canli-mcp / The desk, built into canli-mcp

Every common way a backtest misleads, checked in one call

Read a backtest's daily returns for every common way a backtest misleads, each with evidence and what would clear it: the search behind it (deflated Sharpe, overfitting), factor exposure sold as skill (alpha after the Fama-French five factors and momentum), lookahead in the weights, smoothing and selling tail risk, a few days carrying the result, an edge that fades, numbers too good for real trading, and, with the Lab built, near-copies and the luck bar among a million strategies over the same months. Flags are reasons to look, not verdicts.

The tool is backtest_forensics in canli-mcp 0.22.0, free and MIT-licensed, with no account and no API key. It is part of canli-mcp itself, so it is there as soon as canli-mcp is added.

Arguments

As the server declares them
ArgumentRequiredType
returnsyesnumber[]
datesyesstring[]
trials_triednointeger
variantsnonumber[][]
weightsnonumber[]
pricesnonumber[]
factorsnoboolean

How a model reaches it

Ask in plain words. The model finds backtest_forensics with canli-mcp's find_tool, reads its arguments with describe_tool if it needs to, and runs it with run_tool. The call has this shape, with your values in place of the angle brackets:

{
  "name": "backtest_forensics",
  "arguments": {
    "returns": "<number[]>",
    "dates": "<string[]>"
  }
}

Words that find it

Besides its name, title and description, find_tool matches a question to backtest_forensics on these words: backtest, lie, detector, forensics, audit, too, good, to, be, true, overfitting, lookahead, factor, exposure, alpha, smoothing, tail, risk, concentration, decay, red, flags, due, diligence.

More

Nothing here is investment advice.