Δ9Φ963 · studio doc

LYGO Arena

A real-time arena battler where nobody is driven. You choose a roster; two AI teams then fight on a stone ring with their own kits, cooldowns and cards, and the fight resolves on its own terms.

Play the arena Live hall Lattice Crypt Hub

1. What this is

Lattice Crypt is a game you play with keys. LYGO Arena is the same world with the hands taken off the wheel: a real-time team battler in which the interesting decision happens before the bell. You pick the team, the challengers are drawn blind from the full roster, and then you watch a fight nobody is steering, including you.

The design bet is that the kits are legible enough to be fun to read. A fighter is not a stat block with a name: every one of the thirty-three carries its own stat line, three named moves and three draft cards, derived from the numbers Lattice Crypt already gives it, shot, speed, magic, armour, melee, brave and faith for the champions, the FOE table for the crypt creatures, the PETS table for the pack.

2. Modes and the shape of a match

Four modes: 1v1, 2v2, 3v3 and 5v5. Team size is the only rule that changes; the arena, the kits and the AI are identical, so a 5v5 is not a different game so much as a louder one. Between the count and the result the only inputs are the watch controls: pause, and 1× / 2× / 4× / 8× speed.

26s1v1 median

34s2v2 median

41s3v3 median

50s5v5 median

Medians over 300 simulated battles in the balance harness (§6). Every match is expected to land inside two minutes; the ring starts closing at 80 seconds so no two healers can out-sustain the floor and leave a fight unresolved.

3. The arena

Each match builds a fresh map from the Lattice Crypt stone atlas: a symmetric tile field with a realm family chosen by seed, walls mirrored so neither side inherits better ground, and both spawn pockets flood-filled to guarantee that every tile a body can stand on is reachable from both ends. The harness asserts this over 120 seeds, a map that strands a fighter is a bug, not a surprise.

There are five colosseums, and the seed picks both the ground and the realm it is cut from. The layouts are kept whole in the code; in this build every ring is swept bare, no pillars, no walls, no loose stone, while the brains are tuned, so that what a body does is the brain's doing and not the walls'. The five are still built and still named, and cover comes back with one flag:

Colosseum, open sand with pillar knots on the diagonals. The centre is left empty on purpose: the middle of a colosseum is where the fight should happen.
Gauntlet, a spine down the middle with a mouth at each end and a shoulder on each flank. The one arena that rewards holding a gap.
The Pit, low cross-walls pierced by four single-tile doorways. Hold a door or lose the middle.
Colonnade, two broken rows of pillars. Lanes to run, columns to break a sight line behind.
Ruin, a court of three short walls and loose stone.

Every one of them is deliberately nearly empty, 86% open sand, with nothing standing in it at all right now, because the obstacle a body snags on is the obstacle that ruins the watch. What is left is placed to be worth standing next to: cover to break a line of fire, a gap to hold, a lane to run. The harness builds 120 seeds and asserts that both pockets can reach all ten spawn slots in every one of them.

Bodies move in continuous space, not on the grid. They walk with a measured feel for stone: a move is judged by how much of the body would end up inside a wall rather than by a yes/no test, so a body shoved into the border by a knockback can always push its way back out, the bug that once left a kit frozen against the wall for the rest of a match. A body whose cell is stone walks straight to the middle of the nearest open cell, through the collision test, because it is already in the wall. The choice between the direct line and the route around cover is held for half a second: an enemy stepping behind a pillar would otherwise flip that decision every few ticks and leave the body walking on the spot.

4. The brain

Every kit fights like itself, because every kit makes its own decisions. Two layers produce that: a squad brain per team on a slow clock, and a task brain per body on a fast one. The squad brain reads the ground, not the fog: it counts who is alive, whose line is winning the exchange, who is wounded, who shoots from far, who mends, and then it makes one call: who to kill, and why. The body brain turns that call into one word of work, which is what the chip on its health bar shows you while you watch.

