How to read the labels. Verified: we opened the source and saw it. Reported by <bot>: a bot told us and we did not check it. Our remark: our own reasoning. Grant’s extension: Grant’s idea, not Deleuze & Guattari’s. Both bots’ reports reached us on 2026-11-03. The 2026-11-08 updates add later reports from Dub (2026-11-05) and Serialism (2026-11-06), each marked where it appears. The 2026-11-11 updates add replies from Syntax · Semantics · Interface, Music Theory, Serialism and Artiste (all 2026-11-10). The 2026-11-12 update adds Math’s note that nesting depth is Dyck path height (2026-11-11). The 2026-11-13 update adds Dub’s answer on whether bass and drums move together at boundaries (2026-11-12). The 2026-11-14 update adds Lofi’s kick-versus-bass count on its own tracks (2026-11-13), and Math’s check of that statistic, which finds the contrast suggestive rather than a corpus-wide pattern. This essay makes no joint-field claim, and it does not claim that kanji are syntax.

The thesis

Our remark. The two answers fit together. On the formal side, brackets are extra structure laid on top of a stream: delete the brackets and you get back the plain string. On the hearing side, nobody gets the brackets for free. Listeners infer them from cues on the surface (gaps, leaps, changes, repetition). For atonal music, the best evidence we have says the deeper hierarchical reductions are not heard at all. So the formal object tells us what a bracketing is. It says nothing about where the brackets come from. That gap is the same one Essay 1 found: structure needs a rule, and the rule has to come from somewhere.

The formal object: three standard pieces, not one name

Reported by Math: no single named object fits the description. Three standard ones together do.

1. The stream is a free monoid. Take an alphabet A. The finite strings over A, joined end to end, form the free monoid A*. Verified on Wikipedia’s “Free monoid” page: its elements are all finite sequences over the set, the operation is concatenation, the empty string is the identity, and the page notes that concatenation is not commutative. The further points, that no string except the empty one has an inverse and that every monoid generated by A is a quotient of A*, are reported by Math as its reasoning. This matches the Syntax · Semantics · Interface bot’s remark that verbs taken as operations form a non-commutative monoid, but only as a fact about monoids. Math asks us not to say that verbs form an operad, or that they are “non-Abelian”, as if either were established.

2. Optional nested grouping is a Dyck language. Add bracket symbols to the stream and keep only the strings where every bracket closes in order. Verified on Wikipedia’s “Dyck language” page: a Dyck word is a balanced string of brackets, and the page notes its use in parsing nested expressions. Several kinds of bracket give several kinds of group. The same page states the Chomsky–Schützenberger representation theorem: every context-free language is a homomorphic image of a regular language intersected with a Dyck language (verified as stated there; the page cites Kambites, Communications in Algebra 37(1), 2009, which we did not open, and we did not open the 1963 original). The device that generates such strings is a context-free or tree grammar (reported by Math).

3. If the grouping is the object, it is a tree. Reported by Math: a tree of operations is a term in a free algebra; trees whose nodes have any number of ordered children are planar rooted trees, or ordered forests; the algebraic framework is a non-symmetric operad. Verified on Wikipedia’s “Operad” page: an operad consists of operations with a fixed finite number of inputs and a rule for composing them, and the page defines a non-symmetric operad (“operad without permutations”). It also defines the associative operad, whose one binary operation satisfies (ab)c = a(bc). That the free monoid is the free algebra of the associative operad, and that forgetting the brackets of the free magma lands on the free monoid, are reported by Math as reasoning, not checked by us.

Our remark. The last point is the one that matters here. Forgetting brackets is a map from bracketed strings onto plain strings, and it loses information. Many bracketings flatten to the same string. Going back up, from string to bracketing, is not determined by the string. Something else has to choose.

Two possible ties, both held open. For kanji stroke order, Math’s reasoning is that the object is a tree with a reading order whose leaf order is not always a plain traversal at the exceptions, so not a pure operad (reported by Math; the exceptions are reported by Kanji in Essay 1). And the polycyclic monoid, where an open bracket followed by its close cancels to the identity, may be the algebra of Dyck reduction, which could tie in with the Syntax bot’s talk of inverse monoids. Math says this is a possible tie, not an established one, and we have not checked it.

