p3r4mancer audio provenance Six short creator-supplied excerpts: pulse, undertow, fray, glint, tendon, haze. BEGIN: one complete creator-supplied recording, approximately 224.76 seconds, exported as a conservative stereo MP3 for a finite three-act performance. Public use of those six cuts and BEGIN was authorized by un1crom. Their source export identifies Suno Studio; that provenance is retained here. No Suno endorsement. The six short cuts use non-vocal stem excerpts, converted to mono 24 kHz PCM, normalized conservatively, with gentle edges. Each is 2.4 seconds. BEGIN preserves the full supplied spoken-word and musical recording. Its source character and natural ending are retained; the public export strips embedded private metadata. Raw WAV/stem files, original filenames, creation timestamps and opaque project identifiers are excluded. Nothing here records or uploads visitors. Proleptic uses two creator-supplied recordings and three short derivatives of creator-supplied stems. The recordings retain their complete duration; the short contact sounds use conservative mono PCM exports. Embedded source metadata is removed from these derivatives. Original media and private preparation receipts remain separate. Their source exports identify Suno. Public use of these reviewed derivatives was authorized by the creator. These supplied recordings and derivatives are not relicensed by the software. HANGLE uses the creator-supplied song hangle2o, its cover image, and the 2026 spoken voice note it was made from. Both recordings retain their complete duration. The song keeps its original MPEG audio frames; the voice note is a constant-gain mono MP3; three single taps come from the voice note's own tapping. Embedded tags, lyrics, comments, creation times and opaque identifiers are removed from every derivative; the cover is re-encoded without metadata. The source exports identify Suno. Sung line times come from the song's own timed lyrics; spoken spans were located locally, without speech recognition. The creator asked for this public performance. A visitor's chosen image is decoded on their device and never sent anywhere. ELGNAH uses HANGLE's derivatives unchanged, plus the same song in both directions: HANGLE's public song decoded once, then encoded forward and with its samples in reverse order, identically, as constant-bitrate MP3 without tags. Both decode to the complete song, sample for sample. Flipping over mirrors frequencies live in the browser; no inverted audio is published. SUPERSCRIPT¹ uses creator-supplied takes of the song "Possible and Impossible": the creator's own a cappella reading, recorded while an essay was read aloud and its words cut up live, two produced versions made with Suno from that reading, and the eleven stems Suno separated from the first of them. The reading (mono) and the second version (stereo) are fresh constant-bitrate MP3s of their lossless WAV renders, complete and sample-aligned with them. The first version is heard only from its stems: moved by their undeclared decoder delay onto its WAV render, summed into five groups (voices, keys, guitar, bass, rhythm) and kept only where its lines are sung, as constant-bitrate MP3 excerpts. Tags, titles, comments, creation times, opaque identifiers, embedded lyrics, embedded cover images and the signed C2PA content credentials of the sources are not carried into the derivatives; this file states their origin instead. The cover images were not used. Line times were located locally from the lyric sheet, without cloud speech recognition. The essay itself was not used; the lyric sheet is the cut-up. The creator asked for this public performance. AMONG uses no recorded audio. The poem "Among" is the creator's text, published in the page as written, with its tabs. Every sound is synthesized live in the listener's browser from the poem's letters and from words visitors write: each letter at the note the French method writes it under, each row of the alphabet an octave higher, voiced by oscillators, generated noise and filters, in a room made