Changelog

All notable changes to the Magnetic Agnostic Structure (MAS) specification are documented in this file. The format follows Keep a Changelog and the project adheres to Semantic Versioning.

Versioning rules for MAS:

• MAJOR — backwards-incompatible schema changes (removed/renamed fields, tightened constraints, semantic changes to existing fields).
• MINOR — backwards-compatible additions (new optional fields, new schemas, new enum values that consumers are required to ignore-on-unknown).
• PATCH — clarifications, documentation, examples, bug fixes that do not alter the validation surface.

A change to the bundled component database (data/*.ndjson) follows the same rules: adding a material/shape/wire is MINOR, removing or renaming one is MAJOR.

[Unreleased]

The next release is a MINOR one: everything below is additive (new optional fields and new records), with one requirement relaxed. No existing document becomes invalid.

Added

• Saturation current on datasheet transformers (MAS-RFC 0017). Optional saturationCurrentPeak and saturationCurrents on magnetic.json#/$defs/magneticDatasheetTransformerElectrical, identical to the fields the inductor and coupled-inductor variants already carry. Energy-storing (flyback) transformer datasheets state I_sat on the primary; the closed variant had nowhere to put it.

• Lead-to-pin assignment (MAS-RFC 0013, ABT #1178). Two optional fields on coil.json#/$defs/connection — end (start | finish | tap, which end of the winding terminates on this terminal; start is the dot end, a series junction is a tap on both sections) and parallel (integer >= 0, which parallel strand terminates here; absent means all of them together) — and one on bobbin.json#/$defs/pin: removable (boolean, default false), for the corner pins bobbin makers offer cut off to open up creepage. The RFC also fixes the default pin numbering used when a bobbin brings no pin names of its own: counter-clockwise from row 0, seen from the mounting side (owner decision 2026-09-12). Documented in docs/magnetic/coil.md, including the industry assignment rule (rows are isolation groups, start and finish adjacent, taps shared).

• Multi-chamber bobbins and toroid bases (MAS-RFC 0014, ABT #1179). Four optional fields on bobbin.json. processedDescription.dividers[] describes the interior walls of a split former — thickness along the column axis and coordinates are required, height (radial reach; absent = as far as the flanges) and crossingSlot (width, depth, angle) are optional — and functionalDescription.numberChambers (integer >= 1) is the count a catalogue row can state before anybody has transcribed the wall drawing. functionalDescription.base describes the moulded base a wound toroid is mounted on, which is a bobbin of family t (decision D1 of the 2026-09-12 survey, so it reuses pinout / pins[] / connections[] rather than inventing a second pin vocabulary): mounting (horizontal | vertical), length, width, height and standoff required, pocketInnerDiameter, pocketDepth, boatWidth and the selection limits maximumCoreOuterDiameter / maximumCoreHeight optional. The bobbin's existing orientation is unchanged and keeps its former meaning. Documented in docs/magnetic/coil.md.

• Lead sleeving (MAS-RFC 0016, ABT #1181). coil.json#/$defs/connection gains sleeve — material (an insulation-material record or the name of one), wallThickness and innerDiameter required, overlapIntoWinding and numberLayers (default 1) optional — so the sleeve a winding shop slides over a lead is data rather than shop practice. magnetic/insulation/material.json gains form (tape | film | sleeve | varnish), which is what lets a coordinator pick sleeve stock instead of tape, and cti, the measured comparative tracking index per IEC 60112 in volts (the material group of the same name in insulation coordination follows from it per IEC 60664-1; neither is renamed). The RFC's leadCreepage[] output field was landed and then withdrawn the same day (Alf, 2026-09-13): the local outputs.json#/$defs/insulationCoordination it was added to is referenced by nothing (the insulationCoordination property points at the PEAS mirror), so the field was unreachable; per-lead creepage is reported by MKF's insulation-coordination result instead, with no schema field.

