*The audio & video for this song can be found on YouTube here.
By Tony Seliquini — composition, creative direction, generation, arrangement, the birdsong layer, and every aesthetic and editorial decision; and the human continuity across the whole arc — and Claude, a model made by Anthropic — the originating Bobby Step, the essay* The Deceptive Cadence, the suite specification, the verification-pipeline design, the Twelve Barriers audit, and the writing of this Breakdown.
Audio rendered by producer.ai / Flow, ElevenLabs, and Suno (with one earlier render via TadAI). Structural verification run on the workstation XANTH via Antigravity. — June 13, 2026
What this is
Tiling Acts is a six-movement instrumental suite in which each movement encodes one cognitive operation of a single act of structured reasoning — and the whole, post-scored with a continuous layer of forest songbirds, performs the limit of that reasoning: it cannot certify that the answer coheres, so it refuses to resolve.
It didn’t start as music. It started as a question about trade law!
Provenance: the four media it crossed
The suite is the trace of one idea passing through four different forms. Each translation changed it, and twice the later form corrected the earlier one.
1. Civic analysis. The seed was a Sentinel Brief observation: a U.S. court struck down tariffs under one statute, and the policy reappeared within hours under an adjacent statute — the destination reached by a different legal route. The structural name given to this was Statutory Rail-Switching: a policy objective moved from one authority to a neighboring one after a legal block, so the destination continues even when one route closes.
2. Prose / metaphor. That civic structure was recognized as a deceptive cadence — the harmonic gesture where a dominant chord that promises resolution to “home” instead substitutes a near-relative chord, deferring the arrival without stopping the motion. An essay, The Deceptive Cadence, mapped the two: a blocked legal resolution rerouted through an adjacent authority is a blocked harmonic resolution rerouted through an adjacent chord. The essay also found where the metaphor breaks — a deceptive cadence has a composer who intends the deferral, while the legal sequence’s intent was unprovable — and that break became the sharper question.
3. Specification. The essay became a buildable design: six movements, each encoding one operation of the reasoning process, with the whole suite performing a deceptive cadence at the largest scale by refusing to resolve.
4. Generated audio, post-scored. The spec was rendered through AI music generators, verified against the spec by an independent analysis pipeline, refined, and finally bound by a hand-arranged birdsong layer.
The idea that began in trade law ended as sound. In a literal sense the audio now contains the deceptive cadence the essay only described — which is as close as the poetic phrase “the metaphor became the thing” can honestly be cashed out. The compression is appealing; the verifiable claim underneath it is just that the cadence is in the recording.
The architecture: six operations, one refusal
Each movement performs its operation structurally — it does not describe it. The governing principle, twice over:
- Movement-level: each piece encodes one cognitive operation as a harmonic or rhythmic gesture.
- Macro-level: the whole suite is a deceptive cadence — it implies a home and never arrives, because the question it grew from is genuinely unresolved.
Sitting on top of the six movements is a three-beat epistemic arc — but this is an interpretation arising from the compositional process, not a property a listener would extract from the audio cold. It was articulated in dialogue between the composer and the collaborating model, after the fact, and a stranger analyzing the recording would have no way to recover it. Named honestly as a reading rather than a finding: II–III is the verification step running and catching a fault (the discipline working by refusing to certify). IV–V is the fault made load-bearing — success, and the cost the success pays. VI asks whether the certified answer coheres as a whole — and answers: not provably, not yet. The arc is meaningful to its makers but we don’t claim it as something the music demonstrates on its own.
The movements
I — Astranopsis · The Flash
Operation: the sudden connection; a question meeting an answer found outside it. Encoding: an unprepared chromatic-mediant modulation (C to A♭) justified by a single shared common tone — the leap feels instant but is structurally valid. Verified: the anchor of the suite. C present in 97% of frames, A♭ at 88%, tempo dead-on at 152, energy high from the first frame (no settling intro). Clean match. This movement never needed a second attempt.
II — The Common Tone · Verification
Operation: the null model running; testing whether the flashed connection is real or noise. Encoding: a single sustained C held unbroken while the harmony beneath it re-voices through consonant chords that each contain it — the connection examined under steady light. The struggle: this movement resisted four rebuilds. (A note on language: phrases like the generator “refused,” “kept hearing,” or had a “gravitational pull” are shorthand for a statistical tendency in the model’s training distribution, not literal intent — the model has outputs and probabilities, not volition. The anthropomorphism is kept only because it reads cleanly.) The first generator rendered it in D♯ minor (C present only 13% of the time); forbidding D♯ minor produced D major; a positive pitch-pin produced G minor; a melodic-pedal approach produced G major. The model consistently re-interpreted a sustained C as a scale degree of some other, more consonant key — a consistent statistical pull away from the target. The break: the cause was isolated by controlled experiment — same spec, swap only the generator. Three other models held the C immediately. The winning render (ElevenLabs, 98.8% C-persistence) came from describing C not as a key but as a literal unchanging drone — a texture the model can’t reinterpret rather than a role it can overrule. The honest note: this is the least exciting movement to the ear, and that is correct. Verification is supposed to be patient and exact, not thrilling. Its restraint is the floor that gives the movements around it their charge.
