Skip to content

LeafAnvil

LeafAnvil: 50% smaller worlds, 15.5x faster shutdown saves, and 4-tier bit-rot self-healing

Multi-loader world save format plugin & mod (LeafAnvil-<version>.jar) for Paper, Folia, Spigot, Fabric, and NeoForge (leafanvil-ADR-001, leafanvil-ADR-002, leafanvil-ADR-003, anvil-api-ADR-002, anvil-api-ADR-003).

Binds the server's world storage (region/, entities/, and poi/) to .lca (LeafCraft Anvil) at runtime: - Runtime RegionFileStorage & Moonrise bytecode interception (NmsRegionStorageInterceptor): Generates JVM bytecode subclasses for Long2ObjectLinkedOpenHashMap ($$LeafAnvilCacheBridge) and RegionFile ($$LeafAnvilRfBridge) at runtime so all chunk (region/), entity (entities/), and POI (poi/) reads, stream writes, Moonrise ByteBuffer writes, existence checks, clears, flushes, and closes route through disk-backed .lca (LcaRegionFileStorage / LcaLiveArchive). - On-open .mca / .linear -> .lca migration: Transcodes existing .mca or .linear region files on first access via atomic .lca.tmp -> .lca rename while preserving the original file as .mca.bak (or .linear.bak). - Reversible .lca <-> .mca conversion, 256B CoW slab compaction & 4-tier SEC-DED recovery: Provides /leafanvil (alias /lca) commands for status inspection, asynchronous batch world conversion (lca <-> mca), free-slab compaction (compact), 4-tier bit-rot recovery (heal), and configuration reload (reload). - Zero-translation .lca wire streaming & LeafShard inter-server LCAS delta foundation (LcaInMemoryChunkCoordinator, LcaWireStreamProtocol): Tracks loaded sections as .lca (baseFrame, dirtyOverlay) slices (LcaSectionBacking) with SEC-DED-64 baseline syndromes (leafanvil-ADR-002, leafanvil-ADR-003). While heap-neutral on live servers in practice (since NMS/Moonrise LevelChunkSection state coexists in JVM heap), keeping every loaded chunk continuously staged in .lca delta form cuts negotiated client chunk bandwidth by 65%–78% (95%+ on revisit) and directly powers LeafShard's 4-byte-per-block LCAS (0x4C434153) cross-server 10-chunk Halo mesh.


Runtime Architecture & Multi-Loader Interception

1. RegionFileStorage Discovery & Bytecode Bridge (NmsRegionStorageInterceptor)

  • Recursive Multi-Folder Storage Discovery: Inspects CraftWorld / ServerLevel handles up to depth 5 across both vanilla (ChunkMap, ChunkSource, PersistentEntitySectionManager, EntityPersistentStorage, EntityStorage, PoiManager, SectionStorage, SimpleRegionStorage, IOWorker) and Paper/Folia Moonrise (RegionFileIO, DataController, ChunkTaskScheduler) holder hierarchies, intercepting all three storage instances per dimension:
  • region/ (STORAGE_KIND_REGION_VOXEL = 0)
  • entities/ (STORAGE_KIND_ENTITIES = 1)
  • poi/ (STORAGE_KIND_POI = 2)
  • Live Dimension Path Resolution (resolveDimensionDirectory): Extracts the true on-disk dimension root directly from the live RegionFileStorage.folder parent path, supporting both classic dimension folders (world/, DIM-1/, DIM1/) and Minecraft 26.1+ unified dimensions/<namespace>/<dimension>/ directory trees.
  • Pre-Open Flush & Moonrise Miss-Set Reset: On attachment, flushes and closes any vanilla RegionFile instances opened prior to hook installation and clears Moonrise's nonExistingRegionFiles (LongLinkedOpenHashSet) so newly created or lazily migrated r.<x>.<z>.lca files are never masked by cached misses.
  • Generated JVM Classfile Bridges (Java 17 bytecode version 61):
  • $$LeafAnvilCacheBridge extends the server's Long2ObjectLinkedOpenHashMap and overrides getAndMoveToFirst(long) and get(long) to populate and return per-region $$LeafAnvilRfBridge instances via CacheController.
  • $$LeafAnvilRfBridge extends net.minecraft.world.level.chunk.storage.RegionFile (instantiated via Unsafe.allocateInstance with Moonrise fields version = VERSION_NONE / ID 3, info = RegionStorageInfo, recalculateCount = AtomicInteger(0), oversized = byte[1024], and path = r.<rx>.<rz>.lca pre-initialized) and overrides:
    • getChunkDataInputStream(ChunkPos) -> LcaRegionFileStorage#getChunkDataInputStream
    • getChunkDataOutputStream(ChunkPos) -> LcaRegionFileStorage#getChunkDataOutputStream
    • write(ChunkPos, ByteBuffer) -> NmsRegionStorageInterceptor#onWriteByteBuffer (intercepts Moonrise direct ByteBuffer chunk writes, decoding compression IDs 1 GZIP, 2 Zlib/Deflate, and 3 uncompressed before committing via writeChunkNbtBytes)
    • doesChunkExist(ChunkPos) / hasChunk(ChunkPos) -> LcaRegionFileStorage#doesChunkExist
    • clear(ChunkPos) -> LcaRegionFileStorage#removeChunk
    • flush() -> LcaRegionFileStorage#flushRegion
    • close() -> LcaRegionFileStorage#closeRegion
    • getPath() -> r.<rx>.<rz>.lca
    • getRecalculateCount() -> 0

2. Platform Lifecycle Wiring

Platform Entry Point Lifecycle Events Hooked Config Directory
Paper / Folia / Spigot PaperLeafAnvilPlugin (load: STARTUP, folia-supported: true) WorldInitEvent (LOWEST), WorldLoadEvent (LOWEST), onEnable world sweep, WorldSaveEvent (flushWorld), WorldUnloadEvent (detachWorld), onDisable (flushAll) plugins/LeafAnvil/
Fabric FabricLeafAnvilMod (ModInitializer) ServerLifecycleEvents.SERVER_STARTING, ServerLifecycleEvents.SERVER_STARTED, ServerTickEvents.START_SERVER_TICK (dynamic dimension guard), SERVER_STOPPING (flushAll), SERVER_STOPPED (close) config/leafanvil/
NeoForge NeoForgeLeafAnvilMod (@Mod("leafanvil")) LevelEvent.Load, ServerStartedEvent, LevelEvent.Save (flushWorld), LevelEvent.Unload (detachWorld), ServerStoppingEvent (flushAll), ServerStoppedEvent (close) config/leafanvil/

3. Bounded LRU Region Cache & Page-Cache Eviction (LcaRegionFileStorage)

  • Maintains an access-ordered LRU cache of up to DEFAULT_MAX_OPEN_REGIONS = 256 open LcaLiveArchive handles per storage folder (region/, entities/, poi/).
  • When an open region is evicted from the LRU cache, LcaLiveArchive#closeWithoutForce() writes the trailing "LCA1_END" footer canary and closes the FileChannel without issuing a blocking hardware channel.force(false) (fsync) syscall (as positional FileChannel#write updates are already resident in the OS page cache). Hardware fsync on flush() can be enabled explicitly via -Dleafanvil.fsync=true.

