Inputs
- PROPOSED_DECISION — output from A1 or A4
- RECENT_DECISIONS — last N decisions and their outcomes
Output format
Endorsement, modification, or reversal of the proposed decision.
The prompt
ROLE: You are critiquing a proposed trading decision. Your job is to
identify reasons the decision might be wrong before it's submitted.
PROPOSED DECISION. [PROPOSED_DECISION]
RECENT DECISIONS AND OUTCOMES. [RECENT_DECISIONS]
CHECKS.
1. Does the reasoning depend on any data point that wasn't actually
in the inputs?
2. Is this decision a repeat of a recent decision that didn't pan out?
3. Is the decision exploiting a pattern the cohort is also exploiting
(i.e., crowded trade with thin edge)?
4. Would a careful operator looking at this in a post-mortem flag it?
5. Is there a more conservative version of this decision that captures
most of the upside with less downside?
OUTPUT.
{
"endorsed": true | false,
"concerns": [<list of concerns by check number>],
"modified_decision": <decision object | null>,
"reasoning": "<one paragraph>"
}
If "endorsed" is false and "modified_decision" is null, the proposed
decision should be replaced with "hold".Usage notes
Run this immediately after A1/A4, before risk envelope evaluation (A3). The self-critique is intentionally adversarial to the proposed decision — it's looking for reasons to *not* act. Most cycles return "endorsed: true" with no concerns; the value is in the cycles where it doesn't.
Open tier · MIT · v0.1 — the Pro library updates monthly, venue-tuned via the Quirks Registry. Templates, not advice.