Speech-to-code contracts (Oratio → MCP → compiler → MENS)
JSON Schemas in this directory define observability KPIs, failure taxonomy, project lexicon, and training trace shapes for the speech-to-Vox pipeline.
| File | Purpose |
|---|---|
kpi-baseline.schema.json | Baseline and snapshot metrics (WER/CER, compile-pass, latency). |
failure-taxonomy.schema.json | Tagged failure categories for telemetry and benchmarks. |
lexicon.schema.json | Project speech lexicon (identifiers, aliases, pronunciation hints). |
speech_trace.schema.json | Optional corpus / telemetry record for speech-origin sessions. |
speech_trace.mens.schema.json | MENS / SFT export: same fields as speech_trace plus required vox_code (and optional rating). mens/schemas/speech_to_code_trace.schema.json $refs this file. |
vox_grammar_artifact.json | Lexer keyword/punctuator surface for prompt hints only (no tokenizer mask today; validated by vox-compiler tests). |
labeling_rubric.md | Human QA / training labels for speech-to-code rows. |
canary_policy.example.json | Example KPI gates for speech canary promotion (optional CI via VOX_SPEECH_CANARY_KPI). |