MethodologyKilled

An AI that reads price charts like a human

Stopped early

nothing in the markets we trade for it to learn from, so we stopped before spending

In plain English

We wanted an AI to read price charts like a human, but the only trader it could have learned from traded coins we don’t touch, so we stopped before spending on it.

What was claimed

An AI reading charts the way a human trader does will spot edges that code misses.

The bar it had to clear

a visual-discretionary edge inside our asset universe

What the data said

the only candidate edge lives on alts we don't trade

Why it's dead

The vision-on-candlestick methodology needed a trader whose edge is both visual/discretionary and inside our universe (BTC/ETH/HYPE/HIP-3). The only attribution available (the only trader whose documented edge the model could have learned from) lived on alts (smaller coins outside that universe), structurally outside scope, so it was killed before spending engineering on calibration.

The detail

Killed pre-data (operator decision). The pilot render shipped, but the four-way ablation (the planned test that strips out components one at a time to see which, if any, carries the edge) never ran: the premise required a signal source mapping onto our asset universe and none existed. Killing pre-data preserved the workstream slot for a class with a viable signal source.

Kill date: 2026-05-01
Sample: pre-data
Method: Documented kill
Verdict: no signal source