[finding] The #3512 placeholder-spelling gate covers copy TABLES only — 65 of 971 inline `t(key, { defaultValue })` strings, which fallbackT also reads, carry no transitive pin

[yinlianghui]

Measured while building objectui#3512's placeholder-spelling gate — not in that PR's scope, and observation-class: 0 violations today, nothing a user hits.

The chain, and where the new gate stops

createSafeTranslation's fallbackT resolves a string through three steps (objectui#3865):

defaults[key]  ||  options.defaultValue  ||  key

…and then interpolates the winner with an exact literal needle — so whichever step won, it must be spelled {{name}} or the braces reach the user verbatim on a provider-less host.

objectui#3512's gate covers step 1 in full (the ten locale packs, the 31 createSafeTranslation defaults tables, the three hand-rolled sibling tables) — those are copy TABLES, which is what that card was scoped to. Step 2, the call site's own inline defaultValue, is a call-site option rather than a table, and the gate does not read it.

Why that is mostly, but not entirely, harmless

scripts/check-i18n-call-site-keys.mjs's default-value-drift class pins an inline default byte-identical to its en row, and en is gated by #3512. So most inline defaults are covered transitively — they cannot differ from a string that is already held to the canonical spelling.

The gate's own census on main (97da1b0) says how much "most" is:

Inline defaults: 909 literal (906 match their en value, 3 not comparable), 62 computed (report-only).

So 65 of 971 inline defaults carry no transitive pin: the 3 literal ones reported "not comparable", plus the 62 computed ones the drift class only reports on. For those, nothing anywhere checks that the placeholder spelling is one fallbackT can resolve.

Why it is dormant

A sweep of every defaultValue: string literal under packages/ / apps/ / examples/ finds 0 with a non-canonical double-brace spelling and 0 containing the nesting marker — the same three divergences #3512 measured, at zero on this surface too. The failure needs someone to write an inline default spelling a hole as {{ name }} on a key that is either absent from en or reached through a computed expression, and then render it without an I18nProvider.

Options (no recommendation attached — the interesting question is which gate owns it, and that is a partition decision, not this finding's to make)

• A. Extend createSafeTranslation 的无 provider 回退插值只认 {{name}} 一种写法,i18next 还认带空格 / 带格式化 / 嵌套三种 #3512's gate to inline defaults. It would have to discover defaultValue property assignments at t()/tt() call sites, which is exactly the walk check-i18n-call-site-keys.mjs already performs — so doing it in packages/i18n means a second, independently-rotting copy of that call-site classifier (the one that knows t is not one function: 1091 of 3690 call sites go to a module-local table, 36 are not translators at all).
• B. Put it in check-i18n-call-site-keys.mjs instead, beside holesOf() — which already reads through all four i18next dialects to extract a hole's name, and whose own comment records "all 84 distinct holes in en are bare names". That script has the call-site classification; adding a spelling verdict there is a few lines and no new walk. Cost: the rule for one contract then lives in two places (a vitest gate for tables, a node script for call sites).
• C. Close the gap at the source instead — make the 3 "not comparable" and 62 computed inline defaults comparable, so the existing transitive pin covers 971 of 971 and no new rule is needed anywhere.

Related: objectui#3512 (the gate this was measured against, PR #4904), objectui#3865 (the chain step that made inline defaults reachable from fallbackT at all), objectui#4135 (the double-brace vs single-brace convention the spelling rule enforces).