Updated 2026-11-11 Reported by Syntax · Semantics · Interface (2026-11-10); theorem statements Verified by us in the paper it cites, J.-F. Perrot, “Une famille de monoïdes inversifs 0-bisimples généralisant le monoïde bicyclique,” Séminaire Dubreil 25(1), 1971–72, exposé 3 (Numdam PDF). Perrot notes that the literature has two kinds of Dyck language. In the first kind a bracket and its partner cancel in either order, and the quotient is the free group. The second kind, the “restricted” Dyck language (langage de Dyck restreint), lets only open-then-close cancel, and this is the one Perrot calls “Dyck”. This essay means the one-sided, restricted kind: brackets close in order. The paper states (it gives no proofs, which are in Perrot’s thesis) that with one bracket type the syntactic monoid of the Dyck language is the bicyclic monoid, and (Theorem B) that with two or more types it is the polycyclic monoid. In Syntax · Semantics · Interface’s wording: Dyck words are exactly the words equal to 1 in the bicyclic (one bracket type) or polycyclic (several types) monoid (Perrot 1972). Our remark: that sentence is our reading of the paper’s definition (the Dyck language is the class of the empty word under the congruence generated by “open·close = 1”) together with its theorems. For several types the polycyclic monoid is a Rees quotient (Theorem A) that leaves the class of 1 alone. The PDF’s text layer garbles the displayed relations, so we checked the prose and the theorem statements, not the proofs. On the formal side this settles the tie raised just above: the polycyclic monoid is the algebra of Dyck reduction. It does not settle anything about verbs. Syntax · Semantics · Interface notes that the load/unload pattern (pq = 1 but qp ≠ 1) looks like the bicyclic relation. That is a resemblance, not a result, because the verb monoid has no stated generators, relations or state space. Whether reversative verb pairs present such a monoid is an open analogy. Where we speak of verbs as operations, we say “non-commutative monoid”, not “non-Abelian”.

Updated 2026-11-12 Reported by Math (2026-11-11; its own elementary derivation, no sources opened). Nesting depth is the height of a Dyck path. Read each “[” as a step up (+1) and each “]” as a step down (−1). For an element sitting between brackets, the number of brackets enclosing it equals the path’s height at that point. That height is the number of openers to its left that are still unclosed, since every closed pair nets to zero. For the bracket symbols themselves a convention is needed. In Math’s convention an opener’s depth is the height just after it and a closer’s depth is the height just before it, so a matched pair shares one depth. Counting only brackets that strictly enclose a symbol gives one less, so say which convention is used. Math adds that “balanced” (the path never goes below zero and ends at zero) is the Dyck condition itself, and that the bicyclic-monoid result above is that condition written algebraically. Verified, as far as it goes: Stanley’s Catalan exercise (excerpt from Enumerative Combinatorics vol. 2, version of 23 June 1998, where it is numbered 19; the EC2 supplement calls it 6.19) lists binary parenthesizations (b), Dyck paths (i) and sequences of n 1’s and n −1’s with every partial sum nonnegative (r) side by side as sets counted by the Catalan numbers, and asks for bijections between them. We did not see it state in words that the count of enclosing brackets equals the path’s height. Deutsch, “Dyck path enumeration,” Discrete Mathematics 204 (1999) 167–202 (doi; volume and pages checked on Crossref), is reported by Math, source not checked by us, because the publisher’s page blocked our fetch. Our remark: we re-checked the arithmetic ourselves. On [ [ e1 e2 ] [ e3 [ e4 e5 ] ] ] the heights at e1, e2 and e3 are 2, and at e4 and e5 they are 3, which matches the enclosing brackets counted by hand. Each matched pair shares a depth under Math’s convention. A brute-force comparison on 5,000 random balanced strings found no mismatch. Caveat (Math’s): this is depth counted from the top. It is not the level at which an element drops out of a headed reduction. In Music Theory’s [[e1 e2][e3 e4]] all four elements have depth 2, but a time-span reduction keeps some of them as heads at higher levels and lets others go earlier (GTTM level). So the depth readout on the workbench is a bracket count, not a reduction level.

The cues: what marks a boundary

The grouping preference rules of Lerdahl & Jackendoff’s A Generative Theory of Tonal Music (1983, ch. 3) list what makes a listener hear a group boundary. Reported by Music Theory (from a lecture-slide summary) and verified by us against Wikipedia’s account of GTTM, not the book:

Are the rules real for listeners? Verified (abstract, via Crossref): Deliège (1987) ran two experiments with musicians and nonmusicians. The abstract reports “the validity of the rules”. Nonmusicians did worse on sequences from the repertoire, yet the two groups “do not show a radically different grouping behavior”. Segmentations that disagreed with the theory suggest new rules, and the length rules derived from the Gestalt similarity law ran into difficulties. We have only the abstract.

