Your LLM returned almost-valid JSON. Stop hand-patching it.

说明与用途

LLM 通常返回几乎有效的 JSON，但与预期模式不匹配。coerce-json 库自动清理、强制类型、填充默认值并记录更改，将验证成功率从 13.5% 提升至 86.5%（或开启 fuzzy 后为 100%）。

问题

LLMs often output JSON that is syntactically valid but semantically mismatched to the expected schema, causing runtime failures and manual patching.

结果

Coercing almost-valid LLM output with coerce-json increases validation success from 13.5% to 86.5% (or 100% with fuzzy matching) on a representative corpus, eliminating manual patching and providing an auditable change log.

工作流

• 1. Receive raw LLM output (often fenced JSON).
• 2. Pass the raw string to coerce from coerce-json along with a Zod schema.
• 3. coerce extracts JSON, removes fences, coerces primitive types, applies defaults, and records each change.
• 4. Inspect the changes log to audit modifications.
• 5. Use the coerced value for downstream processing or further validation.

证据

Every field is *almost* right. id is a string. active is a string. There's a role you never asked for.

来源

coerce is a zero-dependency library that takes almost-valid model output and makes it fit your schema — reporting every fix:

来源

Under plain Zod, 13.5% of them validate. Run them through coerce with defaults, and 86.5% validate. Flip on fuzzy: true and it's 100%.

来源