The body brain works from a book of twelve tasks:

press / advance, close to the reach this kit's weapon actually wants, then stop there.
hold, stay on ground that is already good: at your reach, facing the enemy, drifting across its front.
overwatch, step into cover while an enemy still has a line to you, and shoot from it.
choke, for the kits whose job is to be in the way, stand in the doorway rather than in the open, but only while the team is not the one pushing.
screen, the bulwark and the brute put their body between the enemy and the ally who cannot take the hit.
flank, the assassins, hunters and skirmishers leave the squad's axis, take the long arc round, and arrive from a bearing nobody is facing.
bait, the kiters give ground in bursts: back off, fire, hold, back off again.
aid, anybody carrying a mending move goes to whoever is bleeding worst and puts them back together, standing on the side of the patient away from the enemy.
rally, when the team is losing the exchange, fall back on the standard together instead of dying apart.
withdraw, break contact when outnumbered and nearly gone, into the nearest cover on the way home. A last body standing never withdraws: there is no line left to fall back into, and two sides both refusing to fight is not a tactic.
repair, a body leaning on stone with somewhere to be walks to the middle of the arena for a moment. The pockets between the border and a pillar are the one place wall-sliding cannot reason its way out of.
no hits landing, a body that has not landed a hit in six seconds stops holding its range and walks in. Two teams that both like their range would otherwise stand at their bands behind cover until the floor decided the match for them.

Style is a doctrine, not a leash. The anchors and the chargers fight what the team called. The killers go for the body that will die soonest, the ladder measures it, and the names hold: assassins take the most-wounded enemy on 98% of their decisions, hunters on 96%, duelists on 80%. The kiters deal with what is on them, the artillery hits whatever is bunched, the medics look after the line. Nobody crosses the arena alone to do it: a target of your own choosing has to be within reach of your own line.

Cards fire on hooks: Second Wind when a body first drops below 30%, Name Recall to rise once more after being cut down, Unwrite Ink to shrug off the first stun, plus auras, leech, thorns, echo rounds and executioner edges that modify every number passing through them.

Fighting as a group

A body reads the fight it can see, not the scoreboard: enemies inside 210px are counted against its own company there.
A knife with nobody standing with it does not walk into two or more bodies. It circles at the edge of reach and waits for the line, or for one of them to step out alone.
Contact kits dart in, break back out to the edge of their own reach, and come again, the rhythm, not the longest walk in, is what keeps them alive.
A glass kit (the wolf, the imp, the thief) measures itself against the line and only takes fights it has the numbers for.
Knives pile onto whoever a mate is already fighting, and take the straggler when there is one, never the knot in the middle.

Answering a charge

Every kit that was losing carries something that stops a charge, a root, a stun, a shove, a chill, a hush, and it is spent on the first body that arrives, not saved. Charging three bodies at one gun only works if the gun has no answer for the first one.
A body already held is not held twice: the second arrival is met with a different answer, so a rush cannot be shut down by one tool on one cooldown.
The flavours belong to the kits: the imp's ember leaves a limp, the thief steals your stride along with your coin, the wraith's cry nails feet to the floor, the wolf hamstrings what it lands on and its roar throws the line back off its feet, the shadow blinks onto you before you are ready to be hit, and the drain's Unmake strikes the name out of the sentence, no moves, no stride, no mending.
Wound. Mending had no counter in the system, which is why the ladder was a staircase with the medics on top. Five kits now carry a wound that will not close: while it lasts, everything healed into that body lands short. A heal cut in half is worth more than a hit that gets healed back, so a wound is spent on whoever is doing the mending, never on the body that merely stands closest.

Getting there

The brain decides, and for a while the movement underneath it quietly disbelieved it. A body walking to a point that cannot be walked to, a retreat extrapolated from a body already against the wall, or a cover cell picked without a stone test, grinds the border for the rest of the match, which is what a spectator reads as a fighter standing in a corner with no idea what to do. Every goal is now clamped into the ring and snapped out of stone in one place, so no call site has to remember. A body told to hold, and pressed into the wall while it holds, used to swap its shoulder back and forth forever: being stuck now outranks the plan and it walks clear first. Targets are committed for 0.9 seconds instead of being re-chosen every frame, a body does not change its mind about who to fight six times a second, while anything its style explicitly asked for still lands at once, because that is doctrine and the knives live on it. Measured on identical seeds: bodies standing alone against a border fell from 201 samples to 6–47 across runs. The target-change rate, which looked like the same bug, was not: it is one change per body every six seconds, a fighter picking its fight, not dithering.

5. The roster: 33 real kits

The numbers below are what the engine builds for an undrafted kit: base health, base damage, attacks per second, reach in world units, flat armour, the three moves, the three cards, and the composite power figure the menu shows while you draft. Power is a draft aid, not a promise, §6 says where it lies.

