era-accuracy

name	era-accuracy
description	Use when a change claims 1.7.10/1.8.9 authenticity or could affect combat feel — the era ground truths, what is client-side and therefore untouchable, the known myths, and the anti-features the community history warns against.

Era accuracy: ground truths, contracts, and anti-features

Ground truths (decompiled, in-repo: project/docs/SYNOPSIS.md + project/docs/legacy-combat.md)

• 1.7.10 and 1.8.9 knockback MATH is byte-identical; the era differences are DELIVERY, measured on the real servers (project/docs/research/2026-06-05-era-wire-measurements.md, esp. ADDENDUM 2): 1.7.10's tracker wire was JOIN-ORDER BIMODAL — a victim who joined before their attacker received the FULL stamp (identical to 1.8.9: standing (0.4, 0.3608) ≈ 1.99 blocks, sprint (0.9, 0.4607) ≈ 4.95), a victim who joined after got one decay tick less (standing (0.2184, 0.2751) ≈ 0.99 blocks, sprint (0.4914, 0.3731) ≈ 2.54 — and "relog for less KB" folklore). 1.7.10 never restores → combos compound in BOTH orders; 1.8.9 melee sends in attack() then restores (flat, order- independent — why 1.8 KB is remembered as consistent). Mental's tracker ships the full stamp; tracker-decayed is the opt-in later-joiner wire. A "players only move ~1 block" report means the decayed wire is shipping — that is a misconfiguration or a bug, NOT era truth (this verdict was reversed once already; see addendum 2). Era combo verticals DECLINE (the jump stamp free-falls: 1.8.9 combo hit two ships vy ~0.25–0.28 depending on stamp phase); victims stayed LOW in era combos (apex dips to ~0.5 every other hit). Standing-hit vertical is 0.3608 (equilibrium baseline), never 0.4.
• The era's within-tick attack ordering (addendum 4): an attack was processed in the attacker's connection slot BEFORE the victim's same-tick movement packets — a combo hit thrown the instant its victim touches down (the spam norm) reads the PRE-landing flight and ships the declining vertical, never a grounded 0.3608 re-stamp. Mental replicates via tick-stamped packet records + the MotionLedger's end-of-previous-tick read (was VictimMotion.currentExcludingTick; the published PlayerView IS that freeze), the +1 staleness allowance in the view's immune check (was Snapshot.isDamageImmune), and the auto feedback window at (max/2 − 1) ticks. "Combos feel floaty / too much vertical" = boundary hits shipping grounded re-stamps = one of those three broke.
• Legacy damage: sharpness 1.25/level, crit-before-enchant with no sprint exclusion, pre-1.9 tool tables (diamond sword = 8), rod/projectile knocks from the SHOOTER'S POSITION, probabilistic armor resistance.
• The combat compendium (project/docs/research/2026-06-06-combat-compendium.md) is the authoritative mechanic-by-mechanic reference: knockBack ignores its strength param (0.4 hardcoded); sprint and KB-enchant are LEVELS of one extra (0.5 h/level + flat 0.1 v — KB II ships 1.4, sprint+KB II 1.9); crits add ZERO knockback; blocking halves damage AFTER knockback (era blocked hits knock FULL — Mental cancels only FULL blocks); a stronger hit mid-invuln deals difference damage with NO knock and no flinch; PURE-vanilla snowballs/eggs/rod bobbers never knock PLAYERS (the EntityPlayer zero-damage gate, both eras) — era rodding is CraftBukkit- lineage behavior, which Mental's rod/projectile modules target; the era jump stamp includes Jump Boost; ice nearly doubles 1.7 knockback (slipperiness × 0.91 ground drag); attack() ×0.6's the attacker's own server fields. Each claim is decompile-cited and wire-measured there.

The client-side technique contract (NEVER touch server-side)

These work because movement is client-authoritative; Mental preserves them by construction and must keep doing so:

• The 0.6 self-velocity multiplier per sprint attack (why high CPS reduces received KB — 0.6^clicks — and why straightline locks/sumo reducing exist)
• W-tap / s-tap sprint re-arming; jump-resets; A/D strafe redirection
• A real victim's walk/jump never entered the legacy server's motion fields — held input is integrated by the client ON TOP of the knock packet

The one exception that needs server-side reconstruction — block-hitting's sprint reset. In 1.7/1.8 starting a block dropped the sprint flag (the client sent STOP_SPRINTING), and the re-engage on release re-armed the sprint KB bonus — that's why block-hitting maintained the bonus like an easy w-tap. Modern clients (1.17+) KEEP the sprint flag through an item-use block (you're slowed but still flagged sprinting — visible as sprint particles while blocking), so that STOP/START never crosses the wire and block-hitting silently stops resetting sprint. Mental reconstructs it ALWAYS-ON since 2.6.0: the BlockResetTap (knockback core — moved out of default-OFF SWORD_BLOCKING, whose absence on defaults was half the 2.6.0 "resets never re-arm" report) re-arms the connection InputLedger on the block right-click, gated on the RAW client sprint flag (the only signal that survives the ledger's own post-hit clear) so a stationary defensive block never gains a phantom bonus. The item gate: shields always; SWORD_BLOCKING contributes its decorated-sword test while enabled. The re-arm is EXACTLY a START-shaped gesture — one hit per reset (2.6.0, the owner's directive; the 2.5.1 sticky held-chain is retired): the accepted bonus hit spends it, the next block-hit takes a fresh right-click, and an UNSPENT re-arm survives the block release (era-exact — the era client's release re-engage is what re-armed the bonus, so a block-tap earns the same single fresh hit a w-tap does). The right-click that re-arms it is the born-cancelled RIGHT_CLICK_AIR PlayerInteractEvent (the listener is ignoreCancelled=false, self-filtering on useItemInHand() != DENY) or the victim-aimed PlayerInteractEntityEvent — both reach the re-arm (2026-07-10). NOTE the local-player sprint-particle visual during a block is client-authoritative rendering and cannot be removed server-side (same family as the right-click "item rises" cosmetic) — only the FUNCTIONAL reset is restored.

