UNDERTOW

PERFORMANCE COMPILER 1 / 9 OCT 2026 / RESEARCH

Between
records.

A rhythm might survive its source. An echo might outlive a voice. A second of silence might say more than a seamless blend. Twenty-four attempts to find out.

The first Performance Compiler turned authored musical relationships into bounded source-verified operations. The 24 recordings were rendered over six pairs with 114 causal-control records, including broken controls and disputed reviews. Nothing in a successful render proves it sounds good.

24 original takes6 source pairs114 control recordsArtistic acceptance remains open

WHAT THE FIRST LISTEN REVEALED

Something worked. Something sounded awful.

11-inherited-contour-foreground carried Ghost's rhythm into Dauerversprechen's material; an attributed source-informed reviewer recognised the inheritance but found degradation and potential fatigue. 05-rupture-r1 gave Ghost → Mouthful one second of actual silence; a paired reviewer preferred it to two seconds. The model's own gap estimates were unreliable. 07-ghost-dauer-overlap remained technically functional without a confirmed audible echo contribution. Those qualifications are part of the instrument, not footnotes to erase.

The next experiment: Riding the pitch ↗

SOURCE PAIR 1 / 7 TAKES

A Mouthful of Roll → Leaning Cell

01-carry-r1 · 28s · Agent-rejected

A Mouthful of Roll loses its dry body; its last gesture becomes a bright finite afterimage. Leaning Cell takes bass auth…

A Mouthful of Roll loses its dry body; its last gesture becomes a bright finite afterimage. Leaning Cell takes bass authority beneath the descendant, then opens only after the residue has expired.

The early high-band carry was heard as an abrupt, inconsequential cue.

Exact retained identity and control count

WAV SHA-256: 245a26f9b974bc9d1e2f86d2d18d6034bb3f58ecd696858767f771ac87df689e

6 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r2 · 28s · Agent-rejected

A Mouthful of Roll loses its dry body; its last gesture becomes a bright finite afterimage. Leaning Cell takes bass auth…

A Mouthful of Roll loses its dry body; its last gesture becomes a bright finite afterimage. Leaning Cell takes bass authority beneath the descendant, then opens only after the residue has expired.

The alternate early carry did not establish a convincing relationship.

Exact retained identity and control count

WAV SHA-256: bd88baa3138e51f384dd0e951fba7763cb9222470b370e69ac1d8cbd13d385a9

6 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r3 · 28s · Agent-rejected

The outgoing mouth gesture returns six times as a mid-band counter-rhythm after its dry source stops. Leaning Cell enter…

The outgoing mouth gesture returns six times as a mid-band counter-rhythm after its dry source stops. Leaning Cell enters below it at reduced level. The echo owns the upper foreground until expiry; only then does the new world open.

The mid-band residue was heard, but the low-only bridge and later entry felt disconnected.

Exact retained identity and control count

WAV SHA-256: da5121ec2c0423903af7988e73fbbe8f2e8617b38d0582bd01b87a873822e4cf

6 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r4 · 28s · Rendered; not selected for listening approval

The outgoing mouth gesture returns six times as a mid-band counter-rhythm after its dry source stops. Leaning Cell enter…

The outgoing mouth gesture returns six times as a mid-band counter-rhythm after its dry source stops. Leaning Cell enters below it at reduced level. The echo owns the upper foreground until expiry; only then does the new world open.

Its deterministic render and causal controls are retained. No independent musical verdict is inferred.

Exact retained identity and control count

WAV SHA-256: 108f84994aa9aa43ddedd9ec2c1d4579691fe9bc3fad1ebac9b6e58b9a9d08a9

6 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r5-answer · 28s · Agent-rejected

The outgoing mouth gesture becomes a finite answering object. Leaning Cell takes the bass, then its own mid-band phrases…

The outgoing mouth gesture becomes a finite answering object. Leaning Cell takes the bass, then its own mid-band phrases answer only in the measured gaps between the descendant's returns. These existing sources alternate the foreground before the incoming spectrum opens.

Conditional answers executed, but the listener heard an unintegrated effect and drop.

Exact retained identity and control count

WAV SHA-256: 0bae6176e794ff91c304893583a9a4e86845acd6a2f06ab39545d6c3545d2197

6 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r6-dialogue · 28s · Agent-rejected

The mouth cell becomes a half-second returning voice. The incoming source answers with its own retained attack in the me…

The mouth cell becomes a half-second returning voice. The incoming source answers with its own retained attack in the measured gaps, while its low pulse quietly establishes the new floor. After four returns, the spectrum opens over350ms rather than appearing as a blunt cut.