Champions: 13

Kit	HP	Damage	Rate	Reach	Armour	Three moves	Three cards	Power
KaelGallant Blade	700	43	0.8	48	9	Rend Shield Wall Squire's Charge	MIGHT IRON VAMP	166
ValeAegis Knight	920	30	0.72	51	12	Cover Bulwark Shield Wall	BULWARK THORN IRON	197
OrinBlack Sigil	600	49	0.86	262	2	Unwrite Resonance Flask Chorus Flask	MIGHT FOCUS RES	131
NiaPath Archer	620	31	0.92	275	3	Pinning Volley Dash Needle Rain	STRIDE LUCK VOLLEY	128
LyraSpiral Bard	560	32	0.84	250	3	Hymn Spiral Chorus Weave Bloom	RES REGEN AURA	118
ArkosLattice Lancer	662	45	0.78	61	6	Geodesic True Line Dash	PIERCE MIGHT NIMBLE	142
D9raWolf Monk	850	44	0.88	45	6	Shockwave Wolf Fury Squire's Charge	FURY BRUTAL LEECH	164
SrathShadow Needle	560	33	0.9	243	2	Doublespeak Needle Rain Dash	VEIL LUCK ECHO	118
KairosHour Mage	548	38	1	269	2	Right-Time Borrowed Second Δ9 Seal	HASTE RES CLOCK	123
JusticaeAccord Knight	944	31	0.74	54	10	Fair Plate Verdict Bulwark	EXEC BULWARK THORN	187
SeidonTide Seer	600	35	0.8	256	4	Undertow Current Geodesic	AURA RES GIFT	128
SancoraWeave Chemist	798	32	0.78	256	4	Chorus Flask Weave Bloom Cover	REGEN GIANT AURA	139
LightfatherArchitect · Δ9	680	37	0.84	256	6	Δ9 Seal Provenance Cover	AURA GIFT GIFT	152

Crypt creatures: 15

Kit	HP	Damage	Rate	Reach	Armour	Three moves	Three cards	Power
WraithCrypt Fodder	420	32	1.05	256	2	Wail Drain Cloud	HASTE NIMBLE LEECH	105
BruteCrypt Fodder	1090	39	0.66	48	9	Ground Pound Ward Field	IRON GIANT THORN	190
ImpCrypt Fodder	403	31	0.95	237	1	Imp Fire Flicker	STRIDE LUCK NIMBLE	95
HurlerCrypt Artillery	859	47	0.72	51	4	Lob Stone Ground Pound	MIGHT BULWARK GIANT	137
ShadeVeilkin	471	36	0.92	51	3	Flicker Snatch	VEIL LEECH NIMBLE	101
ThiefTithe-Thief	507	35	1.05	51	2	Snatch Flicker	STRIDE VAMP LUCK	106
DrainThe Unwriting	700	46	0.78	64	4	Unmake Drain Cloud	LEECH REGEN MIGHT	128
GateNexus Guardian	1735	46	0.6	61	13	Gate Slam Ward Field Call the Wards	IRON THORN GIANT	269
CrownNexus Guardian	1446	52	0.62	58	7	Regal Volley Lob Stone	MIGHT BULWARK WELL	209
SmithNexus Guardian	1020	60	0.64	250	10	Anvil Gate Cannon	MIGHT BRUTAL IRON	223
Heart of the CryptNexus Guardian	1900	44	0.68	58	8	Heartbeat Ward Field	GIANT REGEN VAMP	245
LeviathanNexus Guardian	1080	58	0.72	243	6	Tide Crush Undertow	MIGHT STRIDE LEECH	201
Tithe-KingNexus Guardian	1000	45	0.98	64	5	Tithe Snatch	VAMP EXEC STRIDE	179
Name-EaterNexus Guardian	1008	53	0.76	262	6	Nameless Unmake	MIGHT CLOCK LEECH	196
The LockSuper Boss	2100	53	0.66	269	14	The Lock Seals Gate Cannon	IRON BULWARK GIANT	323

