Locale-aware prompts: keep values typed until rendering

Operational case

An order notice for OR-958 contains 1,234.50 units of CAD and a dispatch instant of 2026-11-04T18:30:00Z. The assistant should not rewrite the amount as a string before the application knows whether the customer requested English Canada or French Canada. Nor should it assume that a date displayed to a Montreal user has the same calendar day as the UTC instant. The prompt returns the message key and typed fields; the UI formats them after the customer's locale and time zone are resolved. A reviewer checks that the translated text does not change the amount or promise a delivery day the source did not state.

Order OR-958; amount_value=1234.50; currency=CAD.
Dispatch_instant=2026-11-04T18:30:00Z.
Message key=dispatch_notice; locale=fr-CA; zone=America/Toronto.
Model output: wording and typed field references only.
Renderer: number, currency, date, and plural rules for resolved locale.

Performance and operating cost

Formatting N typed fields is O(N) application work for a fixed message, while model generation adds variable latency and cannot guarantee locale rules. Store canonical values once and render per recipient; this avoids duplicating formatted strings in data records. Testing each locale and plural category adds cases, but it catches errors that a fluent sentence review may miss. A cache key must include locale, timezone where relevant, message version, and any value affecting plural choice. Reusing a rendered string across recipients can expose the wrong date or currency format even if the translation itself is sound.

Common Mistakes

• Do not use the model as the sole currency or date formatter.
• Do not equate locale with timezone.
• Do not build plural messages by concatenating translated fragments.

Connected lessons

• Prompt engineering applications
• Prompt Engineering
• Prompt templates: bind variables without changing instruction structure
• Prompt caching: reuse stable context with a versioned expiry
• Python amount parsing: define a wire format before Decimal
• Translation prompts: preserve terms and machine placeholders
• Mixed-language requests: separate intent from reply language
• Cross-language retrieval: keep policy meaning attached to its source
• Cross-locale evaluation: compare decisions, not word-for-word text
• Project: release a multilingual order notice safely
• Multilingual prompt workflow decisions

Continue with: Location prompts: validate coordinates, units, and precision.

Continue with: Localization prompts: render typed dates, numbers, and units.