Horizon server @PACK_VERSION@ — Minecraft 26.3, Fabric 0.19.5, Java 25

Install Python 3.10+ and native Java 25. Extract this ZIP and open a terminal here.

  python3 tools/server.py prepare

Read Minecraft's EULA at https://www.minecraft.net/eula. Create/edit
.local/server/optimized/eula.txt to contain eula=true if you agree, then:

  python3 tools/server.py run --heap 4G

Windows: use python in place of python3 if needed. Server downloads and assets
stay in .local. No game code, third-party mod JARs or credentials are bundled.
Lithium and FerriteCore optimize the original simulation/storage; tick rate and
vanilla gameplay rules are retained. No Paper-style gameplay configuration.

From the server console:
  horizonserver bench populated-50 120
  horizonserver stop

From an operator's in-game chat, prefix each command with /. JSON/CSV reports
appear in the instance's horizon-benchmarks directory. They record actual peak
players, tick execution durations, observed tick count per elapsed second,
JVM heap/GC and platform metadata. meanRate is tick execution capacity, NOT TPS.
An idle server does not establish performance for 50 active players.

Baseline:
  python3 tools/server.py prepare --baseline
  python3 tools/server.py run --baseline --heap 4G

Use separate world copies with the same seed, settings, Java and maximum heap.
Only run one at a time unless you give them distinct ports/resources. When
benchmarking 50 players include normal exploration, entity farms, redstone and
chunk loading; repeat with 1, 10 and 100. Do not change simulation/view distance
between matched runs or turn off game logic to manufacture a speedup.
This first release has not been validated under a 50-player real workload.

Optional Java 25 HotSpot experiment: add --compact-headers to the run command.
Reports capture the actual compact object header flag and collector names.
Compare against the same workload with the flag omitted; object allocation
changes do not establish a whole-server RAM/TPS gain.

Optional collector comparison: select --gc g1 or --gc zgc on the run command.
Match world, player activity, heap, Java and header mode; use one collector at a
time. gc_time_ms includes collector cycle accounting, not a measured pause total.
Use JFR pause events for pause analysis and compare RSS/CPU/TPS together.

Compare generated reports:
  python3 tools/compare_benchmarks.py baseline-ticks.json optimized-ticks.json --strict

Baseline must be a separate instance: --baseline adds no optimizers but doesn't
delete mods from an existing optimized instance. Use clean directories.