LaunchGate
A release verdict before your users see the release
Not an AI summary of charts, but a deterministic criterion: load-test results checked against your SLOs and baseline, with a GO / CONDITIONAL_GO / NO_GO verdict right in the pull request.
“Unless your verification process is mature, do not rely entirely on Pass/Fail results to guarantee the reliability of releases. Frequently, you'll encounter 'false positives' and 'true negatives' when testing with distinct types of load.”
The k6 documentation explicitly warns that static Pass/Fail thresholds in CI are noisy and prone to false alarms. LaunchGate turns raw metrics into an authoritative, noise-resistant release verdict (GO / CONDITIONAL_GO / NO_GO) evaluated against historical baselines.
What a verdict looks like
Demo data. This is how the report lands on a pull request after the k6 run.
- service
- checkout-api
- release
- v2.14.0 · PR #812
- baseline
- v2.13.2 (last 5 green runs)
| check | observed | target | severity | result |
|---|---|---|---|---|
| p95 latency | 284 ms | ≤ 300 ms | critical | ✓ pass |
| p99 latency | 612 ms | ≤ 800 ms | critical | ✓ pass |
| error rate | 0.4% | < 1% | critical | ✓ pass |
| p95 vs baseline | +3.1% | ≤ +10% | critical | ✓ pass |
| saturation point | 1,850 RPS | ≥ 2,400 RPS | minor | ! fail |
- hint: db pool exhausted at stage 4 (pool=20): raise pool or add read replica before peak
1. SLOs live in your repo
slo.yaml defines p95/p99, error-rate, allowed regression vs baseline, and saturation thresholds. One source of truth for the test and the gate.
2. k6 on the pull request
Your CI runs k6 on every PR. The load generator is replaceable: LaunchGate reads the JSON summary, not terminal output.
3. A reproducible verdict
Same input, same verdict. A critical violation is NO_GO and blocks the merge; a minor one is CONDITIONAL_GO with a hint on what to fix.
GitHub 17.08 Outage (7h 47m): Why the Autoscaler Went Blind
A live LaunchGate validation report: why HPA watched host CPU (48%) instead of Istio sidecar concurrency limits, how client retries amplified traffic 10x, and how to verify your cluster in 1 hour.
Three outcomes, no dead-end report
Team runs it alone
The gate, slo.yaml and k6 suites stay in your repo. No further paid engagement required.
Periodic review
Monthly review of the SLO budget and gate false positives.
Paid next step
Fractional Performance Engineer: monthly, cancel anytime, $9,000/mo.
CI Gate setup: 1–2 weeks, fixed $7,500. Details on /pricing