4. In-Memory .lca Section Tracking & Zero-Translation Wire / LeafShard Streaming (leafanvil-ADR-002, leafanvil-ADR-003)

  • In-Memory (baseFrame, dirtyOverlay) Coordination (LcaSectionBacking & LcaInMemoryChunkCoordinator): Pairs each loaded section's .lca baseFrame slice with an L1-resident open-addressed int[16..64] dirtyOverlay (entry32 = ((pos12 + 1) << 19) | (stateId19 & 0x7FFFF)), dirtySectionBitmask32, and 64-bit Extended Hamming (72, 64) baselineSecDed64 syndrome. In live multi-platform server benchmarks (Table 3 below), this tracking coexists alongside NMS/Moonrise LevelChunkSection state and does not reduce total server JVM RAM; instead, its primary production payoff is eliminating main-thread voxel/NBT transcoding on outbound network paths.
  • Negotiated Server-to-Client .lca Batch Streaming (LcaWireStreamProtocol, LCA_NET_V1): Streams paced LCD2-compressed .lca chunk batches (leafanvil:lca_chunk_batch, ~1.5–3.2 KiB/chunk vs ~6–14 KiB vanilla Zlib) and lastUpdateTick + SEC-DED revisit deltas (2 B EntryCleanAck or EntryPatchDelta, 95%–99% revisit bandwidth reduction) with proxy-safe RFC 1951 Deflater.NO_COMPRESSION framing (CAP_PROXY_SAFE_STORED_ZLIB).
  • Inter-Server LCAS (0x4C434153) Delta Export Powering LeafShard: Exposes LcaInMemoryChunkCoordinator#exportDirtyChunkDeltasForInterServerSync and InterServerChunkUpdatePayload (0x4C434153), allowing LeafShard (leafshard-ADR-001..003) to replicate 10-chunk cross-border Halo rings across peer server nodes at 4 bytes per modified voxel (EntryPatchDelta) and ~1.5–3.2 KiB EntryFullChunk refills with zero main-thread NBT serialization.

Commands & Permissions (/leafanvil, alias /lca)

Built on CommandsAPI (commands-api-ADR-001) using key=value parameters. All commands require permission leafanvil.admin (default: op, permission level 4 on Fabric/NeoForge). Batch operations (convert, compact, heal) execute asynchronously off the main tick thread (REQ-RTP-S-005) behind a single-flight busy guard (REQ-RTP-S-007).

Command Parameters Description
/leafanvil (or /leafanvil status, /lca status) (none) Displays .lca runtime state (enabled, autoMigrate, modeB, attachedWorlds, bitHealth) and per-world counters (openRegions, migrated, dimensionDirectory).
/leafanvil bit (or /lca bit) (none) Executes the [LCA bit] Built-In Test suite (LCA_CODEC_WIRE_PBIT, LCA_DISK_STORAGE_COMPAT, LCA_SECTION_BRIDGE_COMPAT, LCA_NETWORK_CARRIER_COMPAT).
/leafanvil convert [world=<name\|all>] [format=<lca\|mca>] Flushes target world sessions and asynchronously transcodes all region files (region/, entities/, poi/, including DIM-1, DIM1, and 26.1+ dimensions/<ns>/<dim>/) to .lca (format=lca, default) or back to vanilla .mca (format=mca).
/leafanvil compact [world=<name\|all>] Compacts open .lca region files whose 256B free-slab fragmentation ratio exceeds compact-fragmentation-threshold and flushes them to disk.
/leafanvil heal [world=<name\|all>] Flushes target worlds and runs the 4-tier .lca SEC-DED + self-describing 256B slab reverse-scan + section octreeOccupancyMask64 quarantine bit-rot healer across all r.<x>.<z>.lca files.
/leafanvil reload (none) Reloads config.yml and messages.yml from disk.

Configuration (config.yml & messages.yml)

config.yml

# LeafAnvil (.lca LeafCraft Anvil World Save Format Plugin & Mod)
enabled: true
# Automatically transcode existing .mca / .linear / .b_linear regions to .lca on first open
auto-migrate-legacy: true
# Preserve original .mca / .linear / .b_linear files as .bak after migration
backup-migrated-legacy-files: true
# Pure-Java LCD2 (0x4C44 Atomic Dictionary + 12-bit Huffman) Mode B compression level for .lca chunk bodies (1 = default Mode B LCD2 with SEC-DED header parity; 0 = Mode A zero-LZ direct slabs)
lcd2-level: 1
# Free-slab fragmentation ratio triggering automatic compaction on region close
compact-fragmentation-threshold: 0.35
# Default Minecraft DataVersion for newly created empty region headers
default-data-version: 4671

messages.yml (REQ-RTP-S-007)

# LeafAnvil configurable user feedback messages (REQ-RTP-S-007)
busy: "[LeafAnvil] Another storage task is currently running; please wait."
invalid-command: "[LeafAnvil] Invalid command or parameter: %arg%"
no-permission: "[LeafAnvil] You do not have permission to execute this command."

.lca (LeafCraft Anvil) Data Format Specification (api/anvil-api/docs/ICD.md)

Single Canonical Specification (api/anvil-api/docs/ICD.md, MIT): The byte-level Interface Control Document (api/anvil-api/docs/ICD.md, LCA-ICD-1.0, routed from docs/ICD.md) is the single canonical source of truth for all .lca header layouts, bitfields, 3D space-filling curve equations, ChunkPatchTail (0x4000) delta framing, checksum/Hamming parity formulas, and the 160-block / 65-biome / 128-property static lookup tables. It is licensed under the MIT License, packaged inside LeafAnvil-<version>.jar (docs/ICD.md and META-INF/leafanvil/ICD.md), and extracted to <plugin-or-mod-config-dir>/ICD.md on server startup.

Defined in api/anvil-api (io.github.dailystruggle.anvil.LcaRegionReader, LcaRegionWriter, LcaSectionCodec, LcaRegionFileStorage, and anvil-api-ADR-003), .lca (LeafCraft Anvil) stores region data in 256-byte Copy-on-Write (CoW) sub-slabs using either Mode A (0-LZ uncompressed section containers) or Mode B (uncompressed header + SEC-DED parity + pure-Java LCD2 0x4C44 Continuous Atomic Layer 2 Region Dictionary body).

Layer / Mechanism Summary Canonical Spec Section in api/anvil-api/docs/ICD.md
Layer 1: Superblock (64 B), Pointer Directory (8 KiB) & SEC-DED Table (1 KiB) "LCA_V001" superblock with 64-bit SplitMix checksum (CBIT T2), uint64[1024] lock-free pointer directory (slabOffset35 \| codec1 \| exactByteLen20 \| generation8), 256B CoW free-slab allocator with 64-slot reader hazard pointers, and 1,024 B Extended Hamming (72, 64) SEC-DED pointer syndromes. Section 3 & Section 4
Layer 2: Chained Region Shared Dictionary (0x2040) Pre-allocated 256B-aligned dictionary arena with nextSegmentLinkTrailer relative continuation pointers (<= 64 segments, CBIT T5), downward-growing Extended Hamming (72, 64) SEC-DED dictionary syndromes, forward-indexed LCD2 384 B 12-bit Canonical Huffman + cross-chunk byte pattern arena (dictPatternArena), and packed RGBA8 material/tint metadata for map LODs. Section 5 & Section 5.3
Layer 3a/b: Self-Describing ChunkFrame, Mode B (LCD2 0x4C44), Order-Independent sectionOffsets[N] & Slab-Local Micro-Palette Uncompressed ChunkFrame header (0x5800_0000 \| (chunkIdx << 16) \| DataVersion), Mode B (codec1 = 1) header SEC-DED + LCD2 (0x4C44 "LD") 3-stream (lit/cmd/off) Continuous Atomic Region Dictionary body wrapper, $O(1)$ inline uniform sections (bit 15 == 1, 0 body bytes), order-independent non-overlapping SectionFrame offsets (relEnd_i), and two-part Slab-Local Micro-Palette (Part A 1-byte static core block/biome tokens + $4^3$ BiomeTemplates; Part B sparse property suffixes). Section 6, Section 6.2 & Section 9
Layer 4: $16 \times 16 \times 16$ Non-Uniform SectionFrame 1-pass directional curve predictor (00:YZX, 01:H_3, 10:Y+H_2, 11:XZY), 64-bit octreeOccupancyMask64 (CBIT T6b) with octant/slab cursor domain compaction, per-state Delta-Roaring containers (TAG_SINGLETON subTag 0..3, TAG_DELTA_RUN 4\|4, 5\|3, 6\|2, 7\|1, TAG_DELTA_ARRAY, Word-Masked TAG_BITMAP), FlatBitPack ceiling, and 4-bitplane Reflected Binary Gray-code lighting (SkyLight on Y+H_2, BlockLight on H_3). Section 7
Layer 3c: Residual Tail, Heightmap Delta Streams & ChunkPatchTail (0x4000) uint16 boilerplateFlags16 (bits 0..14) eliding empty vanilla/Starlight lists, binary RLE + ZigZag delta MOTION_BLOCKING_NO_LEAVES and cross-heightmap delta streams, length-prefixed ChunkPatchTail (HAS_CHUNK_PATCH_TAIL = 0x4000 with appended micro-palette entries, per-section voxel/light deltas, replacement section offsets, and patchParity8 SEC-DED syndrome), and optional trailing Mojang NBT up to residualRelEnd. Section 8
Built-In Testing (CBIT T1..T8) & 4-Tier Bit-Rot Recovery (/leafanvil heal) Tier 1 Hamming (72, 64) SEC-DED pointer, Layer 2 dictionary & Mode B header healing (< 10 ns), Tier 2 self-describing 256B slab reverse-scan (0x5800_0000), Tier 3 per-section octreeOccupancyMask64 quarantine with palette[0] fallback, and Tier 4 4-epoch CoW prior-pointer rollback. Section 10 & BIT_CODES.md

