A pass is keyed by the content of the harness's dependencies, so it is
also valid in any worktree with the same files. Passes are published to
parity-shared/ beside the shared vanilla data, and a worktree adopts one
whose key matches its own files instead of running the harness: an
agent's gate only runs what its own changes affect, instead of the whole
suite from scratch. Every runner on the machine also shares
PARITY_MACHINE_JOBS slots, so several agents' gates can't overload it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Desert, snowy, swamp and jungle nights, lush, sulfur and dripstone caves,
warm and frozen oceans, crimson forests and slime chunks. A cube mob grown
by finalizeSpawn takes its constructor's head rotation from the entity's
unseeded random (AbstractCubeMob.onSyncedDataUpdated), so that yaw alone is
not compared.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/test_natural_spawn_parity.py compares whole chunk spawn passes,
spawnCategoryForPosition from chosen sites and chunk generation packs
against vanilla's NaturalSpawner on generated box worlds.
Divergences fixed on the way: the reduced water ambient tag is
reduce_water_ambient_spawns, and the room check uses each type's spawn
dimensions scale (4 for slimes and magma cubes, 2 for sulfur cubes).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fixes from 3076 to 3459 (game event, cat, frog, painting, goat, goat
horn, filtered text, POI, blending, spawn egg, effect duration, archaeology,
double-sided signs, the 1.20 experiment and chunk statuses), the block
entities their schemas add, and block entities' remainders (a decorated
pot's shards stay next to its sherds). The sample world generator logs in
to 1.18.2 to 1.19.4 servers and gates the scene by version.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Opening a world saved by an older version runs the file fixes, level.dat and
the DataFixers upgrades of every file instead of refusing it. The save
round-trip harness treats a lit campfire's cooking progress as a timer in the
vanilla views: Horizon saves it as loaded (stage 3 checks it exactly), and it
moves on by however many ticks each vanilla server ran before data get.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/parity_shim.py stands in for java and javac in harness runs: it
records each oracle run's outputs and the content of every file under the
paths its arguments name, and replays the run while those are unchanged.
After a Rust-only change a harness only pays for its Rust side
(block_update 53 s -> 5 s). A harness that fails while reusing oracle
results is rerun fresh, and stops reusing them if that passes; --all and
--fresh never reuse them.
Harness probes built in the dev profile are optimised (overflow checks
and debug assertions kept), and the chunk generation harness runs its
vanilla and Horizon sides concurrently.
HORIZON_NO_AUDIO keeps the game off the speakers; tools/parity.py sets it,
since the save round-trip harness launches the game on a virtual screen.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DataFixerUpper reads typed fields without taking them out of a record's
remainder, so a typed field a fix drops comes back on write with its value
from the record's last read: entities keep their old ArmorItems, HandItems,
body_armor_item, saddle, Trident and Particle next to the new fields, as
vanilla 26.3 writes them. Fixes that write and read entities refresh that
map. The area effect cloud particle moves to custom_particle (4420), boats
keep Type (BoatSplitFix's remainder update misses), and region files with a
short header read as empty, as vanilla's RegionFile does. The sample worlds
now use the item formats each version accepts and enable its experiments.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/parity.py traces each harness (its Python, Java oracle sources and
the Rust sources of the probes it runs, from Cargo's dep-info) and reruns
it only when one of those, the vanilla artifacts or the toolchains change.
Harnesses run in parallel within the available memory, and their probes
build without LTO and with 16 codegen units into .local/parity-target: a
one-line change rebuilds every horizon-world probe in 3 seconds instead of
122. The full suite takes about 20 minutes instead of an hour.
tools/gate.py is the check a branch passes before integration, and
tools/supervise.py manages the per-area worktrees: status, new worktrees,
saving unfinished work, and integrating a branch only when the gate passes.
