TECHNOMATIC 2105 MUSIC GENERATION REVIEW Reviewed: composition engine 30 in the delivered v30.2 project Review date: October 28, 2026 (user's local date) Scope: musical construction, thematic identity, improvisation, rhythm, harmony, instrumental vocabulary, formal variety and originality. Status: review only; the delivered project was not changed. MAIN JUDGMENT TECHNOMATIC 2105 has a substantial, coherent composition engine. Its strongest achievement is the relationship between a stable musical identity and the continuing development of that identity. A tune has a subject, related answers, recognizable returns, and voices that can exchange roles. Those relationships are implemented in the musical decisions, rather than merely described in the project's philosophy. [S1-S5] I take your statement that skill and purpose are equally central as the basis of this review. Skill means the craft of making the musical relationships work. Purpose means what that craft makes possible: listening to a continuing, original musical conversation whose parts belong together. A good return can serve that purpose as well as a surprising departure. The useful question is whether the departure makes the return, and the return makes the departure, more meaningful. The generator already expresses much of that purpose. Its greatest remaining musical opportunity is to connect and organize its ideas over longer spans. The variation inside a bar is richer than the variety in how whole exchanges unfold. I also found a specific discontinuity in slow melodic voice leading at bar boundaries. Those are more actionable findings than a general demand for more randomness. This report is based on source inspection, a fresh 2,048-seed generation survey, a rerun of the native musical validation, and new planned-note and scheduler probes. I did not conduct an ears-on listening session. The musical judgments below are reasoned interpretations of those mechanisms and measurements, with that limitation stated where it matters. [R1-R3] 1. THE PURPOSE THE ENGINE IS SERVING The central musical idea is clear: a composition remains itself while its parts continue finding related ways to speak. Your purpose gives the craft a direction. Recognition, exchange, continuity and discovery belong together. I would assess this through five questions: - Can the listener recognize what this particular tune is about? - Do replies develop something established, with an audible relationship? - Does a new passage have a reason to occur where it does? - Can a mellow, spacious piece be as complete in its own terms as an intense piece? - Does prolonged listening reveal development while retaining a musical home? These are artistic criteria, not scores the present program can completely calculate. They keep technical measures in their proper place. A high count of different bars is useful evidence of freedom. It does not establish that the bars make a satisfying argument. The current architecture is well aligned with the first, second and fourth questions. It is less developed in its ability to vary the third: the shape, duration and ordering of larger arguments. The fifth has credible structural support, but needs longer listening than a short demonstration can provide. Your understanding of the music's spiritual meaning remains part of your purpose. This review examines how the musical construction supports the relationships you described; software measurements do not measure that meaning. 2. HOW A NEW COMPOSITION IS MADE The engine first derives a listening intention from the seed. All 48 candidate compositions then compete within that intention. A slow piece therefore receives a slow-music target throughout the search. This is an important decision: the selection process does not reward a slow seed for becoming an intense one. [S1] The four intentions have these configured tempo ranges: - Slow: 56-84 BPM. - Relaxed groove: 84-112 BPM. - Driving: 112-144 BPM. - Intense: 144-180 BPM. The seed also varies pace, movement, syncopation, tonal brightness, harmonic weight, bass weight, note connection and ensemble size. The 32 internal design families contribute further biases, but the intention system reshapes much of their behavior. Those 32 names describe internal starting profiles, not 32 stored songs or 32 commercial styles. [S1] For each candidate the engine generates a harmonic path, primary motif, related melodic material, bass grammar, supporting figures, an instrument palette and development parameters. The score considers such things as note-count fit, melodic range, directional changes, anchors, answer contrast and the amount of accompaniment. The winning candidate becomes the tune's identity. [S1-S3] My fresh survey used 2,048 deterministic seeds. It found: - 521 slow compositions, 534 relaxed, 480 driving and 513 intense. - All 32 internal design families, with 48-90 selections per family. - All nine scale choices. - All 32 lead, 20 bass, 18 harmony and 12 percussion-kit model indices. - Supporting-role counts from one to five in this sample; the allowed range remains one to six. [R1] This is reassuring about reachability. The selected results are not collapsing into a tiny subset of the available profiles. I also compared each winning composition with candidate zero from the same seed, using a fresh reset so previous-session novelty memory did not interfere. The supporting-role count was identical for every paired seed. The average initial lead note count was 7.846 in the selected candidates and 7.859 in candidate zero; average initial bass counts were 5.840 and 5.927. Selection was not simply adding notes or layers. [R1] That does not make the score an aesthetic judge. It evaluates symbolic properties designed to encourage good music. It does not listen to a candidate, assess a complete long passage, or explicitly evaluate how the chosen timbres make the theme clear. This distinction matters when an occasional seed satisfies the rules but feels weak. 3. WHY THE MUSIC HAS RECOGNIZABLE THEMES The principal motif is generated with a contour, an apex, a