MJ playbook — how an automatic motion jockey moves sound on the A³ dome

v2, 2026-10-08 (final, 06:10). Written from motion perception (v1, 02:58), then checked against the MJ Lab simulation (report.html, log.md). Where the sim confirmed, refuted or refined a rule, it says so.

Tags: [R: …] = from the research note (the row it rests on); [practice] = club/stage practice; [MJ hypothesis] = my own rule; [sim ✓] = the MJ Lab model supports it across its robustness checks; [sim ✗] = the model contradicts it; [sim ~] = the model cannot settle it. The sim is a model, not a listening test.

The one fact behind every rule: the crowd does not study the motion, it notices it. A move that has to be listened for is a move that did not happen.

1. Speed

  1. Cruise at one lap per 4 bars. Without stops the best cruise is ~50-75 °/s (2 bars below ~110 BPM); once the ships breathe (rule 18), 4 bars at every tempo, because a fast rabbit turns every restart into a sprint. [sim ✓ r43] [R: B1, MAMA ~5° front] [sim ✓ against 8 bars and 1 bar in 18/18 listener variants]

  2. One lap per bar is the ceiling for a path, and only with a wide band above 3 kHz. [R: Roggerone 2019] [sim ✓: at 1 bar the ships cannot keep up (speedMax) and the floor gets 4 events a bar]

  3. Faster than one lap per 2 beats is an effect, not a path — swirl, Leslie, tremolo. Use it for 1-2 bars, never as the cruise. [R: Féron 2010] [sim: a swirl costs a little, like every effect]

  4. Slower than one lap per 8 bars is a drift, not motion. [R: Saberi 1990 + B1] [sim ✓: half the floor misses it]

  5. A speed change must be ×2 to be heard as faster, ×½ as slower. Double or halve; never nudge. [R: Carlile & Best; Locke 2016] [sim ✓ for doubling (6/6 seeds); halving hurt in the model]

2. The band (FREQ / Q of the 3D filter)

  1. Move the top of the mix: 6-8 kHz at 2 oct, or 3 kHz at 4 oct. [R: B4, Yost & Zhong, Langendijk & Bronkhorst] [sim ✓: 8 kHz/2 oct > 3 kHz/2 oct in 18/18 listener variants; caveat: the sim does not know the track’s spectrum]

  2. Never narrower than 1 octave for a move that should be followed. [R: Yost & Zhong 2014, Roggerone 2019] [sim ✓ 18/18: 1 oct loses ~45 % of the engagement, ½ oct ~90 %]

  3. Never move anything below ~250 Hz as a path; avoid centres at or below 1 kHz. [practice: 4DSOUND, Rex Club] [sim ✓: 250 Hz and 1 kHz bands scored ~0.6 of 3 kHz]

  4. Elevation moves need 6-12 kHz in the band. When gravity takes a ship over the crowd, the band needs that energy or the dive is silent. [R: Langendijk & Bronkhorst 2002] [sim ~: built into the model, but r35 showed it is not what makes the planets cost; the 8 kHz band wins for front/back and width, not for elevation alone]

  5. Pick a high or wide band and keep it; switching FREQ up only while a ship is over a group buys little. [sim ✓, MJ hypothesis refuted in its strong form]

  6. A narrow high band sits at its own height — do not try to move it vertically. [R: Middlebrooks 1992]

  7. The narrower or lower the band, the slower the ship. [R: B1/B4 coupling]

3. How much moves at once

  1. Two ships move; the rest wait. [practice: 4DSOUND “less becomes far more”] [sim ✓ again with the breath: 3 ships fatigue 0.41] [sim ✓: 2 > 1 in 18/18 variants, 2 > 4 in 16/18; with holds and doublings on, a third ship already tires the floor]

  2. Never turn several ships the same way at the same speed (vection). [R: Väljamäe 2009] [sim ~: no vection at 1 lap / 4 bars; counter-rotation itself cost a little because ships meet head-on]

  3. No more than one audible event per ship per 2 beats. [MJ hypothesis — it is built into the model’s fatigue, so the sim cannot test it; it is the main thing a listening test must check]

4. When to bend, when to hold still

  1. A bend that should be heard must leave the path by ≥ 30° as seen from the floor. [R: Grantham 1986; B2]

  2. The beat link is a move on the one, not gravity breathing. If gravity is on, gate it to beat 1 of the bar: the bends then land on the one and cost a sixth as much. [sim ✓ r37] The plan’s pulse moves a ship by < 1° per beat; 20-30° is needed. [R: B5] [sim ✓: no effect at any depth or rhythm tried]

