Detect BS · Last updated 12 August 2026
This document says plainly what this software is, what it is not, and what it cannot tell you. It is written to be read rather than to be skipped, because the limits below are the point of the product rather than an afterthought to it.
Detect BS does not provide financial, investment, trading, medical, legal, engineering or any other professional advice. It produces no recommendations. Nothing it outputs should be treated as guidance to buy, sell, hold, diagnose, prescribe, build or decide anything.
The author is not a licensed financial adviser, broker, investment professional or physician, and nothing supplied with this software constitutes a professional relationship of any kind.
A number produced by a statistical model fitted to data you supplied. It is an estimate. It is frequently wrong.
A neural network fits noise as readily as it fits a real pattern, and it cannot tell the difference. That is not a flaw in this implementation — it is true of every neural network. The verification layer exists precisely because of it.
It can show you that a model performs worse than a naive baseline, worse than guessing the mean, or no better than the same model trained on deliberately shuffled targets. When it shows you that, believe it.
It cannot establish that a model will work on data it has not seen. No method can. Walk-forward scoring, negative controls, permutation importance and agreement between engines together make it much harder to fool yourself. They do not make it impossible, and they say nothing about whether the relationship you found will still hold tomorrow.
A model that passes every check may still fail in use. Conditions change, relationships break down, and past behaviour constrains future behaviour far less than people expect.
If the engine beats a naive baseline it means the model carries information the baseline does not. It does not mean a strategy built on it would be profitable. Transaction costs, spreads, slippage, liquidity, taxes, timing and execution are not modelled anywhere in this software and are not in any number it reports.
Results depend entirely on the data you give it. Data that is incomplete, mislabelled, revised after the fact, survivorship-biased, or that leaks information about the future into the past, will produce results that look valid and are not. The software cannot detect most of these problems and does not claim to.
The optional market and economic feeds retrieve data from Yahoo Finance and the St. Louis Federal Reserve. Their accuracy, availability, timeliness and terms of use are theirs, not the author's. Data may be delayed, revised or wrong. Verify anything that matters against a source you trust.
Any decision you make using this software — financial or otherwise — is your own decision and your own responsibility. The author accepts no liability for the outcome of any such decision. If a decision is significant, consult a qualified professional who is licensed to advise on it.
The software is provided "as is", with no warranty. Liability is limited as set out in LICENSE.txt, which governs where these documents overlap.
micahforstein@gmail.com