Real-time decisioning engine · Aug 2026
Slipstream
A decision engine that classifies an anonymous storefront visitor and picks their page variant in microseconds — with a live storefront and merchant console anyone can open.
How it works
Rules live in an immutable in-memory snapshot, hot-swapped atomically — the read path never blocks and never allocates, even under concurrent load.
Durability is kept off the hot path entirely: every decision batches to an async write path instead of blocking on disk.
Outcomes per variant are attributed against a protected holdout with 95% confidence intervals — the interval is a 95% Wald interval on a fixed sample, not valid under repeated peeking.
The client ships as a dependency-free JavaScript snippet, designed from the integrating storefront's side in.
Fine print
- All traffic and revenue on the public demo is simulated.
- Decision time is measured server-side; the network round trip is separate and mostly physics.
- Single instance, one machine — no replication or failover, and no real network between client and service in this benchmark.