Never hit a free-tier quota

Architect for headroom, never hit quotas

Never hit a free-tier quota

Architect every layer with enough headroom that we never approach the free-tier ceiling under realistic load.

Why

The point of staying on free tiers is predictable failure mode: when we hit a quota, the service stops — it does NOT silently bill us. But hitting a quota is still a customer-visible outage. Outages are design failures, not billing events. We should never be surprised.

The layered defense:

1. Pick services whose free tier is 10× our forecast load, not 1.1×.
2. Cache aggressively at one layer down (build-time JSON, KV, browser localStorage, static HTML) so the upstream quota is never the hot path.
3. Push work to build-time GitHub Actions when possible — cron-rebuilt static JSON files have effectively infinite read quota.
4. Spread load across data-shape-specific stores so no single quota becomes the bottleneck (see layer-4).
5. Monitor approach to quotas, not just hits. If usage crosses 50% of any free-tier ceiling, that’s the trigger to add the next layer.

Exceptions

None. If a feature can only work by riding a quota wall close, the feature is wrong.

See also

• no-card-on-file.md — the failure-mode pair
• AGENTS.md “five non-negotiable rules” §1

Edit on GitHub · Back to index