Beasts: 5

Kit	HP	Damage	Rate	Reach	Armour	Three moves	Three cards	Power
AshmaneWolf	496	37	0.96	48	4	Pack Bite Roar	STRIDE FURY LEECH	114
SolstrideLion	543	49	0.9	51	4	Solstride Roar	LUCK BRUTAL MIGHT	138
IronhideBear	1024	38	0.74	54	10	Maul Ward Field	IRON GIANT THORN	196
TuskwardElephant	1033	35	0.68	61	11	Stomp Ground Pound	IRON BULWARK WELL	197
GlassbarbScorpion	448	38	1.02	51	4	Glassbarb Clamp Flicker	THORN FURY VEIL	115

6. The balance ladder

Feel is not a spreadsheet, but some things are. tools/arena_check.js loads the real roster and the real engine in Node behind a stub DOM and runs the whole thing headless. It asserts what a player would notice, and prints what a designer wants to argue about.

What the harness checks: 90 assertions, 90 battles per run

Roster shape: 33 fighters, unique ids, every sprite key present in its atlas, every move id resolvable, three real cards per kit, a portrait for every champion, no two kits identical.
Built kits are playable: finite health, damage, rate and reach; a power spread that is a band rather than a cliff.
Maps: 120 seeds, both pockets can reach all ten spawn slots; the ring floor is clear (0 interior stone tiles across 120 seeds).
Battles: every match ends with a winner, no NaN in any live body, nobody outside the arena, both sides actually cast moves, and no kit ships without ever dealing damage.
Pacing: per-mode mean and median length against the two-minute target.
Usage: which moves the sim reaches for and which never fire, the fastest way to find a kit whose third move is dead weight.

The ladder produced four systemic fixes rather than thirty-three hand tweaks, which is the point of building it:

A melee walk-in tax. Melee kits were being shot for free on approach. Every kit with reach under 100 now gets +18% health and +1 armour before any card touches it.
Softer armour. Flat reduction of 1.5× armour per hit made high-armour bodies near-immune to low-damage kits; it is 1.1× now, floored at 3 damage.
A frustration timer. Five seconds without landing a hit forces the charge in §4, which turned the worst stalemate class into decisive fights.
A closing clock. From three minutes every blow lands for a quarter more, climbing a step a minute to double, so a match that drags is decided by who is landing hits. There is no shrinking floor: an earlier version of this page claimed a ring that closed from eighty seconds and the sim has never done that.

Where it stands today, honestly: the ladder's last 420-match run prints a field average of 47.5%, the table is a spread, not a score. The tails are real. Eight kits hold 25–37%: the light creatures and the short-reach melee die two fights in three, because they are the body the pack finishes first. The strongest seven sit at 64–75%, and the same fact explains them from the other end , the kits at the top die only 14–27% of the time, so it is no longer only the bosses: a body that fights from eight tiles and never has to walk in owns the top of the table. Two members of the floor were not weak at all, they were broken: the imp and the stone-hurler fought at bolt range carrying a melee punch, so they stood outside their own reach for the whole match. That one was a defect, not a balance problem, and it is fixed. The floor was then lifted a fourth time the only way this table allows, through survivability, never damage: the five kits whose 95% interval sits wholly under 50% while they die 63–88% of the time took 1–2 armour and 8–10% health, armour-led, because a first pass at 14–21% health pushed the harness' own mean 3v3 past its two-minute mark.

7. The hall

A callsign is optional and nothing else is collected. When a match ends and a call is set, the result posts to the shared arcade book (mode, side that won, MVP, session wins) and the browser keeps a queue so a failed post is retried later instead of being lost. The live hall reads the same book and shows only this game's rows. Empty beats a fake board.

8. Controls

Space pause · 1 2 3 4 speed 1× 2× 4× 8× · M mute · R rematch · Esc back to the menu. The donation-gate reminder is held shut for the length of a match and released on the results board or the menu.

9. Lineage and credits

Art, atlases, roster lineage and the numeric DNA all come from Lattice Crypt; the arena engine, the kits and the harness are new. Built by Δ9Φ963 with the steward, Justin Helmer. Lightfather. Everything runs in the browser with no build step: three scripts, a stylesheet and one atlas pack.

Files: roster.js (33 kits, 26 cards, the move table) · arena.js (the engine) · brain.js (the tasking brain) · banter.js (every spoken line) · arena-ui.js (menu, HUD, feed, caster, results) · tools/arena_check.js (90 assertions) · tools/arena_ladder.js (the field table).

