I'm using Claude to draft translations for a small web app. Some strings contain placeholders such as {userName} and {remainingDays}, and a few use ICU plural syntax. The surrounding words need translation, but the application still needs to parse the result.

There are also short button labels with character limits. I'd rather have Claude flag a translation that won't fit than quietly remove part of its meaning.

Is it better to provide the strings as a resource file, or separate each entry into its key, text, context, and length limit? What instructions have worked for preserving translation placeholders, especially when the target language needs a different word order or different plural categories?

We will still have a fluent speaker review the wording. I'm mainly looking for a practical way to validate ICU message syntax and catch broken variables before that review.