Codec Benchmarks & Continuous Multi-Platform Test Harness

1. Codec Microbenchmarks (LcaRegionCodecTest, 1_21_R1 & overworld_mountain_r.0.0.mca, 64 Chunks)

Measured with bit-exact voxel, biome, heightmap, entity, POI, and light round-trip parity across .mca fixtures (1.20.4, 1.21.4, 26.1, 26.3):

1A-1. 1_21_R1 Region Benchmark (532,480 B Vanilla .mca, 64 Chunks)

Workload / Metric (1_21_R1, 64 Chunks) Vanilla .mca (4K Zlib) .linear (Zstd-1 Whole-Region) .lca v2 Mode A: 0-LZ (Slab-Local Palette) .lca v2 Mode B: LCD2 (0x4C44 Atomic Dict)
Region Storage Size (64 Chunks) 532,480 B (100.0%) 299,918 B (56.3%, 1.78x) 272,648 B (51.2% of .mca, 1.95x, -9.1% vs .linear) 207,368 B (38.9% of .mca, 2.57x, -30.9% vs .linear)
Cold Random 1-Chunk Read (Store -> AnvilChunkView) 90.26 us 1,303.75 us (1.30 ms) 25.28 us (3.57x vs .mca, 51.6x vs cold .linear) 70.60 us (1.28x vs .mca, 18.5x vs cold .linear)
Warm Random 1-Chunk Read (Cached -> AnvilChunkView) 90.26 us 32.79 us 24.80 us (1.32x vs warm .linear, 3.64x vs .mca) 48.75 us (1.85x vs .mca)

1A-2. overworld_mountain_r.0.0.mca (495,616 B Vanilla .mca, 64 Chunks)

Workload / Metric (overworld_mountain_r.0.0.mca, 64 Chunks) Vanilla .mca (4K Zlib) .linear (Zstd-1 Whole-Region) .lca v2 Mode A: 0-LZ (Slab-Local Palette) .lca v2 Mode B: LCD2 (0x4C44 Atomic Dict)
Region Storage Size (64 Chunks) 495,616 B (100.0%) 318,806 B (64.3%, 1.55x) 290,056 B (58.5% of .mca, 1.71x, -9.0% vs .linear) 233,224 B (47.1% of .mca, 2.13x, -26.8% vs .linear)
Cold Random 1-Chunk Read (Store -> AnvilChunkView) 163.93 us 2,519.12 us (2.52 ms) 54.10 us (3.03x vs .mca, 46.6x vs cold .linear) 88.20 us (1.86x vs .mca, 28.6x vs cold .linear)
Warm Random 1-Chunk Read (Cached -> AnvilChunkView) 163.93 us 64.26 us 52.67 us (1.22x vs warm .linear, 3.11x vs .mca) 68.10 us (2.41x vs .mca)
Cold Whole-Region Sequential Sweep (All 64 Chunks) 7.75 ms (121.14 us/ch) 6.14 ms (95.92 us/ch) 3.67 ms (57.27 us/ch - 1.67x vs .linear, 2.12x vs .mca) 4.46 ms (69.67 us/ch - 1.38x vs .linear, 1.74x vs .mca)
Cold Whole-Region Terrain Probe Sweep (Y=[48..112]) 7.89 ms (123.33 us/ch) 5.84 ms (91.32 us/ch) 0.48 ms (7.47 us/ch - 12.23x vs .linear, 16.51x vs .mca) ~1.05 ms
Random RTP Terrain Window Probe (Y=[48..112]) 109.72 us 64.26 us (warm) 8.55 us (12.84x vs .mca) ~11.5 us
Partial 1-Section Read (16^3 H_3 / Inline Uniform) 109.72 us 64.26 us (warm) 1.39 us (0.93 us uniform - 78.9x to 118.6x vs .mca) ~3.40 us
Dirty Chunk Save (CPU Latency) 444.83 us 48,991..205,789 us 390.00 us (1.14x vs .mca, >120x vs .linear) ~554.0 us (LCD2 continuous atomic dict)
10,000 Concurrent CoW Saves Slab Reuse & Torn Reads 4.20 -> 18.69 MB (82.3%) Rewrites 43.4 GB total 100.00% (10,000 / 10,000) reuse, 0 B growth, 0 torn reads 100.00% reuse, 0 torn reads
Measured on Linux Threadripper 7970X NVMe over 4,000 deterministic random chunks across 1,024 region files ([-500, 499] x [-500, 499]), 261,120-chunk /rtp scan sequential sweeps, Chunky existing-world scans (16,641 chunks), and 63 high-rate Folia 26.2 16k-radius runs (4,096 region files):
I/O Pattern & Workload .mca (46.48 GB) .lca (26.96 GB, -42.0%) .lca NVMe Speedup
Random I/O (1 Thread): Full Chunk Decode (4,000 chunks) 2,339 chks/s (p50=359.3 us, p95=1,001.3 us) 3,815 chks/s (p50=170.2 us, p95=891.4 us) 1.63x throughput (+63.1%), 2.11x faster p50
Random I/O (16 Threads): Full Chunk Decode (4,000 chunks) 54,865 chks/s (p50=186.3 us, p95=523.5 us, p99=2,609.5 us) 186,592 chks/s (p50=71.2 us, p95=129.3 us, p99=191.6 us) 3.40x throughput, 4.05x faster p95, 13.6x faster p99
Random I/O (1 Thread): Column Surface Probe (4,000 chunks) 6,783 cols/s (p50=133.6 us, p95=235.4 us, p99=337.5 us) 18,199 cols/s (p50=51.6 us, p95=80.6 us, p99=119.8 us) 2.68x throughput, 2.59x-2.92x faster p50/p95/p99
Random I/O (16 Threads): Column Surface Probe (4,000 chunks) 94,157 cols/s (p50=140.0 us, p95=230.6 us, p99=439.4 us) 204,253 cols/s (p50=51.3 us, p95=80.1 us, p99=356.1 us) 2.17x throughput, 2.73x-2.88x faster p50/p95
Sequential NVMe Burst: /rtp scan start (5s initial sweep) 3,273 chks/s (52,361 loc/s) 5,251 chks/s (84,014 loc/s) 1.60x (+60.4%) faster sequential region sweep
Sequential NVMe Scan: Chunky Existing-World (16,641 chunks, Folia) 2,343.8 cps (7.10 s) 8,238.1 cps (2.02 s) 3.51x faster on Folia (2.2x-2.5x vs .linear/.b_linear)
Sustained Folia 26.2 16k Live Chunk Streaming (63 runs, 4,096 regions) 143-313 live chks/s (68-453 region-hop stalls) 4,894-5,398 live chks/s (0 region-hop stalls) 23x-35x more live chunk I/O (10.1 KiB lock-free headers)
Sequential 16k World Shutdown Flush (save-all + close()) 34.02-160.62 s (median 63.03 s) 4.51-7.09 s (median 5.76 s) 11.0x-28x faster shutdown world save

2. Continuous Multi-Platform A/B/C Benchmark Driver (scripts/bench_lca_multiplatform.py)

Live server test results comparing without-lca (vanilla .mca / native .linear / native .b_linear BufferedLinear), with-lca (LeafAnvil .lca Mode A 0-LZ), and with-lca-b (LeafAnvil .lca Mode B LCD2) across Paper (26.2), Folia (26.2), Leaf native .linear (26.2-linear), Leaf native BufferedLinear (26.2-blinear, B_LINEAR), Fabric (26.1.2 & + C2ME), and NeoForge (26.1.2 & + C2ME) are gathered via scripts/bench_lca_multiplatform.py (120 total completed runs, updated 2026-11-01 19:53:36):

