Stressmark

New Features

• Intelligent stress test report - When a stress test ends, the recap now opens with a clear verdict (PASS / WARN / FAIL), a 0-100 score, and an answer to the question that actually matters - how many players can this server handle - instead of only raw min/max TPS and memory numbers. The console report, the in-game ops summary, and a saved file all share the same analysis.
• Honest player-capacity estimate - Reports an equivalent real-player range (e.g. ~9-17 real players) derived from the measured per-bot tick cost and the 50 ms tick budget, with an explicit caveat: stress bots only exercise chunk loading, so real players - who also add networking and entity-tracking load this test does not measure - may lower that figure. No more reading "5 bots were fine" as "5 players is my limit".
• Bottleneck diagnosis - Identifies the real limiting resource - the single-threaded main loop, overall CPU, or memory - so the advice fits the actual wall. Because Minecraft's tick loop is mostly single-threaded, it tells apart "one core maxed out" from "the whole machine is maxed out", and won't tell you to add RAM when the main thread is the bottleneck.
• Trend & memory-leak detection - Samples performance every ~10 s during the run and reports whether TPS held stable, degraded, or collapsed over time, rather than only the worst instant. Memory is judged by the post-GC live-set, so a healthy sawtooth heap (which briefly spikes near 100% before each GC) is no longer mistaken for trouble, while a steadily climbing live-set is flagged as a suspected leak.
• Prioritized, hardware-aware recommendations - Advice is now ranked by impact, de-duplicated, and quantified against your actual hardware (cores, max heap, GC type) - for example view-distance 10→8 ≈ -18 ms/tick or increase -Xmx to ~6G, rather than fixed, generic threshold messages.
• Saved Markdown report - Every run also writes a timestamped report to <world>/stressmark-reports/ with TPS/MSPT sparklines, the full per-sample time-series table, and a playbook for each recommendation (probable cause, remedy, exact flag/command) - so results survive the console scrollback and can be kept or shared.

Bug Fixes

• Fixed misleading Young-Gen GC time in reports - The "Young Gen total ms" figure was a fabricated proportional estimate (total pause × young / (young + old)) because young and old GC pause times were never tracked separately. Reports now sum the real per-collection pause time for each generation.

Improvements

• Verified analysis engine - The new report logic (trend, memory, bottleneck, capacity, grading, recommendations) lives in a self-contained, Minecraft-independent module covered by unit tests, so its verdicts are checked against synthetic scenarios rather than only observed in passing.