• Two sleeve-stock records in data/insulation_materials.ndjson (ABT #1174), the first with form: sleeve, so an insulation coordinator can pick the tube a lead is sleeved with instead of hard-coding a name:

  record	maker, family	walls recorded (m)	dielectric strength	temperature
  PTFE extruded tubing	Zeus, AWG tubing	the 9 nominal walls of the Standard/Thin/Lightweight classes, 1.3e-4 to 5.1e-4	457 V/mil (ASTM D149, lower bound of 457-483)	260 C
  Polyolefin heat-shrink tubing	Alpha Wire, FIT-221 (2:1)	the 10 nominal recovered walls, 4.1e-4 to 1.40e-3	500 V/mil	135 C

  Both datasheets publish dielectric strength per mil with no specimen thickness, so the one published value is recorded at each published wall; that pairing is stated in each record's description. The walls are a list of what the maker sells, not of what fits a given lead: which wall exists depends on the tube size (Zeus's 0.41 mm standard wall is only offered from AWG 20 to 10), and nothing in the record ties a wall to a size.

  Silicone-rubber-coated glass-fibre sleeving (UL 1441) was not added. Varflex, Techflex and Omerin publish a grade breakdown voltage for the whole sleeve (e.g. Grade A 8 kV) but no dielectric strength per thickness, and Varflex and Techflex no wall either; dielectricStrength is required, and dividing a grade voltage by a wall would be our number, not theirs.

• Magnetic shunts (MAS-RFC 0015, ABT #1180). magnetic.json gains shunts[], the pieces of permeable material — a ferrite plate, a ferrite-polymer or flexible-ferrite sheet — put deliberately where the leakage flux runs, which is how an integrated-leakage transformer gets the leakage inductance it is designed for. placement (inWindow | betweenSections | onColumn | outsideWindow), coordinates, dimensions and material (a core-material record or the name of one) are required; name, gapToColumns (inner / outer) and segments[] (length / gap, for a shunt built from several pieces in series) are optional. A shunt belongs to neither the core nor the coil — it is added at assembly — and deliberately not to core.geometricalDescription, which is regenerated on autocomplete. Documented in docs/magnetic.md.

  No new core-material records ship with it, and that is a finding, not an omission. The four sheet grades surveyed on 2026-09-12: TDK FPC film C350 / C351 are already in data/core_materials.ndjson (µi 9 at 1 MHz, Bs 255 mT at H = 25 kA/m, both from the FPC datasheet). TDK Flexield IFL04, Fair-Rite's flexible sheets M1…M6 and 3M EM15TF cannot be filed: core/material.json requires saturation (Fair-Rite publishes no material data sheet for the M grades at all, and 3M publishes no saturation figure) and requires permeability.initial, while all three publish only µ′ at an RF measurement frequency, which is a different quantity. Filing them would mean inventing a saturation figure or a measurement condition. Relaxing either requirement is a separate, owner-approved change.

Changed

• resistivity is no longer required on a core material. It stays a defined, documented property and every existing record keeps it; it is simply no longer in the required list of schemas/magnetic/core/material.json.

  Rationale (ABT #1039). Two facts made the requirement untenable. First, the field is inert for the loss model most EMI-suppression ferrites use: in MKF, core-material resistivity is read only by CoreLossesRoshenModel's eddy-current term and by CoreMaterialCrossReferencer's similarity scoring, never by the lossFactor path. Second, it cannot be estimated when a maker does not publish it. Across NiZn grades at one permeability (mu_i 800-850) the filed and published values span five decades — Ferroxcube 4A11 1e2 ohm-m, Ferronics J 1e3, Meiwa J / ACME F80 / ACME M80 1e6, Fair-Rite 43 / Shinn Der N4S / Encore N1 1e7 — because resistivity is set by Fe(II) content and sintering, not by permeability. Requiring it therefore forced a fabricated value, of the one property the material's own loss model would not read.

  This unblocks real, shipping grades whose makers publish permeability, saturation and a loss curve but no resistivity.

  Consumers of the generated bindings must regenerate and adapt. The quicktype type changes from std::vector<ResistivityPoint> to std::optional<std::vector<ResistivityPoint>>, so call sites break at compile time rather than silently. In MKF there are THREE, found by compiling rather than by grepping — an early grep of mine was truncated and reported two:

  • ResistivityCoreMaterialModel::get_resistivity — throw MISSING_DATA on a disengaged optional, as it already does on empty data. No fallback.
  • CoreMaterialCrossReferencer::MagneticCoreFilterResistivity — mirror the Curie-temperature filter: no reference value skips the dimension, a candidate without one is culled from that filter. Nothing substituted.
  • StrayCapacitance::core_image_factor — the conduction-only branch is an else if, so a material without resistivity falls through to the no-data case already documented below it. Guard the optional, nothing else to decide.

