Real direction-accuracy/P&L measurement + BULLISH trigger fixes

Add NVIDIA NIM as free-tier LLM provider; research + UI updates

llm: bench provider on DAILY quota 429, not just per-minute

Fix AI conflicts: persistent retry (no terminal 'AI unavailable'), consistent direction/reasoning, live-anchored 1D/3D targets + gap note

ML model retrain + research updates + news/llm tweaks

fix: live price AI context, top5 IST midnight reset, remove Primary Call UI

feat: Gemini/SambaNova providers, ATR power-law range model, crash-guard triggers, backtest retry rounds

fix: Ollama timeout 70s, 2-min backoff, safe float in ATR clamp, script fixes

fix: realistic AI ranges, Ollama cold-start warmup, price-target prompt

fix: cache Ollama health result, auto-retry top5 on empty cache, fix 3-col tf-grid

feat: AI-owned targets, BUY/SKIP chip, top picks expansion, Ollama last-resort fix

fix: dynamic provider reorder + OHLCV cache path + all-TF banner

refactor: Ollama 2nd in chain, AI all TFs, max concurrency, JSON mode

fix: resolve AI unavailable cascade on watchlist predictions

fix: increase Ollama timeout to 25s for HF Spaces reliability

fix: improve Ollama setup debugging and add configuration guide

fix: repair Ollama fast-path — wrong model name, wrong API endpoint, latent NameError

perf: move Ollama to fast-path, raise concurrency limits for watchlist/top5

feat: add Ollama fallback, request context, and cached OHLCV helpers

feat: add loophole detection and specialist analysis

fix: prevent cache poisoning of watchlist predictions + OHLCV thundering herd

feat: persistent OHLCV cache, per-TF loading, news sort, UI fixes

refactor: extract shared llm_client.py + per-day cache + LLM self-learning + HF Spaces paths