KILL Context / Nexus

Historical Analog Retrieval Added No Information (NX-A1)

Does retrieving "similar historical market situations" add predictive information over a simple baseline?

Sample

300,000+ evaluations/queries against historical data

Period

2026, preregistered Discovery evaluation

Costs

Not applicable — killed before any product use

Verdict

KILL

"This setup looks like March 2020" is one of the most seductive ideas in trading — and one of the easiest to ship as a beautiful AI feature. Before building it, we tested it: does analog retrieval over our historical data actually add information beyond a baseline that ignores the analogs?

Result: killed in Discovery. Across more than 300,000 evaluations, analog retrieval added no measurable information over the baseline. The pretty "AI Analog Card" would have been decoration.

What this means: the discipline worked — an attractive feature died in evaluation instead of shipping as marketing. Historical context in AlphaDash will only ship when a mechanism survives this exact test.

What this does NOT mean: that historical context is worthless in principle — only that this implementation, on this data, measurably added nothing. The verdict is recorded and any future attempt must beat it.

← All findings · How we measure

This page describes a research verdict under our published methodology. It is not trading advice, and information claims are not profitability claims. Past results do not guarantee future outcomes.