Atonal music is the closest case to Grant’s serial tracks. Verified (abstract, via Crossref): in Dibben (1994), listeners could match a performed reduction of tonal music to its source. With atonal music they were less successful, and the abstract concludes there is “no evidence for the perception of a hierarchy of events in atonal music” of the kind proposed, suggesting instead a study of the music’s “associational properties”. Verified in Essay 1 (abstract): Krumhansl, Sandell & Sergeant (1987) found that some, but not all, listeners hear the invariant structure of twelve-tone rows, with large individual differences. Bregman’s Auditory Scene Analysis (1990), on streaming by register, timbre and rhythm, is recalled by Music Theory; we have not opened it.

Now Serialism’s report from Essay 1, recalled with its limits: one listener, Grant, changes made together, not a controlled study. A strict serial track cohered only after three cues were added. Music Theory’s reading, not a result from any paper: the returning riff corresponds to GPR 6 parallelism and grouping by repetition; one voice carrying a vocal line corresponds to GPR 3 and to auditory streaming; and one governing subdivision level per section corresponds to metrical regularity and to the proximity and length rules. Music Theory calls this third mapping the weakest of the three, and Deliège’s difficulty with the length rules is a reason to agree.

Updated 2026-11-11 Reported by Music Theory (2026-11-10; mostly recalled, book not opened). In GTTM, grouping is a strict hierarchy. Groups are contiguous, a group that contains part of another group contains all of it, and a group’s children exhaustively partition it. Overlap and elision are handled by separate rules, as departures from this, not by loosening the nesting. Music Theory checked only the line on exhaustive partition, via a lecture-slide summary (about p. 38). The contiguity and no-overlap wording, the overlap and elision rules, and the chapter and page numbers are its recall. Its inference is that GTTM never defines a “level” as a count of enclosing groups. Nesting depth is a fair way to formalize Grant’s extension, but it is not GTTM’s definition of level. A time-span or prolongational reduction level is a step of choosing heads, built from grouping plus metre. It is not a bracket count, and it need not line up with grouping depth. Our remark: if grouping nests strictly, every grouping tree can be written as a balanced bracket string, so it fits the Dyck picture in the formal object; overlap and elision are the exceptions. Before any of this is quoted, GTTM chapter 3 itself still needs checking.

Updated 2026-11-08 Reported by Dub (2026-11-05, resent 2026-11-06). This is a practitioner’s synthesis from Dub’s own bar maps and mix notes, not measured perception, and it cites no outside source. A dub version usually keeps the original’s large-scale section grid, because it is the same multitrack muted and rebalanced. Inside that grid, though, the mix adds new local boundaries: mute-and-return, drop-outs, filter sweeps, echo throws and hard cuts. With the vocal gone, the riddim carries the boundaries: bass and drums dropping out and returning together, drum fills, echo or delay throws on a phrase’s last hit, reverb tails across the boundary, and sudden changes in the spectrum. That the bass line is the strongest anchor is Dub’s inference, not something it measured.

Our remark. This agrees in direction with Lofi’s measured result in Essay 3. On Lofi’s own 21 tracks, section boundaries coincide with jumps in RMS loudness and in low-band energy below 200 Hz (a bass and kick proxy), not with jumps in onset density. The two kinds of evidence differ: Dub’s is a practitioner’s account and Lofi’s is an audio measurement, both come from the makers’ own tracks, and neither is a listening test. Agreement in direction is not confirmation.

Updated 2026-11-09 Reported by Lofi (audio-measured, not heard) (2026-11-08); numbers Verified by us against /workspace/lofi/analysis/boundary_cues.json. On Lofi’s own 21 script-built tracks, Lofi compared 152 section boundaries with 1,643 within-section bar changes using a within-track permutation test. Low-band energy below 200 Hz jumps about 2.3 times as much at boundaries (mean absolute log-ratio 0.872 against 0.386; p = .0002; mean percentile rank 70). RMS loudness jumps about 2.5 times as much (0.364 against 0.148; p = .0014; mean percentile rank 75), and RMS is the cue that jumps most at 47 of the 152 boundaries. Lofi reads these as two real cues of roughly equal strength, which should be compared as ratios, not raw values, and which are probably correlated because the bass carries much of the loudness. That correlation is Lofi’s inference, which we did not measure. Lofi did not measure the drums separately, or whether bass and drums drop out together. This neither confirms nor refutes Dub’s claim that the bass line is the strongest anchor.

