers. PTQ ranks them on every bar by % up since yesterday's close or by volume, applies your price band and minimum volume, and only enters the top few. That's how you test something like "the top gainers priced $5 to $10 between 07:00 and 09:00".

A few limits, stated plainly:

• The finest bar is 1 minute. PTQ doesn't test on ticks.
• PTQ only ranks the list you give it. A stock that wasn't on your list can't be picked.
• Watchlist and pre-market strategies are for testing only, not practice or live trading, because that data is at least 15 minutes old.
• Each ticker on the list costs 1 credit per backtest.

Dark mode, and easier to read

PTQ now follows your device's appearance setting. Switch your computer or phone to dark and the app and the site go dark with it. Gains, losses and warnings were recoloured so they stay readable. There's no separate switch inside PTQ.

We went through every screen against Apple's design guidelines:

• Text is easier to read and grows when you raise your browser's text size.
• Every risk slider takes an exact typed number.
• Steps that matter, like allowing trading again after a stop or locking a competition entry, ask you to confirm and say what will happen.
• PTQ follows your device's settings for higher contrast, less transparency and less motion.
• Screen readers can read the charts and checks, and each chart has a "Show the numbers" table under it.

The see-through glass is now only on the bar at the top. Cards and panels are solid, so text always sits on a plain background. We also cut about one word in six across the app. Nothing you need to act on went, and every risk warning and simulated label is still there.

Take a strategy to MultiCharts

If you trade in MultiCharts, you can now export a saved strategy as a PowerLanguage signal. Open it on the Research page and expand Export to MultiCharts, or ask Claude. The top of the file lists the settings MultiCharts needs and every place it will behave differently from PTQ.

If a strategy uses something the export can't copy exactly, like a percent stop or your own indicator, you get a plain message saying which rule and why. It never changes a rule to make it fit.

Run it in MultiCharts, save the List of Trades, and bring it back. PTQ puts those trades next to its own backtest so you can check they match. Exports and imports use no credits.

The Academy course

If you're working through the course in the Academy:

• Days 3 to 6 now handle yesterday's high, low and midpoint built from intraday bars, a trading-hours window, stops at the last swing low or high, targets as a multiple of the risk (2R, 3R), and 1-minute data from your own CSV.
• Forex costs now start at 1 basis point a side. The old 6 basis point floor was about 7 pips a side on EURUSD, which made forex strategies look worse than they are. Everything else stays at 6.
• Gold and silver data now comes from the right feed, so Day 3 no longer comes back blocked on XAUUSD. 5-minute gold goes back about 8 months on a basic Tiingo key, so use 1-hour bars for a two-year test. The feed has no volume.
• Day 3 now works out how much of the account one stopped-out trade loses. Above 2% it warns you and shows what a losing run does to the account. Above 10% it won't go on unless you confirm. It also warns when one trade holds the whole account, because a gap can jump past the stop, and when your stop is far wider or tighter than one bar of your market normally moves.
• The Day 6 paper drill now sizes in fractions, so a 0.02 BTC order fills instead of rounding to nothing. The same risk rules apply there.

Download the newest prompt and tools from the Day 1, 3, 4 and 5 lessons, re-send the prompt in the same chat, and re-run Day 3.

The Jev trading loop

The Jev prompt builds a better loop now:

• It asks Jev for a judgment at most once every 4 seconds by default, reuses the last one in between, and backs off on rate-limit errors. Through the Vercel AI Gateway the old half-second loop kept failing with "access frequency is too high". Re-run the prompt to get it.
• Start and Stop buttons on the dashboard. Stop pauses new orders and keeps the data updating. Kill still closes positions.
• Every install and run step has a Windows form, and it works out which system and shell you're on.
• Ask for a market it can't trade, like forex or MT5, and it stops before building anything and offers the nearest one it can.
• You can set the bar size, from 1 minute to 1 hour.

The Academy price

The Part-Time Quant Academy is $29 a month from 1 October. If you joined at $9, you keep $9. If you saw $9 somewhere and were charged $29, reply to any email from us and we'll sort it.

Fixes

• New backtests and validations had been labelled "Historical result" on the Research page since 21 September, which also stopped strategies earning practice access. Fixed on 1 October. Results from before 21 September stay historical, because crypto risk-adjusted figures now use a 365-day year. Re-run the backtest and validation to bring one up to date.
• On the competition entry page, the pencil next to the strategy name now lets you rename it. If a submit fails, the confirm box stays open and shows the reason next to the button. If you've already entered, the page opens on "You're in".
• Academy members who opened the PTQ key screen got no code and no reason. It now says straight away that the Academy plan has no software key, and links to the competition page, where your access is.
• Seven bugs in the Jev loop are fixed. The position cap now counts buys still waiting to fill, resting orders are cancelled on every exit, old inventory is closed instead of freezing the loop, and fees, rounding and timeouts are handled properly.

If something here doesn't work the way it says, or there's something you want PTQ to do next, tell us.

A quieter week in the app and a busy one on the competition. Here's what changed.

Setting risk limits from Claude no longer breaks the home page

If you asked Claude to set your risk limits before you'd finished setup in the app, the home page could crash when you opened it. Now the app fills in anything the chat didn't set with PTQ's defaults as your profile loads. Set your limits from Claude whenever you like and the app opens normally. Limits set from the chat don't count as finishing setup, so you can still go through it in the app.

Every screen is now tested against the half-filled profile the chat leaves behind, so this one stays fixed.

