What changed since July

The 0.8 readiness test: a real dataset end to end

• 95 BDF parquet files, IntelLiGent 101069765, CC-BY-4.0, six contributors by ORCID
• Datasets point at the existing Zenodo files; md5 verified 95/95
• The 11 known issues from the metadata sheet are encoded as conformance on the tests
• Every material spec now carries a lot; silicon-graphite blends are three distinct records
• Half-cell: working electrode + Li counter, typed HalfCellDevice

Parameter sets are records now live

• Each claim is exactly one of quantity, curve (OCP with hysteresis branch) or expression (BPX)
• Per-claim provenance class: measured › fitted › literature › derived › assumed
• Deterministic id from (target, scope, name), so re-import is idempotent
• GET /parameters/resolve?material_kind=graphite cascades spec › kind and reports what's missing
• The cell design tool now draws its OCPs from these records, DOI and license attached

Materials and electrodes became first-class

• Material kind is a curated vocabulary (key, family, formula, EMMO IRI, aliases), not a record
• Material spec is a record with deterministic id, contributor, license, funding
• Electrode has a kind, design values and its own emission; a cell links to its electrodes, a value links to the sample behind it
• Half-cell electrodes are named by role (working, counter), never by polarity
• Kind vocabulary topped up; cell_configuration added; test protocols emitted

Why it mattered: authoring the OCV set exposed that create_material_spec minted a random id per call, so re-runs silently accumulated duplicates. That was the H1 blocker for 0.8.

Proposal: BattINFO-viewer docs/internal/MATERIALS-MODEL-PROPOSAL-2026-08-03.md

Collections: one citable thing over many datasets

• series_id on a dataset; emitted as dcat:inSeries and schema:isPartOf, type gains dcat:DatasetSeries
• Series-flavored datasets are exempt from the missing-cell-link rule
• Registry renders the forward "Series" link and the reverse "Used by datasets" panel with no new code (generic *_id rule)
• Next: mint the Flores collection record anchored on the deposit DOI, then corpus v5

Three releases and the duplicate cleanup

• chemical-substance 0.15.0, electrochemistry 0.37.2, domain-battery 0.20.2
• 17 duplicate labels resolved. Rule: foundational domain wins cross-domain, best-documented wins same-domain
• 9 quantity classes fleshed out (ChargeRecovery pairs with ChargeRetention, IEV 482-03-35)
• Electrode chemistry classes un-deprecated: they were defined classes still doing inference work, deprecation only removed query vocabulary
• 6578 BattINFO references re-verified against the new closure

Open defect worth saying out loud: ElectrochemicalHalfCell is a subclass of ElectrochemicalCell, but the parent is elucidated as two electrodes. We type half-cells as HalfCellDevice until that's fixed.

Also: electrode roles Working/Counter/Reference are not disjoint, so a dual-role Li counter-reference is expressible.

Every predicate a record carries is now defined

• Hosted, generated, versioned records context; records can reference it by URL instead of inlining
• Audit: 121 of 168 distinct predicates in emitted artifacts had no defining subject (43% of occurrences)
• Retiring the served ns# namespace: every canonical key mapped; 426 preview artifacts re-rendered with zero ns# IRIs PRs open
• battinfo.org/validate: JSON-LD workbench with summary, table, tree and graph views plus three validation layers, fully offline

Hardening, browsing by kind, richer record pages

• Two review rounds (red-team 28 Jul, blind five-persona 31 Jul) came back green; fixes to silent-loss packaging, 409 identity conflicts, rehydration, record-level license, long path names
• Kind pages and linked-resource panels: browse a material kind and see its parameter sets and cells
• Record dossier merged on the registry; platform record console (KPIs, QR, related records, ORCID contributors, print-to-datasheet) still in PR
• Equipment: MC3000 chargers published as spec + units + channels, BLE MAC as identity

Two rulings we want input on today: who may propose a correction (any verified ORCID on the record?), and whether a maintainer self-edit should still create a visible change record.