Configuration and reporters

decisium run loads configuration exactly like cucumber-js: your cucumber.{js,mjs,cjs,json,yaml}, -p <profile>, and cucumber's --format name[:path] and --import <file> flags, added to the file's values. Decisium adds its own step definitions to import.

Decisium options

In worldParameters.decisium, so raw npx cucumber-js reads them too. Flags (--url, --record, --strict, ...) and DECISIUM_* env (DECISIUM_URL, DECISIUM_RECORD, DECISIUM_STRICT=1, ...) override them.

// cucumber.mjs
export default {
  worldParameters: { decisium: { record: "none", strict: true } },
  // url, record, headed, judged, strict, verbose, debug, agentReport,
  // reportDir, confidence: { target, answer }, checkTimeout, actionTimeout, maxTurns
};

Precedence: flag > DECISIUM_* env > worldParameters.decisium > default. A value of the wrong type or out of range fails at startup (exit 2). --parallel workers read the same values.

Tuning

// cucumber.mjs
export default {
  worldParameters: {
    decisium: {
      url: "http://localhost:3000",
      reportDir: "e2e/runs",
      confidence: { target: 0.6, answer: 0.85 },
      checkTimeout: 10000,
      actionTimeout: 10000,
      maxTurns: 12,
    },
  },
};

Option	Default	Flag	Env	Affects
`reportDir`	`"runs"`	`--report-dir`	`DECISIUM_REPORT_DIR`	Where `<reportDir>/<timestamp>/` (log, trace, screenshots, `report.md`) is written.
`confidence.target`	`0.5`	`--confidence-target`	`DECISIUM_CONFIDENCE_TARGET`	How sure Jev must be of the element an action acts on (and a drag's drop target). In (0, 1].
`confidence.answer`	`0.8`	`--confidence-answer`	`DECISIUM_CONFIDENCE_ANSWER`	Every other answer: a check's required state, the value to type or select, which earlier value a check refers to, a dialog answer. A pass under `answer` + 0.1 is marginal. In (0, 1].
`checkTimeout`	`5000`	`--check-timeout`	`DECISIUM_CHECK_TIMEOUT`	Ms a check waits for the page to reach the expected state.
`actionTimeout`	`5000`	`--action-timeout`	`DECISIUM_ACTION_TIMEOUT`	Ms after an action to wait for requests and DOM changes to settle.
`maxTurns`	`8`	`--max-turns`	`DECISIUM_MAX_TURNS`	Turns an action Step may take before `STEP_TIMEOUT`, 1-20.

reportDir

Default "runs"; --report-dir, DECISIUM_REPORT_DIR. Non-empty path containing timestamped run folders. Change it to match your CI artifact upload path.

confidence.target

Default 0.5; --confidence-target, DECISIUM_CONFIDENCE_TARGET. Top probability in (0, 1] for action and drop targets. Reword ambiguous Steps first; tune this only when the intended control is clear.

confidence.answer

Default 0.8; --confidence-answer, DECISIUM_CONFIDENCE_ANSWER. Probability in (0, 1] for other answers and plausible check readings. Marginal-pass line is this value + 0.1. Lowering it below 0.6 warns once per run; prefer rewording.

checkTimeout

Default 5000 ms; --check-timeout, DECISIUM_CHECK_TIMEOUT. Positive integer; how long checks wait for the expected state. Raise for slow pages.

actionTimeout

Default 5000 ms; --action-timeout, DECISIUM_ACTION_TIMEOUT. Positive integer; how long actions wait for requests and DOM changes to settle. Raise when slow actions finish too early.

maxTurns

Default 8; --max-turns, DECISIUM_MAX_TURNS. Integer 1–20; action Turns before STEP_TIMEOUT. Raise for long wizards, or split them into smaller Steps.

Only the top probability is configurable; the lead over the runner-up stays fixed (0.1 target, 0.3 answer): it is what catches "two controls look the same". Confidence affects live answers only (a replay reuses recorded resolutions with their recorded confidence; older cassettes use the original defaults); timeouts and maxTurns apply to replays too. The effective values go into the run event in trace.jsonl, and report.md says when confidence was not the default.

When to change them:

Slow app: raise checkTimeout and actionTimeout.
Long flow in one Step (a wizard): raise maxTurns, or split the Step.
AMBIGUOUS on a clear page: reword the Step first, then raise or lower confidence.target.
Marginal passes: reword. Lowering confidence.answer below 0.6 prints a warning: a wrong reading can then pass.

Reporters

Any cucumber formatter: cucumber's own junit, json, html, message, or a package such as the official Qase reporter. Only one formatter may write to stdout: one configured without a path takes it; otherwise Decisium's console output does.

decisium run features/login.feature --format junit:reports/junit.xml

Qase

Decisium ships nothing Qase-specific; wire the official reporter as in any cucumber project.

npm install --save-dev cucumberjs-qase-reporter
echo '{"mode":"report"}' > qase.config.json   # or mode "testops" with QASE_TESTOPS_API_TOKEN, QASE_TESTOPS_PROJECT
decisium run features/login.feature --format cucumberjs-qase-reporter:/dev/null

Or in your configuration, which raw npx cucumber-js uses too:

// cucumber.mjs
export default {
  // Raw cucumber-js only: Decisium's steps and formatter (decisium run adds both itself).
  import: ["node_modules/decisium/dist/steps.js"],
  format: ["decisium/formatter", "cucumberjs-qase-reporter:/dev/null"],
};

Tag scenarios @QaseID=<id>. A failing Step's message starts with its outcome; app failures (ASSERTION_FAILED, JUDGED_FAILED, BLOCKED, ACTION_FAILED) report as failed, every other outcome as invalid. A marginal pass reaches every reporter as a text/plain attachment marginal-pass.txt on its Step (Qase stores it as file.txt).