Open the 2.2 cycle's version: 2.2.0dev

[derek73]

The tree has said 2.1.0 all cycle. Two costs, one of which prompted this and one which is arguably the bigger.

Read the Docs "latest" was stuck on the last release

docs/conf.py:66 reads nameparser.__version__ directly, and .readthedocs.yaml installs path: . from the default branch — so "latest" rendered the version we shipped rather than the one we are building.

The differential harness's version tell was running on one cylinder

The tell exists so a worker that silently resolved to the checkout cannot pass as the baseline. With the tree and the 2.1.0 baseline wheel both reporting 2.1.0, only the __file__ path told them apart — which is exactly the weak state test_tell_rejects_a_module_loaded_from_the_checkout records in its own docstring:

"when the tree and the baseline share a version the version half of the tell agrees and only the path gives it away"

That is the condition the worker_v1.py invocation trap exploited. A bare run now reads:

tree:     nameparser 2.2.0dev (.../nameparser/__init__.py)
baseline: nameparser 2.1.0    (.../uv/environments-v2/baseline-worker-.../nameparser/__init__.py)
corpus: 751 names; intentional diffs: 0; unexplained: 0

Both halves discriminate again.

dev, not a bare bump

PRE_RELEASE = "dev" rather than VERSION = (2, 2, 0) with PRE_RELEASE = "", because master must not claim to be a release that does not exist on PyPI. Verified:

2.2.0dev  ->  normalizes 2.2.0.dev0   > 2.1.0 ✓   < 2.2.0 ✓   is_prerelease ✓

So pip install git+… gets something that sorts below any real 2.2.0.

Known cost

VERSION is a pure numeric tuple by design — its docstring invites nameparser.VERSION >= (2, 0, 0) — so it cannot carry the dev marker, and VERSION >= (2, 2, 0) is now true while 2.2.0 is unreleased. The constant says so and points at __version__ for callers who need the distinction. Nothing in-repo gates on the tuple.

CITATION.cff deliberately holds at 2.1.0

It names the last citable released artifact, so it is the one version string that does not track the tree. It lags for the whole cycle and catches up at release. Nothing enforces that — no test, no CI — so the checklist now states it rather than implying the two are kept equal.

Checklist

Release step 2 becomes "clear PRE_RELEASE" instead of "bump VERSION", and a new step 9 opens the next cycle's version next to step 8's ledger. Both are cycle-open work, and pairing them is what stops the tree and the docs drifting apart again.

Full suite 3096 passed; mypy and ruff clean; harness clean at 0 unexplained.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.50%. Comparing base (d910709) to head (e30c1cd).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #362   +/-   ##
=======================================
  Coverage   98.50%   98.50%           
=======================================
  Files          44       44           
  Lines        2883     2883           
=======================================
  Hits         2840     2840           
  Misses         43       43           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.