[derek73]

Opens the 3.1 cycle after 3.0.0 (release checklist steps 8–9), in the same shape as #936, and fixes the comparison harness defect that step 8's check turned up.

• Version. VERSION = (3, 1, 0) and PRE_RELEASE = "dev", so the tree reports 3.1.0dev, which normalizes to 3.1.0.dev0: above 3.0.0 and below 3.1.0. The _version.py comments name the new versions.
• Release log. A * 3.1.0 - Unreleased entry says 3.1 refuses a Lexicon, Parser or locale pack pickled by 2.x, which 3.0 loads with a DeprecationWarning (the 3.0 deprecation bullet).
• CITATION.cff stays at 3.0.0: it names the last released version.
• Baseline. compare.py's DEFAULT_BASELINE is now "3.0.0".

The harness fix. The first bare compare.py against the 3.0.0 wheel reported 669 of 7,107 entries moved, in a tree whose nameparser/ is the release's plus the version bump. The baseline worker recorded the seven fields, the reports, initials() and the case views, but not the generation, credentials, honorifics, patronymic and surname2 views snapshot.read() records since 3.0, so every name filling one read as moved. Against 2.8.0, the baseline of the whole 3.0 cycle, that was invisible: no 2.x wheel has those views, so each move was a real new key.

• The worker reads each view as the tree side does (getattr(p, key, "")) and its version tell names the views the wheel's ParsedName has.
• unrecorded() leaves out the views a baseline lacks, as it already did the alternatives before 2.7: a name the tree fills a new view for is not a name that moved. --baseline 2.8.0 now reports 1,155 entries where the old harness reported 1,669 (2026-11-08); no committed figure was taken from a bare 2.8.0 run.
• test_the_worker_views_are_every_view_a_snapshot_records ties the worker's views to snapshot.RECORDED_LATER, so a view added to the snapshot and not the worker fails there. With the pre-fix compare.py, six of the new and changed tests fail; with the fix all pass.

Checks. Bare compare.py: 0 of 7,107 entries read differently from 3.0.0. The full suite (40,191 passed), the serial benchmark tests (60), the HTML docs build, mypy and ruff pass.

Reviews.

• Code review of the harness fix and the cycle bump: nothing blocking. It rendered the worker for the 1.4.0, 2.8.0 and 3.0.0 wheels and checked each tell and row against snapshot.read(), checked unrecorded() for each kind of tell, ran the new tests against the pre-fix compare.py, and reproduced every figure above (0, 1,155, 1,669, 669). One point taken: the three view rows now assert they still reach their view, as the case-view rows do (fix commit, checked by listing each row's views and a name without one). Two left as they are: the 3.1 release-log line states the cycle's plan, as Open the 3.0 cycle after 2.8.0 #936's did for 3.0, and the assert in unrecorded() reads a tell our own worker writes.

🤖 Generated with Claude Code