stzOutputSchema

The declaration is compiled -- and judged -- HERE. Nothing about a schema is discovered at call time; a defective declaration never reaches a model.no extension

readsno extension

(no comment in the source: the name reads as a sentence) set nameQCSXTZZZIBWU

(no comment in the source: the name reads as a sentence) fieldsno extension

(no comment in the source: the name reads as a sentence) number of fieldsno extension

(no comment in the source: the name reads as a sentence) field namesno extension

(no comment in the source: the name reads as a sentence) required field namesno extension

(no comment in the source: the name reads as a sentence) has fieldno extension

(no comment in the source: the name reads as a sentence) fieldno extension

(no comment in the source: the name reads as a sentence) refuses unknown fieldsno extension

the one knob ------------------------------------------------ Closed-world: a field nobody declared REFUSES the answer instead of being reported and dropped.QCSXTZZZIBWU

(no comment in the source: the name reads as a sentence) allow unknown fieldsQCSXTZZZIBWU

the court --------------------------------------------------- An ALREADY-PARSED value against the declaration. [ :ok, :value, :findings ]no extension

The family's rule face: run me, hand me back the findings.no extension

(no comment in the source: the name reads as a sentence) holdsno extension

(no comment in the source: the name reads as a sentence) number of findingsno extension

RAW MODEL TEXT to a validated whole, or to a refusal that names the rule it failed. [ :ok, :value, :findings, :shape ]no extension

(no comment in the source: the name reads as a sentence) acceptsno extension

goldensno extension

(no comment in the source: the name reads as a sentence) diffno extension

saying it out loud ------------------------------------------ One sentence per finding, the rule named in every one. Long lists are cut, and the cut is stated rather than silent.no extension

THE CLAUSE IS SHORT BECAUSE A LONGER ONE MEASURED WORSE, and that is the only reason. Against the shipped 135M model, ten structured prompts validated 2/10 with the wording below. Adding two instructions aimed at the observed failures -- "do not explain the structure", and "replace every <...> with a real value" -- took it to 0/10: a small model told not to explain explained more, and the negation was the thing it echoed. Reverted on the measurement. Do not re-add instructions here without re-running base/test/neural/_measure_structured.ring; taste is not evidence.no extension

The GBNF text for this declaration, or a raise naming the construct that cannot be expressed. Nested structures are REFUSED rather than flattened: a grammar that accepted what this schema rejects would put the two layers into disagreement.no extension

TRUE when this declaration can be expressed as a grammar at all.no extension

One line per constraint the grammar does NOT carry. Empty means the grammar carries everything this schema declared.no extension

Does anything actually CONSTRAIN DECODING with this grammar? Ask before reporting an answer as grammar-constrained. It answered 0 for as long as that was true; gbnf_machine.zig is the rung underneath it, so it answers 1 since 2026-08-20 -- the ENGINE is the authority on that number, never this comment.no extension

(no comment in the source: the name reads as a sentence) decoding statusno extension

The declaration read back as prose, for a narrated test or a reader who wants to see what was actually promised.no extension