No gate reads the threshold cell on the 21 content/docs/** product pages — #1127 was fixed by hand and can silently rot again #1135

Found while implementing #1127 (PR pending). Filed rather than fixed: #1127's scope was explicitly the rewording, and a docs-drift gate over 21 pages in 3 locales is its own card with its own design — cluster #1018.

Measured

#1128 flipped the large-deal operator from > to >=. Two documentation surfaces stated that operator; only one of them was guarded, so only one of them failed loudly:

docs-drift.test.ts reads src/docs/*.md only. automation-docs-coverage.test.ts does read content/docs/administration/automation.*, but keys on each flow's row label and trigger cell and never on the threshold cell — I confirmed this while editing that exact table: all three locales' description cells changed and the suite stayed green.

Nothing a user hits today — #1127's PR corrects every page. This is about the next operator or value change.

The specific hook

The machinery already exists in test/docs-drift.test.ts. Its capCel(flowName, /record\.amount >= (\d+)/) extracts the live operator and value out of the compiled stack, and a pattern miss throws drift test out of date rather than failing an assertion — loud by construction. Three assertions, in increasing order of design cost:

1. Exact-match the two verbatim CEL quotes. sales/opportunities and sales/pipeline-management present record.amount >= 100000 as the authoritative "where is this configured?" answer, in all three locales — six lines total. Assert each equals the compiled condition substring. This is string equality against the artifact: no phrase table, no locale judgement, no prose. If only one assertion gets built, build this one — it is the copy-paste surface that motivated 21 product pages will state "> $100K" for a gate that cuts at ">= $100K" once #1087 lands, and no gate reads them #1127, and it costs a capCel call plus a toContain.

2. A per-locale phrase table for the prose pages, keyed on the derived operator, so flipping the source flips which table must match: inclusive en: ['$100K or more', '≥ $100K'] / zh-Hans: ['$100K 及以上', '达到'] / zh-Hant: ['$100K 及以上', '達到'], exclusive ['over $100K', '超过 $100K', ...]. Assert the inclusive set matches and the exclusive set does not.

3. The reverse-direction grep as a test, which is what 21 product pages will state "> $100K" for a gate that cuts at ">= $100K" once #1087 lands, and no gate reads them #1127 used for its verification and is cheap to keep: zero matches for the exclusive phrasings across content/docs/**.

Why this is a card and not a rider

Assertion 2 is the part that needs design, and it is the reason this should not be bolted onto #1127. A phrase table over prose in three languages is a rewording tax if it is too strict, and a rubber stamp if it is too loose — and Chinese carries no word-for-word "or more" (超过 is strictly exclusive; the inclusive reading needs 及以上 / 达到), so the table is not a mechanical transliteration of the English one. Getting that boundary right is #1018's subject matter.

Refs #1127 · #1128 · #1087 · cluster #1018