dynamic

Experimental dynamic runner profiles

These adapters build an explicit invocation and parse the runner's own result format. They do not discover or execute tests during analysis. The common executor admits the executable, immutable inputs, generated reporters, report manifest and bounded process. Complete means a structurally complete observed report, never adequate tests, a covered requirement, a verified mutation kill, browser/application identity or release qualification.

Runner matrix

Missing runtime/device qualification blocks a release support claim. Fixture success does not qualify other runner versions, operating systems, browsers, workers, child dependencies, all lifecycle cases or full application execution. All profiles remain experimental. No generic JUnit parser acceptance is counted as execution: the retained XML fixtures were actually emitted by the specifically named pytest, Bun and Deno versions.

Input and outcome boundaries

Executable must name the actual already-installed runner entry point or runtime. There is no npx, installation, shell expansion or caller argv passthrough. Config is the runner's configuration except for Rails, where it is an explicit boot helper. Legacy Storybook uses native config discovery: pin those files in InputFiles; custom Config is rejected. Ruby Minitest, Rails and Test::Unit require exact selected files; legacy Storybook accepts a complete story set only. Playwright literal file selections are regex-escaped and anchored. Bun paths receive an explicit relative-path prefix. Vitest substring selection cannot prove exact-file selection and is rejected; use pinned configuration for its test set. Browser/device configuration is explicit in Project. Configurations, lockfiles, executable scripts and their interpreter/package dependencies need the common caller's identity binding; pinning one entry-point script alone cannot establish the full dependency closure.

WebdriverIO configuration must direct its JSON reporter into the fresh report directory. ReportFiles lists every expected worker/module report. Nightwatch uses the same explicit manifest. Cypress uses the profile-owned cypress-*.json pattern under the fresh directory, subject to the common 32-report limit; Mocha uses the corresponding mocha-*.json pattern. Missing/extra expected workers, collection failures, unknown status, inconsistent counts, absent completion markers, no tests, contradictory exit, timeout, interruption and overflow cannot become a complete successful result.

The parsers reject duplicate/case-aliased JSON keys, unrecognized XML structure, contradictory category counts and terminal-attempt conflicts. Native errors cannot disappear behind passing rows. Nightwatch assertion lastError is accepted only with a retained failed attempt.

Playwright and Jest preserve reported retry attempts. Jest retry cardinality must match native invocations; missing history remains a problem. Vitest 5.0.3 can emit passed plus earlier failure messages without attempt ordinals: this is FLAKY with a missing-history problem. Bun/Deno/pytest native JUnit, AVA TAP, RSpec JSON, TestCafe and WDIO JSON do not prove complete retry history, and retain NOT_REPORTED. Nightwatch retains supplied retry data, checking its cardinality. A final passing report never retroactively erases process errors.

Original format evidence

• Node native reporter events
• Jest JSON CLI
• AVA TAP CLI
• pytest JUnit output
• unittest result hooks
• RSpec JSON formatter
• TestCafe JSON reporter
• WDIO JSON mapping
• Nightwatch JSON reporter
• Cypress reporter selection
• Mocha native reporter events
• Detox runner forwarding
• Storybook test runner CLI

Context7 verified the applicable official API docs during implementation. Ruby reporter hook APIs were also checked against the exact installed framework source. Mocha execution support does not imply Mocha affected-analysis routing: that separate capability remains unimplemented. Tests retain actual raw runner reports in testdata; browser schema cases are explicitly labelled synthetic contract tests. TestProfileOwnedReportersLive runs disposable source only and reports a skipped test when an optional runtime/framework is absent. It does not silently claim qualification.

Cypress and Rails runtime qualification

testdata/runtime-qualification.json retains the exact local runtime and tool pins. Cypress 16.1.1 was invoked through the common executor using a pinned task-only launcher to locate its isolated binary/cache and explicit Node, Chrome and macOS arch tools. The native run subcommand precedes --posix-exit-codes. Cypress's retry event is a serialized object; the fixed reporter consumes test:after:run reconstructed tests for every native attempt instead. Spec identity comes from invocationDetails.absoluteFile. Synthetic root collection tests have no spec source and originate under /__cypress/runner/; they remain incomplete collection errors. Other missing source metadata also remains unresolved.

Rails booted an actual application with an HTTP integration route in a network-disabled container derived from the pinned official Ruby image. Its deliberate collection exception produced exit 1 without a report, retained as incomplete. The raw mixed reports and collection report/stderr are fixtures. Full dependency closure remains NOT_OBSERVED; a tool or image pin does not establish test adequacy or qualify other runtimes/applications.