  Verified end to end on a clean origin/main worktree with this schema and those three patches: MKF builds with 0 errors, and the catalogue plus one resistivity-less material parses as 1079 records, 1078 with resistivity and 1 without. The same catalogue on the unpatched engine is rejected outright.

  Until a consumer regenerates, it will REJECT a record that omits the field — nlohmann::json raises [json.exception.out_of_range.403] key 'resistivity' not found and, in MKF's case, the whole load_core_materials call fails, not just that record. Land the regeneration before writing any record that omits it.

Added

• Seven bobbin plastics in data/insulation_materials.ndjson (ABT #1164), so a bobbin's functionalDescription.material resolves to a record carrying a relativePermittivity. Every material value used by data/bobbins.ndjson now names a record: PA66, PBT, PPS, A3X2G10, x2g5, Zen.6130L and SKYT.5220FR (PET was already present). All eight are NORWE coil-former materials; the vendor short codes are resolved to their grade and base polymer through NORWE's published material list, and each permittivity is taken from that grade's own datasheet:

  record	grade	polymer	eps_r (1 MHz)
  PA66	Envalior Durethan AKV30H2.0	PA66-GF30	4.0
  PBT	Envalior Pocan B4225	PBT-GF20 FR(17)	3.6
  PPS	Toray Torelina A504X90	PPS-GF40	4.2
  A3X2G10	BASF Ultramid A3X2G10	PA66-GF50 FR(52)	3.6 (dry)
  x2g5	BASF Ultramid A3X2G5	PA66-GF25 FR(52)	3.7 (dry)
  Zen.6130L	Celanese Zenite 6130L	LCP-GF30	4.0
  SKYT.5220FR	SK Chemicals Skytra 5220FR	PET-GF30 FR	4.7

  These are the GLASS-FILLED values, which is what a bobbin is moulded from; the neat polymer is materially lower. The two BASF polyamides publish a dry and a conditioned figure (3.6/5 and 3.7/5) and the dry, as-moulded figure is recorded — PA66 is hygroscopic and its permittivity drifts with humidity, so the recorded value is the lower bound of its range.

  SKYT.5220FR is the one record whose permittivity is not from its own maker's datasheet: SK Chemicals publishes no dielectric data for the grade. Its value is that of Rynite FR530L NC010, the PET-GF30 FR grade NORWE replaced with it (NORWE material-change note 09/2022) and which SK Chemicals states the grade corresponds to in mechanical and electrical properties. The substitution is recorded in the record's manufacturerInfo.description.

  Consumers that counted the bobbin wall as air in a winding-to-core dielectric stack can now read t / eps_r for it instead.

• Seven bobbin plastics get a relative permittivity, so a bobbin wall can enter a dielectric stack as t/eps_r instead of being counted as air (requested by MKF, ABT #1164): PA66, PBT, PPS, A3X2G10, x2g5, Zen.6130L and SKYT.5220FR. Every functionalDescription.material in data/bobbins.ndjson now resolves under an exact-name lookup; previously only PET did. Values are the glass-filled grades' own, from the manufacturers' datasheets, with the NORWE material number, grade reference and datasheet URL recorded on each.

  Two sourcing decisions are called out because they are judgement, not datasheet:

  • The two BASF polyamides (A3X2G10, x2g5) carry their conditioned permittivity of 5.0 at 23 C / 50% r.h., not their dry as-moulded 3.6 and 3.7. PA66 is hygroscopic and a bobbin in service equilibrates toward the conditioned figure; the dry value would under-state winding-to-core capacitance and so over-state the predicted self-resonant frequency. The dry figure is recorded in each record's description.
  • SKYT.5220FR (SK Chemicals Skytra 5220FR, 131 bobbins) carries 4.7, which is Rynite FR530L NC010's value, not its own — SK Chemicals publishes no dielectric data for the grade. It is carried across on NORWE's material-change note 09/2022, which records 5220FR replacing Rynite FR530L in the same bobbins, and on SK Chemicals' statement that the grade corresponds to that DuPont PET-GF30 FR grade in mechanical and electrical properties. The substitution and both sources are written into the record's description.

  PA66 was also removed from Nylon 6.6's aliases: it now names a record of its own, and the wire-serving polyamide at 3.4 is a different material from the glass-filled bobbin resin.

