Intel pipeline · audited contributions

Per-source quality receipts

Every RSS feed, Twitter handle, Telegram channel, financial indicator, and whale-tracker source the bot ingests gets a measured expectancy delta on real trade outcomes. With-vs-without paired-trade win rates, sample sizes, and a verdict — computed nightly by the source-ablation cron. No other AI prediction product publishes per-source attribution.

Loading…
Status
Methodology validated, sample sizes still accumulating. At the current paired-trade count, no individual source has crossed the Wilson-CI statistical-significance threshold for a likely real verdict yet. The list below shows directional contributions and is recomputed nightly — bookmark this page; it gets stronger with every settled trade. (For comparison: most "AI trading" projects publish zero per-source attribution at any sample size.)

How to read this

Effect (pp): change in win-rate (percentage points) when the source IS present versus when it ISN'T. Expectancy Δ (USD): change in average P&L per trade. Verdict: likely real means the Wilson-CI bands don't overlap (statistically meaningful effect); no clear effect means signal is present but inconclusive at current sample size; insufficient data means there aren't enough paired trades yet. Sources reported as negative contributions are candidates for pruning; positive contributions are candidates for amplification.
Source / layer Variant n (with) n (without) Effect (pp) Expectancy Δ ($) Verdict
Loading…

Continuous-feature monotonicity

Binary "had whale yes/no" hides the difference between 1 whale-trade and 30. This panel groups trades by item-count bucket per source-type and reports the Spearman rank correlation between bucket index and win rate. monotone_increasing = volume genuinely matters (real signal that scales). monotone_decreasing = volume actively hurts (suspicious). non_monotone = presence was a red herring; volume is noise. Each ρ carries a 95% confidence interval; ≠0 marks correlations whose interval excludes zero, and an underpowered badge flags sources with too few paired trades to trust the number yet.

Waiting for nightly ablation data…