[mayur-dhavan]

Fixes #423.

Problem

metric() absorbs a rounding carry (9999 -> 10.0 k) by bumping the exponent, but the carry check was guarded by exponent < 30, so at the quetta boundary the mantissa gained a digit instead:

>>> humanize.metric(9.996e30)
'10.00 Q'    # before -> '10.0 Q' after
>>> humanize.metric(9.996e31)
'100.0 Q'    # before -> '100 Q' after
>>> humanize.metric(9.996e32)
'1000 Q'     # before -> '1.00 x 10³³' after (no prefix above quetta)

Fix

Drop the exponent < 30 guard so the carry logic applies to every bucket, and when the bump pushes the exponent to 33 (past quetta, where no larger SI prefix exists), fall back to scientific(), matching the documented behavior for values beyond the prefix range. The negative-exponent side (quecto) already worked and is covered by a new regression case.

Tests

Five new parametrized cases in test_metric: 9.996e30 -> 10.0 Q, 9.996e31 -> 100 Q, 9.996e32 -> 1.00 x 10³³, 9.999e-30 -> 10.0 q, -9.996e31 -> -100 Q. tests/test_number.py passes locally (246 passed) and the full suite passes except the pre-existing test_benchmarks.py collection errors, which require the optional pytest-benchmark plugin and fail identically on unmodified master.

AI disclosure: this contribution was prepared with AI assistance (an automated coding agent acting on behalf of the author); the change and tests were validated locally against the project's test suite as described above.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

[mayur-dhavan]

Closing this as a duplicate of #424 (Oct 1), which fixes #423 with the same boundary handling and identical expected outputs (10.0 Q, 100 Q, 1.00 x 10³³). I missed it when checking for existing work. Sorry for the noise.