Handbook
intelligence_math_knowledge_ingest_v1 v1
Purpose: Convert bounded source slices into partitioned MathGenesis candidate objects without self-validation or promotion.
Updated
Inputs
Required: run_mode, target_partition, source_bundle, existing_library_index, clean_room_policy, prompt_pack_version.
Outputs
Required: batch, route_trace, status.
Substrate reuse
- Generic tasks:
extract_schema_from_text,contradiction_scan,decompose_problem,llm_enum_route,llm_multi_label,plan_decision_pack - Operators:
seq,branch,fallback_chain,for_each,try_catch,until_ok
Verification policy
- input schema validation
- partition guard
- native object schema validation
- provenance present
- no direct promoted_discovery write
Failure modes
- missing source metadata
- parse failure
- schema failure
- partition violation
- contamination risk
- unresolved source ambiguity
Runtime status
Proposal only. Repository-native implementation, fixtures, validators, maturity registration, and allowlist entry are required.