Harnesses find probes through lab.cargo_target() instead of a hard-coded
target/, the save round-trip harness builds the binaries it launches, and
the mob AI harness now fails on divergences like every other harness
(--report-only keeps the old behaviour). Debug builds keep line tables
only, so worktrees stay small.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/rea_vanilla.py runs REA's JADX provider over the pinned client and
server JARs: class search, class members, method bodies and callers
(including lambdas). REA and the JADX engine are pinned in
research-lock.json.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Uncommitted changes the agent had in progress when it was stopped; may not
build or pass its harnesses yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mth.atan2 builds its tables with StrictMath.asin (fdlibm, ported) and
HotSpot's cos; the controls multiply by the folded RAD_TO_DEG float
constant. Mob.checkDespawn runs before the tick and, in 26.3, persistent
mobs only reset their idle time near a player. NearestAttackableTargetGoal
always requires a line of sight and ignores mustReach when searching.
MeleeAttackGoal starts its canUse clock at game time 0. The harness now
records every random draw of the vanilla mobs with its caller, the random
state per tick, and starts each scenario at game time 1000.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Send the join packets in vanilla's order with the permission level entity
event, recipe book, server data, ticking state and post effects; announce
the newcomer only to the others; list players with every info action;
use vanilla's level key order and border warning time; sync the game time
every 20 ticks without clocks and the overworld clock only when it jumps.
Build the command tree from the commands report for the player's
permission level and server kind, with vanilla's suggestion providers
(VanillaCommandSuggestionsExporter), and answer command suggestions.
The tree matches what a vanilla integrated server sends node for node.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mobs::spawn_rules ports SpawnPlacements over a SpawnLevel trait the game's
MobWorld implements: the placement types with per-type isValidSpawn
(leaves, ice, magma for fire-immune types) and isValidEmptySpawnBlock with
each type's dangerous and immune blocks, and every registered predicate in
vanilla's draw order: the dimension's monster light test and block light
limit, thunder, peaceful, spawn reasons, guardians' sky from below water,
drowned, bats, ghasts, ocelots, slimes (surface chance and slime chunks),
strays under powder snow, striders, turtles, polar bears, tropical fish,
nautiluses, silverfish and endermites near players, patrols and the
spawnable-on tags.
Divergences fixed on the way: nether monsters ignored light, hoglins and
piglins had extra checks, slimes spawned outside slime chunks, ocelots and
ghasts skipped their odds, bats ignored the surface and their draw order,
and several predicates read the wrong blocks.
tools/test_spawn_rules_parity.py: 18,669 cases over 87 entity types match.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ToggleKeys holds the sneak, sprint, attack and use ToggleKeyMappings with
the dispatch of KeyboardHandler.keyPress (releases always reach them;
presses and SDL key repeats only without a screen), MouseHandler.onButton,
Gui.setScreen (releaseAll / restoreToggleStatesOnScreenClosed) and
LocalPlayer.respawn (resetToggleKeys, which Horizon lacked). Presses behind
a screen no longer flip a toggled key. tools/test_toggle_keys_parity.py
drives the real Options mappings through the real static KeyMapping
functions (8,000 sequences).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Situation captures the screen music, the menu without a player, End boss
music, creative abilities and underwater selection and the music volume;
the harness calls the real getSituationalMusic/getMusicVolume with a real
camera attribute probe for every dimension and biome.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Horizon had the Auto-Jump option but never jumped. MotionWorld::move_self
now runs LocalPlayer.updateAutoJump after the player's own move (behind
canAutoJump: option, autoJumpTime, ground, shift, input, block jump
factor), the next aiStep presses jump for one tick (input.makeJump) without
toggling creative flight, and the game feeds the option each tick.
tools/test_auto_jump_parity.py runs the real updateAutoJump on a LocalPlayer
shell in an independently built copy of the resident test scene with real
block states (10,568 cases, 1,200 jumps).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the beam geometry into block_entity_render::beacon_beam so the
block entity render harness can replay it against vanilla submit calls
(positions, UVs and colours of the solid beam and glow per section).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The oracle runs the real updateIsUnderwater, UnderwaterAmbientSoundHandler
and BiomeAmbientSoundsHandler over the real SoundEngine, comparing every
level random draw, loop fade, mood sample and playing instance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Entity loot tables now run on the faithful loot engine with an entity
context: the victim, the killer and direct attacker, the damage type and
its tags, the last player to hurt it and their luck, entity predicates
(flags, equipment, components, sheep, cube size, raider captains,
vehicles, entity type tags), damage source predicates, killed_by_player,
random_chance_with_enchanted_bonus, enchanted_count_increase, furnace_smelt
and tag entries. Draws come from the world's random sequence for the table.
mobs::drops ports dropAllDeathLoot: Monster loot for babies, equipment
drops in EquipmentSlot order with Looting's equipment_drops effect and
durability, the fox's mouth item, the wither's star, the charged creeper
head hook, horse inventories and chests, and getBaseExperienceReward with
the equipment bonus and baby zombie reward, split into orbs.
