Precompile at build vs runtime compile for low-traffic Handlebars

A low-traffic Node.js service renders Handlebars templates and needs a cost-conscious decision about when compilation happens. Requests are infrequent, template changes are rare, and the main constraints are per-request CPU, memory churn, startup latency, and bundle size.

One documented approach is precompiling templates at build time with the Handlebars precompile step. This removes runtime parsing cost but increases the initial bundle and requires regenerating output whenever a template changes. The alternative is compiling templates on demand at runtime, which keeps the initial payload small (potentially helpful for cold starts) but adds parsing overhead on every request and may cause latency jitter under sparse traffic.

Two documented options complicate the choice: the knownHelpers flag can shrink generated JavaScript by omitting helper lookup code when helpers are statically known, and whitespace-stripping mode changes the generated AST size but risks removing meaningful spaces in layout-sensitive output such as plain-text email.

For a very low-traffic workload, does precompilation's per-request CPU saving outweigh the larger bundle and rebuild requirement, or does runtime compilation win on cold-start cost? Does knownHelpers change this trade-off enough to justify the reduced flexibility? Should whitespace stripping be evaluated together with precompilation, or treated as an independent correctness risk?