• Optional coreElectricalReference on magnetic. Records how the core is referenced electrically in the assembled component: floating (not bonded), grounded (bonded to a circuit reference, with an optional isolationSide naming whose local ground; absent = protective earth / chassis), or tiedToWinding (bonded to the start or end terminal of a named winding). A ferrite or powder core has no terminal of its own and follows the conductors around it unless a mounting clip, copper strap / flux band or conductive tape bonds it. The choice sets how the winding-to-core capacitances appear at the terminals (C0/12 floating vs C0/3 tied to a winding end, for a linear potential ramp) and whether the primary→core→secondary common-mode path closes through the core or is diverted. Absent means floating, which is what every model assumed before the field existed, so no existing document or result moves. Non-breaking. Requested by MKF (ABT #1167) after the stray-capacitance review of 2026-09-12; examples/02_flyback_efd25_3c95.json shows it in use.

• Optional permittivity on core materials. Adds a permittivity object to the core material schema with a complex part (ε′ + j·ε″, same sign convention as complex permeability — the imaginary part is a positive loss magnitude), mirroring the complex entry under permeability: a real and an imaginary sub-object, each a new permittivityPoint (value, plus optional temperature and frequency). Enables modelling of dimensional effects (dimensional resonance) and displacement current in high-permittivity ferrites, and the effective usable bandwidth of a given core size. Non-breaking: the field is optional and existing materials are unaffected.

• Complex permittivity data for 7 MnZn power ferrites — ML91S, ML95S, ML27D (JFE), DMR51W, DMR96A (DMEGC), P63 (ACME) and 3F36 (Ferroxcube). ε'(f) and ε''(f) over 0.1–10 MHz, digitized from Fig. 23 of A. Nabih, F. Jin, R. Gadelrab and F. C. Lee, "Characterization and Mitigation of Dimensional Effects on Core Loss in High-Power High-Frequency Converters," IEEE Trans. Power Electron., 2023, doi: 10.1109/TPEL.2023.3285633 (https://ieeexplore.ieee.org/document/10149511). Values are read-offs from log–log figures — expect ~5–10 % uncertainty.

Breaking

• Removed masVersion and masConformance top-level fields. MAS documents are the polymorphic payload of the shared PEAS container, and the PEAS root object was closed (additionalProperties: false) to reject junk keys. Rather than carve MAS-specific metadata into the shared root, both fields were removed.
• Retired the conformance-classes feature (RFC 0002). With masConformance gone, the Class A/B/C bundles (schemas/conformance/), their validator (scripts/validate-conformance.py), the test vectors (tests/conformance/) and docs/conformance.md were removed. RFC 0002 is marked Withdrawn.
• Removed inputs.converterInformation (unused; the topology seed files under inputs/topologies/ remain for future use).
• TMFD (μi=14) replaced by TMFD 60 / TMFD 90 (data, ABT #185). TDG's own material workbook shows TMFD is an Fe-Si powder series in 60μ/90μ only — the two "duplicate" DC-bias fits were the per-grade coefficients (60μ: b=7.8e-7/c=1.70; 90μ: b=2.95e-7/c=2.23), and the μi=14 single record was a botched merge of them. No cores referenced TMFD. The per-grade fits reproduce TDG's published 82% μe@100 Oe (60μ) and the series loss coefficients reproduce the published 450 kW/m³ @ 50 kHz/100 mT.
• Generic Metglas / Finemet records renamed to their actual alloy grades (data, ABT #221). Each was a single record flattening a whole alloy family. Migration map for consumers that referenced a material by these bare names: Metglas → Metglas 2605SA1 (Fe-based amorphous, the workhorse; Tc 395 °C matches 2605SA1), Finemet → FT-3M (FeCuNbSiB nanocrystalline; Bs 1.23 T / Tc 570 °C match FT-3M). The fitted μ(T) and loss curves are unchanged, only re-attributed. No cores or examples in MAS referenced the old bare names. Additional grades (Metglas 2605S3A/2605HB1M/2714A, Finemet FT-3K50T, …) are added separately (MINOR).

Note: scripts/migrate-to-1.0.py still writes masVersion and is now obsolete.

