Flakiness analysis

Not every failure is a bug. Tynkr knows the difference.

Most test monitoring tools show you red and green. Tynkr shows you why— by analyzing the root cause of every failure before deciding whether it's real or noise.

How it works

Two scores, one verdict

Every time a test fails, Tynkr runs it through two layers of analysis before surfacing a verdict to your team.

01

Root cause classification

Tynkr looks at what actually happened — not just that the test failed. Was it a network timeout? A missing element? A 500 error from your backend? Each signal maps to a root cause category.

Frontend runtimeBackend errorNetwork transportResource pressure
02

Flakiness verdict

Based on the root cause, Tynkr assigns one of three verdicts so your team knows exactly what to do next.

CI Failures3 real · 12 flakes
auth_login
FLAKE
checkout_flow
FAIL
search_results
FLAKE
password_reset
FLAKE
api_timeout
FAIL

Signal detection

What signals Tynkr reads

Tynkr collects and correlates five categories of evidence from every execution before forming a verdict.

Failed network requests (4xx, 5xx, timeouts)
Browser console errors and page crashes
Locator failures and assertion mismatches
Slow requests (>1.5s) and execution phases (>4s)
Failure patterns across multiple executions

Flakiness score

The flakiness score (0–100)

For workflows with 5+ executions, Tynkr calculates a score based on pass rate variance. A test that fails 50% of the time scores highest — that's the definition of flaky. A test that always fails scores low — it's just broken.

Pass rate → flakiness score

0% (always fails)50% (most flaky)100% (always passes)

Score peaks at 50% pass rate. Drops to near zero when a test consistently passes or consistently fails.

Confidence

Confidence included

Every verdict comes with a confidence level (0–100%) and a plain-English rationale.

Transport or environment instability appears intermittently

72% confidence

Client-side runtime issue usually reproduces consistently

84% confidence

No black box. You see exactly why Tynkr called it a flake — the evidence is right there alongside the verdict.

Output

The output your team actually uses

Each analyzed execution includes a structured report your team can act on immediately — no digging through raw logs.

  • A one-line summary of what happened
  • The probable cause and which phase it affected
  • Supporting evidence (network trace, console log, step that failed)
  • Recommended next steps — specific actions, not generic advice

Stop triaging noise. Focus on what actually broke.

Tynkr classifies every failure automatically. Your team gets verdicts, not just red lights.