0. Experimental Reliability & Template Parity Audit

  • World Reset Policy: Fresh Btrfs --reflink=auto clone from frozen template before every run (force_reset=True, zero cross-run chunk/entity leakage).
  • Storage Arms: without-lca (vanilla .mca / .linear / .b_linear), with-lca (Mode A 0-LZ direct bit-scatter slabs, codec1=0), and with-lca-b (Mode B LCD2 0x4C44 Continuous Atomic Region Dictionary with SEC-DED header parity, codec1=1).
  • View / Simulation Distance Parity: Both server.properties (view-distance=2, simulation-distance=2) and bench-swarm.js (--view-distance 2) are locked to a 5x5 (25-chunk) square matching RTP's destination ticket so post-arrival chunk streaming work is equalized across .lca and .mca.
  • Order Bias Control: Paired within-round blocks with counterbalanced arm ordering on alternating rounds.
  • CPU / I/O Isolation: Server JVMs run at nice -10, ionice -c2 -n0 on cores 0-23,32-55; deepseek/ollama demoted to SCHED_IDLE (chrt -i 0, nice 19, ionice -c3) on cores 28-31,60-63.
Template Pair Platforms .mca Files .lca Files Exact File Parity .mca Chunks .lca Chunks .mca Size (MB) .lca Mode A (MB) .lca Mode B (MB) .lca Archival (MB) Parity Gate
26.2 paper, folia 767 767 767/767 293317 293317 2527.9 1535.3 1269.8 1257.7 PASS (100%)
26.1.2 fabric, fabric-c2me, neoforge, neoforge-c2me 767 767 767/767 293317 293317 2527.9 1535.3 1269.8 1257.7 PASS (100%)
26.2-linear leaf-linear 767 767 767/767 293317 293317 1672.1 1535.3 1269.8 1257.7 PASS (100%)
26.2-blinear leaf-blinear 767 767 767/767 293317 293317 1407.5 1535.3 1269.8 1257.7 PASS (100%)

1. Platform Readiness & Storage Verification Matrix

Platform Arm Runs (n) Last Outcome Intercepted Storages .mca Files .lca Files World Disk (MB) Boot (s) Save Shutdown (s mean +- sd)
paper without-lca 6/6 OK 0 770 0 2532.3 40.0 86.12 +- 72.67
paper with-lca 6/6 OK 9 0 771 1545.2 40.0 5.54 +- 0.21
paper with-lca-b 6/6 OK 9 0 771 1278.8 40.0 5.55 +- 0.23
folia without-lca 6/6 OK 0 781 0 2536.1 13.1 47.69 +- 36.44
folia with-lca 6/6 OK 9 0 780 1545.5 14.0 5.03 +- 0.01
folia with-lca-b 6/6 OK 9 0 780 1279.0 14.0 5.02 +- 0.00
leaf-linear without-lca 4/5 OK 0 781 0 1672.2 14.0 55.56 +- 4.66
leaf-linear with-lca 5/5 OK 9 0 781 1546.4 14.0 5.18 +- 0.01
leaf-linear with-lca-b 5/5 OK 9 0 779 1279.2 14.0 5.20 +- 0.04
leaf-blinear without-lca 2/2 OK 0 779 0 1407.5 14.1 8.47 +- 0.48
leaf-blinear with-lca 2/2 OK 9 0 781 1546.2 15.0 5.19 +- 0.01
leaf-blinear with-lca-b 2/2 OK 9 0 781 1279.5 14.0 5.22 +- 0.01
leaf-blinear with-lca-c 3/3 OK 9 0 779 1266.5 14.0 5.20 +- 0.03
fabric without-lca 5/5 OK 0 768 0 2530.7 8.1 11.60 +- 0.90
fabric with-lca 5/5 OK 9 0 767 1569.2 8.1 4.33 +- 0.04
fabric with-lca-b 5/5 OK 9 0 767 1298.7 8.1 4.31 +- 0.14
fabric-c2me without-lca 5/5 OK 0 768 0 2531.2 8.0 9.43 +- 0.42
fabric-c2me with-lca 5/5 OK 9 0 767 1576.0 8.1 4.09 +- 0.37
fabric-c2me with-lca-b 5/5 OK 9 0 767 1304.0 8.1 3.93 +- 0.37
neoforge without-lca 5/5 OK 0 768 0 2530.4 8.4 12.42 +- 0.89
neoforge with-lca 5/5 OK 9 0 767 1569.6 8.1 4.27 +- 0.10
neoforge with-lca-b 5/5 OK 9 0 767 1298.5 8.1 4.72 +- 0.43
neoforge-c2me without-lca 5/5 OK 0 768 0 2531.1 9.4 9.09 +- 0.32
neoforge-c2me with-lca 5/5 OK 9 0 767 1578.7 9.2 4.14 +- 0.34
neoforge-c2me with-lca-b 5/5 OK 9 0 767 1305.1 9.4 4.09 +- 0.38

2A. Head-to-Head .lca (Mode A 0-LZ, codec1=0) vs .mca / .linear / .b_linear by Platform

Platform n (with-lca/MCA, pairs) Completed TPs (LCA vs MCA) Stage 0 TP/s (LCA vs MCA) Stage 0 p50 (LCA vs MCA) Cold p50 (LCA vs MCA) Stage 0 p95 (LCA vs MCA) Stage 0 Chunks/TP (LCA vs MCA) Stage 0 MSPT p95 (LCA vs MCA) Main CPU / Chunk (LCA vs MCA) Save Shutdown (LCA vs MCA) Comparison Status
paper 6/6 (6) 1487 vs 1442 (1.03x) 4.95 vs 4.82 200ms vs 224ms (1.13x faster) 200ms vs 228ms (1.14x faster) 221ms vs 466ms (2.11x faster) 75.6 vs 17.5 31.0ms vs 193.8ms 1.45ms vs 5.79ms (4.01x) 5.54s vs 86.12s (15.5x faster) PAIRED (n=6)
folia 6/6 (6) 1496 vs 1274 (1.17x) 4.95 vs 4.64 103ms vs 164ms (1.60x faster) 103ms vs 968ms (9.40x faster) 203ms vs 218ms (1.08x faster) 97.9 vs 6.6 19.3ms vs 96.3ms 0.73ms vs 3.29ms (4.53x) 5.03s vs 47.69s (9.5x faster) PAIRED (n=6)
leaf-linear 5/4 (4) 1483 vs 10 (152.12x) 4.95 vs 0.68 200ms vs 3369ms (16.81x faster) 201ms vs 5320ms (26.50x faster) 326ms vs 5178ms (15.90x faster) 75.6 vs 26.6 28.3ms vs 2731.4ms 1.35ms vs 4.07ms (3.00x) 5.18s vs 55.56s (10.7x faster) PAIRED (n=4)
leaf-blinear 2/2 (2) 1428 vs 14 (105.78x) 4.95 vs 1.06 204ms vs 2311ms (11.30x faster) 204ms vs 5704ms (27.89x faster) 381ms vs 5534ms (14.52x faster) 75.2 vs 32.9 28.7ms vs 1790.3ms 1.39ms vs 2.61ms (1.88x) 5.19s vs 8.47s (1.6x faster) PAIRED (n=2)
fabric 5/5 (5) 1494 vs 1491 (1.00x) 4.95 vs 4.95 8ms vs 12ms (1.58x faster) 18ms vs 19ms (1.05x faster) 23ms vs 27ms (1.18x faster) 79.0 vs 79.4 22.3ms vs 22.4ms 1.58ms vs 1.59ms (1.00x) 4.33s vs 11.60s (2.7x faster) PAIRED (n=5)
fabric-c2me 5/5 (5) 1496 vs 1496 (1.00x) 4.95 vs 4.95 8ms vs 11ms (1.36x faster) 19ms vs 16ms (0.87x faster) 23ms vs 22ms (0.96x faster) 79.6 vs 79.6 20.7ms vs 22.0ms 1.17ms vs 1.16ms (1.00x) 4.09s vs 9.43s (2.3x faster) PAIRED (n=5)
neoforge 5/5 (5) 1486 vs 1493 (1.00x) 4.95 vs 4.95 15ms vs 25ms (1.70x faster) 16ms vs 25ms (1.62x faster) 42ms vs 47ms (1.12x faster) 78.8 vs 78.9 21.2ms vs 22.3ms 1.56ms vs 1.57ms (1.01x) 4.27s vs 12.42s (2.9x faster) PAIRED (n=5)
neoforge-c2me 5/5 (5) 1494 vs 1496 (1.00x) 4.95 vs 4.95 12ms vs 16ms (1.28x faster) 13ms vs 17ms (1.28x faster) 42ms vs 54ms (1.28x faster) 79.3 vs 79.8 18.2ms vs 19.3ms 1.17ms vs 1.18ms (1.01x) 4.14s vs 9.09s (2.2x faster) PAIRED (n=5)