Added

• cableCore datasheet electrical subtype (new oneOf branch on magnetic.manufacturerInfo.datasheetInfo.electrical[]) — clamp-on / cable ferrite cores (clip-on ferrites, cable rings, split/snap-on suppressors): a 1-port common-mode suppression core the cable is threaded through. Electrically a 1-port impedance like chipBead, but a distinct part class — retrofit/threaded onto a cable rather than reflow-soldered — so a consumer can query and prefer real cable cores for cable-level CM mitigation instead of overloading chipBead. Carries impedancePoints (with the existing per-point current for the DC-bias-derating curve), numberTurns, dcResistance, ratedCurrents/ratedCurrentPoints, impedanceTolerance, selfResonantFrequency, mountingForm (solidRing / snapOn / split / screwable — retrofit clamp vs build-time ring) and maximumCableOuterDiameter (the inner-diameter cable-fit limit, a primary cable-core selection param). Multi-turn curves reuse the existing one-electrical-entry-per-configuration idiom (a datasheet's 1/2/3-turn |Z| tables become one entry each, discriminated by numberTurns), and the toroid/ring geometry (inner/outer diameter, height) and ferrite material stay in the shared core description — no geometry or material fields are duplicated into the electrical block. Purely additive: every existing part still matches exactly one oneOf branch (verified against a per-subtype sample of the bundled catalog), and unknown-subtype-ignoring consumers are unaffected. Motivation: Hertz's radiated cable-mitigation picker currently selects from chipBead SMD beads for want of a cable-core class; this gives clamp-on cores a first-class home once the parts are ingested.
• saturationCurrentPeak on the commonModeChoke electrical variant (ABT #279) — peak core-saturating bias current in Amperes for current-compensated CMCs, mirroring the inductor variant's field. Motivation: 427 catalogued CMCs (WE-CMB/WE-LF/WE-CMBNC among others) carry a datasheet I_sat that previously had no schema slot, blocking their retag from the mistagged inductor variant.
• Magnetics (Mag Inc.) power ferrites (data, ABT #213): L, R, P, F, T — first Magnetics ferrite materials in the DB (their powder cores were already covered). Constants from the 2021 ferrite catalog (μi, Bs@1194 A/m 25 °C, Br, Tc, ρ, density, cross-checked against the parts database); loss model = Magnetics' official per-range equations P[W/m³] = a·f^x·B^y·(b−cT+dT²) mapped verbatim onto Steinmetz ranges (R/F/T reproduce the catalog's typical-loss table to ±3%); the catalog typical points are stored in advanced_core_materials. Known caveats: P's official equation overpredicts ~3× at 100 kHz vs the (internally inconsistent) typical table — kept official, discrepancy tracked in ABT #213; saturation has only the 25 °C point (Magnetics publishes no tabular 100 °C Bs); J/W/M/C/E/V (high-perm/filter grades) still pending loss-factor data.
• 6 more TDG MnZn ferrites (data): TPF26, TPW23, TP5H, TP5-B, TP5R, TP6 — measured μi–T curves and Pcv points from TDG's material-performance workbook with house-pipeline Steinmetz fits (13–31% mean error; TP6 split into [1,3 MHz]+[3 MHz,1 GHz] ranges). These grades have no public datasheet; Bs/Tc/resistivity/density are anchored on the closest documented family grade (TPF26←TP4A, TPW23←TPW30, TP5H/TP6←TP5E, TP5-B/TP5R←TP5) and must be replaced when TDG characteristic sheets arrive (ABT #196).
• 5 TDG MnZn ferrites (data): TP4, TPW33, TPG33B, TPB16, TPB22 — built from TDG's measured material-performance workbook (μi–T curves as the initial permeability, manufacturer Pcv points) plus per-material datasheet constants (Bs/Br/Hc/Tc/ρ/density; TPG33B constants from the 2025-11 automotive brochure, ρ/density carried from the TPG33 base grade). TPW33/TPG33B get house-pipeline Steinmetz fits (6.2% / 3.4% mean error, points in advanced_core_materials); TP4/TPB16/TPB22 carry their measured Pcv points inline (single (f,B) sweeps — a Steinmetz fit is not identifiable from them).
• TP4A initial permeability upgraded from the single spec point (2400 @ 25 °C) to the measured 24-point μi–T curve; TP4A/TP5 gained their measured Pcv–T sweeps in advanced_core_materials.
• frequencyFactor on pocoPermeabilityMethodData (optional {a, b, c, d}): percent-of-initial logistic rolloff (a / (1 + (f/b)^c) + d) * 0.01, matching the implementation MKF ships since ABT #169. Legalizes the 48 POCO V2026 records that already carried the fitted factors.

