Run portable capsule tests on Windows

The native Windows test host executes already compiled WebAssembly components without WSL, a Linux node or a VM. It is a controlled development test process, with a closed set of imports and execution controls. It is not a Windows production node and does not certify Linux isolation, admission or performance.

Get the developer tools. The authenticated Windows developer bundle contains latent-portable-test-host.exe, and its support files. Separately obtain your application project and its compiled component, capsule manifest and contracts. Use the exact three files produced by that application's build, preserving the paths listed under artifacts in latent.project.json. A build may run on a separately selected Linux/WSL/CI host; the portable invocation itself does not contact that host or start a VM.

The compiler's package assembly and admission evidence are separate from these test inputs. A successful portable test does not authorize publishing or deploying a package. Source inspection, a checksum and execution authority are different observations.

Select the native host and application

Use the authenticated frontend and independent verification inputs from workspace setup. Load its downloaded script with -SessionOnly to define the command shortcut without provisioning, connecting or installing a Linux workspace:

. "$HOME/Downloads/setup-lsf-workspace.ps1" -SessionOnly

Acquire the Windows bundle into the private host cache, then explicitly request portable execution. $Artifacts is the root under which the descriptor's artifact paths exist; it is not a node state directory.

$PortableBundle = (dev acquire --bundle-directory (Join-Path $Inputs 'windows') `
    --publisher-policy $DeveloperPolicy --trusted-root $TrustedRoot `
    --verifier $WindowsVerifier --verifier-sha256 $WindowsVerifierSha256 `
    --version $Version --target windows-x86_64 --allow-candidate | ConvertFrom-Json).result
$Artifacts = 'C:\My application build'
$Tests = (dev test --workspace test-native-greeting --environment portable `
    --project $Project --artifacts $Artifacts --portable-bundle $PortableBundle.bundle `
    --controlled-development | ConvertFrom-Json).result
$Tests.passed
$Tests.results | Select-Object id, status, category

This workspace is a private test-report location, not a Linux node connection. Portable test processes are bounded and reaped when the command finishes; there is no foreground up process to keep running. The report identifies portable, the native host and selected profile. Unsupported required scenarios make coverage fail; they do not count as successful skipped tests.

For a greeting, success and the declared empty-name error should agree with the same component on the Linux node. Select a focused case with --select CASE_ID. The selected language's bounded engine profile still applies on Windows.

Read support and differences

Scenario	Portable scope
Typed success, declared error and trap	Actual component execution with fresh activation state
Fuel, memory and deadline failures	Bounded native test execution; no production performance claim
Cancellation before execution	Supported closed prestart control
Cancel an observed running node activation	Requires the real node's Invoke/Cancel receipt workflow
Clock, random, logging and context	Only the declared native imports and explicit test provider configuration
Metrics and buffered HTTP	Explicit bounded providers/controlled peer; record the fixture substitution
Node-local services, immutable blobs, secrets and immediate events	Use their real Linux-node fixtures
Linux PSI, cgroups, sandbox and external-capsule admission	Not established by the portable host

A capability denied by the native host may have a different platform error code from the public Linux node's outer CLI. Shared comparisons retain both actual codes and allow only the documented distinction; unrelated errors do not pass. No requested node test silently runs here, and no unsupported portable import silently gains a production provider.

Return to the application development choices or the edit/watch walkthrough when a test needs the actual node lifecycle or recovery protocol.