AIPE-Magnetics-Database

Provenance, licensing and scientific limitations

What a record guarantees

• The value is the upstream value (converted to SI with the original retained); the upstream file is identified by URL, commit / DOI, locator and SHA-256.
• The record validates against schema v0.2 and its references resolve. This is not an experimental verification. review_status is unreviewed for every third-party record.

Data classes are never mixed

Experimental measurements, manufacturer datasheet curves, manufacturer-tool exports, values derived from measurements and fitted parameters are separate records with an explicit data_class; canonical materials copy properties from a single member, never a blend. No model prediction is stored in the Database.

Known limitations and flags

Topic	Handling
MAS values come largely from manufacturer datasheets (digitised by OpenMagnetics)	MAS is Apache-2.0, but that licence cannot grant rights the manufacturers never gave; records keep the datasheet URL and third_party_unreviewed. Rights holders may request removal (open an issue).
MAS Curie temperatures	often a datasheet bound (“>= 210 degC”); curie_temperature_may_be_bound
MAS Steinmetz fits	no single fit temperature (temperature polynomial ct0-ct2): kept as model_metadata, not converted into AIPE Steinmetz sets
MAS shapes	tolerance ranges only; no midpoints, no Ae/le/Ve derived; 3 inverted ranges kept verbatim (upstream_dimension_range_inverted)
MAS parts	no effective parameters published (no_effective_parameters)
MAS 3C94 B-H loop		B	up to 509.7 indicates mT stored as T upstream; kept and flagged suspected_unit_error_flux_density
MAS complex permeability	9,552 rows with negative components kept, record flagged
MAS origin = MagNet loss points	27,588 rows; each matches a Princeton row (match fraction 1.0) -> duplicate
Princeton MagNet	frequency is a 10 Hz Welch estimate (query ranges); temperature is the oil-bath set point; DC bias round(mean H) with small negative values (5,307 rows, flag dc_bias_small_negative_values)
Paderborn LEA	GPL-3.0-only subtree with LICENSE + NOTICE; MagNet-derived permeability = derived_view; TDK MDT copies in datasheet/ not imported; B-H files N27/N30/T37 in mT converted with unit_reinterpreted_mT; 46 isolated mT rows excluded; T37 B-H flagged as probable copy of N30; upstream pyproject says MIT, LICENSE says GPL-3.0 (LICENSE treated as authoritative)
HEFMAG, Zenodo 21626116	temperature and DC bias not stated (temperature_not_stated); loss as J/m^3 per cycle
Identical tables in different MAS materials	14 groups kept as primary and flagged identical_table_in_other_material
MAS B-H loops (1.5 M points)	not committed; regenerated by refresh (records keep row count and SHA-256)

Never use these data beyond their recorded condition ranges without your own validation; the Database neither interpolates nor extrapolates.

Licences of committed data

Directory	Licence	Obligations
data/normalized/mas	Apache-2.0 (MAS)	keep attribution and data/raw/mas/LICENSE.upstream.md
data/normalized/princeton-magnet	MIT	keep the MIT notice; cite the MagNet TPEL papers
data/normalized/paderborn-lea, data/raw/paderborn-lea	GPL-3.0-only	copyleft subtree: LICENSE, NOTICE, source offer
data/normalized/hefmag, data/normalized/zenodo-smc-ferrite-triangular, matching data/raw/	CC-BY-4.0	attribution (citation in the dataset record)
curated catalogue (package)	NOASSERTION	transcribed datasheet values with locators