Fixes found on the way: sheep dropped their wool twice, baby monsters
dropped nothing, armour bonus experience and the equipment slot order were
wrong, hoglins were fire immune, and fishing treasure was unreachable (the
hook predicate format changed).
tools/test_entity_death_loot_parity.py: 1,578 cases over 87 entity types
match vanilla.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add horizon_client::hud_layout (crosshair and attack indicator, hotbar
with selection, off hand by main arm, item slots with the pop squeeze,
armour, hearts including frozen and absorbing kinds, food, air bubbles,
mount hearts capped at 30, experience bar and level) and the
test_hud_parity harness whose oracle runs vanilla Hud methods on a
stand-in client. The game now draws these through hud_layout; the
experience level is centred with integer division like vanilla,
spectators get no crosshair and mount hearts show in creative too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The game now ends each frame with FramerateLimiter.limitDisplayFPS below
260 fps (sleep, overshoot averaging, spin-wait) instead of waking the event
loop at a computed deadline, and the limiter honours Thread.interrupted.
tools/test_frame_limiter_parity.py runs the real limiter bytecode with its
clock calls rewritten to a scripted virtual clock (5,288 cases).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Horizon spawned mobs whatever the spawn_mobs and spawn_monsters game rules
said; natural spawning now follows ServerChunkCache.tickChunks (the chunks
are still shuffled every tick) and the custom spawners tick only with
spawn_mobs. The slimes this put in the test world were what woke its sculk
sensor.
Falling blocks keep their UUID and loaded compound while they move, read
NoGravity, DropItem and a signed Time, and drop themselves as an item
after 600 ticks off the ground instead of landing. A crafter's saved
triggered flag only follows the block state when the state changes
(CrafterBlock.neighborChanged), and a horse saves the attribute instances
it was loaded with, as AttributeMap.pack does.
The harness compares vanilla log lines without their time stamps, accepts
extra entities only in chunks vanilla never generated, treats a falling
block's Time as a timer, and keeps the lit furnace's slots out of the
vanilla view. Its scene starts the falling block far below the 600-tick
limit and keeps the campfire from finishing during the run. The known-gap
list is empty.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The oracle drives MusicManager over the real SoundEngine with the situation
as input; the script covers delays, replacements, frequencies, fades,
loading screens, pauses, silent starts, toasts and unplayable events.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a decoding capture proxy (tools/net_capture.py), an offline vanilla
client launcher (tools/vanilla_client.py) and a vanilla codec oracle that
decodes every captured clientbound packet with vanilla's StreamCodecs.
Fix the entity data serializer ids (vector3 onwards were off by one), the
sheep wool and piglin baby indices, send mob variants, villager data and
equipment, and send entity movement at each type's update interval and
tracking range with position syncs, like ServerEntity.sendChanges.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
VanillaMobAiOracle runs vanilla mobs in a real headless flat server level,
ticking them as ServerLevel does while the rest of the level stands
still; mob_ai_probe replays the same scenarios on Horizon. Both print each
mob's AI-visible state every tick (position, motion, rotations, goals,
targets, navigation, controls, input, brain activities and memories,
sounds) and tools/test_mob_ai_parity.py reports the first divergence.
Fixes found: float entity dimensions, Mth.sin/cos in moveRelative and the
strafe/swim controls, Mth.atan2 and the float thresholds of the controls,
RandomPos float constants, the ambient sound clock starting at zero, hurt
and death sounds drawn from the mob's own random (also when burning),
knockback for every damage type not tagged no_knockback, Entity.move's
collision flags, restitution and footstep/swim sounds, and mob sound
volumes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Camera.tick, tickFov, alignWithEntity with getMaxZoom over visual-shape
clipping, setRotation and move with JOML quaternion arithmetic,
calculateFov/calculateHudFov, getFieldOfViewModifier (now including
flight), ClientAvatarState walk/bob history (the backwards-extrapolated
walk distance and velocity-based bob) and GameRenderer.bobHurt/bobView as
the JOML pose matrix with property flags. The game renders through them,
adding hurt tilt, the camera distance and scale attributes, the mirrored
front camera and the sleeping camera. tools/test_camera_parity.py compares
18,502 cases against the real classes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a differential harness (test_block_entity_render_parity) whose Java
oracle runs each vanilla BlockEntityRenderer on a real block entity and
captures the submitted model vertices, textures, light and text calls.
