Subject. A frozen implementation of true union semantics for restricted day-of-month/day-of-week cron fields under the # (nth weekday) and W (nearest weekday) special forms, a feature added to the widely-used croniter library (upstream commit pinned in harness/departments/croniter_fixtures/, MIT licensed, vendored so the refereed subject cannot drift).
Why this department exists. ROADMAP.md known gap #1: every earlier department was authored inside this repository, so "the protocol is domain-agnostic" rested on in-house evidence. This subject was born outside, a codebase this repository had never touched, and its calibration has an unusual provenance: an external referee suite was first authored blind by a party independent of the implementer, then proven against five planted mutants (5/5 caught, 2026-08-08). That external referee is not vendored here. The instruments below, oracle, claims, decoys, were re-authored by the implementing process in this repository's own format, and are accepted only because they reproduce the independent calibration's verdicts: the same five mutants, still all rejected, now through the unchanged harness/protocol.py. Who authored what: implementation and in-repo instruments by the implementer; the blind external referee and the five mutant classes by the independent party; the mutant diffs by the implementer to the independent party's specification.
The battery.
| Role | Instrument |
|---|---|
| rivals | the two semantic calibration mutants: an inverted forward combiner, and a union whose sides stay contaminated by each other's special form, same API, same claim, wrong dates |
| decoys | degenerate probe expressions (0 0 * * *), and the special forms stripped from the probes, the substance under test removed, shape kept |
| surrogates | fixed-stride and first-of-month date generators, schedules with no cron semantics |
| lesions | the other three calibration mutants as exact source patches, magnitudes measured at import as the share of a bounded behavioral fingerprint each changes |
The distinguishing reference claim is exact agreement with an independent calendar-arithmetic oracle that never calls the subject; the rejected reference claim is "accepts the API flag," true of every rival, presence of an API proves nothing about its semantics.
First command, the audit, which is parametrized over all departments:
.venv/bin/python -m pytest -q -o addopts='' tests/test_department_conformance.py tests/test_harness_croniter_department.pyHonest scope. This department measures the ingestion half of known gap #1: a foreign subject, refereed by the unchanged protocol, with independently-authored battery content. What it does not measure: adoption by an outside team, every step here was still orchestrated from this repository's own process. That axis stays open, and saying otherwise would be the exact promotion this tree exists to refuse.