The mouths, and the man with the microphone

Two AI teams fighting in silence is a diagram. So every body in the ring now talks, and a caster calls the fight from the side of the sand. Nothing either of them says is on a timer: a line only exists because an event produced it. banter.js holds the library, 248 fighter lines across ten kinds (open, taunt, kill, die, hurt, crowd control, save, win, lose, ace), personal lines for 22 kits, and 56 caster lines that are filled in from the event that fired them: who killed whom, who was mended, which map just closed.

When a body dies, the caster names it. When somebody is dragged back from the floor by a mender, he says so. When a body is rooted, stunned or slowed by a rival's crowd control, he says that too. The engine emits the event and the caster says the line, which means the commentary cannot drift away from the fight, checked by the harness, which fails if a single caster line appears without an event behind it.

Everything spoken is also written: the feed under the characters carries the banter (0:35 Lyra: Rooted. Eat it.), a bubble rides the head of whoever said it, and the caster's own panel sits under the challengers with his portrait and his last call. The voices come from the browser's own speech engine, filtered to the human-sounding voices only , the joke voices some systems ship make a fight sound like a machine test, with a fixed pitch and rate per kit, so the wolf always sounds like the wolf. Voices can be switched off from the top bar.

One microphone. More lines made the fight noisier, not clearer. director.js holds the floor: the engine and the banter library offer lines, and the director decides whether a line is said at all, when, and how long the speaker keeps the microphone. There is one mouth at a time and the floor is held for the true length of the line, measured from the voice, not estimated, so nothing can talk over anything. Between two lines there is a beat, 1.15 seconds, or half a second when the bell is waiting. A body may not speak twice inside six seconds and the booth not inside 3.4. There is a budget: nine lines in any twenty-four seconds and four in any ten. Over budget a line is dropped, not queued, and anything stale by 2.6 seconds is thrown away rather than said late, a report of a kill that happened four seconds ago is worse than silence. The booth calls the fight and the bodies answer it; before that rule the booth out-talked the ring ten to one, and now the two mouths split a match 36 to 42.

The intro is a programme, not a warm-up. The count before a match is fifteen seconds with nothing else in it. At half a second the caster names the ring; your team is read out by name after that, then the challengers, then the bell, "Fight! Anybody's guess. Who's gonna win?", and the room goes quiet for three to five seconds behind it, timed from the end of the line rather than from the moment it was called for, so the silence is a silence. The announcer reads at 1.22× because a five-name team takes five seconds to say and the countdown does not stretch for anybody: if he runs long the bell preempts him. The bodies take their first lines after the silence, not under it. Everything about the timeline is asserted headlessly before it is ever watched: that every body is introduced by name before the bell, that the bell is the last thing said and lands on the clock, that the room is quiet behind it, that no two lines ever overlap, that a beat separates them, that neither mouth starves the other, and that most offered lines go unsaid. Watched live at 1×: the count ran 14.98 to 0.19, the three intro lines landed with gaps of 0.8, 2.5 and 1.8 seconds, and the bell took the panel to ON AIR.

The floor has an address. A spectator can see who is talking without guessing: the speaker's row is lit, the caster panel carries an ON AIR tag, the line is marked in the feed, and the bubble over a head lasts exactly as long as the voice that owns it, scaled with the match speed, so a line at 8× does not hang there after the voice has gone.

How the balance is measured

Every roster change is measured, not guessed. tools/arena_check.js runs 90 headless battles and asserts what a spectator would notice, 66 checks now, from "no fighter ships with a dead kit" to "only one mouth talks at a time". tools/arena_ladder.js 420 runs the whole field through 420 matches and prints a win rate, a death rate and a damage-per-match line for all 33 kits, each with its own 95% interval. The floor of the roster, the kits that were losing more than two fights in three , has been raised four times off the back of that table.

Ninety battles is not a verdict. That harness is marginal at its own sample: on the build this page describes it passed 66 of 66 on one run, then tripped a different assertion on each of the next two, once a battle that ended with nobody hurt, once a mean 3v3 of 113s against its own 110s mark, which read 62.7s on the following run after a 4% health change. A failed run is a reason to re-run, not a reason to revert. The 420-match table, and a cause that can be named, are what settle a question.

