Skip to content

Latest commit

 

History

History
46 lines (32 loc) · 1.92 KB

TRUST.md

File metadata and controls

46 lines (32 loc) · 1.92 KB

Engineering trust

What has been independently compared?

Six cases execute against official UVA HotSpot v7.0 at pinned commit 9f92256f7a45db6057c809e94fcd224ddad9157a. Four synthetic cases use an equation-level detailed-grid mapping (Class A); EV6 and the accelerator package study are Class B because package models differ. Three Class-A inverse power limits and seven Class-B interventions are also compared. HotSpot is independently authored software, not physical ground truth.

What has been experimentally validated?

Nothing at Grade A. No sufficiently specified matched public hardware holdout was found, so hardware_validated=false. The strongest public candidate is Grade B and was not used to manufacture an accuracy statistic.

What is only analytically verified?

Plane-wall, generated slab, package-series-resistance, and transient lumped cases; the manufactured solution and mesh studies verify discretization and conservation. These tests are repository-authored and are not independent.

What has not been validated?

Detailed package structures, fluid flow, radiation, boiling/phase change, leakage feedback, temperature-dependent properties, thermo-mechanical stress, reliability, JEDEC theta/psi interpretation, and hardware absolute accuracy.

What assumptions dominate error?

Package/contact/spreading resistance, spatial power knowledge, coolant boundary conditions, material condition/temperature, and the omission of local package features. Class-B peak-rise disagreements quantify how strongly package model form can matter.

Appropriate use

Use Aethermor for architecture-stage feasibility screening, inverse headroom, sensitivity, and intervention ranking. Do not use it for product, package, or reliability sign-off.

How can I challenge a result?

Follow the challenge bundle, retain the raw artifacts, and file the model-discrepancy issue template with a fully specified benchmark.