Fixed

• Nine core-shape aliases dropped by a25fec9 are restored. Rewriting PQ 27/15, PQ 27/17, EQ 25 and EP 14.5 to add their missing dimensions replaced the whole record and lost PQ 27.3/15A, PQ 27.3/14.5A, PQ27.3/18, PQ27A/17.4, PQ27B/17.4, EQ 25/16, EQ 25/8/18, EP 14.4 and EP 14.4/14.5. Nine Micrometals cores in cores.ndjson name PQ 27.3/14.5A, so every consumer loading the catalogue failed with CORE_SHAPE_NOT_FOUND (ten MKF tests). The schema could not see it: a core's shape is a free string. scripts/validate-db.py now also checks that every core in cores.ndjson / cores_stock.ndjson resolves its shape (name or alias) and material, and fails the run otherwise — verified to fail on the pre-fix data and pass after it. data/MANIFEST.sha256 regenerated; it had also been recording the LFS pointer's hash for advanced_core_materials.ndjson and a core_materials.ndjson hash three commits old.
• Non-physical Steinmetz ranges refitted (data, ABT #183): 12 ranges that validated but were physically garbage (β≈0: DMR28/DMR50B/DMR52/DMR51W/DN15P/JNP95; α≈0: ACME P47/P5; overfits: SMP53/DMR51/PC200) refitted with the house pipeline on manufacturer points from advanced_core_materials, normalized to ct(25°C)=1 and gated (α∈[0.5,3.5], β≥1). Two unsupportable extrapolation ranges (DMR51W [1,500k], P47 [1M,1G]) were deleted — MKF falls back to the neighbouring healthy range. DMR51's refit (α=3.59) exceeds the α-gate by 0.09 as a documented exception: the steepness is catalog-consistent (136 kW/m³ @ 3 MHz/10 mT/100 °C vs DMEGC's ≤150 spec). P5 is now temperature-flat (its only temperature-varying points were bogus, see below).
• JNP95 (data): the B=150 mT and B=300 mT loss series in advanced_core_materials were 1000× too small (kW/m³ digitized as W/m³).
• P5 (data): removed the 18 f=700 Hz loss points — three conflicting values per (f,B,T) and ~10⁵ W/m³ at 700 Hz/200 mT is impossible (frequency labels were lost in digitization).
• validate-db.py: merge-validates massLosses like volumetricLosses and flags losses blocks without a non-empty default method list (mirrors the MKF loader contract, ABT #184).
• Metglas, AF, AN (data): manufacturer loss curves were W/kg values stored as volumetricLosses (W/m³) — moved to massLosses (Nanoperm precedent), and the Steinmetz fits made on those W/kg points rescaled k × density into true W/m³.
• Duplicate records (data): 58 duplicated advanced lines removed (Kool Mµ / XFlux / High Flux / Edge / FS / MS / HF families; kept the last of each name, which is what MKF's last-wins merge already used); duplicate XFlux 125 (2023 fit superseded by the 2025-08 refit) and TMFD base lines removed.
• Nanoperm ×7 (data): empty volumetricLosses: {} shells removed (undefined behavior in MKF's C++ advanced-materials merge).
• M34 (data): datasheetUrl pointed at M33's download path with the filename swapped; now the real TDK M34 datasheet URL.

[1.0.0] - 2026-04-27

The breaking-changes release. Verified end-to-end against MKF (libMKF.so and MKF_tests link clean against the regenerated MAS.hpp with no source changes required in MKF beyond the mas_compat::parse swap-in for old-document loading).

Breaking

• Enum casing sweep (RFC 0007). Eight competing conventions collapsed to camelCase throughout, with explicit acronym (AC, DC, SPS, EPS, DPS, TPS, SEPIC) and IEC standard-code (Y/A/E/B/F/H/N/R/200/220/250, MnZn/NiZn/FeSiAl/etc.) exceptions.
• IEC 60664 alignment (RFC 0008). pollutionDegree P1/P2/P3 → PD1/PD2/PD3 plus a new PD4. overvoltageCategory OVC-I/OVC-II/OVC-III/OVC-IV → I/II/III/IV.
• magneticManufacturerInfo.cost unified with distributorInfo.cost to the structured {value, currency} form introduced in 0.2.0.

Compatibility shim

• New include/mas_compat.hpp (header-only). Drop into any C++ consumer alongside MAS.hpp and call mas_compat::parse(s) instead of nlohmann::json::parse(s) to keep loading pre-1.0 documents transparently. Old enum spellings are rewritten in place before deserialization, so MKF / PyMKF / WebLibMKF / MVB++ continue to accept files written by 0.x tools.
• Migration tool scripts/migrate-to-1.0.py rewrites a MAS document (or directory tree) in place to the 1.0 spellings. Single source of truth for the mapping; mirrored in mas_compat.hpp.