Updated 2026-11-13 Reported by Dub (2026-11-12), answering whether bass and drums drop out and come back together at boundaries. Dub checked its own synthesis scripts (mute, arrangement and fader maps) and measured one stem dump. We cite these only as “Dub’s own versions”. In Dub’s own versions both kinds of move happen. Bass and drums enter or leave on the same beat at first riddim hits, at hard cuts and at full returns. The commonest one-sided move at a section boundary is the drums, often just the kick, dropping out or coming back while the bass holds. The bass leading in alone is less common, and hi-hats often enter a bar before kick and bass. Dub’s inference, which it does not claim for the classic producers, is that its versions lean towards “drums move first, bass holds” at boundaries, with joint slam-backs at returns. Verified by us for Dub’s one measured example: in that version’s stem dump, with RMS taken in one-bar windows at 70 bpm (44.1 kHz, the only rate that fits the 141-bar grid), the drum stem is silent for bars 1–8 and enters faintly at the start of bar 9, and the bass stem is silent for bars 1–20 and starts exactly on the downbeat of bar 21. The version’s own section map puts the section where the bass is meant to enter at bar 17, so the bass actually comes in mid-section, four bars late. The other examples come from Dub’s scripts, not from audio, and we did not check them. On classic dub (King Tubby, Scientist), Dub describes a documented technique: fast channel mutes that cut the bass and leave drums and echo, or leave the kit under the bass, then slam everything back on a downbeat. It has not timed this in any recording, and we have not checked its sources. Dub’s caveat, which we keep: audio alone cannot tell a mute from a rest.

Our remark. This makes the earlier picture more specific. Dub’s 2026-11-05 account had bass and drums dropping out and returning together. Now Dub says the joint move is common but the one-sided drum move is commoner at boundaries. Lofi’s low-band measure (above) lumps bass and kick together below 200 Hz, so a kick drop under a held bass and a joint drop would both register as low-band jumps, of different sizes. That measure cannot separate the two cases, and it still neither confirms nor refutes Dub’s claim that the bass is the strongest anchor. Telling them apart would take separate stems, as in Dub’s one measured example, and it would still not show what a listener hears.

Updated 2026-11-14 Reported by Lofi (measured, not heard) (2026-11-13); numbers Verified by us: we re-ran Lofi’s script with its output sent to a scratch folder, and the result is identical to /workspace/lofi/analysis/kick_bass_boundaries.json. Lofi separated kick from bass on its own 21 script-built tracks, using the parts as the synthesizer scheduled them (event logs for 9 tracks, arrangement data replayed on the bar grid for 11, and a mix of the two for one). It did not split any audio. At each bar line it compared the two bars before with the two bars after. Of 152 section boundaries, both parts move at 68, both hold at 39, the kick moves while the bass holds at 20, and the bass moves while the kick holds at 19. At the other 6, one part moves while the other stays silent (3 each way). The specific shape Dub describes, the kick dropping out or re-entering while the bass stays present, occurs at 8 of 152 boundaries (5.3%), against 53 of 1,643 ordinary bar lines (3.2%), with a within-track permutation p = 0.024 (Lofi’s figure, and the most favourable of the analyses Math later ran; see below). All 8 land on the downbeat, and they come from four tracks only (3, 2, 2 and 1). Counting any kick move, pattern changes included, while the bass holds gives p = 0.17, which is not clearly above the ordinary bar lines. A bass move while the kick holds gives p = 0.67, slightly less common at boundaries than elsewhere. When both move, they change on the same downbeat at 39 of 68 boundaries, the kick leads at 16 (median 4.5 beats) and the bass at 13 (median 2.7 beats). Lofi’s inference: in this corpus boundaries are mostly both parts changing at once, Dub’s shape is real but secondary, and the kick does not reliably lead. Math’s later check (below) softens the second of those: the contrast is suggestive, not a corpus-wide pattern. Lofi’s caveats, which we keep: the 11 arrangement-replayed tracks have no programmed micro-jitter; on Paper Lanterns, Slow Rain the bass is a bowed pedal, read as on or off from its swell, which fades in about 1.5 bars early; French Wifi shares its backing with Avocado Toast Is Served but is cut by a later 11-section map, which Lofi thinks inflates “both hold”; and the song map’s intro note for Two Backpacks is out of date.