Horizon now animates shulker lids, bell shakes, active conduits (cage,
wind and the camera-facing eye), skull jaws and ears, pot wobble and
sherd sides; signs wrap lines like StringSplitter and measure spaces
with the font; books use the Mth sine table and player heads the slim
default skin.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The oracle runs vanilla SoundManager/SoundEngine on OpenAL Soft null device
(inline executor, silent buffers) and replays scripts of plays, stops,
category gains, option volumes, channel ends, pauses, queued, delayed and
looping sounds and tickable ambient instances, printing every AL source
parameter, isActive, subtitle notification and pool count per step.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The item model harness ran out of memory, so it now streams both outputs and
looks Horizon's rows up by offset. With the vanilla side no longer crashing on
swap-animation lookups (properties now come from the baked set), on armour
(trim sprites are generated by the items atlas's paletted_permutations source
once SpriteSources is bootstrapped) or on compasses (item tags are bound from
the data pack), 197,000 more stacks were compared and 121,841 differed.
Horizon now composes transformations as JOML 1.10.9 does: rotationXYZ and
axis-angle quaternions with cosFromSin, and Matrix4f operations that take the
identity, translation, affine or generic path by matrix properties, with
unfused a * b + c. Float RGB tint constants are scaled by 255, the shulker box
renderer always encodes its texture, and a new palette module ports
PalettedPermutations and PaletteMapping so trim sprites exist on the Horizon
side too.
196,290 resolved stacks match; 2,582 cases are still rejected by vanilla's
stand-in client and not compared.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The chunk generation harness found 70 differing chunks across three seeds in
the overworld, Nether and End. Proto chunks, biomes and decoration now span
the dimension type's build range rather than the noise height, so features
above y=128 in the Nether and End city upper floors survive; the decoration
level reports no sky light, as unlit vanilla chunks do, so mushrooms and crops
pass or fail their light checks the same way; beehive decorators store their
bees; iceberg ellipses skip NaN distances; double slabs never waterlog;
mineshaft spawners save an empty SpawnPotentials list; and the dump example
forgets structure starts between rectangles, as the oracle starts each with
fresh chunks.
The oracle prints absent block entity keys as vanilla loads them from an empty
tag, and the harness treats vanilla's save-format equivalences (plain sign
text, absent empty banner patterns, unfiltered sign messages) as equal. All 2,352
chunks over 225 structure sites match with zero block, biome or block entity
differences.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add SoundRegistry (sounds.json with replace, subtitles, file validation and
event references weighing as much as their event) and a SoundEngine model,
with a differential harness resolving every sounds.json event against the
real SoundManager.prepare and SimpleSoundInstance.getOrResolve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ports AbstractNautilus, Nautilus, ZombieNautilus and their brains (taming
with pufferfish, saddles and nautilus armour, riding with a dash, breeding,
spawning below sea level, drowned riders), SulfurCube on the slime code
(swallowed blocks and their archetypes, shears, TNT cubes, splitting) and
CamelHusk on the camel (its own saddle and sounds, husk jockeys). Camels now
save their saddle and stay tame after loading, and horse armour layers load.
The entity model harness compares 150 layers, all matching, including the new
mobs and previously unmapped armour, harness, fish and villager layers; mob
attributes pass with the new types.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The warden's walk, idle wobble and pulsating spots now use Mth's sine table
as WardenModel and WardenRenderer do, and baby sniffers no longer apply
SnifferAnimation.BABY_TRANSFORM: SnifferRenderer wraps the baby geometry in a
plain SnifferModel, so that animation never runs in 26.3. The harness now
drives the illusioner through explicit arm poses like the other illagers, and
the oracle shows its hat as IllusionerRenderer does. All 132 compared layers
match over 3,876 cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>