Merge: portfolio re-evaluation action

feat: portfolio re-evaluation GitHub Action + structured LLM timeout

Merge sprint9/quality-execution into main — Sprint 8+9 complete

docs: update README and showcase page with dashboard, online evals, prompt versioning

feat: VPS dashboard + LangSmith online evaluators and prompt versioning

fix: VPS schema migration, EU noise word, global Pydantic warning filter

fix: force websockets>=13 after alpaca-trade-api to prevent yfinance crash

feat: increase max_tokens to 65536 + add project showcase HTML page

feat: install alpaca-trade-api in CI + add Alpaca secrets to GitHub

feat: widen market cap to $5M-$500M, dynamic amnesia, more noise words

fix: Kelly per-verdict caps, more noise words, higher token limit, Pydantic warning

docs: update README for Sprint 9 — debate, Alpaca, catalyst, evaluators

fix: bump structured output token limit to 8192 + add noise words

fix: make alpaca-trade-api optional to prevent CI dependency conflicts

feat: Sprint 9 — Execution & Quality Control

🧠 Update Agent Memory Ledger

🧠 Update Agent Memory Ledger

fix: cron performance — stale seeds, Kelly cache, ticker noise, SEC timeouts

fix: Brave 429 rate limiting + garbage ticker extraction

docs: update README with SEC EDGAR and Kelly Criterion features

feat: SEC EDGAR ground truth + Kelly Criterion position sizing (Epic 3 & 4)

docs: update README with LangGraph refactor, VPS layer, and project structure

feat: LangGraph best-practices refactor (Sprint 8 Epics 2-3-4)

🧠 Update Agent Memory Ledger

🧠 Update Agent Memory Ledger

fix: UK pence double-conversion bug + Graham Number normalization for .L tickers

🧠 Update Agent Memory Ledger

🧠 Update Agent Memory Ledger

feat: integrate LangSmith observability (Sprint 8 Epic 1)

🧠 Update Agent Memory Ledger

docs: add TEAM_GUIDE.md explaining the full v7.0 architecture and usage

🧠 Update Agent Memory Ledger

fix: add LLM model fallback chain to prevent 429 rate-limit failures

🧠 Update Agent Memory Ledger

feat: major architecture refactor with yFinance screener, scoring, and insider feeds

fix: revert to stable natural-language search queries, remove freshness filter to prevent API timeouts on Canada/Australia

fix: inject local exchange acronyms into region searches to increase foreigner ticker density and fix No Ticker Found bugs

feat: upgrade cron intelligence with seen-ticker caching and multi-candidate LLM extraction loops

chore: default LLM to gemma-3-27b for superior comprehension and stability

chore: default LLM to llama-3.1-8b after performance evaluation

fix: swap OpenRouter LLM to Llama-3.1-8B to prevent connection timeouts during extraction

refactor: improve social scout regex and search query for better ticker extraction

fix: normalize UK pence currency before gatekeeper price check

feat: include WATCH verdicts in the paper portfolio tracker

feat: implement Graham-aligned Sector Logic gatekeeper and live Portfolio paper-tracking memory

fix: resolve docstring syntax error causing cron failure

fix: restore missing base dependencies (chainlit, langgraph, openai) overwritten by local pip freeze

fix: add HuggingFace Spaces metadata block to README

docs: rewrite README to fully document the PrimoGreedy LangGraph and ReAct architectures

feat: upgrade agent to ReAct tool calling using Finnhub API for deep fundamental research