The prompt library
Every prompt the lab has published: what each one builds, what it costs to run, and the lab's own result it ships with.
Every prompt the lab has published, in one place. Each one builds a tool with Claude, ChatGPT or Cursor: a tester, a paper bot, a fill report, a dashboard. None of them is a trading signal, and none of them makes money on its own. What each one gives you is the honest version of a check the lab paid real money to learn, with our own result for that strategy printed next to it so you know what you are testing against.
How to use one. Open the post, copy the prompt into your assistant, and run what it builds on Kalshi's free public data. No account is needed until the course's last lesson, and real money is never required. Every prompt carries the same three rules the lab uses: count a paper fill only when a real trade printed through your price, split any result into halves before you believe it, and distrust a result that comes from one lucky day.
Free with a Pondletter account
| Prompt | What it builds | The lab's own result it ships with |
|---|---|---|
| Test Kalshi's favourite strategy yourself | An honest backtester for the "buy the favourite" rule on Kalshi's free history, at prices you could really have traded | 53,317 paper trades, 91% won, about 1.2¢ a trade short after the fee |
For members
| Prompt | What it builds | The lab's own result it ships with |
|---|---|---|
| Kalshi longshot strategy: both sides, honestly | One tester that compares the longshot at the ask, the favourite at the ask, and a resting bid proven by the trade tape, on the same markets | 38,565 longshot paper trades, 2.9% hit rate, about 1.8¢ a trade short on all twelve markets |
| Kalshi perps: a liquidation and funding dashboard | A small dashboard that shows your liquidation price, funding cost and break-even move before you click Up or Down, from free public data | The 7% rule as it really pays, and the trap we hit on our own hedge |
| Course lesson 1: read a market correctly | A market reader: price, book, fee by series, close time, and what the market actually settles on | The three things most first bots get wrong, from our own early mistakes |
| Course lesson 2: test a rule on free history | A downloader and tester for months of settled markets in any series, with a one-minute picture of the bid and ask | Reproduces the lab's paper bot to within a tenth of a cent |
| Course lesson 3: a paper bot that records every decision | A paper bot that rests orders the way a real one would, counts a fill only on a through-price print, and logs every decision including the ones not to trade | Half of the lab's 10,969 real orders never filled; median fill 1.7 seconds |
| Course lesson 4: honest fills | A fill report for your paper bot or for Ping that measures the touch-counting lie in cents per contract | 32% of our paper fills were touch-only; the paper twin ran 4.8¢ a contract ahead of the real bot |
| Course lesson 5: grade a result | A grader that reads a pass mark you wrote before the first trade and prints sample, halves, best-day share, a day-resampled confidence interval, fill rate, and PASS or FAIL | Block 1's one day at 118% of profit; block 2's split halves; the weather simulation's +1¢ that was a zero |
| Course lesson 6: small blocks with written stops | A wrapper around your paper bot that places real orders only inside a written block: arming file, stops from the exchange's own fills, reconciliation by client id, a halt when blind | The live bot's block files, its four halts this autumn, and the NO-buy-is-"sell" bug that froze it for 14 hours |
| Advanced lesson 1: what the last minute is really worth | A settlement watcher for the 15 minute Bitcoin markets, which settle on an average, not the price on your screen | 635 real settlements reproduced exactly; our live last-minute record alongside |
What is coming
The six core lessons are complete. The advanced track continues (speed, fees in depth, testing without fooling yourself, a stop-loss live case, risk controls, reading the book), one lesson at a time, each with its prompt. New prompts follow new studies: when the lab publishes a tested strategy, the prompt that lets you test it yourself comes with it.
The rules every prompt follows
Prompts teach methods: reading the book, proving a fill from the tape, counting the fee. They never contain the lab's own order placement, size or caps for the market-making seats, which is the one family of strategy where copying spoils the edge for everyone. Members' prompts and the course are included in the $15 membership. Nothing here is advice.