Paramus — Lab Result Analysis

Overview

Turn raw run outputs into a decision: QC the data, reduce it, analyze against the objective, and recommend the next action. The value is the QC→reduce→analyze→decide pipeline, not one statistics call.

Guidance vs execution: parsing, statistics, and modeling run through tested Paramus tools. Do not silently drop outliers or invent fits in ad-hoc code.

When to Use

• Analyzing completed experimental runs before the next campaign iteration
• Converting instrument output into a decision-ready summary Do not use for: proposing the next experiments (use the active-learning loop) or dispatching runs.

Workflow

Raw run outputs + objective + acceptance criteria
  ↓ 1. Ingest & parse            → structured measurements
  ↓ 2. Quality control            → flag failed/out-of-range runs
  ↓ 3. Reduce & analyze           → aggregate, fit, compare to target
  ↓ 4. Assess vs objective        → met / not met / inconclusive
  ↓ 5. Emit decision summary      → result + recommendation + provenance

Procedure

1. Discover tools via search/get_schema ("parse results", "statistics", "curve fit").
2. Ingest and parse raw outputs into structured measurements.
3. Apply QC rules; flag (do not silently delete) failed or out-of-range runs.
4. Reduce and analyze against the objective; quantify uncertainty.
5. State whether the objective is met, not met, or inconclusive.
6. Emit a decision-ready summary with recommendation and provenance.

Domain Judgment

• Flag anomalies explicitly; never silently discard data points.
• Distinguish "target not met" from "inconclusive" — they drive different next actions.
• Carry uncertainty through to the recommendation; a point estimate alone is misleading.

Fallbacks

• Analysis tool unavailable → deliver parsed data + a manual analysis outline, clearly scoped.
• Endpoint unreachable → stop and report; do not fabricate fits or QC verdicts.