legacy-motion-physics

name	legacy-motion-physics
description	Use when reasoning about knockback, velocity, trajectories, player movement, or any motion math in Mental — the exact Minecraft motion model (identical 1.7.10 → modern for knocked players), its constants, integration order, and the traps that produce block-scale errors.

The Minecraft motion model (1.7.10 ≡ 1.8.9 ≡ modern, for knocked players)

Constants

Thing	Value
Gravity	0.08 / tick
Vertical drag	× 0.98 (after gravity)
Horizontal drag, airborne	× 0.91
Horizontal drag, on stone	× 0.546 (= 0.6 slipperiness × 0.91)
Walk / sprint ground accel	0.1 / 0.13 per tick
Air accel (held key)	0.02 / 0.026 (sprint)
Jump impulse	0.42
Velocity packet clamp	±3.9 per axis (motion × 8000 as shorts)
Terminal walk/sprint speed	accel × 0.546/0.454 stored ⇒ moves 0.22 / 0.286 per tick (4.3 / 5.6 m/s)

Integration order per tick (every version)

1. Input acceleration (held keys) — client-side for real players
2. Friction factor chosen from the PRE-move ground state
3. Move by motion (collisions land the entity; vertical collision zeroes motY)
4. motY = (motY − 0.08) × 0.98; horizontal × the factor from step 2

The traps (each cost a debugging round)

• Pre-move friction: a knock's LAUNCH tick decays at ground drag even though the move lifts the victim. Modeling friction post-move overshoots a 0.4 knock by a full block (3.06 vs the real ~2.06).
• onGround only refreshes while moving downward. Zeroing motY on a standing entity makes it read airborne; the next launch then takes air friction. Clear horizontal axes only; leave the −0.078 gravity equilibrium.
• Send-then-restore: melee knockback to player victims is sent as a packet and the server's own motion copy is restored immediately (1.8.9+ vanilla, still true today). The CLIENT integrates the trajectory. Server-side motion of a player victim after a melee hit is NOT the trajectory.
• Knockback math never changed; the WIRE did (measured on real vanilla: project/docs/research/2026-06-05-era-wire-measurements.md). 1.7.10 shipped knocks via the next tick's tracker, and the wire was JOIN-ORDER BIMODAL (addendum 2, measured both orders): the tracker runs BEFORE the per-connection phase, whose slots run in join order with each player's physics fused to their own slot — a victim who joined BEFORE their attacker received the FULL stamp (identical to 1.8.9), a victim who joined AFTER got it one decay tick late (ground hits lose ×0.546 horizontal — the era's "relog for less KB" folklore). 1.8.9 melee sent inside attack() then RESTORED the pre-hit fields — order-independent, hence "consistent" in era memory. 1.7.10 never restored → combos compounded (both orders). Rod/projectile rode the tracker on BOTH eras. KnockbackDelivery models this per profile: tracker ships the full stamp (the dominant mode); tracker-decayed opts into the later-joiner wire; never both — determinism is the product.
• The era vy baseline is a state machine, not the delivered knock: servers ticked player physics input-free, so standing victims park at the −0.0784 equilibrium (standing-hit vertical = 0.3608, NOT 0.4), and the movement handlers' jump bookkeeping OVERWRITES motY = 0.42 at any rising liftoff (+0.2 facing push if sprinting) — a knocked victim's baseline one tick later is jump-impulse FREE-FALL (1.8.9 combo hit 2 ships vy 0.2477–0.2862 = 0.42 eight-or-nine gravity steps later, stamp-phase dependent, wire-measured). The MotionLedger + GroundFsm (was VictimMotion + GroundTransitionWatcher; packet-fed via the parse rim, was GroundPacketTap, for real clients, session tick sampler for packetless ones) replicate exactly this; era combo verticals DECLINE, which is why combo victims stayed low.
• Within a tick, the attack was judged BEFORE the victim's movement packets (the attacker's connection slot ran first): a combo hit thrown the same tick its victim touches down — the spam-cadence norm — reads the PRE-landing flight. Mental: tick-stamped packet records + the MotionLedger's end-of-previous-tick read (was VictimMotion.currentExcludingTick) give the published PlayerView the end-of-previous-tick view. A same-tick landing smuggled into the attack read ships a grounded 0.3608 re-stamp the era never sent — apexes stop dipping, combos read floaty (the live symptom that found this, addendum 4).
• Reference flights (flat ground, measured): full stamp plain ≈ 1.99 blocks, sprint ≈ 4.95 (1.8.9 always; 1.7.10 victim-joined-first); decayed wire plain ≈ 0.99, sprint ≈ 2.54 (1.7.10 victim-joined-after only — an artifact-prone half; a "players move ~1 block" report means the decayed wire is shipping). Jump impulse alone rises ≈ 1.25.
• Ground drag is slipperiness × 0.91 from the block UNDER the feet (compendium §2): stone 0.546, ICE/packed ice 0.8918, slime 0.728 — a 1.7.10 packed-ice lane ships hit 1 at 0.4 × 0.8918 = 0.3567 (decayed wire) and residuals compound between hits (settle 5.37 vs stone 2.99 — ice nearly DOUBLES era knockback). The ledger reads it per segment (GroundFriction); the bot harness does NOT model floor slip, so its SETTLE under-reads on ice — trust the per-hit wire values there.
• A knock's residual takes TWO grounded pre-move decays (knock tick + liftoff tick) before the air segment: era chains pin 0.4 × slip² × 0.91^k exactly (ice hit 2 = 0.4821 = 0.4×0.8918²×0.91⁷). Ledger implementation: submit-time launch state (the live flag flips on the hit tick!), launch-tick drag in recordLiftoff, grounded seed on a player's first observed state.
• The era jump stamp includes Jump Boost: motY = 0.42 + 0.1×(amp+1) + the 0.2 sprint push (decompiled jump(), both eras; measured: a boosted victim's combo hit 2 ships vy 0.3286). The watcher caches the impulse per player.
• attack() also multiplies the ATTACKER's server fields ×0.6 on every bonus-knockback hit (beside the sprint clear, both eras): a player knocked mid-trade who counter-hits compounds the next received knock off the smaller residual. Ledger: scaleHorizontal on accepted bonus-KB hits.
• Mental's friction knobs are SURVIVING-FRACTION multipliers (vanilla ÷2 ≡ 0.5, the NachoSpigot convention). Forks publishing divisors (Panda/Sport/ Wind, typically 2.0) port as 1/d — pasting a divisor unchanged inverts the feel. Full research: project/docs/research/2026-06-04-improved-knockback.md.