aggregate-codify-proposals

To aggregate YAML artifacts into a master proposals index, you recursively scan the inference output tree, parse strict metadata, validate schemas against taxonomy tiers, and compose a single proposals.yaml file with deterministic proposal_id generation from content hashes.

Deterministic proposal aggregation generates stable proposal_id values from content hashes during YAML parsing. It enforces strict failure modes for required metadata fields, confidence tiers, and tier/path alignment to prevent silent data loss during the composition process.

You classify inferred proposal artifacts into a two-level learning taxonomy by applying schema validation against taxonomy and tier expectations. The process assigns both learning_category and sub_category labels to each proposal within the master proposals.yaml index.

No, this proposal aggregation pipeline requires an existing STM evidence tree. It applies to codify-phase workflows where many infer-*-from-code skills emit per-target proposal files under a shared STM evidence output path for recursive directory walking.

The best way to prepare inferred artifacts for product LTM promotion is to validate metadata fields, classify proposals by learning category, and output a strict master proposals.yaml index. Downstream systems like the enrich process then reliably consume this exact artifact.

Taxonomy classification fails during proposal aggregation when strict schema validation detects missing required metadata fields, misaligned confidence tiers, or incorrect tier/path alignment. These strict failure modes intentionally halt the process to prevent silent data loss.