Validation & claims

cufemlab labels every analysis with exactly what its result means and how far you can trust it. This page defines the validation tiers, lists which of the 15 live analysis types fall in each, and states precisely what "internally validated, external passport not yet attached" is - and is not.

The newest type, cufem3d_gpu_field_analysis, is experimental: GPU-accelerated 3-D field analysis is available on compatible GPU workers, and runtime metadata reports the actual execution backend and fallback state. Every result carries an analytic cross-check against the magnetized-sphere reference, and an optional CPU/GPU numerical parity check runs when run_reference_check is enabled (off by default); the type has no external validation passport, and it is not an independently validated 3-D torque prediction.

The six tiers

A result carries one of six tier labels. The tier tells you what the number was checked against and what an engineer is entitled to conclude from it.

Where the 15 live types fall

Fifteen public analysis types: two externally cross-checked, five internally validated, one experimental (the GPU 3-D lane), one experimental surrogate, two engineering estimates and four utilities. Two plus five plus one plus one plus two plus four is fifteen. The scope column below marks each type's tier.

Standalone eddy-current solvers exist in the engine and worker, but on the customer API eddy physics reaches users only through eddy_thermal_coupled. They are not separately exposed as customer analysis types.

What "internally validated, external passport not yet attached" means

All five new physics types return result.v1 with validation_card.status = "not_validated" and claim_allowed_public = false. That wording is precise and deliberate. Read it exactly:

Internal card vs external oracle card

The difference is the source of truth the result is checked against:

An external passport for the new physics types is a planned follow-on. Until it is attached, the honest statement is: "validated internally against its analytic / energy identity; external validation passport not yet attached."

The experimental surrogate, stated plainly

Data-rights note: TEAM-7 and H(curl)

Reading a result card in the API

Every job result is result.v1. The validation_card is where the tier and status live. Fetch it and read it before you rely on a number.

import os
from cufemlab_client import Client

client = Client(api_key=os.environ["CUFEMLAB_API_KEY"])  # default base https://app.cufemlab.secrotec.nl
project = client.create_project(name="thermal-check")

job = client.analyze(
    project_id=project.id,
    analysis_type="thermal_conduction_steady",
    input_params={},          # sensible defaults for every parameter
    gpu=True,
)
job.wait(poll_interval=10)
result = job.result()

print(result.verdict)                       # pass / fail
print(result.value)                         # result_summary.value
# validation_card.status == "not_validated" for the five new physics types:
# internally validated, external passport not yet attached.

In-app Jupyter injects a per-session scoped key, so Client.from_env() works with zero paste. Direct HTTP: POST /api/v1/jobs {analysis_type, input_params, project_id} with a Bearer access token or X-API-Key; poll GET /api/v1/jobs/{id}; read GET /api/v1/jobs/{id}/result. Errors: 422 invalid domain params, 402 INSUFFICIENT_CREDITS, 429 CONCURRENCY_LIMIT_EXCEEDED (max 2 concurrent jobs per user).

Summary

• Two types are externally cross-checked (Tier 1) and ship an independent oracle card.
• Five new physics types are internally validated (Tier 2) - correct by their own analytic and energy identities, external passport not yet attached (validation_card = not_validated).
• One type is an experimental in-distribution surrogate (Tier 3), not a general-purpose neural operator. One is the experimental GPU 3-D field lane (cufem3d_gpu_field_analysis, Tier 4) — cross-checked against an analytic sphere every run, with an optional CPU/GPU parity check (run_reference_check, off by default), and no external passport.
• Two types are engineering estimates (Tier 5); four are utilities (Tier 6).
• TEAM-7 / H(curl) internal validation exists; third-party benchmark data is not published or bundled.
• cufemlab makes no accuracy or speed claim against any commercial tool.