Guide: the conjecture funnel, and its hit rate

For you if you want to generate leads automatically — and, more usefully, to see what fraction of them survive.

First command:

.venv/bin/python scripts/13_discovery_run.py --dry-run
.venv/bin/python scripts/13_discovery_run.py --report

The premise

Most numerical "discoveries" are already known or trivial, and a system that does not measure its own hit rate is measuring its operator's enthusiasm.

So the funnel is built to make the unflattering outcomes cheap to record and impossible to skip:

• already-known is a first-class terminal state with its own detector, run before any screen, because "this is in the literature" is the most likely true answer and the cheapest to obtain;
• survives cannot be claimed without naming the three checks that ran;
• a refutation is not accepted unless it survived a precision increase.

The seam, and why it is strict

ontology/schema.py, funnel.py, metrics.py and registry.py contain no subject-matter knowledge whatsoever. They would work unchanged for a chemistry lab or a compiler-optimisation search. Everything that knows the subject lives in ontology/domains/.

Four tests enforce it: an AST import scan, a clean-interpreter sys.modules check, a lexical scan for subject vocabulary, and a subprocess that runs the whole pipeline with the laboratory made unimportable and still requires it to generate, screen and report.

If the seam leaks the design is worthless — a bookkeeping layer entangled with its subject cannot be trusted to report a number its operator does not want.

Read ontology/README.md before touching any of it.

The ledger is private, on purpose

The funnel writes to conjectures/, which is gitignored. It is a notebook of unreviewed leads, and by the package's own premise most of them are known, trivial or wrong. A list nobody has checked, published in a repository that is otherwise checked, would be read as a set of claims.

Publish ontology.metrics.render_text. Never the log. An empty conjectures/ in a fresh clone is the rule working, not a bug.

Related: refute.md — the funnel generates leads, the battery refutes them.