Skip to content
runward
runward
Product
Who it is for
Trust
Documentation
News
Try it on an example Talk to us

Docs · The doctrine · Section 13

Summary decision matrix

Everything above fits in one decision table. Each row is a trade-off, with its recommended default and the signal that commands a change.

The trade-off matrix

DecisionRecommended defaultEvolve when
Core languageTypeScript (orchestration and interface)Never without a reason; polyglot goes through a sidecar or a service.
Specialized capabilitySidecar in the library's languageAs soon as a mature library lives in another ecosystem.
Component on the critical pathCompiled service behind a contractLatency or throughput proven insufficient.
DecompositionHexagonal modular monolithLoad, lifecycle, isolation, or high availability demand it.
Integration with legacy systemsAnticorruption layer behind the portProgressive replacement, feature by feature, never in one block.
Number of agentsSingle orchestrator directing specialistsParallelizable complexity, never for sequential reasoning.
Model abstractionSingle port, three tiers, direct SDKNever a heavy chaining framework by default.
StateStateless agent, immutable journal plus derived working viewMulti-instance: externalize the in-memory structures.
Crossing the process boundaryAs late as possible, on an objective triggerThen accept eventual consistency, compensation, and explicit causal order.
Contract evolutionVersioned, additive, read with a tolerant readerBreaking change: expand then contract, under consumer-driven verification.
Memory and forgettingValue score, invalidation (not deletion), consolidation with pointers to the source factsPhysical deletion reserved for invalidated derivatives under retention.
EvaluationTests, plus a continuous, guarded evaluation loopBounded self-tuning only, guarded, under human override.
Multi-agent budget and discoveryCentral aggregated counter, in-memory registryBlock allocation and shared capability cards as soon as you leave the process.
Model unavailableMulti-provider failover behind the portAvailable from real go-live onward.
Changing the model (new or updated)Validate in shadow deployment, then in stagesNever in a single move; rollback at the slightest divergence.
Waiting for a human approvalSuspend and release (durable state), rehydrate on decisionNever block the process; low-urgency requests batched in a queue.
Non-critical serviceFail-open, signaled degraded modeSensitive action: explicitly fail-closed instead.
Delegation autonomyBounded: templates, budgets, approvalsWiden only if the outcome actually improves.
Untrusted inputsRetrieved content untrusted, least privilege, approval on the actionHigh privilege: dedicated patterns (frozen plan, quarantine of the untrusted).
Agent identityIts own principal, distinct from the user served; explicit delegation, bounded in time and scopeNever borrowing the user's full rights; an agent outside the registry = a security anomaly.
Runtime secretsPlaceholder on the agent side, real key attached at the network boundary, allowed destinationsNever a key in the model's context or in the execution environment.
Shared building block (connectors, skills, cross-application memory, delivery)Consumed via a port; adapter = discovery client via an index (section 15)The index stays an index; delivery delivers; the orchestrator decides.

Quick decision tree

Does the capability depend on a mature library in another language?YESsidecar in that languageNOstay in the TypeScript coreDoes a component have its own load, lifecycle, or isolation?YESextract it as a serviceNOstay in the modular monolithIs the task parallelizable and highly complex?YESseveral agents under one orchestratorNOsingle orchestrator, one single threadMust the load be carried across several instances?YESexternalize the state, then replicateNOsingle instance, in-memory state ok

Talk to us

Tell us who you are, and we will get back to you.

Your details are recorded by Another Framework, publisher of runward, to get back to you. They are neither sold nor shared, and are deleted three years after our last exchange. To see or delete them: contact@runward.dev.