12d • General discussion
Added to levz
1. Session validity window learning
Tracks how long each shadow trade takes to resolve per session. After 10 shadows per session she calculates average resolution time and adjusts valid_for_hours automatically with a 50% safety buffer. London resolving in 2h average → validity window tightens to 3h. NY resolving in 5h → extends to 7.5h. Range: 1.5h–8h.
2. Per-pair RR minimum
Tracks actual RR achieved per pair. After 5 shadows per pair she calculates average delivered RR and adjusts min_rr. Pairs consistently delivering 2:1+ get their minimum raised. Pairs struggling to reach 1.3:1 get it lowered. Adapts in 0.1 increments. This ensures she only fires on pairs where the reward genuinely justifies the risk based on evidence.
3. Per-pair × session confidence multiplier
After 5 shadows per pair per session combination, she calculates a win rate and assigns a multiplier (0.85–1.15). Strong session → multiplier raises, grade threshold effectively lowers. Weak session → multiplier drops, threshold effectively raises. Injected directly into the Anthropic analysis prompt so the AI is aware of the historical performance context.
4. Per-pair watcher trigger sensitivity
Watcher escalation trigger defaults to 2 conditions. After 10 watcher escalations per pair, she checks the hit rate. If XAUUSD grades well 65%+ of the time after just 1 trigger condition, she lowers it to 1. If EURGBP rarely grades well after 2 conditions, she raises it to 3. Range: 1–4 conditions.
5. Per-pair volatility threshold
Extends the global vol_threshold to be independent per pair. Each pair gets its own ATR contraction threshold that adapts in 0.05 increments after 8 contracting shadows. EURGBP might settle at 0.60 while XAUUSD stays at 0.75 because they have different natural volatility profiles.
6. Per-pair watcher cooldown
Default 60 minutes. If a pair consistently resolves fast (under 30 minutes), cooldown tightens toward 20 minutes — she can re-engage the pair sooner. If a pair resolves slowly (over 4 hours), cooldown extends toward 120 minutes — no point watching it again so soon. Adjusts in 5-minute increments.
7. Time-of-session bias (news windows excluded)
Each pair's performance is tracked in 30-minute buckets throughout the day. 15 specific UTC windows are permanently excluded — all major news release times (NFP, CPI, FOMC, ECB, BOJ, ISM etc.). After 5+ observations in a clean bucket, if a time window has 70%+ win rate she injects a positive bias note into the Anthropic prompt. If 30% or below she injects a caution note. The AI sees her own historical performance context for that specific pair at that specific time.
0
0 comments
Added to levz
powered by
skool.com/ascending-with-ai-agent-levz-8813
From survival mode to funded trader. Learn to use AI Agent Levz, pass prop challenges & ascend.
Suggested communities
Powered by