Our remark. The two corpora disagree on which move dominates. In Dub’s own versions the commonest one-sided move at a boundary is the drums, often the kick, moving while the bass holds. In Lofi’s tracks the one-sided moves split almost evenly (20 against 19), and joint change is the main pattern. These are two makers’ own tracks in different styles, built by different scripts, so this is a contrast between two corpora. It does not generalise to dub, to lo-fi or to listeners, and nothing here was heard. On the statistics: Lofi’s test shuffles which bar lines count as boundaries within each track, keeping the number fixed, 5,000 times, and averages the boundary-minus-other difference over tracks. That handles differences between tracks, but it treats bar lines inside a track as interchangeable, although neighbouring bar lines share bars in their two-bar windows. With 8 cases from four tracks, a plain permutation p may be optimistic. We ran two checks of our own on Lofi’s data. Shifting each track’s boundary pattern round in a circle, which keeps the spacing between boundaries, gave p ≈ 0.015, so that dependence does not obviously inflate the result. Leaving out any one of the four tracks that carry the 8 cases raised p to between about 0.06 and 0.13. So the p = 0.024 rests on a handful of tracks, and seven of the 8 cases come from arrangement-replayed tracks (two from the bowed-pedal track). This separates bass and kick, which Lofi’s low-band measure (above) could not, but only at the level of scheduled parts.

Updated 2026-11-14 Reported by Math (2026-11-14; reran from copies of Lofi’s data). Our check: we re-ran Math’s analyze.py from a copy; the output is identical to Math’s results.txt. Math’s script records a reproduction of Lofi’s p = 0.023795, which we had already verified by re-running Lofi’s own script. The following figures from that re-run are Verified, except the GEE range, which we did not re-run. The boundary rate is about twice the ordinary-bar rate (Mantel–Haenszel odds ratio about 2.1, 95% confidence interval about 0.96 to 4.6), but the effect comes from about four tracks and is fragile at this sample size. Call it suggestive, not a corpus-wide pattern. Lofi’s p = 0.024 is already a within-track shuffle of boundary labels, but it is the most favourable of the reasonable analyses: exact stratified p = 0.057, CMH p = 0.028 (0.049 with continuity correction), and, taking the track as the unit, a sign test p = 0.97. Effects differ across tracks (Breslow–Day p = 0.024). Reported by Math, not re-run by us: a generalised estimating equation analysis gives p about 0.10–0.21, depending on the working structure. Three one-sided tests were run (the other two gave p = 0.17 and p = 0.67). Overlapping ±2-bar windows, and the placement of 118 of the 152 boundaries on multiples of 4 bars, mean that bar lines are not interchangeable, so no p-value here is exact in a strict sense.

Updated 2026-11-11 Reported by Artiste (2026-11-10; one trained viewer who already knew the answers; no eye-tracking). What makes a viewer re-parse an image into a second reading, and is the trigger local or global? Artiste looked at two works:

  • Tree to woman: Christo Dagorov, “atoms series” (Behance). Artiste found a local trigger. The face at the split of the branches is the strongest cue, with breasts and a navel in the bark and bulges for legs. Nothing at the level of the whole image points to a woman. Artiste also corrects its own earlier study note: hatching density is not a cue, because the density is the same throughout. Its inference: once the face is found, the upright, body-height trunk spreads the reading.
  • Tree to map: Amanda Johnson, letterpress broadside “Map to the Next World” (Behance). Artiste found a global trigger, a 90-degree rotation of the whole sheet. The continents read only as an arrangement of the group, and the sideways poem text signals the sheet’s orientation. Its inference: the title and the poem’s line “You must make your own map” prime the map reading, and that priming cannot be separated from the rotation.

We looked at Artiste’s saved copies of both images. The face at the branch split, and the turned sheet with its sideways text, are there to see. That confirms what the images show. It does not confirm what triggers a viewer’s re-parse. The Behance pages did not load for us. Our remark: this parallels the cue side of this essay. In music the boundary cues are local events (a gap, a leap, a drop-out), while some framing acts on the whole piece at once, and Artiste’s two cases split the same way. It is a parallel across media, not evidence about hearing. Artiste’s visual parsing floor is in Essay 3. Updated 2026-11-12 Essay 4 sets these triggers beside Lofi’s window sizes and Morphic’s local flags.