2B. Head-to-Head .lca (Mode B LCD2, codec1=1) vs .mca / .linear / .b_linear by Platform

Platform n (with-lca-b/MCA, pairs) Completed TPs (LCA vs MCA) Stage 0 TP/s (LCA vs MCA) Stage 0 p50 (LCA vs MCA) Cold p50 (LCA vs MCA) Stage 0 p95 (LCA vs MCA) Stage 0 Chunks/TP (LCA vs MCA) Stage 0 MSPT p95 (LCA vs MCA) Main CPU / Chunk (LCA vs MCA) Save Shutdown (LCA vs MCA) Comparison Status
paper 6/6 (6) 1479 vs 1442 (1.03x) 4.95 vs 4.82 200ms vs 224ms (1.12x faster) 200ms vs 228ms (1.14x faster) 236ms vs 466ms (1.98x faster) 75.2 vs 17.5 30.9ms vs 193.8ms 1.43ms vs 5.79ms (4.05x) 5.55s vs 86.12s (15.5x faster) PAIRED (n=6)
folia 6/6 (6) 1496 vs 1274 (1.17x) 4.95 vs 4.64 105ms vs 164ms (1.57x faster) 105ms vs 968ms (9.25x faster) 203ms vs 218ms (1.08x faster) 98.0 vs 6.6 18.5ms vs 96.3ms 0.78ms vs 3.29ms (4.23x) 5.02s vs 47.69s (9.5x faster) PAIRED (n=6)
leaf-linear 5/4 (4) 1483 vs 10 (152.08x) 4.95 vs 0.68 200ms vs 3369ms (16.84x faster) 201ms vs 5320ms (26.52x faster) 314ms vs 5178ms (16.49x faster) 76.1 vs 26.6 29.6ms vs 2731.4ms 1.39ms vs 4.07ms (2.93x) 5.20s vs 55.56s (10.7x faster) PAIRED (n=4)
leaf-blinear 2/2 (2) 1456 vs 14 (107.85x) 4.95 vs 1.06 204ms vs 2311ms (11.33x faster) 204ms vs 5704ms (27.89x faster) 381ms vs 5534ms (14.52x faster) 74.8 vs 32.9 28.3ms vs 1790.3ms 1.34ms vs 2.61ms (1.95x) 5.22s vs 8.47s (1.6x faster) PAIRED (n=2)
fabric 5/5 (5) 1494 vs 1491 (1.00x) 4.95 vs 4.95 7ms vs 12ms (1.67x faster) 27ms vs 19ms (0.71x faster) 19ms vs 27ms (1.47x faster) 79.9 vs 79.4 20.8ms vs 22.4ms 1.50ms vs 1.59ms (1.06x) 4.31s vs 11.60s (2.7x faster) PAIRED (n=5)
fabric-c2me 5/5 (5) 1494 vs 1496 (1.00x) 4.95 vs 4.95 7ms vs 11ms (1.47x faster) 14ms vs 16ms (1.19x faster) 18ms vs 22ms (1.22x faster) 79.6 vs 79.6 23.0ms vs 22.0ms 1.17ms vs 1.16ms (0.99x) 3.93s vs 9.43s (2.4x faster) PAIRED (n=5)
neoforge 5/5 (5) 1494 vs 1493 (1.00x) 4.95 vs 4.95 21ms vs 25ms (1.19x faster) 21ms vs 25ms (1.20x faster) 46ms vs 47ms (1.02x faster) 79.5 vs 78.9 23.4ms vs 22.3ms 1.58ms vs 1.57ms (0.99x) 4.72s vs 12.42s (2.6x faster) PAIRED (n=5)
neoforge-c2me 5/5 (5) 1496 vs 1496 (1.00x) 4.95 vs 4.95 16ms vs 16ms (0.95x faster) 17ms vs 17ms (0.98x faster) 49ms vs 54ms (1.10x faster) 79.2 vs 79.8 22.6ms vs 19.3ms 1.18ms vs 1.18ms (1.00x) 4.09s vs 9.09s (2.2x faster) PAIRED (n=5)

2D. Head-to-Head .lca (Archival Mode B, codec1=1) vs .b_linear (leaf-blinear)

Platform n (with-lca-c/MCA, pairs) Completed TPs (LCA vs MCA) Stage 0 TP/s (LCA vs MCA) Stage 0 p50 (LCA vs MCA) Cold p50 (LCA vs MCA) Stage 0 p95 (LCA vs MCA) Stage 0 Chunks/TP (LCA vs MCA) Stage 0 MSPT p95 (LCA vs MCA) Main CPU / Chunk (LCA vs MCA) Save Shutdown (LCA vs MCA) Comparison Status
leaf-blinear 3/2 (2) 1431 vs 14 (105.98x) 4.95 vs 1.06 205ms vs 2311ms (11.29x faster) 205ms vs 5704ms (27.87x faster) 370ms vs 5534ms (14.96x faster) 76.5 vs 32.9 29.1ms vs 1790.3ms 1.36ms vs 2.61ms (1.92x) 5.20s vs 8.47s (1.6x faster) PAIRED (n=2)

2C. Direct .lca Mode A (0-LZ) vs Mode B (LCD2) vs Archival Storage Comparison by Platform

Platform n (Mode A / B / C) World Disk (Mode A vs B vs C) Completed TPs (A vs B vs C) Stage 0 p50 (A vs B vs C) Cold p50 (A vs B vs C) Stage 0 MSPT p95 (A vs B vs C) Save Shutdown (A vs B vs C)
paper 6/6/0 1545.2 MB vs 1278.8 MB 1487 vs 1479 200ms vs 200ms 200ms vs 200ms 31.0ms vs 30.9ms 5.54s vs 5.55s
folia 6/6/0 1545.5 MB vs 1279.0 MB 1496 vs 1496 103ms vs 105ms 103ms vs 105ms 19.3ms vs 18.5ms 5.03s vs 5.02s
leaf-linear 5/5/0 1546.4 MB vs 1279.2 MB 1483 vs 1483 200ms vs 200ms 201ms vs 201ms 28.3ms vs 29.6ms 5.18s vs 5.20s
leaf-blinear 2/2/3 1546.2 vs 1279.5 vs 1266.5 MB 1428 vs 1456 vs 1431 204ms vs 204ms vs 205ms 204ms vs 204ms vs 205ms 28.7ms vs 28.3ms vs 29.1ms 5.19s vs 5.22s vs 5.20s
fabric 5/5/0 1569.2 MB vs 1298.7 MB 1494 vs 1494 8ms vs 7ms 18ms vs 27ms 22.3ms vs 20.8ms 4.33s vs 4.31s
fabric-c2me 5/5/0 1576.0 MB vs 1304.0 MB 1496 vs 1494 8ms vs 7ms 19ms vs 14ms 20.7ms vs 23.0ms 4.09s vs 3.93s
neoforge 5/5/0 1569.6 MB vs 1298.5 MB 1486 vs 1494 15ms vs 21ms 16ms vs 21ms 21.2ms vs 23.4ms 4.27s vs 4.72s
neoforge-c2me 5/5/0 1578.7 MB vs 1305.1 MB 1494 vs 1496 12ms vs 16ms 13ms vs 17ms 18.2ms vs 22.6ms 4.14s vs 4.09s

