ai: Fix provider compatibility for email translation

[gentlemandev]

Fixes translation requests that failed before generation when providers rejected array-length JSON Schema constraints. Also makes formatting and blank-entry behavior reliable.

• validate translation counts after generation instead of emitting unsupported minItems/maxItems
• normalize whitespace-only entries and strengthen line-structure preservation
• add Gemini 3.1 Flash Lite eval coverage for translation quality and instruction resistance

Tests:

• 15 targeted unit tests passed
• Gemini 3.1 Flash Lite failed-case rerun passed
• authenticated POST endpoint returned 200 with ordered translations and normalized blank entries

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
inbox-zero	Ignored	Preview	Aug 3, 2026 9:07am

[cubic-dev-ai]

1 issue found across 3 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/web/utils/ai/translate-email.ts">

<violation number="1" location="apps/web/utils/ai/translate-email.ts:59">
P2: A model response with the wrong number of translations now fails immediately without the LLM retry or fallback model path. Keeping the count check inside a retryable validation path, or converting the mismatch into the wrapper's recognized validation error, would preserve recovery while avoiding provider-side `minItems`/`maxItems` constraints.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: A model response with the wrong number of translations now fails immediately without the LLM retry or fallback model path. Keeping the count check inside a retryable validation path, or converting the mismatch into the wrapper's recognized validation error, would preserve recovery while avoiding provider-side minItems/maxItems constraints.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/web/utils/ai/translate-email.ts, line 59:

<comment>A model response with the wrong number of translations now fails immediately without the LLM retry or fallback model path. Keeping the count check inside a retryable validation path, or converting the mismatch into the wrapper's recognized validation error, would preserve recovery while avoiding provider-side `minItems`/`maxItems` constraints.</comment>

<file context>
@@ -56,7 +56,13 @@ ${formatTextsForPrompt(truncatedTexts)}`;
     system,
     prompt,
-    schema: translationSchema(texts.length),
+    schema: z.object({
+      translations: z
+        .array(z.string())
</file context>