Skip to main content

​
Product fit and non-fit

Responsibility: prevent inappropriate adoption and architecture drift.
Authority: product boundary.
Owner role: product/architecture. Change policy: a change requires product sign-off.

​
Strong fit

Akter fits when the domain has many addressable things whose mutations should serialize and whose behavior combines relational data with one or more of:
  • durable commands and retry-safe receipts;
  • realtime connections and cursor-based events;
  • timers or recurring commands;
  • external calls that need retry and dead-letter recovery;
  • finite workflows owned by the same identity;
  • tenant-aware placement and caller attribution.
Typical identities include users, workspaces, rooms, documents, devices, agents, orders, subscriptions, and control-plane resources.

​
Conditional fit

  • Read-heavy systems fit when writes have clear actor ownership and SQL serves cross-actor reads.
  • Transient coordination fits with activation-local values when forgetting state after hibernation is acceptable.
  • Batch work fits when it belongs to an actor job or workflow; independent bulk computation may need a separate compute system.
  • High throughput fits when it partitions across many actor identities rather than concentrating on one hot identity.

​
Poor fit

  • one globally hot mutable object that must scale linearly across workers;
  • workloads dominated by atomic mutation across many unrelated actors;
  • analytics systems whose core operation is broad scans rather than identity-local behavior;
  • hostile or untrusted code execution;
  • applications unwilling to operate or consume Postgres-compatible storage;
  • systems that require exactly-once external outcomes from providers with no idempotency or reconciliation;
  • teams primarily seeking a mature standalone workflow ecosystem rather than an actor framework.
The framework should reject architecture drift: no separate public primitive for background tasks, no broker subsystem without a real consumer and retention model, and no AI-specific layer. See boundaries and competitive positioning.