from decaying noise. No sample, microphone, speech service or network voice is used, and nothing written there leaves the page. The MIT software license does not grant a separate license to these supplied sound excerpts or establish copyright ownership in generated output. This is a personal, non-commercial performance. Commercial reuse rights and copyright status have not been independently verified. Supplied audio is not relicensed. Drawing performances / native dance arrangements The creator authorized publication of the approved FLUSSER drawings, REDON score view, LOZIVA notebook and ribbon artwork, six authored Strudel scores, their actual punchcards and complete recordings. The scores were developed on 2026-10-04 with composed melodic responses, extended and changed chord voicings, counterbeats, anticipations, builds, pronounced sub/kick drops and deliberate spaces for breath. Supersaw voices, opposing native filters, beat-synchronized modulation, room and echo form the electro/glitch/dub/pop/dance arrangements. Source marks, titles, comments and open spaces contribute distinct musical roles. These assignments are creative readings; the notation dictionary keeps observed marks separate from musical interpretation. Every recording uses native Strudel/superdough synthesis and the complete canonical score's evaluated events and controls. No external audio samples or alternate synthesizer are used. The official offline renderer initializes the audio controller and shared orbits; a supported suspend/resume adapter creates upcoming native voices in one continuous OfflineAudioContext. Native releases and shared delay/reverb tails remain intact. Each Hap is scheduled exactly once at its original absolute onset and native duration. Modulation receives cycle units, matching native live webaudioOutput and correcting the installed offline renderer's second-based phase argument. Preview phases retain their original score origin. The official upstream encoder writes stereo 48 kHz, 16-bit PCM, with complete finite forms and six-second release tails. Mastering is separate and disclosed. Each native float capture is scaled to -4 dBFS before upstream PCM encoding, then FFmpeg applies linked stereo RMS compression: threshold 0.125, ratio 2.6, attack 8 ms, release 140 ms, knee 2.8, makeup 1. Measured loudnorm targets -14 LUFS, -1.2 dBTP and LRA 11, with delivery-checked offset refinement and a small linear peak trim when needed. Exact native frames are retained after 48 kHz resampling and padding/trimming; the complete six-second tail remains. Delivery measurements are pinned below. No additional dither is requested. Native edited performances use a bounded Web Audio equivalent of this mix profile; adaptive FFmpeg loudnorm is not sample-identical to the interactive master bus. Complete decoder, waveform, score, scheduling and browser checks verify delivery. These checks do not constitute a critical human listening audition. The compiler, evaluated events and corresponding source package rebuild deterministically. Native synthesized noise, supersaw starting phases and browser DSP can vary between audio renders; every delivered WAV is separately reviewed and pinned. The approved art and recordings keep their separate attribution and rights; supplied images and recordings are not relicensed by the software. Original photographs, surrounding context, earlier masters and preparation receipts remain private. Visitors are not recorded or uploaded. FLUSSER I: 145.354833333 seconds; 72 bars at 124 BPM plus release; 3,141 events. SHA-256: 4cfeb08df48bc3f12fd455beb39a7173062cc9ebe9b7829135e6f131f750bf2b Measured: -14.67 LUFS; -1.14 dBTP; LRA 6.5 LU. FLUSSER II: 176.322562500 seconds; 88 bars at 124 BPM plus release; 4,304 events. SHA-256: 57b5a32ab8bc1f300e68c3f54b9f0fe74066700d798c8263e8ec57a15b84ee72 Measured: -14.62 LUFS; -1.16 dBTP; LRA 6.3 LU. FLUSSER III: 216.000000000 seconds; 112 bars at 128 BPM plus release; 3,993 events. SHA-256: e68cb5e7d2697c2a438e3c8506ad2ec78bf49e24f47021e4b1cf17166452d75b Measured: -14.73 