Can a chart fingerprint forecast the next move?

The verdict

Recognition worked. The original hourly forecast did not clear the comparison or cost tests.

What we tried

Shazam recognizes songs from short fingerprints. We tried the corresponding idea on price charts: retrieve a historical pattern, then inspect its continuation.

What happened

In the original hourly experiment, matching recognized a coin’s own history 99% of the time. Forecast ranking correlation was only 0.005, with a t-statistic of 1.9. It lost to plain nearest-neighbour matching and 24-hour reversal, and did not clear costs.

What this means

Recognition and prediction are different tasks. Later causal fixes have not yet been followed by a complete corrected hourly rerun. The quoted numbers belong to the original experiment, not a newly validated version. Chart matching remains useful as a descriptive research tool.

Evidence record

Published figures are from the original experiment summaries and the approved research handoff, dated October 7, 2026. They describe the implementation tested, not every version of the underlying idea.

Keep exploring