Naur's compiler case tests one historically bounded documentation-and-consumption system
Type: kb/types/note.md · Tags: foundations, context-engineering
Peter Naur's compiler case reports a strong transfer failure. A motivated successor group received the full program text, annotated sources, extensive written design discussion, and personal advice, yet proposed extensions that the original group judged to be patches damaging the compiler's structure. The case therefore rules out a simple answer in which more prose of the same kind automatically transfers the program theory.
It does not test every possible artifact-and-interpreter arrangement. The supplied package and its use belonged to the documentation and knowledge-consumption technology of its time: people had to organize, find, select, and activate the relevant material through their own reading and work practices. The case did not test rationale linked to the decisions it affects, ADR-like records with explicit scope and status, machine-maintained indexes, semantic retrieval, dependency-aware context assembly, or automatic loading of the relevant record at the decision point.
These later mechanisms create an empirical possibility, not a rebuttal by definition. They may make previously supplied premises easier to find, preserve rejected alternatives and applicability conditions, or place relevant rationale where a person or LLM can use it. None holds a program theory by itself. The theory-holding claim belongs to the interpreter-plus-artifact composite and requires causal evidence that the retained state shapes coherent modification across novel demands, including implications not stated verbatim and recovery after delayed contradiction.
The appropriate comparison therefore varies both representation and consumption while holding source evidence and task demands fixed. Ordinary documentation, structured rationale with indexing and context assembly, and raw records from which an interpreter reconstructs a theory are different transfer pathways. Naur's case establishes failure of one historically important pathway. Whether a newer pathway transfers more of the required capacity, to people or to an LLM-based composite, remains an experimental question.
Relevant Notes:
- Programming as Theory Building — abstracted-from: supplies the compiler case, the package given to the successor group, and Naur's interpretation of its failure
- Naur binds program theory to humans by equating machine execution with formulated criteria — parallels: the human-only inference and the transfer evidence are both bounded by the computational and documentary technology being considered
- Holding a program theory means sustaining coherent search under delayed feedback — supplies: the longitudinal bearer test that newer transfer machinery must pass
- Design rationale must preserve decision premises its interpreter cannot regenerate — sharpens: what a rationale package must add beyond implementation and generic competence