Concept note · the lab vocabulary
Code provenance
Whether code is what it claims to be — and proving it with an audit trail.
What it is
Whether a piece of code is what it claims to be. Not whether it runs — counterfeit code runs; that is the entire reason it survives review. Provenance asks the examiner’s questions: does this implementation actually do the work its name and docstring promise, do its inputs influence its outputs, and can every verdict about it be traced afterward? In an era when code is produced faster than any human can read it, provenance is the difference between a codebase you own and one you merely occupy.
In the lab
- CodeRadar →
- The working instrument: three daemons run 97+ discrete rules over the four invariants, at file-save speed, and a live circuit graph makes agent damage visible in seconds.
Our position
Provenance must be provable, not felt. A reviewer’s intuition does not scale to agent speed, and asking a second model to grade the first only moves the trust problem. Our position: legitimacy reduces to mathematical invariants — inputs influence outputs, computation is captured, iteration accumulates, functions produce — and violating any one is sufficient proof of counterfeit. Verdicts are binary, checks are pure predicates on the syntax tree, and every repair writes a before/after to a state mirror, so the audit trail outlives the fix. Deterministic, local, offline — no API, no tokens, no model-version drift.
Honest limitations
Structural invariants catch code that fails to do real work — they do not catch code that does the wrong real work. A correct-looking implementation of the wrong business rule passes every check, so provenance bounds review; it does not replace it. The rule set is AST-level and language-specific, and its statistical bounds are calibrated against known-good code — a codebase with unusual conventions will need its profile relearned before the outlier checks mean anything.
Related in the vocabulary
- Instrumentation →
- Meets it in CodeRadar — building the instrument that makes a hidden state legible.
The experiments are the proof
Every claim in this note is made concrete somewhere in the lab — the idea exists because the machines needed it.
The map of the lab →