The October competition is open

Once a month you can enter one strategy. It locks when you submit it, then it's tracked forward on simulated money for the whole month. No real trades are placed. The ranking is the month's return divided by its biggest drop, so taking more risk doesn't move you up on its own.

October's entries close on 1 October at 00:00 US Central time, which is 06:00 UK, 01:00 New York and 15:00 Sydney. Dates, markets, how entries are simulated and the $500 prize for first place are all in the round notice, and the full rules are here.

Who can enter:

• Part-Time Quant Academy members, through the invite link in the Academy. It makes you an account that's only for the competition.
• Anyone with a paid PTQ subscription, from the same login.

A public leaderboard

The leaderboard is open to anyone, no sign-in. It shows positions and names only. You see your own strategy's simulated results in your competition account, and nobody sees anyone else's numbers. It's also linked from the footer of every page.

Every entry is measured on every bar of the month, so the standings can update through the day.

Entering is clearer

• The builder's button now says "Next: submit your entry".
• Only the round currently taking entries is offered, and it's already selected.
• Before you submit, the confirm screen tells you what locking means. Afterwards you get a link to the leaderboard.
• The competition page has the Report a problem button too.

If you report something

Reports go into the same queue as before. If one needs more detail before we can fix it, you may get an email from PTQ Support. That's an AI assistant working for Lewis. It asks what went wrong and what fixed would look like, and it sends four emails at most. Anything about billing, refunds, your account or security goes straight to Lewis.

Fixes made from your reports now go in this changelog each week, without names.

Fixes

• Academy members could see "membership verification is temporarily unavailable" on the competition page. The member list now updates every hour. Fixed.
• Buying Premium in the Academy now gets you a PTQ key within the hour, without anyone having to step in.
• The Academy tools could fetch crypto and forex prices from the wrong source on Day 3. The Day 1 prompt now installs the fixed tools.
• The Jev trading loop prompt now builds its state from real price history, retries a locked file on Windows, and says clearly that on Alpaca it runs long or flat. A sell closes the position. It can't open a short. It still trades paper by default.

If something in the competition doesn't make sense, or there's something you want PTQ to do next, tell us.

A lot went into PTQ this week, most of it straight from questions on the live training. Here's all of it.

Connect wherever you trade

There's now a list of 101 brokers, exchanges, data sources and tools that PTQ connects to, each with its logo. Search for yours or scroll the whole thing on the integrations page.

Nine are built in: Alpaca, Interactive Brokers, Binance, Coinbase, Kraken, OKX, Bybit, KuCoin and Bitget. You connect them in the PTQ app on your computer and that's it.

Yours isn't on the list? If it has an API, it still works. PTQ signs each order and sends it to a small endpoint you run, and your endpoint places it with your broker. Claude will help you set that up.

If PTQ doesn't have a live price for what you trade, a prop firm's futures for example, the broker you're connected to supplies the price for that order.

Live trading, from your own machine

Live trading runs through the PTQ app on your computer. Your keys stay on it. Nothing goes out until you've switched live trading on and armed that strategy yourself, and every order still has to pass your risk limits. Once a strategy is armed it can keep trading while you're away, inside those limits.

This is new, so start small. PTQ's orders have been tested on Alpaca's paper account and haven't been tested against a live broker yet.

Test on your own data

If PTQ doesn't carry the prices you want, bring them. Prop firm data, a futures feed, your broker's own history, anything with an API. Ask Claude to fetch it or hand it a CSV, and it's saved to your account as its own market. Anywhere from 60 rows to 100,000, and saving it uses no credits.

From there every test runs on it exactly as it does on PTQ's data. Same backtest, same five checks.

Test your own indicator

RSI from your charting platform or a score from your own model. Any number per bar works. Put it in the upload as a column called signal and PTQ works out where to act on it. It tests seven sets of levels so the overfitting and luck checks are real, and it goes through the same quant tests as everything else.

One thing to watch. PTQ can't see how your indicator was built, so if it uses future prices, every test will look better than real life.

Claude talks to you properly

• The first thing it asks is how deep you want to go: keep it simple, in between, or full quant.
• Shorter answers, and the numbers go in tables.
• Every reply ends with numbered options. You pick a number instead of staring at an empty box.
• Any term you don't know links to its page in the glossary.
• After every validation it scores the strategy out of 10 and leads with the score. When an idea has run its course it says so, tells you what it taught you, and offers a new one.
• You can set your risk limits from the chat, like how much one trade is allowed to lose.

The app keeps up with Claude

Make or test something in Claude and the page follows along without a refresh. Run a validation and it opens on the verdict.

Results are drawn now. A backtest opens with the answer in figures before any detail, and so does Performance. Your strategies show as bars you can click, and a research loop shows its progress as a bar.

Report a problem from any screen

Bottom right of every screen there's a report button. Paste a screenshot in with it so we can see exactly what you saw.

Connecting your AI is quicker

Connecting is one step at a time on screen. The check at the end takes two calls and no longer relies on a market being open, which is what tripped us up on the live training. Works with Claude, Codex or any MCP client.

Fixes

• A strategy made in Claude could open on a blank page. Fixed.
• A dip-buy strategy could turn the whole app white. Fixed, and every screen is now tested against the strategies Claude actually writes.
• Deleted strategies stay deleted, and their backtests go with them.
• The quote tool no longer suggests symbols it then refuses.

If the broker you use isn't on the list, or there's something you want PTQ to do next, tell us.