PRD 004: Ralph Research Operator for Pam

Goal

Use Ralph as a bounded operator loop for Pam’s research backlog so progress continues without manual prompting.

Problem

Pam now has a working research loop, but we still need an outer operator loop that can:

1. inspect current state
2. choose the next PRD-sized improvement
3. make a bounded change
4. run validation
5. record what changed

Requirements

1. Ralph must work from a dedicated prompt file, not ad hoc terminal memory.
2. Ralph should read: - research/latest_state.json - research/continuous_loop_runs.jsonl - research/continuous_loop_events.jsonl - this PRD directory
3. Ralph must stay within bounded changes: - prefer one PRD at a time - run focused tests - append findings to generated docs
4. Ralph must not lower deployment thresholds just to make a trade appear.

Success Criteria

1. Ralph can run unattended and leave behind a clear artifact trail.
2. Each Ralph cycle maps to a concrete PRD or sub-task.
3. The output helps Pam move toward a deployable smart-trade system, not just produce activity.