Every trait of a signal, split by outcome — each bar is the average $/trade in that bucket (green = winning trait, red = losing), with win-rate and sample size. Outcome source: broker realized P&L. Small-n buckets are candidates, not rules (n<10 anecdotal · 10–29 early · 30–99 developing · 100+ stronger) — the strand is a hypothesis, not a filter.

The forensic record captured at the instant of every signal — candle pattern, zone confluence, close-location (CLR), the bid/ask spread paid, the execution lag, and the risk flags — split by what each trade actually made. This is the autopsy the bot writes on itself. Outcome source: broker realized P&L + captured instrumentation. Small-n buckets are candidates, not rules (n<10 anecdotal · 10–29 early · 30–99 developing · 100+ stronger).