MetroAI v0.6.0 — KOLAS Compliance OS

A measurement uncertainty MCP server for ISO/IEC 17025 / KOLAS-accredited laboratories, built solo by Yongbeom Kim (KTR ISO 17034 cert. 2024-CM-007).

3 MCP tools

calculate_uncertainty — GUM (ISO 98-3) calculation across 9 calibration templates
pt_analysis — Proficiency Testing (z-score / En / zeta per ISO 13528 + ISO 17043)
reverse_uncertainty — Novel within prior-art search. Given target U, compute per-component standard uncertainty caps. Not found in GUM Workbench, NIST Uncertainty Machine, or major open-source GUM tools as of 2026-05.

9 templates

v0.5: gauge_block, mass, temperature, pressure, dc_voltage
v0.6 (new): tem_lattice (HRTEM d-spacing), sem_eds (ZAF), afm_roughness (Sa/Sq, ISO 25178-2), ocd_scatterometry (RCWA, SEMI MF-1789)

v0.6.0 additions

6 AI agents backbone — semi-intel / job-scout / kolas-monitor / kolas-audit-predictor / orchestrator / schedule (with live/stub data origin flags)
Verifiable audit trail — Ed25519 RFC 8032 signatures + W3C PROV-O provenance graphs
Sobol QMC — verified ±0.003% agreement with GUM analytic on simple linear models
Pydantic v2 input validation across MCP / FastAPI / Streamlit
GradientBoosting audit-risk model with honest synthetic-data metrics (acc 60.6% ± 3.1pp, 5-fold CV; real KOLAS validation pending)

Standards

ISO/IEC 17025:2017 · ISO/IEC Guide 98-3 (GUM + MCM Suppl. 1) · ISO 13528 · ISO 17043 · ISO 18516 · ISO 25178-2 · ISO 22489 · KOLAS-G-002 · SEMI MF-1789 · W3C PROV-O · RFC 8032 (Ed25519)

CI/CD

GitHub Actions on Python 3.10/3.11/3.12 — 36 passing tests.

Streamlit demo

https://metroai-gnbdv7pqq3quqsudb5pwvj.streamlit.app

Honesty

All AI outputs carry data_origin flags (live / stub / synthetic). External accuracy claims tied to real KOLAS validation only. See docs/HONESTY_NOTES.md.