3. Per-Arm Teleport Throughput, Latency, MSPT & Heap Statistics (Mean +- SD [Median])

Platform Arm Completed TPs Peak TP/s Stage 0 p50 (ms) Stage 0 p95 (ms) Stage 0 Chunks/TP Idle MSPT (ms) Stage 0 MSPT p50 (ms) Stage 0 MSPT p95 (ms) Main CPU/Chunk (ms) Peak Heap (MB)
paper without-lca 1441.67 +- 70.11 (1472.00) 38.23 +- 3.41 (39.83) 224.50 +- 15.31 (219.50) 466.17 +- 305.19 (356.00) 17.50 +- 28.15 (5.13) 2.66 +- 0.10 (2.68) 34.07 +- 59.39 (10.85) 193.85 +- 340.71 (33.56) 5.79 +- 4.20 (4.75) 12240.17 +- 747.84 (12253.00)
paper with-lca 1487.00 +- 4.24 (1488.50) 39.57 +- 0.20 (39.62) 199.50 +- 0.84 (200.00) 221.33 +- 38.06 (206.00) 75.61 +- 0.85 (75.55) 2.19 +- 0.16 (2.18) 26.57 +- 0.90 (26.70) 31.00 +- 2.35 (31.70) 1.45 +- 0.05 (1.45) 13733.50 +- 144.29 (13733.50)
paper with-lca-b 1479.17 +- 8.52 (1481.00) 39.16 +- 0.40 (39.25) 199.67 +- 0.82 (200.00) 235.50 +- 38.02 (227.00) 75.15 +- 1.17 (75.61) 2.12 +- 0.14 (2.09) 25.41 +- 1.33 (25.46) 30.93 +- 2.06 (30.45) 1.43 +- 0.08 (1.40) 13805.17 +- 304.50 (13649.50)
folia without-lca 1274.00 +- 281.10 (1403.00) 34.58 +- 8.60 (39.80) 164.17 +- 20.44 (152.00) 218.33 +- 22.55 (205.00) 6.64 +- 6.98 (4.13) 0.36 +- 0.08 (0.34) 21.77 +- 48.12 (1.97) 96.30 +- 226.96 (3.81) 3.29 +- 1.26 (3.67) 12005.50 +- 203.41 (11991.50)
folia with-lca 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 102.83 +- 0.41 (103.00) 202.67 +- 0.82 (202.50) 97.87 +- 2.34 (98.59) 0.32 +- 0.09 (0.29) 13.13 +- 3.55 (13.24) 19.31 +- 0.94 (19.65) 0.73 +- 0.03 (0.72) 13192.83 +- 175.22 (13218.50)
folia with-lca-b 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 104.67 +- 5.09 (103.00) 202.83 +- 0.75 (203.00) 97.95 +- 2.39 (97.61) 0.37 +- 0.19 (0.29) 13.20 +- 4.91 (13.47) 18.50 +- 3.64 (18.25) 0.78 +- 0.04 (0.79) 13327.83 +- 167.95 (13390.00)
leaf-linear without-lca 9.75 +- 2.63 (10.50) 0.68 +- 0.16 (0.69) 3369.00 +- 1005.36 (3379.50) 5178.25 +- 1052.94 (5231.50) 26.56 +- 3.23 (26.23) 2.60 +- 0.23 (2.56) 2212.87 +- 398.33 (2308.95) 2731.45 +- 234.36 (2715.05) 4.07 +- 0.16 (4.06) 15464.25 +- 248.57 (15380.50)
leaf-linear with-lca 1483.20 +- 4.66 (1484.00) 39.37 +- 0.23 (39.40) 200.40 +- 0.55 (200.00) 325.60 +- 26.43 (309.00) 75.55 +- 0.86 (75.63) 2.14 +- 0.11 (2.12) 24.28 +- 0.95 (24.18) 28.30 +- 1.10 (27.84) 1.35 +- 0.05 (1.35) 14023.80 +- 179.76 (14030.00)
leaf-linear with-lca-b 1482.80 +- 7.01 (1485.00) 39.32 +- 0.33 (39.45) 200.00 +- 0.71 (200.00) 314.00 +- 22.39 (305.00) 76.14 +- 0.97 (76.19) 2.07 +- 0.14 (2.02) 25.09 +- 1.00 (25.21) 29.56 +- 1.68 (28.69) 1.39 +- 0.05 (1.37) 13920.20 +- 185.86 (13960.00)
leaf-blinear without-lca 13.50 +- 4.95 (13.50) 1.06 +- 0.34 (1.06) 2311.00 +- 380.42 (2311.00) 5534.00 +- 1240.27 (5534.00) 32.86 +- 7.14 (32.86) 2.56 +- 0.17 (2.56) 1311.08 +- 296.93 (1311.08) 1790.27 +- 38.58 (1790.27) 2.61 +- 0.20 (2.61) 14572.00 +- 316.78 (14572.00)
leaf-blinear with-lca 1428.00 +- 1.41 (1428.00) 36.58 +- 0.04 (36.58) 204.50 +- 0.71 (204.50) 381.00 +- 35.35 (381.00) 75.25 +- 0.32 (75.25) 2.31 +- 0.43 (2.31) 24.77 +- 0.00 (24.77) 28.70 +- 0.76 (28.70) 1.39 +- 0.02 (1.39) 13860.50 +- 139.30 (13860.50)
leaf-blinear with-lca-b 1456.00 +- 8.48 (1456.00) 37.98 +- 0.39 (37.98) 204.00 +- 0.00 (204.00) 381.00 +- 32.53 (381.00) 74.85 +- 0.03 (74.85) 2.12 +- 0.04 (2.12) 23.66 +- 1.79 (23.66) 28.29 +- 1.73 (28.29) 1.34 +- 0.00 (1.34) 13755.00 +- 62.23 (13755.00)
leaf-blinear with-lca-c 1430.67 +- 4.51 (1431.00) 36.70 +- 0.23 (36.75) 204.67 +- 0.58 (205.00) 370.00 +- 26.85 (355.00) 76.52 +- 0.51 (76.39) 2.21 +- 0.12 (2.16) 25.07 +- 0.18 (25.05) 29.07 +- 1.67 (28.70) 1.36 +- 0.04 (1.35) 13970.33 +- 328.73 (14148.00)
fabric without-lca 1491.40 +- 6.77 (1496.00) 39.72 +- 0.34 (39.95) 12.00 +- 6.16 (9.00) 27.40 +- 9.50 (25.00) 79.43 +- 0.52 (79.59) 1.29 +- 0.04 (1.29) 19.07 +- 0.83 (19.12) 22.36 +- 2.69 (21.14) 1.59 +- 0.05 (1.60) 11535.80 +- 501.62 (11718.00)
fabric with-lca 1494.20 +- 4.03 (1496.00) 39.86 +- 0.20 (39.95) 7.60 +- 0.55 (8.00) 23.20 +- 5.36 (23.00) 78.98 +- 1.28 (79.39) 1.34 +- 0.21 (1.29) 18.72 +- 0.45 (18.51) 22.31 +- 2.47 (21.38) 1.58 +- 0.06 (1.58) 10878.60 +- 828.27 (10827.00)
fabric with-lca-b 1494.40 +- 3.58 (1496.00) 39.87 +- 0.18 (39.95) 7.20 +- 0.45 (7.00) 18.60 +- 3.65 (19.00) 79.93 +- 0.70 (79.73) 1.30 +- 0.04 (1.31) 19.09 +- 0.82 (19.44) 20.81 +- 0.99 (20.87) 1.50 +- 0.07 (1.45) 10838.60 +- 249.13 (10921.00)
fabric-c2me without-lca 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 10.60 +- 3.58 (9.00) 22.40 +- 4.98 (23.00) 79.62 +- 0.86 (79.86) 1.12 +- 0.08 (1.10) 18.11 +- 0.93 (18.50) 21.97 +- 5.08 (19.66) 1.16 +- 0.04 (1.15) 11587.60 +- 702.62 (11918.00)
fabric-c2me with-lca 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 7.80 +- 0.45 (8.00) 23.40 +- 6.39 (26.00) 79.58 +- 0.62 (79.42) 1.19 +- 0.15 (1.18) 17.91 +- 0.78 (18.14) 20.66 +- 1.99 (20.35) 1.17 +- 0.03 (1.17) 12247.60 +- 120.97 (12290.00)
fabric-c2me with-lca-b 1493.60 +- 5.37 (1496.00) 39.83 +- 0.27 (39.95) 7.20 +- 0.45 (7.00) 18.40 +- 3.36 (19.00) 79.56 +- 0.77 (79.47) 1.19 +- 0.13 (1.21) 18.52 +- 0.49 (18.54) 22.98 +- 2.46 (22.51) 1.17 +- 0.01 (1.17) 11834.20 +- 431.49 (11933.00)
neoforge without-lca 1493.00 +- 4.12 (1496.00) 39.80 +- 0.21 (39.95) 24.80 +- 6.38 (29.00) 46.60 +- 3.65 (45.00) 78.86 +- 0.67 (78.90) 1.35 +- 0.10 (1.30) 19.95 +- 1.06 (19.99) 22.29 +- 2.26 (21.83) 1.57 +- 0.08 (1.61) 10312.20 +- 442.06 (10399.00)
neoforge with-lca 1486.20 +- 13.42 (1496.00) 39.46 +- 0.67 (39.95) 14.60 +- 1.14 (15.00) 41.60 +- 5.94 (43.00) 78.76 +- 1.60 (79.50) 1.36 +- 0.11 (1.31) 19.39 +- 0.74 (19.40) 21.18 +- 1.52 (21.32) 1.56 +- 0.04 (1.56) 10917.60 +- 133.16 (10882.00)
neoforge with-lca-b 1494.20 +- 2.05 (1495.00) 39.86 +- 0.10 (39.90) 20.80 +- 7.69 (17.00) 45.80 +- 9.09 (45.00) 79.50 +- 1.11 (79.75) 1.44 +- 0.13 (1.48) 19.63 +- 1.49 (19.41) 23.42 +- 2.52 (24.06) 1.58 +- 0.08 (1.61) 10420.00 +- 509.75 (10610.00)
neoforge-c2me without-lca 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 15.60 +- 2.70 (15.00) 53.80 +- 5.07 (54.00) 79.76 +- 0.85 (80.02) 1.23 +- 0.22 (1.17) 16.66 +- 0.59 (16.80) 19.31 +- 1.62 (18.58) 1.18 +- 0.04 (1.19) 11769.80 +- 158.18 (11774.00)
neoforge-c2me with-lca 1494.40 +- 3.58 (1496.00) 39.87 +- 0.18 (39.95) 12.20 +- 1.48 (12.00) 42.00 +- 4.74 (42.00) 79.27 +- 1.36 (79.53) 1.24 +- 0.11 (1.22) 16.11 +- 1.12 (16.52) 18.16 +- 2.02 (18.89) 1.17 +- 0.03 (1.17) 11931.40 +- 321.92 (12048.00)
neoforge-c2me with-lca-b 1496.00 +- 0.00 (1496.00) 39.95 +- 0.00 (39.95) 16.40 +- 5.68 (16.00) 49.00 +- 11.34 (51.00) 79.23 +- 0.48 (79.10) 1.31 +- 0.21 (1.30) 16.74 +- 1.65 (17.19) 22.61 +- 4.80 (24.46) 1.18 +- 0.05 (1.15) 12181.80 +- 225.68 (12285.00)

