Strict-mode diagnostic ordering is now deterministic for unknown properties

[davidahmann]

Problem observed

Strict-mode schema failures that include multiple unknown properties were producing unstable message ordering in parser diagnostics. The result looked functionally equivalent but changed line-by-line across equivalent runs, which made snapshot-style checks and reviews noisy. Teams using strict mode as a gate spent time re-evaluating harmless output churn instead of actual behavioral changes.

Why it matters operationally

When a safety gate emits unstable diagnostics, operators lose confidence in diff signal quality. That affects triage velocity and can hide real regressions among formatting-only noise. Stable warning/error ordering is a low-level reliability property for deterministic compilers, and strict mode adoption increases pressure on reproducible outputs in both CI and human review workflows.

Minimal repro

Use a fixture with multiple unknown fields in one schema failure path and compare normalized unknown-property text between runs.

Fix approach

The parser’s additional-properties error rewrite now normalizes extracted property names into lexical order before rendering “Unknown properties: …”. This preserves semantics while making output deterministic regardless of incoming property-list order. Regression coverage was added around both singular and multi-property cases so the ordering behavior is locked at the parser boundary.

Validation evidence

• passed.
• Formatting checks () passed.
• Full gates surfaced unrelated baseline failures ( and workflow git-patch tests), not in changed parser files.

Open follow-up question for maintainers

Should this deterministic ordering rule also be applied to any other user-facing schema-list diagnostics beyond additional-properties rewriting?

This contribution was informed by patterns from Wrkr. Wrkr scans your GitHub repo and evaluates every AI dev tool configuration against policy.: https://github.com/Clyra-AI/wrkr