Changelog

Unreleased platform integration

• Added ten authenticated Agent Gateway tools for standard/profile discovery, validation, comparison, resolution, report and plan retrieval, and explicit lossy/inference approval.
• Added compatibility:read, compatibility:compute, and compatibility:approve OAuth scopes.
• Compatibility compute can use a scoped Developer API key or a normal signed-in Biosimulant session. Private reports and plans stay with their owner.
• Managed-run preparation can bind an exact compatibility plan ID and digest; the server revalidates the bundle, plan, workspace revision, policy, expiry, and approval receipts.
• Added the optional biosimulant-model-compatibility Agent Skill and routed the general, composition, model-building, runtime-specific, discovery, and publication skills to the same workflow.
• Added an administrator review and revocation screen for adapter and inference releases.

Unreleased standard changes

These docs pin the exact alpha package and bundle shown on the standard overview.

• Python and TypeScript now implement the full rule vocabulary, profile inheritance, locks, deterministic path planning, safety limits and digest-pinned ontology or mapping snapshots.
• The browser-safe entry point includes all public schemas and 650 profiles for offline validation in Web and Desktop.
• The build now calculates review progress and ga_ready from checked, per-profile review records. It does not treat passing tests as scientific approval.
• Cross-language tests compare normalized contracts, reports, plans and hashes byte for byte.
• Reworded the standard and findings in plain English, while keeping stable profile names, URLs, statuses and reason codes.

0.1.0-alpha.2

• Fixed the capitalization of the Biosimulant brand name in package metadata, docs and the port contract schema title.

0.1.0-alpha.1 — implementation draft

The first public alpha of the Biosimulant Model Compatibility Standard. You can use it end to end today, but formats may change before a stable release.

The standard

• 15 JSON schemas, covering manifests, port contracts, profiles, rules, reports, plans, locks and conformance records.
• 650 profiles across 26 domains, built from 266 contract items grouped into 30 packs.
• 18 comparison operators, 8 comparison statuses and stable BMCS_* reason codes.
• Three test cases for every profile, 1,950 in total: a valid contract, an invalid contract, and a comparison that must return UNKNOWN.
• Python and TypeScript reference implementations for validation, hashing and comparison. The Python implementation can also build plans.

Biosimulant CLI

• New biosimulant compatibility commands: validate, normalize, compare, profiles, lock, plan and conformance. Install them with pip install 'biosimulant[compatibility]'.
• Model packages built from a model.yaml with a compatibility block now include payload/compatibility.lock.json.
• A package build fails if the compatibility block is invalid. Loading a model fails if the ports in model.yaml don’t match the ports in its Python code.
• Port contracts are kept when manifests are read and written, and values sent between ports during a run can be checked against their contracts.

Biosimulant platform

• Public endpoints that serve the standard’s schemas, profiles and test fixtures.
• Developer API endpoints to validate, compare and resolve contracts. Reports and plans are stored privately for their owner, and plans that need approval can be approved.
• Runs started from a released lab can be linked to a stored plan. The run’s Passport records the plan, its policy decision and any approvals.
• Ports of released models are indexed for compatibility.
• The Hub shows whether a model declares compatibility, how many of its ports have contracts, which profiles it imports, and each port’s contract.

Documentation

• New pages for the standard, including a searchable catalogue of all 650 profiles and an updated model.yaml reference.

Known limitations

• No profile has been scientifically reviewed yet: 591 are candidate and 59 are draft.
• Ontology and identifier mapping rules return UNKNOWN unless the exact digest-pinned snapshot needed by the rule is supplied.
• A profile being machine-checkable does not mean its field choices or biological limits have been independently reviewed.
• A stable release needs independent scientific review of every profile. See Before a stable release.