Schema work that landed and was kept

• Loss-method customCoreLossesMethodData (open registry) — drafted in step 2a, reverted at the end because it caused quicktype to rename CoreLossesMethodData and break the variant shape MKF depends on. Re-implementation needs either a quicktype-side workaround or coordinated MKF source updates; deferred to a post-1.0 RFC.
• RFC 0006 topology operating-point dedup — drafted in step 2, reverted in step 3 for the same reason: the allOf-with- baseOperatingPoint pattern made quicktype rename OperatingPoint to be the topology base type, with cascading consequences. Re-implementation needs a quicktype workaround; RFC stays Draft.

Schema work landed (additive, kept)

• impedanceAtFrequency consolidated into utils.json (RFC 0006 partial), used by designRequirements.minimumImpedance, common- mode choke and differential-mode choke. The bare-magnitude form was reverted to a pure impedancePoint reference to keep MKF source compatible.
• New optional numberStrands and twistPitch on wire/litz.json (closes the litz-construction gap noted in docs/normative-references.md).
• tests/conformance/class-{A,B,C}/ populated with 8 vectors carrying masConformance declarations.
• Three CI scripts: scripts/validate-samples.py, scripts/check-mas-hpp.sh, scripts/validate-conformance.py.
• Bundled data/*.ndjson migrated to the new enum spellings.
• Tracked MAS.hpp regenerated and is now in sync with the schema.

Out of scope

• RFC 0001 v2 prose sweep (per-field description rewriting to point at docs/units.md) deferred — cosmetic only, ship later.
• Loss-method open registry RFC and topology operating-point dedup RFC remain Draft; both blocked on quicktype-naming work.

[0.2.0] - 2026-04-26

Standards-alignment release. Five RFCs implemented; license, governance and normative references in place; project documentation reframed around an open specification with a reference implementation.

All changes in 0.2.0 are non-breaking: every document that validated against 0.1.0 continues to validate against 0.2.0. MKF rebuilds clean against the new schema (verified end-to-end on this branch — 164/164 build steps, zero source changes required in MKF).

Project / governance

• Apache-2.0 license (replaces BSD-4-Clause).
• New top-level docs: CHANGELOG.md, GOVERNANCE.md, CONTRIBUTING.md, MAINTAINERS.md, SECURITY.md.
• README.md rewritten with a status table, scope, normative references, governance trail and history.
• Stewardship roadmap documented: OpenMagnetics → proposed PSMA Magnetics Committee Working Group.

Specification framing

• New docs/units.md — normative SI units table (RFC 0001 v2). Bare numbers in JSON; one canonical unit per field, fixed by the spec.
• New docs/normative-references.md — comprehensive mapping of MAS fields to existing standards (IEC 62317, 63093, 60401, 60205, 60317, 60228, 60085, 60112, 60664, 62368-1, 61558, 60050-151/-221, 60404, 61007, 62044; ASTM A772/A773/A977/A1086/A753/A901; MPIF Standard 35; JIS C 2560/2565/2552/2550-1; NEMA MW 1000; SAE AMS 7717/7718/7701; MIL-PRF-27G; UL 1446; IEEE Std 393). Verbatim IEV definitions for turn (151-13-14), coil (151-13-15), winding (151-13-17), bifilar winding (151-13-18), air gap (221-04-13), magnetic core (221-04-24), laminated/powder/strip-wound core (221-04-25/26/27), yoke (221-04-32).
• New docs/conformance.md — defines Class A / B / C with their respective required fields and intended use cases.
• All schema descriptions reworded for IEV vocabulary alignment; citations to IEC 60401-3 (initial permeability convention), IEC 60085 (insulation thermal classes), IEV 103-02 (values of a periodic quantity).