The modern-client sprint latch — why Mental no longer runs setSprinting(false). Vanilla's in-attack server-flag clear (the server half of w-tap) is deliberately NOT reconstructed (removed 2026-07-10). On 1.21.2+ a server-side setSprinting(false) is ECHOED to the attacker's own client, which adopts it, drops its local sprint, and confirms with one STOP_SPRINTING — and no START returns (item-use block-hitting blocks a fresh sprint start; the client sends START only on a rising edge), so the raw client flag latches false and every later melee reads plain: a one-way latch. Real vanilla never produced this at spam cadence (its clear sits behind the ≥90%-charge gate, where the client predicts the same clear and re-engages, so the echo is always redundant). The observable era contract is one engagement, one sprint knock (2.5.1, the owner's directive — supersedes the 2.5.0 post-clear re-arm): the era server consumed the sprint flag INSIDE every bonus attack and re-armed only on a client START, so a no-w-tap double flew 7.2 blocks where a w-tap double flew 11.4 (era-wire measurements, the reproduced target). The InputLedger (2.6.0 — subsumed SprintWire, every write ringed for the journal's trail= token) mirrors that — the hit's seq-guarded consume (onServerClear(long)) spends the engagement (sprinting + armed drop, clearedAt opens the SPEND LATCH), and while that latch is open reconcile is BLOCKED from resurrecting the bonus off the stale-high server flag. A held-W modern client's flag stays true forever (its STOP lives inside vanilla's cancelled ATTACK and never crosses the wire), so re-adopting it would hand every held hit a per-hit sprint knock — the 2.5.0 bug the owner ruled out (holding W comboed forever with zero reset skill). Re-arming takes a CLIENT-EXPRESSED re-gesture: a wire STOP→START cycle (w-tap / s-tap / GUI open), the always-on BlockResetTap re-arm for block-hitting (a modern item-use never drops the client flag), or — the 2.6.0 failsafe — the published flag's falling edge (SERVER_FALL: a STOP provably reached vanilla even if the tap missed it; closes the latch, never arms). Same-tick era order (bytecode-pinned 1.8.9 + 1.21.11, observed live): a same-tick w-tap+click crosses ATTACK before START on every client generation, so that click ships PLAIN (note=start-trailed) and the START arms the NEXT hit — pure arrival order is era truth, never reorder. The 2.5.0 auto re-arm assumed a held-W era client kept the bonus; it did not — the era client auto-re-engaged by sending a genuine START, which a modern client at spam cadence never does (its local flag never drops, so it sends exactly ONE START per engagement). PLAYER_INPUT on 1.21.2+ DOES carry a sprint bit (0x40, raw keySprint.isDown() intent — false for double-tap sprinters, true for stationary ctrl-holders; the server ignores it) — the block door's entry-gate corroborator only, never a verdict source (empirical 1.21.11 extraction, project/docs/superpowers/research/2026-07-10-modern-client-sprint-wire.md); the raw byte still rides the ledger ring as KEY_INPUT evidence. Never re-add the deferred setSprinting(false).

Myths (do not implement, do not "fix")

• Victim sprint-key spamming reducing knockback: placebo (the flag changes no server math — pinned live by the era suite).
• "1.7 animations reduce KB": placebo; the real incident was a patched 1.12 Lunar client velocity bug (Nov 2020).

Anti-features (community history says no)

• Randomized knockback (MMC's most-hated era; determinism IS the product).
• CPS-scaled anything / "anti-reduce" normalization (GommeHD: ~80% disapproval, top-player exodus). Fairness features must be opt-in choices.
• Hit-delay knobs (a rejected anti-feature — see combo-hold design).

Verifying authenticity

Vector-level: KnockbackSuite/ProfileSuite (engine expectation == applied velocity). Position-level: EraParitySuite (settled endpoint vs the EraOracle legacy integrator, ≤0.005-block tolerance; combo era gap and held-input effects demonstrated live). Wire-level: the legacy-lab harness against the live netty path (the suites never traverse the parse rim, was HitPacketListener — FakePlayers attack server-side). A claim of era accuracy without a pin in one of these is not yet a claim.

Era trade feel (measured 2026-06-05; corrected same day, addendum 2)

• The consistent era wire is the FULL stamp: standing (0.4, 0.3608), sprint (0.9, 0.4607) — 1.8.9 always, and 1.7.10 whenever the victim joined first. The HALF wire ((0.4914, 0.3731) sprint) was 1.7.10's later-joiner mode only; shipping it universally was a measurement artifact (the lab's fixed connect order) that survived one full release and one "working as intended" verdict. As of 1.5.0 every preset ships the full stamp; tracker-decayed opts back into the half wire.
• The trade-opener opposition: base 0.4 pushes away from the ATTACKER'S POSITION, sprint extra 0.5 along the ATTACKER'S YAW. A victim past the attacker (face-hug) gets ~0.05 h — era-real, identical on actual 1.7.10.
• The community's "comboable 1.7" memory: Kohi/MCSG forks AND vanilla's own full-stamp mode. legacy-1.7 (full stamp + ledger combos) now matches it; legacy-1.8 is the flat 1.8.9 wire; kohi is the fork feel.
• Modern vanilla reference (Paper 1.21.11 bare): standing (0.4, 0.3608), sprint (0.7, 0.4) — what players are calibrated to with Mental off.