The shorter dialogue was heard as a scratch-in cue rather than a compelling interaction.

Exact retained identity and control count

WAV SHA-256: 8d0dc8fe1f7811617a023bfac875dc61138fa5aecb85e641c4968f2d5829b5f2

5 causal-control records. A signal-level difference proves application, not musical perception.

01-carry-r7-continuous-floor · 28s · Agent-rejected

The returning mouth cell and the incoming attack answer across measured gaps over one continuous low floor. The incoming…

The returning mouth cell and the incoming attack answer across measured gaps over one continuous low floor. The incoming upper spectrum arrives over800ms after the final descendant, preserving its already-established bass rather than replacing it with a drop.

A continuous low floor did not earn an audible-relationship pass.

Exact retained identity and control count

WAV SHA-256: 47d87b77b4bf6648784f9b9b33778c64e4289231c71e0b8949e33be21e9294f4

5 causal-control records. A signal-level difference proves application, not musical perception.

SOURCE PAIR 2 / 5 TAKES

Sine, Then Skin → Rigid Cell

02-inherit-r1 · 28s · Review disputed

Sine, Then Skin repeats its exact eight-second event order as a thin high-frequency trace. Rigid Cell supplies the new l…

Sine, Then Skin repeats its exact eight-second event order as a thin high-frequency trace. Rigid Cell supplies the new low body. The inherited trace ends before Rigid Cell takes the whole spectrum; no tempo or pitch correction is invented.

The first critic preferred this version, but its localisation of the difference was inconsistent with the exact audio.

Exact retained identity and control count

WAV SHA-256: 8e16c90f27616cc6cf2137adb5d947b8f3357297ed50d98d75b0ad7566851a88

3 causal-control records. A signal-level difference proves application, not musical perception.

02-inherit-r2 · 28s · Not selected

Sine, Then Skin repeats its exact eight-second event order as a thin high-frequency trace. Rigid Cell supplies the new l…

Sine, Then Skin repeats its exact eight-second event order as a thin high-frequency trace. Rigid Cell supplies the new low body. The inherited trace ends before Rigid Cell takes the whole spectrum; no tempo or pitch correction is invented.

Rejected in the initial comparison; that comparison also contains timing inconsistencies.

Exact retained identity and control count

WAV SHA-256: 216f58f022fcf52cb1884ce61ba70c286b3b649465fca99e450e84bd68a691fd

3 causal-control records. A signal-level difference proves application, not musical perception.

02-inherit-r3-loop · 28s · Agent-rejected

The last two seconds of Sine, Then Skin become a four-cycle high trace with an explicit in-object seam. Rigid Cell suppl…

The last two seconds of Sine, Then Skin become a four-cycle high trace with an explicit in-object seam. Rigid Cell supplies the low body, then owns the spectrum after the inherited cell ends.

The inherited high trace was heard as thin and the incoming entry as unearned.

Exact retained identity and control count

WAV SHA-256: fed2908eb795411c7fc284d21537906e8bafb0f6d2e78abe2d26d565380d88c5

3 causal-control records. A signal-level difference proves application, not musical perception.

02-inherit-r4-body · 28s · Not selected; review partly disputed

Sine, Then Skin's just-heard two-second cell survives in the mid register while Rigid Cell takes low authority. The inco…

Sine, Then Skin's just-heard two-second cell survives in the mid register while Rigid Cell takes low authority. The incoming upper spectrum follows over600ms after the inherited cell ends; the low floor remains continuous.

The critic rejected the handoff. Its claim of eight seconds of silence is contradicted by the PCM; the actual quiet source interval is much shorter.

Exact retained identity and control count

WAV SHA-256: b5750ea8138d99f66f0a34d9acdcda90aa5789106d5ec39db267c423a897f28a

3 causal-control records. A signal-level difference proves application, not musical perception.

02-inherit-r5-cued-body · 28s · Agent-rejected

The just-heard two-second Sine figure retains its mid-register shape while Rigid Cell takes the low body. Cue Rigid afte…

The just-heard two-second Sine figure retains its mid-register shape while Rigid Cell takes the low body. Cue Rigid after its observed source-internal quiet section, then let its upper world enter gradually; preserve a continuous floor without inventing a tempo map.

The revised source cue removed the earlier quiet span, but the listener still heard replacement rather than convincing inheritance.

Exact retained identity and control count

WAV SHA-256: 649c14dd92f0da3ebed282162c950e426d505eca7b7fad212f92b7c1a14c9e3e

3 causal-control records. A signal-level difference proves application, not musical perception.

