Two bugs fixed:
1. Portfolio Manager uses wrong state key (critical):
portfolio_manager.py read state['investment_plan'] (Research Manager's
raw output) instead of state['trader_investment_plan'] (Trader's
memory-refined output). This silently bypassed the Trader agent's
entire contribution — including lessons learned from past trades —
in the final risk assessment and trading decision.
The Trader agent applies FinancialSituationMemory to refine the
Research Manager's plan with insights from similar past situations.
By reading the pre-refinement plan, the Portfolio Manager made its
final decision without this critical context, effectively making the
Trader node a no-op in the decision pipeline.
2. _get_stock_stats_bulk() missing pandas import:
y_finance.py uses pd.isna() in _get_stock_stats_bulk() but never
imports pandas as pd, causing NameError on every call. The fallback
in get_stock_stats_indicators_window() catches this silently, but
it defeats the bulk optimization — falling back to O(n) individual
API calls per date instead of O(1) bulk calculation.
Added 7 tests covering both fixes.
Add _clean_dataframe() to normalize stock DataFrames before stockstats:
coerce invalid dates/prices, drop rows missing Close, fill price gaps.
Also add on_bad_lines="skip" to all cached CSV reads.
voidborne-d
Yijia-Xiao
Edward Sun
luohy15