EdgeAssign

Deterministic Rollouts Without Assignment Storage

Deterministic, reproducible rollouts without an assignment database. EdgeAssign derives decisions on demand and keeps outcomes stable across environments.

That architecture simplifies retention, makes caching predictable, and keeps rollouts consistent. You still own input stability and request hygiene, but the assignment engine stays centralized.

Unified guide: deterministic engine + cookbook recipes.

See pricing Deterministic rollout docs

The core tradeoff

Traditional model: store a per-user assignment row after each evaluation.

EdgeAssign model: derive the decision every time from stable inputs.

This removes assignment storage and sync, and it gives you deterministic reproducibility without a separate assignment database.

What you still need to handle

  • Stable subject IDs across systems.
  • Trait hygiene (avoid raw PII, keep values consistent).
  • Cache TTL and invalidation for fast rollbacks.
  • Reliable request handling and fallbacks.

How deterministic derivation works

// seed + flag + subject => stable bucket
bucket = hash(seed + flag + subject) % 100
eligible = rules(traits)
enabled = eligible && bucket < rollout_percent

Same inputs, same result. No assignment rows are stored.

Why this is different from typical platforms

  • Derived decisions with no assignment database to replicate.
  • Lower retention footprint and smaller audit surface.
  • Deterministic outputs make caching predictable.

For the vendor comparison, see EdgeAssign vs. the usual options.

Who it is for

  • Teams that want deterministic rollouts without assignment tables.
  • Apps that prioritize privacy and minimal data retention.
  • Systems that benefit from cacheable decisions.

Related use cases

For a deeper explanation of why this beats local hashing, see the docs.

FAQ

Is this just a hashing trick?

No. The core challenge is safe seed management and consistent behavior across environments. EdgeAssign centralizes that.

Does this eliminate all operational concerns?

No. You still need stable inputs, trait hygiene, and thoughtful caching.

Can I audit decisions without assignment storage?

Yes. Log safe outputs like flag, enabled, bucket, and request ID in your own observability stack.