SOURCE PAIR 3 / 2 TAKES

Sherbert Oxide → Dauerversprechen

03-obscure-r1 · 28s · Mixed reviews; not selected

Sherbert Oxide remains materially present but becomes a receding high trace. Dauerversprechen tries the low register, wi…

Sherbert Oxide remains materially present but becomes a receding high trace. Dauerversprechen tries the low register, withdraws for two seconds, then returns. The upper trace is not replaced by a new invented object.

Flash retained the energetic transition; the stricter Pro review found its identity interaction superficial.

Exact retained identity and control count

WAV SHA-256: 7c21df734a6eb716a7bf8511d86265a1e37189a43f61c8cec6dd08ccb58a9e0c

4 causal-control records. A signal-level difference proves application, not musical perception.

03-obscure-r2 · 28s · No reliable preference established

Sherbert Oxide remains materially present but becomes a receding high trace. Dauerversprechen tries the low register, wi…

Sherbert Oxide remains materially present but becomes a receding high trace. Dauerversprechen tries the low register, withdraws for two seconds, then returns. The upper trace is not replaced by a new invented object.

The initial critic could not distinguish the two levels and called them duplicates. No winner is inferred.

Exact retained identity and control count

WAV SHA-256: a235ea89fea0ef625536195982d64d334952cf3a0b12af7a3fe54d9adc9a6b1a

4 causal-control records. A signal-level difference proves application, not musical perception.

SOURCE PAIR 4 / 2 TAKES

Antwort ins Leere → Outil 132

04-answer-r1 · 28s · Rendered; not selected for listening approval

Antwort ins Leere stops. Outil 132 may answer only after the rendered outgoing contribution is measured absent for 250ms…

Antwort ins Leere stops. Outil 132 may answer only after the rendered outgoing contribution is measured absent for 250ms, followed by the authored delay. A non-silent observation suppresses the answer rather than playing a scheduled imitation of response.

Its deterministic render and causal controls are retained. No independent musical verdict is inferred.

Exact retained identity and control count

WAV SHA-256: 3203986333abcd8f89dbaa4e6ceb8e2e6f2f811fb7b04286f2cd1bcd27ca019d

3 causal-control records. A signal-level difference proves application, not musical perception.

04-answer-r2 · 28s · Rendered; not selected for listening approval

Antwort ins Leere stops. Outil 132 may answer only after the rendered outgoing contribution is measured absent for 250ms…

Antwort ins Leere stops. Outil 132 may answer only after the rendered outgoing contribution is measured absent for 250ms, followed by the authored delay. A non-silent observation suppresses the answer rather than playing a scheduled imitation of response.

Its deterministic render and causal controls are retained. No independent musical verdict is inferred.

Exact retained identity and control count

WAV SHA-256: 1a725c33a7addf3739dfc2acc8363da07714a614877b90550edad03c0fc3c973

3 causal-control records. A signal-level difference proves application, not musical perception.

SOURCE PAIR 5 / 2 TAKES

A Ghost Declines → A Mouthful of Roll

05-rupture-r1 · 28s · Agent-selected finalist

A Ghost Declines refuses continuity. Its source ends, the exact authored gap is left empty, and A Mouthful of Roll enter…

A Ghost Declines refuses continuity. Its source ends, the exact authored gap is left empty, and A Mouthful of Roll enters as a new fact. No noise floor, riser, tail or compensating layer repairs the rupture.

In a paired listen, the shorter withdrawal held tension like an anticipatory breath and made the incoming world consequential. The exact authored gap is one second; the model’s timing estimates were inaccurate.

Play and retained identity

WAV SHA-256: 90f52492548c8a1da3667bcd1603a05f773c13f480a27c4f663e935b1d2e5066

3 causal-control records. A signal-level difference proves application, not musical perception.

05-rupture-r2 · 28s · Agent-rejected

A Ghost Declines refuses continuity. Its source ends, the exact authored gap is left empty, and A Mouthful of Roll enter…

A Ghost Declines refuses continuity. Its source ends, the exact authored gap is left empty, and A Mouthful of Roll enters as a new fact. No noise floor, riser, tail or compensating layer repairs the rupture.

The longer withdrawal was heard as losing tension and drifting into dead air. The exact authored gap is two seconds.

Play and retained identity

WAV SHA-256: 9a8437ac70f4b10b873433e0c4d83858c9783ac1a1c1f6782004acfa11dc8a93

3 causal-control records. A signal-level difference proves application, not musical perception.