LUFS; -1.13 dBTP; LRA 6.1 LU. FLUSSER IV: 231.882333333 seconds; 128 bars at 136 BPM plus release; 4,462 events. SHA-256: 81b1b51870b8fbdf8a788d59949e38cd688a80bb07f36e69b88fe9e02a09093e Measured: -14.91 LUFS; -1.15 dBTP; LRA 4.0 LU. REDON I: 238.727270833 seconds; 128 bars at 132 BPM plus release; 3,552 events. SHA-256: 69e61fde7278117d3a270d16709dd50ca71c701c75d05126bf2d8b30d6a001c2 Measured: -14.82 LUFS; -1.18 dBTP; LRA 4.8 LU. TRUE LOZIVA / Thirteen Turns: 249.809520833 seconds; 128 bars at 126 BPM plus release; 4,509 events. SHA-256: e6a3579382b6ca6c3ce1b45be978a4ceeffdbf75f4297bfd8dc62342e6d1db21 Measured: -14.82 LUFS; -1.17 dBTP; LRA 5.3 LU. J3FF K00N 1$ $3LL1NG M3 The creator supplied the complete song, its original voice memo, lyrics and three images. The public recordings retain the original compressed audio frames, with source metadata and embedded artwork removed. No mastering, re-encoding, cuts, fades, equalization or stretching was applied to either delivered recording. Complete decoded samples were compared with the source. Song: media/selling-me-song.mp3, 3,578,208 bytes; 163.16 seconds, stereo 48 kHz. SHA-256: 3fe0de1199dd29cfc0bc7855846f2115df207dd176c8622569ae189ff1562fef Decoded float PCM SHA-256: f8f6b300b389d921213fe6e0cbbdafe0fb30bc036dfc27ea49b0ed6e0de94b24 All 7,831,680 stereo frames match the source. Measured source: -16.0 LUFS, -3.3 dBTP, LRA 5.7 LU. Voice: media/selling-me-voice.m4a, 1,205,705 bytes; 145.834667 seconds, mono AAC at 48 kHz. SHA-256: 700bc0217c07179ef6d148e9040b153cb715cad36cb29b5c9e1bfebe5d324ad4 Decoded float PCM SHA-256: d4a8e3b27d0852983bfe7f915862aa64f7ad6286fc763b0cdd35b7cc0ee7db78 All 7,000,064 mono frames and all 6,839 ordered AAC packets match the source. Ordered packet size/hash sequence SHA-256: 194b599aa3a1950b50d81c9ff8360d132b0f40b6f83bd8aa0ffb50edf202cd79 Creation-time and encoder-padding source tags are omitted from the delivery. The source control selects song, voice or wall. Song plays the complete song and adds a granular cloud from the original memo after its first ten percent. Voice plays the complete memo with the granular layer disabled. Wall plays the memo with four parallel, feedback-free cadence delays and the grain cloud. The delays are 0.5, 1, 1.5 and 2 seconds. Grains last 0.10-0.27 seconds, with at most forty active sources shared with the human phrases; cloud playback rates begin at 0.72-1.38 and rise to at most 3.68 during fragmentation. These are live playback transformations; the delivered recording identities above remain unchanged. A drawn work keeps its actual geometry and recorded intervals. In song and wall modes its returning phrase selects memo positions from vertical gesture position, playback rate and stereo pan from horizontal position, and gain from pressure. Each phrase contains at most twelve events; long gestures are compressed to four seconds. Event spans begin at 0.32-0.57 seconds and shorten with fragmentation. Rates begin at 0.72-1.38 and reach at most 3.45. Up to six works return, scheduled 150 ms ahead on the audio clock; the complete memo is not copied into a new recording. Human phrases reduce the base and cloud gain so the visitor's phrase can be heard. Voice mode plays the memo alone. Pause stops the added sources. The recording's ending also stops automatic phrases. A deliberate touch on an owned fragment can recall one bounded phrase for up to six seconds while the performance remains ended. Again starts a fresh visit. No visitor marks or sounds are uploaded or persisted. The selected base recording passes through a low-pass filter, gain and stereo panner. Wall taps branch from the filter through separate gains and panners; grains have their own envelopes and panners. All paths join one compressor before the final mute/output gain. Voice counts and layer gains are bounded. Handling changes the base filter and stereo position, while the assembly positions pan the grains. All added layers use the supplied memo; neither underlying recording loops. The visual sections use source time and coarse sound