Schema Guard: deterministic repair for

Schema Guard: deterministic repair for structured AI output

Independent Python project for validating almost-JSON output before downstream software uses it. The repair ladder handles Markdown fences, trailing commas, Python literals, quoting errors, and related formatting failures, then validates against a schema and records repairs and coercions. On the included synthetic fixtures, parse recovery rises from about 8% to 96%; this is a fixture benchmark, not a claim about every model or real customer workload. Truncation remains a documented limitation. The project includes 38 tests and reproducible results.