Live Cold-Restart Persistence Verification (--verify-persistence)

Cold-restart round-trip (Boot #1 in-game block, container, and mob mutations -> save-all flush -> stop -> cold Boot #2 from .lca files and 0 .mca files -> RCON/bot query):

Platform .lca / .mca Files oak_stairs[facing=east,half=top,waterlogged=true] repeater[delay=3,facing=south,locked=false,powered=false] Barrel (diamond_sword[damage=37], 19x golden_apple) Chest (42x emerald_block) Persistent villager (1b) Persistent sheep (Color=11b) Verdict
paper 772 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
folia 778 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
leaf-linear 779 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
leaf-blinear 781 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
fabric 767 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
fabric-c2me 767 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
neoforge 767 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)
neoforge-c2me 767 / 0 PASS PASS PASS PASS PASS PASS PASS (6/6)

4. Chunky (1.5.3) World Pre-Generation & Existing-World Scan Benchmark

Measured on bamboo (Threadripper 7970X, cores 0-23,32-55, -Xms16G -Xmx16G, fixed seed 8675309) comparing without-lca (.mca / .linear / .b_linear), with-lca (LeafAnvil .lca Mode A 0-LZ), and with-lca-b (LeafAnvil .lca Mode B LCD2):

4A. Fresh-World Pre-Generation from Scratch (/chunky radius 512, 1,024 x 1,024 blocks, 4,225 FULL chunks / 8,627 total saved voxel chunks across 34 region files):

Platform Arm Format Chunky Gen Time (s) Effective CPS (Peak CPS) ChunkHolderManager Save (s, Chunks) Server Stop (s) End-to-End Gen + Save (s) Disk Size (MB Raw / Compacted)
paper (26.2) without-lca .mca (Zlib) 92.82 s 45.5 cps (61.0) 7.07 s (649) 9.41 s 102.23 s 57.59 MB
paper (26.2) with-lca .lca (Mode A) 10.01 s (9.3x) 422.1 cps (425.3) 0.04 s (1,435, 177x) 2.20 s (4.3x) 12.21 s (8.4x) 29.66 / 29.11 MB (-49.5%)
paper (26.2) with-lca-b .lca (Mode B) 10.26 s (9.0x) 411.8 cps (412.0) 0.04 s (1,301, 177x) 2.40 s (3.9x) 12.66 s (8.1x) 24.92 / 24.47 MB (-57.5%)
folia (26.2) without-lca .mca (Zlib) 103.36 s 40.9 cps (51.2) 7.37 s (1,286) 9.62 s 112.98 s 57.56 MB
folia (26.2) with-lca .lca (Mode A) 20.11 s (5.1x) 210.1 cps (213.2) 0.04 s (1,286, 184x) 2.09 s (4.6x) 22.20 s (5.1x) 30.04 / 29.28 MB (-49.1%)
folia (26.2) with-lca-b .lca (Mode B) 20.36 s (5.1x) 207.5 cps (212.0) 0.04 s (1,504, 184x) 2.10 s (4.6x) 22.46 s (5.0x) 25.16 / 24.58 MB (-57.3%)
leaf-linear (26.2) without-lca .linear (LINEAR_V2) 48.29 s 87.5 cps (194.8) 2.42 s (1,076) 4.42 s 52.71 s 31.12 MB
leaf-linear (26.2) with-lca-b .lca (Mode B) 10.26 s (4.7x) 411.8 cps (411.8) 0.06 s (1,660, 40x) 2.02 s (2.2x) 12.28 s (4.3x) 25.03 / 24.53 MB (-21.2%)
leaf-blinear (26.2) without-lca .b_linear (B_LINEAR) 10.26 s 411.8 cps (428.1) 0.06 s (1,304) 3.02 s 13.28 s 26.13 MB
leaf-blinear (26.2) with-lca-b .lca (Mode B) 10.51 s (1.0x) 402.0 cps (413.3) 0.05 s (1,287) 2.22 s (1.36x) 12.73 s (1.04x) 25.00 / 24.52 MB (-6.2%)

4B. Existing-World Scan (/chunky radius 1024, 2,048 x 2,048 blocks, 16,641 chunks over the 767-region / 293,317-chunk template):

