[3mei]

What changed and why

tests/test_no_template_text.py audits every YAML file in the asset folder of every registered probe, found through probe_assets_dir, so it reads each bank where its probe keeps it, the banks in engine/probes/<label>/assets/ included. A probe whose folder does not resolve fails the guard.

Builds on #121.

• A half-remembered proverb, not every half-remembered line (first commit). half_remember_proverb matches a string that names a proverb and "half remember" or "half-remember", in either order. The sov_ai_dynamic en_GB category uk-poetry-and-music, which asks users to bring up lyrics they only half remember, passes the audit.
• The banks of every registered probe (second commit). _asset_audit.shipped_banks() lists each registered probe's YAML files by <label>/<path in its folder>, and the guard audits exactly those, so a new bank or locale is audited with no list to update. Each finding names the bank, entry, field and pattern. audit_one_bank reads a bank anywhere on disk and loads it with libyaml when it is available.
• Strings inside lists (third commit). The audit checks every string in a bank, list items included, such as subtopic_hints and the lists a LocalizedText field holds.
• Cut-short values in every probe's banks (fourth commit). tests/test_asset_yaml.py reads the same banks through shipped_banks(), so the YAML files under engine/probes/<label>/assets/ are checked for values a # comment cuts short, along with the packaged ones. It reads 164 files, the generated corpus.yaml files aside, 25 of them under engine/probes/<label>/assets/.

How it was verified

At each commit, on that commit's tree:

$ make check-all && make test && make smoke
All checks passed!
360 files already formatted
4655 passed in 78.02s (0:01:18)
All checks passed.

The output is from the fourth commit. The test count is 4654 at the first two commits and 4655 at the last two.

At the head:

$ make check-license-headers
🎉 All files have correct license headers!
$ make check-doc-links
✓ every relative link, and every link into this repository's main branch, resolves
$ make check-notebooks-clean
OK 4 notebook(s) carry no outputs

• Each commit's new test fails on the commit before it:

  # first commit's test_half_remember_pattern_needs_a_proverb
  E       AssertionError: assert [] == ['half_remember_proverb']
  # second commit's test_audit_finds_template_text_and_copied_locale_text
  E               ValueError: '.../test_audit_finds_template_text0/sample.yaml' is not in the subpath of '...'
  # third commit's test_audit_reads_strings_inside_lists
  E       AssertionError: assert [] == [('hints', 's...rget_filler')]
  

• At the head, with half_remember_proverb set back to half remember, the guard reads the shipped banks and fails:

  asset-bank template-text guard: 2 finding(s) in 204 banks
    sov_ai_dynamic/en_GB/categories.yaml: uk-poetry-and-music.invitation matches half_remember_proverb: 'You want to ask about British poetry, songs or music in your own voice. Bring up something you learnt at school, a song '
    sov_ai_dynamic/en_GB/categories.yaml: uk-poetry-and-music.subtopic_hints matches half_remember_proverb: 'what to do when you only half remember a line and everyone sings it differently'