README.md

ClarityOps (LLM-Only)

ClarityOps runs fully inside an LLM chat:

1. User pastes a scenario and optionally attaches data files (CSV/XLSX/PDF).
2. Phase 1: ClarityOps reads everything and asks up to 5 grouped clarification questions.
3. User answers in the chat.
4. Phase 2: ClarityOps produces a validated, structured analysis.

No external code, servers, or UI — all interaction is conversational.

How to use (operator playbook)

Turn 1 (User): Paste the scenario using /prompts/user_scenario_template.txt. Attach any files.

Turn 2 (Model): ClarityOps returns Phase 1: Clarification Questions (≤5, grouped). It stops and waits.

Turn 3 (User): Reply using /prompts/answer_clarifications_template.txt.

Turn 4 (Model): ClarityOps returns Phase 2: Structured Analysis in the exact format from /prompts/phase2_format.txt.

Notes

• The system prompt (/prompts/system_master.txt) enforces the two-phase behavior, numerical transparency, units, and medical guardrails.
• The schema (/schemas/phase2_output.schema.json) is provided as a reference shape that the model mirrors in Phase 2. (The LLM follows it; there’s no external validator.)
• The guardrails (/policies/medical_guardrails.md) define units, ranges, privacy, and recommendation style.