Test Retry & Comparable Outcome Evidence Report

Turn archived individual test attempts into a traceable report for weekly QA triage or a retry-policy review. Supply normalized metadata as JSON rows or mapped CSV. The Actor identifies complete retry chains that failed or errored before passing, comparable executions with mixed terminal outcomes, and evidence that cannot support a resolved conclusion.

This is useful when results are retained in exports or a small history database and you need a downloadable handoff. Playwright, Trunk, ReportPortal and CI dashboards already offer test reporting; a SQL query or spreadsheet may also be enough. This Actor provides a bounded one-shot report without installing a persistent reporting service. It does not access CI, execute tests, fix failures, quarantine tests, infer causes or prove stability.

Input

Declare auditId, windowStart, windowEnd, asOf and observationCoverage. Times use real RFC3339 instants with seconds and a known offset, such as 2026-10-02T10:00:00Z or 2026-10-02T12:00:00+02:00. The start is inclusive, the end exclusive, and windowStart < windowEnd <= asOf is required. Observations after asOf reject the input.

Each observation contains:

• recordId, project, testId, revision, environment, logicalExecutionId: literal strings, preserving leading zeros.
• attemptIndex: zero-based integer; status: pass, fail, error or skipped.
• observedAt, chainCoverage: complete, incomplete or unknown.
• sourceReference: an opaque archive reference. It is displayed as text, never fetched.

Use observations OR nonempty observationsCsv, never both. Remove the prefilled array when using CSV. CSV needs the eleven canonical headers or a complete observationColumns mapping. Comma, semicolon and tab delimiters are supported. BOM and quoted fields are accepted; duplicate, empty, unmapped or ragged columns reject. Only normalized metadata is accepted; JUnit XML and raw logs are not parsed. Limit: 5,000 observations and 4 MB total JSON. Strings are at most 160 characters without surrounding whitespace or control characters.

Interpretation

A comparable cohort is the exact project + testId + revision + environment tuple. A logical execution additionally has its own logicalExecutionId; give a fresh ID to an independent execution, retain the same ID for its retries. Environments and revisions are never treated as equivalent automatically.

Resolved chains require complete scope and chain coverage assertions, unique record IDs and attempt identities, contiguous indexes starting at zero, nondecreasing observation times, no skipped observations, no observations outside the audit window, and no attempt after a pass. Duplicate/conflicting rows, gaps, missing attempt zero, skipped rows and unknown coverage remain unresolved. Coverage is your assertion and cannot be verified against CI.

retry-recovered means a complete observed chain ended in pass after fail/error attempts. Mixed terminal outcomes mean complete comparable executions include both passed and failed/error endings. Neither establishes a cause or a flaky probability. All observed passes do not prove stability. No in-window evidence produces an unresolved report.

Attempt counts use raw included observation records, including duplicates. Execution counts use distinct exact cohort/execution groups and include unresolved chains. They are separate denominators; this Actor does not publish a misleading combined failure rate.

Output, billing and data

One dataset item contains the full summary, execution rows, cohort summaries, findings and original observation ledger. Download complete JSON, execution CSV, findings CSV and escaped standalone HTML. References remain traceable to input positions; the HTML has no external scripts. Formula-like CSV values are neutralized for spreadsheet use. Reports reaching 8 MB JSON or 9 MB for any export are refused before the report event is charged; split a large batch.

The price is $0.25 per completed report, including platform usage, with no startup or per-row event. Invalid input or insufficient report budget produces no charged report. A fresh run is a new billable report. Resurrection of the same run preserves one dataset item and one report charge. If export writing is interrupted, partial files may exist without a charge; the run rewrites them on recovery. Billing ambiguity is reported as a failure for inspection, not silently treated as success.

Use authorized, pseudonymous metadata. Cloud run inputs and results are retained by Apify and are accessible under your account's storage permissions. Avoid personal information, credentials, source code and sensitive log bodies.