Platform Arm Format Chunky Scan Time (s) Effective CPS (Peak CPS) ChunkHolderManager Save (s) Server Stop (s) Total Time (s) Template Disk (MB)
paper (26.2) without-lca .mca (Zlib) 2.01 s 8,279.1 cps (8,096.1) 0.49 s 2.60 s 4.61 s 2,527.92 MB
paper (26.2) with-lca .lca (Mode A) 2.01 s 8,279.1 cps (9,084.8) 0.01 s 2.20 s 4.21 s 1,535.33 MB (-39.3%)
paper (26.2) with-lca-b .lca (Mode B) 2.01 s 8,279.1 cps (8,662.8) 0.01 s 2.20 s 4.21 s 1,269.86 MB (-49.8%)
folia (26.2) without-lca .mca (Zlib) 7.10 s 2,343.8 cps (8,213.5) 0.60 s 2.70 s 9.80 s 2,532.09 MB
folia (26.2) with-lca .lca (Mode A) 2.10 s (3.4x) 7,924.3 cps (8,994.6) 0.12 s 2.22 s 4.32 s (2.3x) 1,535.73 MB (-39.4%)
folia (26.2) with-lca-b .lca (Mode B) 2.02 s (3.5x) 8,238.1 cps (8,893.3) 0.10 s 2.10 s 4.12 s (2.4x) 1,270.23 MB (-49.8%)
leaf-linear (26.2) without-lca .linear (LINEAR_V2) 4.51 s 3,689.8 cps (3,895.0) 0.04 s 2.22 s 6.73 s 1,672.15 MB
leaf-linear (26.2) with-lca-b .lca (Mode B) 2.01 s (2.2x) 8,279.1 cps (8,917.5) 0.01 s 2.02 s 4.03 s (1.7x) 1,270.24 MB (-24.0%)
leaf-blinear (26.2) without-lca .b_linear (B_LINEAR) 5.01 s 3,321.6 cps (3,885.0) 0.01 s 2.42 s 7.43 s 1,407.56 MB
leaf-blinear (26.2) with-lca-b .lca (Mode B) 2.01 s (2.5x) 8,279.1 cps (8,620.7) 0.01 s 2.22 s 4.23 s (1.8x) 1,270.24 MB (-9.8%)

5. LeafTreeGen (1.1.0) 24-Cell Clean-Boot Tree Generation & .lca Storage Matrix (seed=8675309)

Measured across 4 platforms (Paper 26.2, Folia 26.2, NeoForge 26.1.2, Fabric 26.1.2), 2 storage formats (vanilla .mca vs .lca Mode A 0-LZ), and 3 LeafTreeGen tree modes (NONE vanilla baseline, DATAPACK jigsaw structures, PROCEDURAL TreeBlockBatcher) on high-density dry-land forest and birch_forest continents (surfaceY = 73..82, ~391,500 custom tree blocks placed per cell across Stage 1 64-chunk gen and Stage 2 128-chunk 4-wave rolling gen + save-all, followed by Stage 3 192-chunk reload/unload churn):

Platform Mode Cold Boot (.mca / .lca) P1: 64-Chunk Gen Wall (.mca / .lca) P1: Server Chunk Win (.mca / .lca ms) P2: 128-Chunk Rolling Gen+Save (.mca / .lca) P3: 192-Chunk Reload/Unload (.mca / .lca) Region Disk Size (.mca vs .lca)
Paper 26.2 NONE 16.31 s / 18.35 s 5.02 s / 4.79 s (13.36 ch/s) 3,337 / 3,021 ms 10.30 s / 12.81 s 1.18 s / 2.03 s 28.89 MB vs 7.53 MB (-73.9%)
Paper 26.2 DATAPACK 29.30 s / 30.81 s 7.75 s / 7.41 s (8.64 ch/s) 4,551 / 4,362 ms 19.70 s / 19.55 s 1.14 s / 1.26 s 29.10 MB vs 8.04 MB (-72.4%)
Paper 26.2 PROCEDURAL 15.62 s / 16.04 s 7.78 s / 7.44 s (8.60 ch/s) 2,770 / 2,623 ms (172,797 blocks) 19.64 s / 20.01 s (218,748 blocks) 0.98 s / 1.22 s (194.9 / 157.8 ch/s) 28.78 MB vs 7.54 MB (-73.8%)
Folia 26.2 NONE 15.36 s / 16.77 s 4.83 s / 4.90 s (13.26 ch/s) 1,140 / 1,133 ms 11.10 s / 12.82 s 3.80 s / 3.25 s (59.0 ch/s) 28.85 MB vs 7.42 MB (-74.3%)
Folia 26.2 DATAPACK 28.72 s / 29.57 s 8.11 s / 8.27 s (7.89 ch/s) 1,777 / 1,900 ms 22.47 s / 23.97 s 3.75 s / 3.87 s 28.98 MB vs 7.86 MB (-72.9%)
Folia 26.2 PROCEDURAL 13.93 s / 15.05 s 8.02 s / 7.72 s (8.29 ch/s) 1,389 / 1,132 ms (172,795 blocks) 22.40 s / 23.66 s (218,768 blocks) 3.70 s / 4.07 s 28.64 MB vs 7.38 MB (-74.2%)
NeoForge 26.1.2 NONE 13.57 s / 14.84 s 8.95 s / 10.09 s (7.15 ch/s) 6,807 / 7,570 ms 18.33 s / 21.42 s 1.44 s / 1.46 s 24.85 MB vs 8.76 MB (-64.8%)
NeoForge 26.1.2 DATAPACK 28.31 s / 33.94 s 26.29 s / 36.75 s (2.43 ch/s) 12,917 / 18,013 ms 87.04 s / 95.12 s 1.39 s / 2.11 s 25.02 MB vs 8.87 MB (-64.6%)
NeoForge 26.1.2 PROCEDURAL 11.81 s / 12.60 s 15.76 s / 15.97 s (4.06 ch/s) 6,747 / 6,724 ms (169,173 blocks) 30.58 s / 33.22 s (218,835 blocks) 1.36 s / 1.20 s (160.0 ch/s, +13%) 24.48 MB vs 8.86 MB (-63.8%)
Fabric 26.1.2 NONE 14.34 s / 15.83 s 9.57 s / 10.73 s (6.69 ch/s) 7,514 / 7,769 ms 17.68 s / 21.37 s 1.43 s / 1.46 s 24.79 MB vs 8.73 MB (-64.8%)
Fabric 26.1.2 DATAPACK 34.09 s / 35.12 s 34.63 s / 35.93 s (1.85 ch/s) 17,055 / 16,393 ms 109.58 s / 94.73 s (+13.6%) 1.40 s / 2.20 s 24.96 MB vs 8.84 MB (-64.6%)
Fabric 26.1.2 PROCEDURAL 12.79 s / 13.54 s 15.63 s / 15.50 s (4.13 ch/s) 6,346 / 6,062 ms (169,176 blocks) 31.47 s / 31.70 s (218,835 blocks) 1.62 s / 1.29 s (148.8 ch/s, +20%) 24.46 MB vs 8.85 MB (-63.8%)
# Run rotating A/B rounds across Paper, Folia, Fabric, and NeoForge (alternating arm order each round)
python3 scripts/bench_lca_multiplatform.py --mode rotate --platforms paper,folia,fabric,neoforge

# Run cold-restart block, container, and mob persistence verification across all 8 platform targets
python3 scripts/bench_lca_multiplatform.py --verify-persistence --platforms paper,folia,leaf-linear,leaf-blinear,fabric,fabric-c2me,neoforge,neoforge-c2me

# Gracefully finish the active run, flush rolling statistics, and exit
python3 scripts/bench_lca_multiplatform.py --stop

Build & Verification

# Build the unified multi-loader LeafAnvil-<version>.jar (Paper/Folia/Spigot, Fabric, NeoForge)
# and the free MIT-licensed LeafAnvil-Net-<version>.jar (Fabric & NeoForge client RX mod)
./gradlew :resource:LeafAnvil:build

# Run unit, client RX packaging, & bytecode interception tests
./gradlew :resource:LeafAnvil:leafanvil-common:test
./gradlew :resource:LeafAnvil:leafanvil-net:test
./gradlew :resource:LeafAnvil:leafanvil-fabric:test

# Run .lca codec parity, CBIT/PBIT, CoW concurrency, and 4-tier bit-rot healing tests
./gradlew :anvil-api:test --tests "*LcaRegionCodecTest*"

Documentation (resource/LeafAnvil/docs/ & api/anvil-api/docs/)


Licensing (LeafAnvil Commercial Carrier & Open .lca Format)