atm10-gc-pause-correlator

Determine whether periodic freezes align with Java garbage-collection pauses rather than ticks, chunk loading, or network waits. Use the real modified profile and measurable evidence instead of assump

SKILL.md

GitHub

ATM10 GC Pause Correlator

Purpose

Determine whether periodic freezes align with Java garbage-collection pauses rather than ticks, chunk loading, or network waits.

Required inputs

Exact Minecraft, loader, Java and modpack versions; actual mod inventory; relevant latest.log/crash report; changed configs; symptom timestamps; and a reproducible workload. Protect the original save/profile before any controlled change.

Workflow

  1. Record current environment, symptom timing, and the smallest reliable reproduction. Preserve original logs and files.
  2. Define the target subsystem and measurement: determine whether periodic freezes align with Java garbage-collection pauses rather than ticks, chunk loading, or network waits.
  3. Collect measurements before, during, and after the symptom; align clocks/timestamps and verify the workload is comparable between runs.
  4. Identify candidate causes from direct measurements and logs. Mark missing signals and avoid inferring causality from correlation alone.
  5. Run a focused, reversible comparison: Capture GC logs or profiler samples with timestamps and compare pause durations against the observed freeze intervals. Use a copied world or staging profile and change one factor at a time.
  6. Repeat the measurement for a baseline and candidate condition; compare distribution/variance as well as averages when relevant.
  7. Roll back changes that do not pass the predefined criterion; request more evidence if test conditions were not comparable.
  8. Produce a decision record linking each conclusion to data, test condition, and confidence level.

Validation checklist

  • Version, workload, and test conditions are documented.
  • Measurements bracket the reported symptom and use consistent timestamps.
  • The comparison includes a baseline and one-variable experiment.
  • Original worlds/configs remain protected and rollback is clear.
  • Correlation is not presented as confirmed causation without a discriminating test.
  • Follow-up checks verify gameplay stability after any mitigation.

Required output

A time-correlated pause analysis and evidence-backed heap/GC follow-up. Include raw measurement references, method, uncertainty, test result, constraints, and rollback plan.

Boundaries and failure handling

Do not recommend destructive entity/chunk cleanup or random JVM flags without evidence. If instrumentation is unavailable, label the suspected bottleneck unconfirmed and choose the next safest observable test.

Trigger examples

  • Positive: A modified ATM10 workload exhibits the performance or persistence symptom scoped by this skill.
  • Negative: No reproducible symptom or measurement problem has been reported.