  3. Breathe: every moving ship stands still for beat 4 of every bar and restarts on the one. A short stop on the grid makes the restart an event for the whole floor. One whole beat (it must outlast the ~336 ms the ear integrates over), once a bar (twice is a stutter), both ships together. [MJ hypothesis; practice: L-ISA “random throws that stop”; R: Chandler & Grantham for the length] [sim ✓, the most robust move of the night: positive in 18/18 listener variants and all 72 of their seeds, 100-174 BPM; every 2nd bar is the gentler variant] 18a. Two ships: call and response. Ship 1 breathes on beat 4, ship 2 on beat 2. [sim ✓, small: +0.006, 6/6 seeds, less fatigue] 18b. For a floor with stamina: hold the last bar of 8 and double for 2 bars in 8. Less gain than the every-bar breath (+0.034 vs +0.056), three times the fatigue. [sim ✓ on 16/18 variants; lost when the floor tires twice as fast]

  4. Do not circle for more than a few laps unchanged. A steady orbit is heard and then tuned out. [MJ hypothesis — the maintainer’s fear] [sim ✓: bare orbit boredom ~0.7, the highest of any condition]

  5. Ease off when the floor tires. [MJ hypothesis] [sim ~: a fixed fatigue threshold cannot serve floors that tire at different rates; an MJ that learns (moves on/off per 8 bars, keep what scores) is a hedge, never the best. The breath made this rule mostly unnecessary: it is cheap enough to leave on.]

  6. No kicks on the one. A velocity impulse towards a group adds events and fatigue faster than interest. [sim ✗ for the MJ hypothesis “kick on the downbeat”; every size tried did worse]

5. Bodies (guest groups) and escort

  1. Gravity must bend, never park. The pull of the groups is never stronger than the ship’s steering back: gravityMax ≤ steerMax, best right at it. [sim ✓, the most robust result: the plan’s cap 1.5 loses in every seed of every listener variant]

  2. Bending halves boredom but does not, on its own, win the floor — and with the breath on, it costs. Use the groups to say where the sound goes, keep them light (half the plan’s masses) and spread; let the breath do the anti-boredom work. [sim ✓ r33-r38, 18/18 listener variants; not caused by the ears (r35-r36) but by the paths: the wells brake the ships and pull them in. Gated to beat 1 at half mass, the planets cost ~2 % (r42)]

  3. Escort wide or not at all: the escort orbit must subtend ≥ 30° from the people escorted. At the plan’s radius (0.15·√m) the ship parks overhead. [R: B2] [sim ✓: 0.3·√m breaks even, 0.15 loses]

  4. Escort is a DJ gesture, not a routine. [sim ✓: every automatic escort rotation did worse than patrol] 25b. Few planets. With the recommended gate, each extra body adds bends and costs a little: one costs −0.004, eight −0.015 (r45). [sim ✓]

  5. Do not feed the biggest group forever. Groups pull ships, ships serve groups, the least-served quarter loses (13/18 variants). Lighter masses (×0.5) give the edges back at the price of some boredom. [sim ✓]

  6. Dead zones are for steering around, not for punishing. [MJ hypothesis; the sim never produced one]

  7. Fly the big path low (elevation reach 0.6-0.7): azimuth is heard better than height. [R: Saberi 1990] [sim ✓, weak]

6. Stereo DJ effects, translated

  1. Auto-pan L↔R → a step between two opposite points, ½-2 beats per side. A step reads better than a glide. [R: B6, Chandler & Grantham] [sim ✓ step > glide; but every effect cost a little: accents, not routines]

  2. Auto-pan circle → rotation, 1 lap per 1-4 bars; 1 lap per beat only as a swirl. [R: B1]

  3. Filter sweep → FREQ sweep of the moving band while it orbits (the band climbs). 4-16 bars. [R: Lee 2015; G]

  4. Riser → elevation climb + doubled rotation every 4 bars + high FREQ, 8-16 bars; drop → snap to front or static on the 1. [R + G; the hold-and-restart result (18) supports the “snap on the 1” half]

  5. Echo throw → throw the return to the opposite side and let it drift back over 1-2 bars. Needs its own path. [G; not in today’s routing]