How Spoken Alpha scores executive language

What we score

For every executive on every call, we build a longitudinal baseline from their prior quarterly calls, then score how the new call departs from it. The output is not a sentiment score and not a guidance number — it is a measure of how a speaker's language has shifted relative to how that speaker has talked before. We surface those departures along a handful of observable signal families:

Three classes of bearish call

A call can be bearish in more than one way, and each way confirms on a different clock. We separate them deliberately, because validating each one requires a different kind of ground truth. Here is one worked example of each, drawn from public earnings calls and SEC filings.

What we validate against

Each class has its own ground truth, so we don't grade them on the same yardstick:

• Clean-miss— the next quarter's reported actuals versus what the Street expected.
• Guidance-cut— the change in a company's own forward guidance, read from its SEC Item 2.02 earnings releases.
• Pre-announce-trailing — a later Item 8.01 pre-announcement filing that confirms what the call language flagged.

Treating these as one undifferentiated “did the stock go down” test would be dishonest — a call can flag real stress and still trade up on a quarter for reasons that have nothing to do with the language. Matching each class to the catalyst it actually predicts is the only way to measure the signal cleanly.

What we don't claim