How nolane-x/nolane-hira works
Nolane HIRA is a research-first, non-autoregressive typed decision engine. Its documented workflow encodes a state once, compiles or reuses a decision schema, scores all logical options while preserving full-option probability mass, and refines uncertain options with a relation core. OOD authority and held-out calibration are explicit safeguards. The tree also shows research branches for typed, semantic, compositional, and transfer decisions, alongside structural and trust-region methods. Runtime wiring beyond the README’s stated design is uncertain; those links are omitted where the excerpts do not establish callers.
Main entry points: runtime.py Decision runtime · semantic.py Semantic encoding · schema.py Decision schemas · contracts.py Option contracts
Components 17
Decision workflow
- Decision runtime
src/nmd/runtime.py - Decision schemas
src/nmd/schema.py - Typed decisions
src/nmd/typed_decisions.py - Option contracts
src/nmd/contracts.py
Decision model
- Semantic encoding
src/nmd/semantic.py - Coarse scoring
src/nmd/sparse.py - Option competition
src/nmd/competitive.py - Relation core
src/nmd/relation_cache.py
Reliability controls
- OOD authority
src/nmd/ood.py - Calibration
src/nmd/calibration.py
Research capabilities
- Factorized decisions
src/nmd/factorized.py - Semantic alignment
src/nmd/semantic_alignment_authority.py - Semantic transfer
src/nmd/semantic_transfer_core.py - Latent composition
src/nmd/latent_composition_authority.py - Structural methods
src/nmd/structural.py - Functional trust
src/nmd/functional_trust.py
Other components
- Requesting client
Connections 14
- Requesting client → Decision runtime: submits request
- Decision runtime → Decision schemas: compiles or reuses
- Decision runtime → Semantic encoding: encodes state
- Decision runtime → Coarse scoring: scores options
- Decision runtime → Option competition: preserves option mass
- Decision runtime → Relation core: refines uncertainty
- Decision runtime → OOD authority: checks OOD
- Decision runtime → Calibration: applies calibration
- Decision runtime → Typed decisions: returns decisions
- Option contracts → Decision schemas: defines options
- Semantic alignment → Semantic encoding: aligns representations
- Semantic transfer → Semantic encoding: transfers representations
- Latent composition → Factorized decisions: composes decisions
- Structural methods → Functional trust: supports repair research