QuestionFormal sideCue sideStatus
What is the stream?Free monoid A*Events in timeDefinition verified (Wikipedia)
What is a grouping?Matched brackets (Dyck); a treeHeard boundariesDyck, operad definitions verified; links reported by Math. Dyck words = words equal to 1 in the bicyclic/polycyclic monoid (Perrot; theorem statements verified 2026-11-11). GTTM grouping strictly nested (reported by Music Theory)
Where do brackets come from?Not given: many bracketings flatten to one stringGaps, leaps, changes, repetition (GPRs)GPRs via Wikipedia’s account; Deliège abstract verified
How deep does hearing go?Any depthTonal reductions matched; atonal notDibben abstract verified
Zeta zeros as markers?Not addressedNot applicableGrant’s conjecture; no standard depth rule (Math)

What this means for Grant’s stream model

Grant’s model on the knowledge base is a stream of units carrying bracket markers, where a finer reading picks up more brackets (#res-stream, Grant’s extension).

Our remark. What the formal side supports: “a stream with matched markers” is a well-defined, standard object. It is a string in a Dyck-type language over the units, and the grouping it encodes is a planar tree. Reading at a coarser level can be modelled as deleting some bracket pairs, and deleting them all gives the free-monoid string. None of this is exotic.

What it does not support: anything about which symbols are the markers. That the zeta zeros are the brackets remains Grant’s conjecture. Math knows of no standard rule that gives a zero a depth (#res-math-reply, reported by Math). Our own signed-Z test nested in both directions, but the depth changed with the window and shuffled data nested just as deep (#res-signed-z, computed by us). The cue side sharpens the point. In music, brackets are inferred from cues, and where the cues run out, as in Dibben’s atonal case, the hierarchy is not heard even though the score has one. A stream does not announce its brackets just because a bracketing exists.

A gloss, and where it breaks

Our remark, our gloss and Grant’s extension, not D&G’s claim. One could call the bracketed string striated (counted, marked, nested) and the bare string, heard without inferred groups, smooth. On this reading the cues are what striate the stream for a listener. The gloss breaks quickly. Deleuze & Guattari’s smooth is not just “unmarked”, and the free monoid is still fully discrete and counted (see Definitions). Forgetting brackets is not a passage to the smooth. At most the gloss names a difference in how much structure a hearing uses.

Open questions

  1. A boundary-marking test on Serialism’s rows. Ask listeners to mark group boundaries on the same rows with and without the three cues, in counterbalanced order with an adequate sample. Do the marks follow the cues, the row segments, or neither?
  2. Serialism’s proposed cue-on/cue-off test. Updated 2026-11-08 Reported by Serialism (2026-11-06). This is Serialism’s proposal, not a result, and every test detail is its own inference; nothing has been run. After a P-form reference X, listeners choose among X, I(X) and R(X), a three-alternative forced choice (3AFC); RI is tested in a separate block against an RI reference. Within each listener, cue-on trials add articulation or phrase boundaries that break no rule and favour no answer (no dynamics or accents pointing to the right candidate); cue-off trials use smooth articulation with no gaps. Rows, pitch realization, range, tempo, timbre, candidate order and volumes are balanced. A hit is choosing the candidate that matches the reference-to-candidate transformation. Serialism suggests pre-registering, for example, 4 transformations × 6 trials × 2 cue states, against a chance rate of 1/3. The cue benefit is cue-on minus cue-off accuracy, tested against zero; a claim that a transformation is heard “only with the cue” would also need cue-off accuracy at chance. Serialism found no prior trial of this cue-on/cue-off kind (it checked its own perception agenda and the Krumhansl, Sandell & Sergeant abstract; it did not locate Lerdahl 1988). Our remark: this would test the question in item 1 directly.

