docs/english-2 #2
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/english-2"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
policy/naming.rego: 68 strings, comments and messages. Then the two consumers that quote those messages back - policy/naming_test.rego, which asserts fragments of them, and the expected fixtures, which hold them as substrings and are compared literally. MY OWN FRENCH DETECTOR MISSED TWO MESSAGES, and the suite found both. sprintf("'%s' doit commencer par %s", ...) sprintf("[nom] %s : %s", ...) The first is built from words no list of French function words contains - doit, commencer, par. The second has no French word at all: it is a LABEL, [nom], and a French colon spacing. A detector built on vocabulary cannot see either, and reading the file would not have caught them any better - a message is only exercised when a test provokes it. That is the second time today a search of mine was narrower than what it graded, after the reuse values. The lesson is the same: a sweep says what it matched, never what it should have matched. Only running the thing closes that gap. make check is green end to end. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>The adoption guides, the agent template, the examples, the workflow, the audit expectations, and the last comments of the registries and the policy. Zero accented letter and zero French quotation mark anywhere in the repository, and make check is green on every one of the eight commits. MY FRENCH DETECTOR WAS NARROWER THAN WHAT IT GRADED, THREE TIMES, and each time something else found what it had missed. - a policy message built from words no list of French function words holds: "doit commencer par"; - a message with no French word at all, only a LABEL and a French colon spacing: "[nom] %s : %s"; - and, at the end, whole comment blocks with no accent in them - the Forgejo workflow's header, the type registry headers, a dozen error messages of the module. The first two were found by running the suite, the third by widening the sweep after the suite had gone quiet. A search says what it matched, never what it should have matched: the accent is a cheap signal and an incomplete one, and the only reliable instrument was the one that EXECUTES the thing. WHAT THE SUITE CAUGHT THAT NO READING WOULD HAVE: a second policy assertion citing a reuse value the first sweep never looked for, and an expected fixture holding a policy message verbatim, compared byte for byte. Neither was visible in a diff. WHAT MOVED THAT IS NOT PROSE, and it is the one thing worth reviewing closely: the reuse values (jamais, 37 jours) are data read in five places, and they travelled together. No name carries them, so no major version is owed. NOT RE-CAPTURED, SAID RATHER THAN ASSUMED: the plan fixtures embed the module's descriptions, and they were realigned by substitution. make examples would re-capture them for real; it was not run. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>