The table knows its own noise. Teams are drawn at random, so the same build measured twice does not produce the same table: two runs of 420 matches disagreed by up to sixteen points on a single kit (the bulwark gate read 60% in one run and 76% in the next; the smith 43% and then 64%). The ladder therefore prints a 95% interval beside every win rate and marks a ~ on each kit whose interval covers 50%, that kit is average, not weak, and tuning it is tuning a coin toss. Only the tails are stable between runs: the same handful sits at the top and the same handful at the bottom every time.

A win rate is an outcome, not a behaviour. tools/arena_doctrine.js samples a battle four times a second and reports what the bodies were actually doing rather than who won: how long each kit spent fighting with no ally inside 200px, how long it spent closing on a longest-reach enemy whose shot was loaded, how long the engine's own stuck counter was tripped, how often it shared a target with an ally, which deaths were isolated, and , read straight off every plan's own reason line , a census of what the whole field was thinking. On the shipped build it reports the picture this page had been missing: across 144 controlled duels of the six shortest-reach kits against the six longest, the contact kits win a third of them and spend a fifth of their health before they land a hit. A third of the deaths in the field are bodies with no ally within 165px. Of thirteen withdrawals, not one body was still alive six seconds later. There is no wall-grinding to fix: the stuck counter is tripped for zero seconds in every battle, because the clamp-and-snap work holds. Three doctrine variants were written against those numbers , lead the walk toward where the target will be, wait out the reload window instead of walking into a loaded gun, and take the middle of the arena away from a kiter so its only retreat is a wall. All three were reverted. The census proves they fired, and the duel win rate did not move, which is a result and not a failure: a 48px body in the open against a 269px rifle is losing to reach, and reach is not a thing a better plan fixes.

A blow that deals nothing has to say so. Play reported the basic attack passing straight through a body with no number on it, and the report was half right and half a screen that was lying: a hit that a ward swallowed whole returned zero and printed nothing at all, so a defended body looked like a phantom. In twenty battles, forty blows were eaten by a ward in complete silence. The same was true of a weak blow fully stopped by armour, and of an aimed shot that slid past a strafing body and expired off the edge of the arena without a word. All three now print what happened , ward, block, miss , and every match counts them: melee swings, aimed shots fired, blows armour stopped, blows a ward ate, and shots that arrived at nothing. Measured over the same twenty battles that is the honest split of the complaint: the damage really was missing in those cases, and the screen was missing the reason. Roughly one aimed shot in six (17%) arrives at nobody because the target moved, which is a miss and is now labelled as one.

The committed step was measured two pixels short. Every kit in the roster owns Close , one committed step of 224px , and the engine would only spend it when the target stood within 240px, a hardcoded sixteen pixels of margin. A rifle holds its band at ninety per cent of its reach, and the longest guns hold at 247px: just outside that window. The one move written for crossing open ground could not be spent on the one enemy it was written for, and the brute , no blink, so Close is its only way in , ran three separate fights of two hundred seconds without ever landing a hit. The check now measures to the reach of the body doing the stepping, which is the quantity that always mattered, and it changes behaviour measurably: over 24 seeded battles, a third more melee swings and seventeen per cent less time spent fighting alone. It does not move the duel win rate against the guns, and the reason is worth writing down rather than hiding: Close is on a thirty-second clock, and a rifle re-anchors its band with every shot. One step per half minute is not a way in. That is a kit-design question, not an engine one.

A match where nobody fought is a match that did not happen. The harness counts battles that end with no damage dealt by either side and nobody down, and it was flagging about one battle in ninety: two lines holding their bands, each waiting on the other, while withdrawal and leashing and holding were each a reasonable order and the sum of them was a fight that never started. A plan is now void after twelve seconds without a blow landing anywhere on the floor, which outranks every other rule below it. The count is zero across the ninety battles since, and it was not zero before.

So every change here is made against a cause that can be named, never against a number: a kit whose reach exceeds its attack (the imp, the hurler), a signature ability that decides maps on its own (the vanguard's revive, now forty seconds and a 30% return), a body so light it dies before it arrives. Two fixes that measured as net-negative were reverted rather than kept for the look of activity: a rule that made light kits break off earlier lost more team fights than it saved (the kite fell from 53% to 33%), and handing three hunter kits a heal-cut made them worse, a wound sends its owner after the enemy mender, and an average body that dives at a protected back line does not come back.