Decision Journal

Decision Journal

Every recommendation the server (or its debate workflow) produces can be logged with its rationale and the price at decision time — then measured later. This is the forward-looking companion to Backtesting: backtest validates the rules on history; the journal validates the live calls going forward.

Write: log_decision(symbol, action, rationale, score)

• action must be BUY / HOLD / SELL
• Price at decision time is fetched automatically and stored
• Example: "Log that I'm HOLDing RELIANCE at 45 with the trend-vs-valuation rationale"

Read: review_decisions(symbol="")

For each open decision: current price, return since decision, and a verdict:

Action	CORRECT	WRONG	NEUTRAL
BUY	return > +1%	return < −1%	in between
SELL	return < −1%	return > +1%	in between
HOLD	—	—	|return| ≤ 5%; BROKEN beyond

Storage

• Primary: MongoDB collection stock_data.decision_journal (configurable via MONGODB_URI/MONGODB_DB_NAME)
• Fallback: if MongoDB is unreachable at call time, entries go to decision_journal.json next to server.py (gitignored). Note: the two stores are independent — nothing migrates automatically if you switch.

Workflow integration

The bull_bear_debate prompt ends with a mandatory log_decision call, so every debated decision is recorded with its full rationale — no untracked opinions.

Honesty rules

• Verdicts use the fixed thresholds above — mechanical, not generous.
• HOLD decisions that drift >5% are marked BROKEN, not quietly excused.
• Nothing is ever deleted automatically; history is append-only.