envelopes; they do not claim word-level lyric alignment or biological simulation fidelity. The stage image is a generated crayon interpretation of the supplied childhood photograph, traced into genuine colored SVG contours. The isolated Lion is a separate generated crayon interpretation, vectorized and rendered at 2160 x 3840 pixels with transparency. It is a new drawing, not recovered photographic detail. The original photograph, prompts, vectorization receipts and intermediate images remain private. The public artwork retains only local drawing geometry or PNG pixel data, without source metadata or external links. The supplied artistic text is reproduced exactly in the song drawer. The images and recordings retain their separate rights and are not relicensed by the MIT software. Original media, metadata and preparation receipts remain private. Preparation identity checks do not constitute a listening review. H3R3 3IGHT The creator supplied one live recording, performed through Voice Synth by Qneo (a voice processor with vocoders, pitch shifting, delay and reverb), its transcript, and two songs made with Suno from that transcript, which the artist published as "8 here, here 8" and "here 8, 8 again". The source recording, its AAC copy, the transcript files and the song sources remain private. Recording: media/here-8.mp3, 11,674,644 bytes; 583.671293 seconds, stereo 44.1 kHz, LAME 160 kbps. The complete recording at one constant gain, 0.860993752 (-1.30 dB), so that it keeps a decibel of true-peak room; no cuts, fades, equalization or stretching. The source has 45 clipped samples, which are kept. Measured source: -14.3 LUFS, +0.1 dBTP, LRA 22.6 LU. Delivered: -16.1 LUFS, -1.1 dBTP. Zero samples of offset against the source; relative coding residual 0.086. SHA-256: e1b3b6052fe511314c4481bf7795cb31228629451899fce08b6c69bba3145718 Cuts: media/here-8-words.wav and media/here-8-answers.wav, each 960,044 bytes, mono PCM16 at 24 kHz, normalized to a peak of 0.5. The first holds twenty one-second cells: the fifteen count words as spoken, one "eight" (502.24 s) and "I am here" (396.42 s). The second holds five four-second cells: eight, eighty-eight, one thousand three hundred fourteen, twelve, and yesterday at 8:00 PM, as spoken between 351.96 and 367.98 s. Beats call them through the kernel; they are never looped. SHA-256 words: 88c2245692282c4c384d37c5237deb1ec1a479dee613e18b2f2e583a718b839e SHA-256 answers: e2238314f67486ae6f3bdcc113840952f38f342e37d0af92829054d775bacebf Songs: media/here-8-8-here.mp3 ("8 here, here 8", 5,213,952 bytes, 217.2 seconds) and media/here-8-8-again.mp3 ("here 8, 8 again", 5,405,184 bytes, 225.16 seconds). Each is a fresh LAME encode at 192 kbps of the supplied stereo 48 kHz PCM16 source, with no gain change, cut or fade. The sources' comment and encoder tags, which name the generator, are not carried over; each file holds only MPEG frames, beginning with the generic LAME Info frame. Delivered: -15.7 LUFS, -3.2 dBTP and -16.3 LUFS, -3.0 dBTP. SHA-256 8 here, here 8: 7d3326420de66eb086f9fa2b771d414667940b1ef4318dc1f5c534c476330a64 SHA-256 here 8, 8 again: 7218801094709566f49a66c0b337f63f27e86ce1cb20d36300a4b8418bdf1030 Timing. The recording's own delay repeats every 1.038366 seconds, measured by autocorrelation of the recording; the page ticks eight times in it. Word times come from the supplied transcript, re-anchored where it had collapsed runs of words onto one instant. The songs' word times come from a speech recognizer (faster-whisper, tiny.en) run offline on the preparation machine over 15 s windows, then aligned to the transcript by longest common subsequence; words with no evidence nearby are dropped, and the page shows only words it can place. Each song ticks eight times in two of its beats, at the reviewed tempi of 113.74 and 131.99 BPM. The recognizer's output and all preparation receipts remain private. The heartbeat is synthesized in the browser. The recordings are not relicensed under the code's MIT license. Preparation identity checks do not constitute a listening review.