MineTuner HUD 1.4.2

Improvements

• Lighter sampling: stats are now sampled once per client tick instead of every rendered frame, and only while something needs them (shown in a list, used by a template token, or recorded by a running benchmark). Ping, CPU, GC and percentile-low readings update at most every 500 ms.
• Graphs and benchmark CPU data now keep updating while the overlay is hidden, F3 is open or the editor is open.
• Benchmark: results are now saved on a background thread, and the screen shows "Saving…" until the write finishes.
• Benchmark CPU is now labelled CPU (process). It is the game process's load, time-weighted over the run, so it will differ from Task Manager's whole-system figure.
• Saved summary.txt files now show the run's actual start time instead of an estimate worked back from the save time.
• Benchmark graphs are cached, so long runs (up to 432,000 samples) no longer cost anything extra per frame to draw.
• 1% Low and 0.1% Low are calculated from a single sort of the samples, and give the same values as before.
• New lists now start with TPS, MSPT and FPS enabled.
• The config file is now checked on load. Hand-edited or corrupted values (nulls, unknown enum names, NaN, out-of-range numbers, duplicate list IDs) are repaired instead of causing problems. The threshold cut-offs in the editor and the Cloth Config screen are validated the same way, so the yellow band can't become unreachable. A file that can't be read is still backed up as minetuner.json.bak-<timestamp> before defaults are used.
• Duplicating a list now copies every setting, including any added in future versions.
• The tick-rate hook no longer crashes the game on launch if a future Minecraft update renames the method it attaches to. Tick-rate display is simply skipped in that case.
• Docs: the README is shorter, with the detailed usage guide moved to the docs site, and there's a new "How sampling works" section.

Fixes

• Fixed left and right click mapping in the HUD editor and the Benchmark screen on Minecraft 26.3.
• Fixed the Avg FPS calculation. It now uses total frames divided by total time, rather than the mean of per-frame FPS, which overstated the average when frame times varied. This applies to both the benchmark and the session Avg FPS stat.
• Fixed MSPT graph colours. Each sample is now coloured by its TPS equivalent instead of being compared against TPS thresholds as raw milliseconds.
• Fixed extra space under the last row of a list, which made the box look bottom-heavy.
• Fixed Backspace splitting emoji in the rename, template and search boxes. Text is now handled by code point, and the 32-character (names and search) and 200-character (template lines) limits count code points.
• Fixed translation keys and stat names being built with the system locale, which could break the UI under locales such as Turkish.
• Fixed the order of saved stat sets in minetuner.json, which could change between launches.