[zero-aska]

This is a follow-up to #3, which merged today. It does the two things I promised in the review thread there, and fixes the author list in bar.toml. Nothing outside tests/, README.md and bar.toml changes.

Planted bugs, now in the repo

The body of #3 gave counts for two planted bugs that I had run by hand, so nobody could trace them. tests/plant_bugs.py now runs them, with eight more. It writes tests/PLANTED-BUGS.md, in the same shape as hostname and punycode.

The script plants each bug as a one-line change to SemVer/Basic.lean in a copy of the project, then runs tests/run_tests.py --count 2000:

Planted bug	Failures
none	0
leading zeros are accepted in numbers	874
identifiers are compared by length only, not in ASCII order	338
a hyphen is not an identifier character	945
an empty build identifier is accepted	257
a numeric identifier comes after an alphanumeric one	309
a version with a pre-release part comes after the release	163
pre-release identifiers are compared the wrong way round	1471
build metadata takes part in precedence	47
numeric identifiers are compared as text	17
a hyphen in the build part starts the pre-release part	1418

All 10 are caught. The first two counts match the ones #3 stated, 874 and 338.

A run counts as a catch only if it finishes. A bug that does not build, a run that dies before it writes a record, and a run that gives up all count as missed. The rule about dying runs fixes a flaw that a review found in the same script in bech32: a run that died could reuse the old record. The fix also went into hostname and punycode. Each bug also breaks a proof. The tests build only the executable, so they still run.

Clearer labels in the record

As promised on #3, the input kind valid is now random version, and random is now random string. Before this, the body's "random versions" meant the record's valid. The counts are unchanged: 3,021,721 inputs, 0 failures.

The spec authors

bar.toml listed the spec authors as claude and fdn-0. The commits and pull requests come from the GitHub account zero-aska. So bar item 1's check would not have refused a spec approval from the account that wrote the spec. spec_authors now names zero-aska too. A self-review of bech32#2 found this, and the same fix went into hostname#2, punycode#2 and bech32#2.

Checks

• python3 tests/run_tests.py: 3,021,721 of 3,021,721 requests agree, 0 failures.
• python3 tests/plant_bugs.py: 10 of 10 caught.
• bar/check.py --offline --no-lean at the pinned mission commit: items 2, 4, 5 and 6 pass. Item 1 waits on a spec reviewer.
• bar.toml now lists tests/plant_bugs.py among the test files.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AovXVgJBHGiN44TxQM4dcd