TickWeave 0.3.0-beta — Forge

For Minecraft Java 26.3, Forge 66.0.9 and Java 25. Install only this loader variant. Client and server; no additional required mods and no TickWeave installation required on the remote peer.

• Extend bounded entity-section queries to clients and integrated servers.
• Reuse immutable vanilla hopper suction geometry when shape and position match; entities, inventories and query results are never cached.
• Forge only: avoid allocation of the supported empty multipart iterator.
• Add startup configuration in config/tickweave.properties, independent optimization switches and safe fallbacks. Existing section settings are imported from the old TOML file when appropriate. Restart after changes.
• Skip optimization Mixins if Lithium, Canary or Radium is detected, or configuration is invalid.
• Preserve tick frequency, hopper cooldowns, redstone, AI, world formats and network protocol.

Validation

All three loader builds passed real integrated-client and dedicated-server startup, hopper/redstone/entity checks, save and reload. Forge passed 23 JUnit tests, 50 initial integrated assertions and 11 reload assertions, multipart checks, dedicated function checks, and real loopback vanilla interoperability in both directions. Fabric and NeoForge have separate builds and validation.

Measured results and limits

18 controlled Forge server runs compared 0.2.0-beta with this version: three runs per variant/workload, 60-second warm-up and 60-second measurement, identical worlds, normal 20 TPS. On an i7-11800H / 16 GB / Windows 11 / Java 25.0.4.1+1 with a 2–4 GiB G1 heap, median server-thread allocation fell 20.90% with 4,096 empty hoppers and 18.22% with hoppers plus 256 inactive markers. Median mean active tick time fell 0.16% and 5.37%. Both versions already maintained about 20 TPS.

GC pause sums were higher, heap results mixed and idle allocation slightly higher. These short workload-specific results do not establish a general speedup and cannot be extrapolated to other loaders.

Three client performance series were excluded after control-path or minimized-window/render-surface failures, including failures in the unchanged old baseline. No partial client series is used as performance evidence. No general FPS, frame-time, physical input latency, chunk-rendering, TPS, memory, GC or network-ping improvement is claimed. Passed client function tests remain valid.

Other loader versions, authenticated Internet play, a second physical LAN machine, other operating systems, every launcher, long-running farms and arbitrary modpacks remain untested. Full source, German/English installation and changelogs, checksums and detailed test/benchmark evidence are included in the release package.