Pam Ralph Loop

You are operating inside the Pam repository.

Your job is to improve Pam’s ability to find smart Kalshi trades through bounded research iteration.

Current source of truth

Read these first:

1. research/latest_state.json
2. research/continuous_loop_runs.jsonl
3. research/continuous_loop_events.jsonl
4. research/README.md
5. docs/prds/PRD_001_eventful_market_observation.md
6. docs/prds/PRD_002_dislocation_research_sleeves.md
7. docs/prds/PRD_003_deployment_gate_and_candidate_review.md
8. docs/prds/PRD_004_ralph_research_operator.md

Mission

Pick the next smallest useful change that improves one of:

1. data quality
2. eventful observation
3. dislocation-focused sleeve quality
4. replay evaluation quality
5. candidate review quality

Do not chase activity. Improve signal quality.

Guardrails

1. Do not lower the hard deployment gate just to create trades.
2. Prefer small, testable changes.
3. Update documentation for anything operationally important.
4. Run focused validation after code changes.
5. If there is no good code change, write a short generated note explaining the bottleneck.

Preferred workflow

1. Inspect current state.
2. Select one PRD or one clear sub-problem.
3. Make a bounded change.
4. Run focused tests or a smoke command.
5. Write a short artifact to docs/generated/.
6. Stop when the change is validated.

Desired outputs

For each Ralph cycle, leave behind:

1. code or config changes if justified
2. a short generated note in docs/generated/
3. clear evidence from tests or commands

Things to avoid

1. broad refactors without validation
2. speculative live trading logic
3. changing unrelated parts of the repo