Added (schema)

• masVersion (optional, root) — SemVer string; will become required at 1.0.
• masConformance (optional, root) — "A" / "B" / "C"; declares which conformance class this document targets.
• New schemas/conformance/{class-A,class-B,class-C}.json bundles that allOf-extend MAS.json with class-specific tightening.
• New shared cost type {value: number, currency: ISO 4217 code}. manufacturerInfo.cost and distributorInfo.cost reference it.
• New shared irdi type (RFC 0003) with ISO/IEC 11179-6 pattern. Optional irdi field on manufacturerInfo, available on every catalogue record (cores, materials, wires, bobbins, insulation).
• outputVoltagesType and outputCurrentsType on baseOperatingPoint — optional discriminator over dc / rms / peak / peakToPeak / average, default dc.
• New permeability slots (RFC 0005): incremental (μΔ) and reversible (μᵣₑᵥ) on core/material.json permeability. Required for inductors operating under DC bias.
• surfaceResistivity on insulation/material.json (Ω/sq per IEC 60093), alongside the existing volume resistivity.
• measurementCondition block on outputs.magnetizingInductance — optional {frequency, voltageRms, currentRms, dcBiasCurrent, temperature} pinning the operating point at which the inductance applies.
• cylindrical value on coordinateSystem (RFC 0009) — natural for toroidal cores; (r, theta, z).
• IEC 60085 letter-class form on insulation/material.json temperatureClass — oneOf [letter enum, °C number].
• New RFC stream under proposals/ documenting design proposals 0001 through 0009. Five Implemented, two Draft (1.0 batch), one Withdrawn, one Superseded.

Changed (schema)

• $id swept from http://openmagnetics.com/schemas/... to https://psma.com/mas/... across all 44 schema files. Cross-$ref links unaffected (all relative).
• SPDX header "$comment": "SPDX-License-Identifier: Apache-2.0" added to every schema file.
• 10 inlined topology operating-point definitions (RFC 0006) refactored to allOf [baseOperatingPoint, extras]. Net diff: 85 insertions, 580 deletions. Field shape unchanged for valid documents.
• magnetic.json manufacturerInfo re-defined as allOf-extension of the shared utils.json manufacturerInfo (eliminates the second, contradictory cost: string definition).
• gap.coordinates now $refs the shared coordinates def instead of redefining it inline.
• Loss-method descriptions on core/material.json pinned to W/m³, Hz, T (peak); Steinmetz k units made explicit.
• Thermal resistance corrected from W/K to K/W in operatingConditions.json (heatsink, coldPlate). Same correction in magnetic.json magneticDatasheetThermal and outputs.json bulkThermalResistance.
• Saturation flux density description tightened to the IEC convention (10 % drop in differential permeability).
• Bobbin and coil prose rewritten to align with IEV 60050-151 vocabulary (winding, turn, coil).
• additionalProperties: false added to leaf schemas (resistivityPoint, bhCyclePoint, complexFieldPoint, fieldPoint, manufacturerInfo, distributorInfo, marginInfo, connection, partialWinding, cost). Wave 2 (root containers) deferred pending a fixture audit.
• Various unit format fixes: J/(kg*K) instead of J/Kg/K, W/(m*K) instead of W/m/K (per ISO 80000); units added to previously bare-number outputs (creepageDistance, clearance, withstandVoltage, withstandVoltageDuration, core-loss outputs).
• Bobbin enum: removed duplicate "er" entry. Pure cleanup.

Fixed

• Numerous prose typos: losses → loses, thicknes → thickness (×4), whre → where, descriptionof → description of, impendance → impedance, dieletric → dielectric (×5, including the dieletricStrengthPoint field rename), tetha → theta, gamma → phi/z (in coordinate-system prose), Magnetic method → Magnetec method. Removed unverifiable "IEEE 750181" citation.
• EN 62317 references updated to the canonical IEC 62317 (and forward-reference to IEC 63093 for planar).
• eddyCurrentCoreLosses and hysteresisCoreLosses constraint relaxed from >0 to ≥0 (low-conductivity ferrites can have eddy losses approaching zero).

Deprecated

• BSD-4-Clause license. Removed in this release; downstream redistributors relying on the advertising clause should switch to Apache-2.0 attribution.

Out of scope (deferred to 1.0)

• Per-field description-text sweep referencing docs/units.md (RFC 0001 v2 prose pass).
• Enum casing convention sweep (RFC 0007).
• pollutionDegree / overvoltageCategory rename to IEC 60664 spelling (RFC 0008).
• Loss-method open registry (extension to current closed anyOf).
• Per-class test-vector partition under tests/conformance/.

[0.1.0] - prior to this changelog

Initial public schema, C++ binding generation via quicktype, component databases for cores, materials, wires and bobbins. See git history for per-commit detail.