The suite as it stands
Validation. A pure-Go implementation of the XML Schema subset ISO 20022 uses. Across the whole catalogue it agrees with libxml2 on 4,746 of 4,746 documents, accepting the same 1,035 and rejecting the same 3,711.
Linting. Checks that need no schema at all: IBAN checksums, BIC structure, currency precision, UETR format and date order. Every finding names the rule, the path, what was expected and what to change.
Scheme rules. Focused profiles that sit above the schema. They cover structural checks, the live CBPR+ SR2025 message set and its cross-message rules, structured-address readiness, enhanced-data expectations, investigations and verification of payee. Exact per-message cardinalities and patterns still come from the operator's Usage Guideline schema. They are not guessed from a base ISO 20022 schema.
Conversion. Seven SWIFT MT types convert to ISO 20022 and back, each with a fidelity report describing what was mapped, derived, shortened or lost.
Generation. A valid sample message for any definition, constructed from a template where one exists and from the schema where none does.
Reference. All 2,845 message definitions indexed offline, with every version and where to obtain the schema.
Agents. Thirteen MCP servers that expose the same work to an AI assistant: initiate, settle, read a statement, reconcile it, remediate an address, and seal the result into an evidence pack an auditor can verify.
Where it runs
The same engine compiled five ways: a command line tool, a terminal interface, a language server for editors, a Model Context Protocol server for AI assistants, and WebAssembly for this website. One engine means one verdict — a message linted here gets the answer it would get in a terminal.
What is being worked on
More scheme profiles. CBPR+ enhanced data is partly covered. The market infrastructure profiles differ enough from one another to be worth separating.
Deeper remediation. Findings currently explain what to change. Where a correction can be derived without guessing, as with IBAN check digits, AskISO now provides it. Extending that safely is the interesting problem.
Broader conversion. Seven MT types are supported in both directions. The constraint is verification rather than effort: a mapping nobody can check against a published source is one this project will not ship.
How priorities are decided
By what fails in production. A message AskISO handles incorrectly goes to the front of the queue. A validator that is wrong is worse than one that is merely incomplete.
Everything else is discussed in the open. The issue tracker is the roadmap, and the reasoning behind a decision sits beside it.
What will not be built
Anything that requires guesswork. Rewriting an unstructured address means deciding which line is the town and which the country. A tool that decides wrongly on a payment has caused real harm. Where correctness cannot be established, AskISO reports the gap and stops.
Why any of this
Vision and mission sets out what this project is working towards, and the circumstances that would make it unnecessary.