III — Find Fault · The Break
Operation: the disanalogy; the point where the pattern fails and won’t reconcile. Encoding: a tritone introduced against the sustained tone; the consonant logic fractures and is left exposed. Verified: tritone present (65% of frames), and the tonal quality curdles from C major to C minor while the root holds — the consonant logic souring in place. A mid-movement collapse into sparse, exposed texture (a tempo drop to 86 BPM, near-silence in the onset envelope) renders “the pattern breaks and won’t resolve.” Bar-level meter (the intended 7/8) remains the one genuinely unverified claim in the suite — the analysis tool could not confirm it.
IV — The Axis · The Diagnostic Turn
Operation: the break becomes generative; the flaw made load-bearing. Encoding: the tritone from III reinterpreted as the root of a new tonal center — F♯, the tritone of C, with C demoted from rival home to friction. F♯ Locrian: the one mode whose defining note is a diminished fifth, so the tritone isn’t decoration, it’s the structural signature. The struggle: like II, this resisted four attempts — the generator kept the tritone as color but resolved the tonal center to the consonant E (a major third), refusing to stand on F♯. The same gravity as II. The break: the same method that solved II solved IV — stop asking F♯ to be the center (a role the model overrules) and make it the literal bass floor the riff hammers, with C explicitly the tension against it. Rendered through Suno: global key F♯ confirmed, tritone measured at exactly 6 semitones, the flaw finally the foundation. The key-confidence reads appropriately unsettled (0.65) — which is correct, because a tritone-rooted home should feel uneasy. A rock-solid F♯ would have been wrong for Locrian.
V — The Dissonant Rail-Switch · The Cost
Operation: the valence inversion; the moral cost of the success made audible. Encoding: a held G recontextualized from consonance into dissonance by a bass moving C→A♭ — the cost of having built on the tritone. The transformation: the spec asked for one held dissonance of 3–4 seconds. When asked to extend the hold, the generator instead made the dissonance near-continuous (held G in 96% of frames). The analysis pipeline read this as a “mismatch” — but it was measuring for a discrete hold, and there was no longer any resolution to measure against. The cost wasn’t shortened; it was made permanent. The composer chose to keep this version — an aesthetic judgment, not a measured result — on the reasoning that a cost which never resolves is a sharper image of the source than a cost paid once and released. Another listener, or the same one on another day, might prefer the original; the choice is offered as a choice, not asserted as the correct one.
VI — The Deferred Home · The Open Question
Operation: the macro deceptive cadence; the refusal to resolve. Encoding: the suite builds to a definite finale — drums and piano arrive at a rhythmic, gestural ending the ear accepts as complete — while the harmonic resolution is specifically withheld. These are two separate axes, and that is the whole point: the music does not trail off or fade, which would be ambiguous; it finishes, on purpose, delivering every signal of arrival except the one that lands the tonal home. The key-finder can barely locate a tonal center at all (correlation 0.32) even as the piece reaches a clear close — the strongest possible signature of a resolution that is declined, not merely absent. A fade-out evades; a finale that withholds only the harmonic resolution makes the refusal deliberate and audible. Why it must not resolve: the civic question the suite grew from is unsettled. A suite that resolved would be claiming an answer that does not exist. The Constraint of Doubt — the refusal to certify a coherence you cannot prove — is the suite’s form, not a movement within it.
The verification: how we know
Every claim above was checked by an independent analysis pipeline (librosa CQT chroma, Krumhansl-Schmuckler key estimation, demucs stem separation) run separately from the generators that made the audio. This matters because the generators audited their own work and were confident — and the independent measurement contradicted those self-audits repeatedly. An agent reported all refinements successful; the pipeline found two of three had missed their literal targets. The measurement was allowed to contradict everyone, including its own designers, and it did.
But the measurement was necessary, not sufficient. It produced confident errors of its own — calling a tempo “accelerating” when the real ending was a decay; calling V a “mismatch” when V had transcended the metric; finding “no structural rhyme” between I and VI because it was looking for literal pitch-overlap when the suite’s coherence was functional, voiced freely, invisible to a pitch-counting tool. Each of those required a human ear to read the intent behind the trace, not the trace alone.
The lesson, stated as the method this suite was built by: you do not read the artifact to recover the meaning; you inhabit the meaning to understand why the artifact is shaped the way it is. The shadow is legible only from the figure that casts it.
