Inputs
- SITUATION — description of what makes this non-trivial
- OPTIONS — list of options under consideration
- CONSIDERATIONS — list of factors the agent should weigh
Output format
Reasoning chain followed by recommendation.
The prompt
ROLE: You are deliberating about an autonomous trading decision that
requires more than a single-shot evaluation.
SITUATION. [SITUATION]
OPTIONS. [OPTIONS]
CONSIDERATIONS. [CONSIDERATIONS]
PROCESS.
For each option, reason through:
- What would happen if we chose it and the situation resolved in the
expected direction.
- What would happen if we chose it and the situation resolved against us.
- What information would we have that would let us update later.
- What this choice closes off.
Then synthesise: which option survives the reasoning best? Why?
OUTPUT.
1. "deliberation": object with one entry per option, each containing
{ "if_right": "...", "if_wrong": "...", "info_we_gain": "...", "closes_off": "..." }
2. "recommendation": chosen option ID
3. "confidence": "low" | "medium" | "high"
4. "what_would_change_my_mind": <a specific observable signal>Usage notes
Run this when the decision-loop scaffold (A1) returns ambiguous output, or when an operator-configured "deliberate threshold" trips (e.g., position size above a threshold, regime classifier says unusual). The "what would change my mind" field is intentional — it primes the agent to revisit the decision if that signal appears.
Open tier · MIT · v0.1 — the Pro library updates monthly, venue-tuned via the Quirks Registry. Templates, not advice.