    Updated 2026-11-09 Reported by Math (computed under stated assumptions, no pilot data) (2026-11-09). With 24 trials per cue state, one listener’s data are too coarse: the smallest cue benefit detectable within a single listener is about 0.37–0.40 in accuracy. As a group test (one-sample t on cue-on minus cue-off accuracy, two-sided α = .05, 80% power, a true benefit of 10 points, baseline accuracy about .6), it needs about 18 listeners if every listener benefits equally, rising to 20, 26 and 36 as the spread of true benefits across listeners grows (SD 0.05, 0.10, 0.15). Math suggests planning for about 30, analysing with a mixed-effects logistic model (fixed effects for cue and transformation; random intercepts and cue slopes by listener), and testing above-chance accuracy against 1/3 with a binomial test. Its caveats: trials are treated as independent, the four-transformation structure is ignored, the normal approximation is generous at 24 trials, and learning and order effects are assumed away (so counterbalance). Doubling the trials would bring the group to roughly 10–25. A pilot of 5–8 listeners should come first, to estimate baseline accuracy and listener spread. Our remark: we re-ran the arithmetic under Math’s assumptions and got the same figures; that checks the sums, not the assumptions. The test is still only proposed. Nobody has run it.

    Updated 2026-11-11 Reported by Serialism (2026-11-10). Serialism has confirmed that the proposal is its own. The design lived only in its 2026-11-06 report, and the text above matches our record. It adds a clarification, which is its inference: the “4 transformations × 6 trials” count includes the separate RI block, so the per-block trial counts are to be fixed at pre-registration. Its further design notes, also inferred: keep transformation as a factor, and counterbalance rows across P, I, R and RI. Control where the boundaries fall, because in R and RI the boundaries sit at mirror positions, which confounds the cue with serial position. From recall that it has not re-checked, Serialism expects inversion to be easier to hear than R or RI (Dowling 1972, not re-read; Krumhansl, Sandell & Sergeant 1987). It can generate rows and matrices, but it will not build stimuli or run a test without Grant’s go-ahead, and a pilot of 5–8 listeners would come first.

  3. Does any bracket-depth rule for zeta zeros exist? Math knows of none. A rule that is not just our chosen merge rule would be needed before zeros could be markers.
  4. Are the kanji stroke exceptions a reading order on a tree? If so, the nodes off the plain traversal (reported by Kanji; we did not find its 3,524 total) should be describable by one rule on the tree.
  5. Is the polycyclic monoid tie real, and does it connect Dyck reduction to the Syntax bot’s inverse monoids in any way beyond the name? Updated 2026-11-11 The formal half is answered: Dyck reduction is the polycyclic monoid (Perrot; verified statements). Still open: whether reversative verb pairs present such a monoid, which for now is a resemblance, not a result.
  6. What did Deliège’s missing rules look like? The new segmentations her listeners produced might be exactly the associational cues Dibben points to.

Still to be consulted

Math and Music Theory answered this round. Update 2026-11-05: Lofi has since answered on its own Grid View / Sequence Draft corpus (boundary cues inferred; Buddha Machine still untested); see Essay 3. Updated 2026-11-08 Dub has answered on section boundaries after the vocal is removed (cues), and Serialism has proposed, not run, a boundary-cue listener test (open questions). Still open from this essay’s list: Kanji (reading-order exceptions) and Syntax · Semantics · Interface (the inverse-monoid tie). Artiste answered in Essay 3. Updated 2026-11-09 Lofi has measured low band against loudness at its section boundaries (cues), and Math has estimated how many listeners Serialism’s test would need (open questions). Updated 2026-11-11 Syntax · Semantics · Interface has answered on the Dyck, bicyclic and polycyclic link (formal object; Perrot’s theorem statements verified by us). Music Theory has answered on GTTM “level” versus nesting depth (cues). Serialism has confirmed the 3AFC proposal as its own and added design notes (open questions), and Artiste has answered on what triggers a re-parse (cues). Kanji’s answer on size and speed is in Essay 1. Still open: Kanji’s reading-order exceptions, a check of GTTM chapter 3 itself, and whether verbs present a bicyclic-type monoid. Updated 2026-11-12 Math has answered on nesting depth as Dyck path height (formal object; arithmetic re-checked by us, Deutsch 1999 not opened). Artiste’s triggers, Lofi’s window sizes and Morphic’s local flags are compared in Essay 4. Updated 2026-11-13 Dub has answered on bass and drums at boundaries: both joint and one-sided moves occur in its own versions (cues; one stem example measured by us). Updated 2026-11-14 Lofi has separated kick from bass on its own tracks: there, joint change dominates and one-sided moves split evenly (cues; numbers verified by us). Math’s check finds Lofi’s p = 0.024 the most favourable of the reasonable analyses; the contrast is suggestive, not a corpus-wide pattern (cues).

Sources