Each headline call is checked against what the price actually did.
┌─[ SUM OF HYPOTHETICAL CALL RETURNS % ]CLAUDE GPT (shadow)
Measured after decisions are available. Long on bullish, short on bearish, 0.2% fees. Overlapping calls; this is not a portfolio return.
┌─[ ROLLING ACCURACY % (last 20 calls) ]CLAUDE GPT (shadow)
Each model may make a different number of calls. Accuracy alone does not establish a profitable trading edge.
┌─[ SCORE vs ACTUAL MOVE ]CLAUDE GPT
Good calls land in the top-right and bottom-left boxes.
┌─[ DECISIONS & REASONS ]
TIME
HEADLINE
CLAUDE
GPT
+4H
+24H
┌─[ CANDIDATE_LEDGER · every breakout and what happened to it ]
SIGNAL CANDLE
PAIR
OUTCOME
CLOSE / BREAKOUT
ADX
AI AT THE TIME
Recorded by the bot itself, hash-chained. Shadow results per outcome come from scripts\8-ledger-report.bat; the AI veto is judged only by the pre-registered H3 test, never by accuracy alone.
┌─[ SYSTEM_HEALTH ]
Is everything running?
Refreshes every 15 seconds.
┌─[ NEWS_FEEDS ]
┌─[ AI_PIPELINE ]
┌─[ RECENT_ERRORS ]
┌─[ DERIVATIVES · OBSERVE-ONLY ]collected for research; never used to trade
COIN
FUNDING / 8H
BASIS
OPEN INTEREST
LONG/SHORT
AGE
┌─[ AI_PROVIDERS ]
Connect the AI Council
Keys are saved only on your PC in C:\dizzycrypto\.env and never shown in full again. Testing lists the provider's models, which costs nothing.
A
ANTHROPIC · CLAUDE
NODE_A · screens every filtered headline
—
O
OPENAI · GPT
NODE_B · second opinion on possible vetoes
—
┌─[ BUDGET & RULES ]
GPT also scores every filtered headline for a fair Claude-vs-GPT scorecard. Never used for trading. ~1–2¢/day.
Both models must agree before any trade is blocked.
Free local filter drops most headlines before any AI call.
GPT is asked only about headlines that could block a trade.
At the cap, AI goes silent; the bot keeps trading on technicals.