PRD 003: Deployment Gate and Candidate Review

Goal

Make Pam decisive about when a research candidate is merely interesting versus actually ready for a small smart trade.

Problem

Pam now has a reliable research harness, but no clean path from a replay-improving sleeve to a live candidate review packet.

Requirements

1. Keep the hard deployment gate: - minimum modeled edge >= 100 bps - minimum hit-rate target > 40% - minimum stability over repeated loops
2. Add a candidate review artifact for the top live opportunities: - ticker - category - sleeve family - modeled edge - eventful score - time to resolution - invalidation reason
3. Separate states: - research_candidate - review_candidate - deployable_trade

Success Criteria

1. Pam can produce a small ranked list of live review candidates.
2. A deployable trade only appears when all gates are met.
3. The loop no longer blurs exploratory sleeves with live execution logic.

Notes

The goal is to make “no trade” an intentional state, not a failure state.