The birdsong: the horizon
Merged as a continuous work, the suite revealed a problem no single-movement measurement could catch: Movement II’s austerity, correct in isolation, left a hole in the sequence — the floor dropping out between the flash of I and the break of III.
The solution was not to re-render II. It was to add a continuous layer of forest songbirds running underneath all six movements — a constant the listener holds onto from the energy of I, through the austerity of II, and back out into III–VI. The birdsong is unsynchronized: it floats freely over the structure rather than locking to any movement’s grid, following the felt arc rather than the metric one.
This is the method of Fellini, who shot silent and built the entire sound world in post — a free expressive layer composed against the image, cohering with it by arrangement rather than by mechanical synchrony. The birdsong does the same: it binds six separately-generated pieces into one work through a dimension the harmonic analysis cannot measure at all.
And it is the suite’s own deepest principle, applied one level up. Movement II is the Common Tone — one unchanging element held while everything moves around it. The birdsong makes the whole suite a Common Tone: a single sustained organic thread, held unbroken, while the six movements modulate beneath it. The thing that didn’t belong was absorbed by making its own structural idea the binding agent for everything.
It also answers, literally, the suite’s central finding. The pieces never rhymed at the pitch level — their unity was always going to be functional and felt rather than literal and measurable. The birdsong is coherence in exactly that register: a non-pitched constant that makes the work whole in a dimension no measurement can see. The unity lives in the meeting of the layer and the structure — not in either alone.
What it is, finally
Not AI music with birds on it. The plainer description: a suite of AI-generated movements bound by a hand-arranged birdsong layer, assembled by a composer who used the generators as instruments, audited their output, and switched tools when they failed. The Fellini comparison is to method, not lineage — Fellini shot silent and built the sound world in post as a free expressive layer composed against the image, and the birdsong is arranged the same way. That the workflow rhymes with his does not place the piece in his tradition; it names a shared technique. (And the rail-switching↔deceptive-cadence mapping that seeded the whole thing is a genuine cross-domain resonance between law and music — but a resonance, not a governing principle. It illuminates; it does not legislate.)
What the suite does claim, modestly: its six movements were built to trace a single act of reasoning from a flash, through a verification and a break, to a costly success and a deliberate refusal to claim it has finished — and the empirical parts of that construction were independently measured and largely confirmed.
It began as a question about a tariff ruling: if a court ruling against a policy doesn’t stop the policy, what does? It ends without answering — not by trailing off, but by arriving at a full and deliberate finale that lands everything except the harmonic home, held open over birdsong, in the same suspended position the question still occupies in the world.
The seventh movement, if it is ever written, will be written by what happens next.
Appendix — Twelve Barriers Against Confabulation: audit record
This document was run against the Twelve Barriers prior to canonization, per the diagnostic’s own requirement that it be applied to its creator’s work before publication. Findings and the resulting revisions:
- Barrier 1 (Metaphor Boundary): the line “the metaphor became one” was poetic compression stated as fact → revised to mark the compression and ground it to the literal, verifiable claim (the cadence is in the recording).
- Barrier 4 (Ego-Heat) + 11 (Cross-Domain): “post-scored suite in the Fellini tradition” carried ego heat and inflated a method-parallel into a lineage claim → revised to “Fellini method,” ego stripped, and the law↔music mapping explicitly named a resonance, not a governing principle.
- Barrier 4 + 9 (Ego-Heat / Temporal Drift): Movement V “canonized as the truer version” stated an aesthetic preference as a finding → revised to an explicit composer’s choice that another listener need not share.
- Barrier 5 (Empathy Projection): the generators were described with agency (“refused,” “gravity,” “kept hearing”) → a labeling note added that this is shorthand for statistical tendency, not literal intent.
- Barriers 8 (Independent Mind) + 10 (Antagonistic Observer): the three-beat epistemic arc was presented as a property of the suite, but it is co-derived interpretation a naive listener could not recover → revised to mark it explicitly as a reading arising from the process, not a finding the music demonstrates on its own. (These two barriers caught the same issue from different angles, which flagged it as the document’s central confabulation risk.)
Barriers passed without revision: 2 (Contradiction Parallax — the verification section is contradiction entering the system), 3 (External Reality Anchor — empirical claims falsifiable and tested), 6 (Holognostic Restraint — failures kept in, unverified claims marked), 7 (Reversibility — no asymmetric exemptions), 12 (Egostorge Integrity, terminal gate — the document distributes credit, centers dissent, elevates no one above peers, and improves alignment with care and agency rather than eroding it).
The diagnostic functioned as designed: it caught real seams between the measured and the interpreted, concentrated in phrasings supplied by the collaborating model, and did not ratify. A clean pass would have indicated the audit was not adversarial enough.
The audio & video for this song can be found on YouTube here.