SOURCE PAIR 6 / 6 TAKES

A Ghost Declines → Dauerversprechen

06-ghost-dauer-exchange · 28s · Agent-rejected

A Ghost Declines leaves one short percussive object behind. Five450ms descendants and Dauerversprechen's own answering a…

A Ghost Declines leaves one short percussive object behind. Five450ms descendants and Dauerversprechen's own answering attack alternate over its new continuous low floor. The incoming upper world opens only after the ghost has expired, without interrupting the floor.

The short exchange was judged abrupt and insufficiently connected.

Exact retained identity and control count

WAV SHA-256: 35dfd867462fbf043052748e6b36d43fdf1a5d5e6bb086aabf42543ef29cf251

5 causal-control records. A signal-level difference proves application, not musical perception.

07-ghost-dauer-overlap · 28s · Functionally retained; musical limit open

The incoming percussive world first teases and withdraws inside Ghost's still-running pattern. A captured Ghost cell bec…

The incoming percussive world first teases and withdraws inside Ghost's still-running pattern. A captured Ghost cell becomes a returning counter-pattern. Bass authority changes once, at8s; the descendant survives that change while the two assessed instrumental upper worlds interact.

The listener found the rhythmic switch coherent and retained it for further review, but could not distinguish the echo from its dry control. No exceptional carry claim is made.

Exact retained identity and control count

WAV SHA-256: 89da358cfb5629799151df49bf2ee04c7be792ba075a46237b47d95644b9524a

6 causal-control records. A signal-level difference proves application, not musical perception.

08-ghost-inheritance-and-memory · 36s · Agent-rejected

Give Ghost sixteen seconds to establish its figure. Dauer teases and withdraws inside that world. At the single bass han…

Give Ghost sixteen seconds to establish its figure. Dauer teases and withdraws inside that world. At the single bass handoff the last four-second Ghost object retains two independent upper-register cycles; a later captured consequence answers after even that loop has expired.

The longer inherited figure was heard as a cluttered rhythmic conflict.

Exact retained identity and control count

WAV SHA-256: e14aeb02453504ada8363a6ecbd4f6190a885a2220b75ebf030243fced3c4669

7 causal-control records. A signal-level difference proves application, not musical perception.

09-ghost-translated-cycle · 36s · Agent-rejected

Preserve Ghost's ordered four-second figure while explicitly accelerating its realised cycle to3.6s. Two brief Dauer att…

Preserve Ghost's ordered four-second figure while explicitly accelerating its realised cycle to3.6s. Two brief Dauer attacks tease the future. At one bass handoff the translated Ghost figure shares the authored450ms cycle with the incoming world, recedes into texture, then leaves one late memory. Pitch changes are intentional; no automatic tempo detection or pitch-preserving stretch is claimed.

The explicit rate mapping executed, but the listener still rejected the interaction as cluttered.

Exact retained identity and control count

WAV SHA-256: 8e839b09c4d4fee7438c7f76fdf0e341119e4a294fb8c08e85e49e27a66de282

7 causal-control records. A signal-level difference proves application, not musical perception.

10-ghost-rhythm-in-dauer-material · 36s · Agent-rejected; research reference

Ghost's ordered energy contour survives the handoff in Dauer's actual upper-register material. The four-second source fi…

Ghost's ordered energy contour survives the handoff in Dauer's actual upper-register material. The four-second source figure is explicitly translated into a3.6s cycle and supplies a causal RMS envelope, while Dauer supplies the waveform and new low body. After two cycles the inherited contour yields to Dauer's own unshaped upper rhythm; its captured transformed consequence returns later.

The source-form to carrier-material transfer is causally verified. The listener did not find audible identity survival; its silence/timing description also has limits.

Exact retained identity and control count

WAV SHA-256: 1443865a22927418eb462113c7dcf5d7d2095a1fee9a070fb955f95d17192f43

7 causal-control records. A signal-level difference proves application, not musical perception.

11-inherited-contour-foreground · 36s · Agent-selected finalist; source-informed review

Give the inherited Ghost contour foreground authority inside Dauer's material. Reduce the new low floor so the translate…

Give the inherited Ghost contour foreground authority inside Dauer's material. Reduce the new low floor so the translated rhythmic detail can be heard; keep the source-shaped upper voice prominent until it yields to Dauer's own contour. Its captured memory answers after that handoff.

The reviewer recognised inherited rhythmic accents across a severe timbral change and heard dialogue between clean and degraded states in the full performance. Roughness and possible fatigue remain musical caveats. The original cell and transformed contribution were supplied alongside the full mix.

Play and retained identity

WAV SHA-256: ea96f983c07ef1b710d44db35a20ecc366fedac8f6108e08bcafe78fd7d1c58d

7 causal-control records. A signal-level difference proves application, not musical perception.