Project nameLoading project…

Saved automatically in this browser. Download a .tradeoff.json copy if you want to keep the project or move it to another machine.

Scenario StudioFirst edition

Build a small system. Put it under pressure. Change the architecture and see what actually changes.

Guided experiments

Add a component

Select a component or connection to edit it. Drag components to arrange; arrow keys move a focused component.

Synchronous · caller waits Asynchronous · caller continuesScroll to explore

Flow connections

Design name and notes

Design nameProject context and assumptions

Component / service

Component name

Work and capacity

Local processing timemsLocal work only. Waiting and downstream calls are calculated separately.

CPU execution slotsReplicas

Waiting queue capacityjobsWaiting jobs are bounded. A full queue rejects new work.Execution modelBlocking workers stay occupied during synchronous calls. Async execution releases workers after local work; suspended requests still wait for their results.Workers per replica

Calls after local work

Dependency dispatch

Each outgoing connection represents one operation. Add a second database connection to model another query. Sequential order is shown in the connection list.

Technology and evidence

Technology labelApply a behavior profileProfiles describe execution behavior. They do not assign technology-specific speed bonuses.Input evidence“Measured” records your input source; it does not independently validate the model.

3 components · 2 connections1 design ready to compare

Your first experiment

Start with the existing flow. Then change one thing.

Run the baseline, create an alternative, and try increasing application replicas or reducing database query time. The comparison will show which waiting time actually changes.

Model the workApply the same pressureCompare the consequence

Model boundaries and project portability

This first version models one acyclic operation flow with up to 12 components and four design alternatives. Local processing times are fixed assumptions; contention and waiting are simulated. Service workers, execution slots, caller-owned database pools, and shared database capacity are separate resources.

Queues can acknowledge admission before work completes. Timeouts do not cancel admitted remote work. Cache hits are a repeatable sequence; cache freshness, database locks, query plans, garbage collection, transactional writes, and duplicate side effects are not modeled. Technology labels do not predict runtime speed.

Project files preserve editable designs, shared workload and failures, target settings, and format/engine versions. Mermaid exports preserve diagram structure, not executable behavior. Browser storage is a convenience; download your project for a durable copy.

View the open project schema