Cloning into '/tmp/simpleenglish-functional'...
== repo's own unit tests ==
----------------------------------------------------------------------
Ran 7 tests in 0.003s

OK

== linter self-test (built-in fixtures) ==
self-test OK: 12 violations in slop fixture, 0 in clean

== fresh-input probes ==
-- slop text (README's real before example):
violations_total: 7 | per_100w: 16.28 | longest sentence: 26 words
-- clean STE rewrite:
violations_total: 0 | per_100w: 0.0 | longest sentence: 12 words
-- edge cases (code-span-as-one-word + trailing conditions):
sentence_over_limit: 0 (9-flag command counted as 8 words) | trailing_condition: 2

== independent reproduction of the headline number ==
model                      base/100w skill/100w reduction
claude-opus-4-5-20251101        2.55       0.57     77.6%
claude-opus-4-6                 2.24       0.40     82.3%
claude-opus-4-7                 2.28       0.42     81.7%
claude-opus-4-8                 1.05       0.62     41.3%
claude-opus-5                   2.13       0.32     85.2%
claude-sonnet-4-6               2.06       0.52     74.9%
claude-sonnet-5                 2.67       0.53     80.0%

raw files scanned: 168; models with both conditions: 7
mean reduction across models: 74.7%  (README claims 74.6%)