fix: retrospective agent evaluated wrong month and crashed; persist its analysis to Reports tab

fix: shift retrospective cron from 5AM to 4AM UTC to stay Sunday in CDT

fix: fill buy orders at current market price instead of analyst entry price

fix: enforce weekend/holiday block even when FORCE_RUN=true

fix: correct notification type for max_hold_days and manual position closes

feat: add max_hold_days exit condition to position monitor (default 30d)

feat: add NYSE holiday awareness to market_hours check

fix: local time, 0-qty guard, mobile card fields, glass UI

refactor: consolidate trading params into strategy; settings keeps only buying_power

feat: purge analyst reports older than 30 days in market analyst housekeeping

feat: purge agent_run records older than 7 days in paper trader housekeeping

feat: define SPY-beating prompts, lock strategy to retrospective agent only

feat: populate all 5 agent prompts in strategy on seed and backfill

feat: add trade_decision prompt to retrospective strategy tuning

feat: enforce hard buying-power cap and prevent concurrent paper-trader runs

fix: remove 30-day time-based position exit for stock trading

feat: store analyst prompts in strategy and tune them via retrospective agent

refactor: replace per-ticker LLM calls with one batch call per analysis type

feat: add LLM qualitative analysis to fundamental and technical agents

feat: add market cap and 52-week performance to fundamental scoring

feat: add StockTwits trending and NASDAQ most-active as ticker sources

fix: enforce tier-based selection and expand noise filter in TrendSpotter

fix(trend-spotter): replace broken Finviz with Yahoo gainers; show full source lists

fix: restore stock picks by raising Anthropic max_tokens and hardening JSON parsing

feat: add Stooq + Google Finance fallbacks when yfinance has no price data

fix: current_price missing from get_fundamentals return dict causing all tickers to be skipped

Set MAX_CANDIDATES to 20

Set MAX_CANDIDATES to 10

Fix 0-pick root cause: larger candidate pool, better noise filtering, skip delisted

Log agent role and model name on each LLM call

Remove all hardcoded GPT-4o references from agent log messages and docstrings

Log active LLM provider and fix misleading GPT-4o docstrings

Consolidate all services around root config files

Clean up env vars and fix retrospective last-Sunday check

Add GitHub Actions CI/CD, HF Space deployment, market hours guard, and static serving