Skip to content

Latest commit

 

History

History
41 lines (33 loc) · 1.64 KB

File metadata and controls

41 lines (33 loc) · 1.64 KB

Guidance for Claude Code / coding agents

What this repository is

42-True is the first public concept of the Large Meaning Model (LMM) described in paper/42-True_LMM_paper.pdf. It is a schema and concept seed, not a training pipeline. The corpus it is defined against does not exist yet and, by design, cannot be scraped — it must be produced. Keep that framing intact in any docs or code you write.

Layout

  • forty_two_true/intent_pair.py — the primitive: the frozen four-tuple ( Id, C, M, O ).
  • forty_two_true/schema.pyClassificationTier, ordered OutcomeGrade, TaxonomyNode.
  • forty_two_true/recipes/ — conceptual declare → classify → match → verify stages. These are reference demonstrations, not real services.
  • data/example_intent_pairs.jsonl — synthetic, clearly-labelled examples.
  • tests/ — schema invariants and example-data validation.

Invariants — do not break

  1. Immutability. All Pydantic models are frozen=True. Construct new records; never mutate in place.
  2. No real declared data. Any example records must be synthetic and labelled as such. Genuine declarations are gathered only under the consented, unobservable protocol — never committed here.
  3. Outcomes are earned, not asserted. A VERIFIED_RESOLUTION outcome must carry a proof_ref; observe_outcome enforces this.
  4. Honesty about status. Don't describe the recipes as a working production system. They run on the example data; the network described in the paper is future work.

Working on it

pip install -e ".[dev]"
pytest
ruff check .