Paper Trading Lab
Three trading experiments, run on paper and graded honestly
- Me. It's a lab, not a product.
- 16 July 2026
- None. Paper only.
- None of this is.
Three strategies, each written down before it ran, each traded on paper only, and each graded against what actually happened. No real money, and nothing here is advice. The journal is the product.
- Divergence Lab tests one well-known chart pattern properly: every historical instance scored on what price did next, with recent data sealed away in code so the final verdict can’t be tuned to fit. It’s now running forward on paper.
- Scuttlebutt reads where public attention is moving (search demand, then what people are actually saying) and paper-trades the extremes. Sentiment tools have no history, so it can’t be backtested; every trade is out of sample by construction.
- Quarterline reads what a company’s customers are doing mid-quarter (app installs and ratings, live ads, search demand), calls the report beat or miss with a confidence score, and expresses each call as a dated options structure. Stand-asides are graded too.
Each one records its own mistakes as carefully as its trades, including the audits that found my own checks couldn’t fail.
Stack
- Python
- Every experiment is plain Python. The scans, the journals and the grading.
- Claude
- Scores the evidence in Scuttlebutt and Quarterline. Divergence Lab uses no model at all.
- Yahoo Finance
- Free price history for Divergence Lab and Scuttlebutt.
- DataForSEO
- Search demand for Scuttlebutt and Quarterline, and app store data for Quarterline.
- ScrapeCreators
- The ads a company is running right now, for Quarterline.
- marketdata.app
- Options prices for Quarterline's paper fills.
Build log
-
Quarterline: Paused. The options data quota ran out. Nothing was open at the time.
-
Quarterline: An audit found every paper fill had been priced before the market opened. The check meant to catch that couldn't fail.
-
Scuttlebutt: Most prices hadn't updated when the weekly scan ran, so it refused to read them. Right call, lost week.
-
Divergence Lab: I planted false claims in a quiz about my own project, then ticked every one of them as right. So I can't be the check on what the machine says.
-
Scuttlebutt: The first paper trade, taken by a scan nobody was watching. A review the same day said not to count it as evidence yet, and found the replay tool said the trade never happened.
-
Quarterline: Started. The rules were written down and signed off the same day.
-
Scuttlebutt: Started. Sentiment tools have no history, so it can't be backtested. The paper journal is the only test.
-
Divergence Lab: An audit went hunting for ways the backtest could see the future. It found no leak. The gaps it did find were fixed or written up for a decision.
-
Divergence Lab: Started. The recent data was sealed away in code before anything was tested.