---
title: "Isolate fixtures when multiple coding agents run QA"
description: "Prevent concurrent QA jobs from sharing mutable records, identities and cleanup scopes, with explicit ownership and fixture manifests for each run."
canonical_url: "https://vostride.com/agent-library/software-factory/isolate-fixtures-for-parallel-qa.md"
md_url: "https://vostride.com/agent-library/software-factory/isolate-fixtures-for-parallel-qa.md"
content_kind: "guide"
topic: "software-factory"
---

# Isolate fixtures when multiple coding agents run QA

Concurrent verification can produce misleading failures when runs share accounts or mutable records. One test changes a subscription while another expects its previous plan; cleanup from a completed run removes a record a still-running test needs. Design isolation before increasing concurrency.

## Define ownership per run

Allocate a namespace and fixture manifest for each execution. Include account identities, organization IDs, object IDs and any sandbox service resources. Seed records using approved application hooks or existing test tooling, and make cleanup use the manifest's exact identifiers rather than a broad name prefix alone.

Use independent actor sessions for collaboration scenarios. If the workflow intentionally shares a document, make the sharing an explicit fixture relationship with synchronized checkpoints. Do not accidentally share the same authenticated browser state across actors with different roles.

## Separate immutable and mutable inputs

Read-only catalog fixtures can sometimes be shared if their revision is fixed. Records affected by purchases, permissions, notifications or inventory generally need separate ownership. For each test, identify both direct writes and indirect side effects, such as a background worker consuming a queue item.

Account for external services. A unique database row does not isolate a shared mailbox, payment customer or device session. The [hooks guide](/docs/agent-qa/guides/hooks.md) documents application-specific setup and checks; supported runtime parallel settings are listed in the [test schema](/docs/agent-qa/configuration/test.md).

## Verify cleanup and collisions

Run two distinct fixtures and confirm each result refers only to its own identifiers. Intentionally schedule their stateful steps close together, then inspect the external persistence evidence. A passing UI flow is insufficient if one run silently consumed the other's record.

Cleanup should tolerate partial setup and remove only resources owned by the run. Preserve failure evidence before deleting artifacts needed for diagnosis. Where an environment cannot provide safe isolation, serialize the affected tests and document the shared dependency rather than treating intermittent collisions as product regressions.

[Collaboration recipes](/agent-library/collaboration.md) · [Network recovery recipes](/agent-library/network-recovery.md)