Platform decision

A migration recommendation should change when the operating constraint changes.

Explore three realistic starting points. Each ledger keeps delivery, operational, and assurance tradeoffs in view.

Greenfield

A new event-processing product can choose its foundations before customer data and operating habits make them expensive to change.

Direction: Start on Northstar, but keep the first data boundary portable.

Delivery
Stay and optimize Build on the familiar queue stack and reuse existing deployment scripts.
Migrate deliberately Adopt managed streams and Northstar’s deployment contract from day one.
Readout Migration carries little sunk-cost penalty before launch.
Operations
Stay and optimize The platform team owns scaling rules, replay tools, and incident runbooks.
Migrate deliberately Scaling and replay are managed; the team retains workload-level controls.
Readout Fewer operational surfaces suit the four-person launch team.
Tradeoff
Stay and optimize Maximum local control with more systems to assemble and observe.
Migrate deliberately Faster baseline with an intentional dependency on the event API.
Readout Use an exportable event envelope to preserve optionality.

High traffic

The existing platform is stable at normal load, but seasonal bursts are exposing hot partitions and costly over-provisioning.

Direction: Pilot the burstiest stateless stream before moving durable workloads.

Performance
Stay and optimize Tune partitions and reserve capacity for the known annual peaks.
Migrate deliberately Use elastic partitions with workload isolation and replay windows.
Readout A representative load test should decide, not vendor limits.
Continuity
Stay and optimize No data movement and no temporary dual-write path.
Migrate deliberately Run mirrored traffic until output parity meets the acceptance threshold.
Readout A staged cutover constrains customer-facing risk.
Tradeoff
Stay and optimize Lower near-term delivery risk, with persistent idle-capacity cost.
Migrate deliberately Engineering lift now, with better burst economics if the pilot holds.
Readout Finance and reliability owners must sign off together.

Regulated

A payments workload can only move after control ownership, data residency, and evidence retention are mapped to the target operating model.

Direction: Complete the control map before committing to a migration quarter.

Controls
Stay and optimize Existing controls and evidence collection are already understood by auditors.
Migrate deliberately Shared controls reduce infrastructure scope but change evidence ownership.
Readout Control inheritance must be documented line by line.
Residency
Stay and optimize Current regions satisfy contractual storage and recovery requirements.
Migrate deliberately Target regions qualify, subject to backup and support-access review.
Readout Residency includes replicas, logs, and operator access.
Tradeoff
Stay and optimize Known audit posture with slower reliability improvements.
Migrate deliberately Stronger managed controls after a longer assurance phase.
Readout The compliance path, not implementation speed, sets the date.