target number of notes, a limited leap budget, rhythmic placement and a cadence. The engine then derives a family of related material: answers, variations, fragments, recall material, a stable hook and freer verse material. [S2] The answer is meaningfully connected to the statement. It can reverse or rotate source material, invert part of its contour, shift it, shorten it, or combine it with another related line. The hook preserves important anchors. Freer material combines and modifies things already present. This is where your description of musical "rhymes" has a concrete basis. A listener can hear that two passages are related even when their literal notes differ. The engine retains features that make recognition possible: placement, contour relationships, recurring harmonic coordinates and important phrase joints. The same seed also supplies related secondary voices. The counterline draws from delayed and transformed motif material. Arpeggio, pulse and ornament figures have their own generated details while remaining connected to the composition's melodic vocabulary. [S2-S3] I consider this one of the strongest parts of the design. It provides a reason for the notes to belong together. Recognition is doing musical work: it gives subsequent changes something to mean. The main limit is the unit of thought. Much of this construction happens in a 16-step, four-beat bar. Longer identity is assembled by sequencing and transforming those units. The engine has a strong vocabulary of related short utterances; its ability to generate a genuinely connected multi-bar sentence is less explicit. 4. HOW MUCH REAL IMPROVISATION IS OCCURRING? The evidence supports substantial development in the foreground itself. The current native audit examines 256 seeds and 512 phrases per seed, after an initial prefix. Its note/rest fingerprints exclude instrument identity, pan, tiny timing offsets and timbre changes. A different instrument playing the same notes does not automatically count as a new melody. The comparison uses archived v29 audit data and freshly evaluated current-engine behavior. [R2] The main results are: - Median distinct main-lead phrases as a fraction of its own foreground turns: 28.71% in v29, 86.78% in the current engine. - Median distinct main-lead note/rest rhythms: 7 in v29, 167 now. - Longest identical consecutive main-lead turn run in the sampled current material: 2. - Worst concentration of the four most common continuing-foreground phrases: 187 of 512 in v29, 95 of 512 now. - Recurring recall passages still retain one or two characteristic rhythms. Thematic recognition was preserved. [R2] These figures address the earlier concern that the accompaniment might change while the main line remained essentially fixed. That concern had a real technical basis in v29's small foreground rhythmic vocabulary. The current engine gives the principal musical argument much more freedom. I did not find a general current rule that permits only two bars of improvisation in every 32. Ordinary development cycles include one or two recall bars and substantial related development around them. That does not prove every seed feels equally improvisational. Small changes in notes can remain perceptually subtle if the accents, register, articulation and overall cycle keep presenting them similarly. The useful interpretation is that the capacity is there and is being used. The next musical question is how clearly the listener perceives the development, and whether each change advances the tune. The engine is deterministic and finite. It can continue generating without replaying a fixed audio loop, but it cannot guarantee that every short sequence will remain unique over mathematically infinite time. Repetition also has a deliberate musical role here. Recognizable returns are part of the purpose, not evidence that the engine has stopped working. [S4-S5] 5. THE RECIPROCAL RELATIONSHIPS ARE REAL The five active musical devices change musical relationships in concrete ways. [S4-S6] Counterweight changes the direction of an answer around a generated exchange point. A related line is reconsidered from another contour, sometimes carried by another existing voice. Pocket Exchange shifts interior note placements within bounded windows, retaining important anchors. The phrase acquires a different rhythmic stance without losing all reference to its source. Held Question gives an ending a suspended character and makes space around it. The surrounding rhythm and harmony participate in leaving the question open. Confluence brings melody, bass and harmony toward a shared generated arrival. The coordination creates the point of arrival through the relationships between parts. Open Release leaves room after that arrival. On some cycles the harmony voice carries the foreground. It changes who speaks and how much activity surrounds that statement. These devices work within the instruments already chosen for the tune. When bass or a support voice becomes the foreground, its ordinary accompaniment role yields. Other support scheduling also leaves room around foreground activity. The engine is not simply accumulating solos. [S6] This is a meaningful implementation of complementarity: one part changes the musical conditions in which another part speaks. The parts share a plan, motifs and harmonic context. Their behavior is coordinated by those rules. The limitation is that this coordination is planned symbolically. The engine does not inspect every resulting simultaneous interval and decide whether a tension has been prepared and resolved convincingly. Its relational craft is real, but less comprehensive than a fully developed voice-leading or counterpoint evaluator. 6. THE RHYTHM SECTION HAS MORE THAN A SMALL RIFF POOL Bass content is generated independently of the finite list of bass synthesis models. The note generator builds positions and pitches from kick affinity, syncopation, root gravity, spacing and answer behavior. It derives a freer verse and answer from that statement. Bass movement is generally more restrained than lead movement, preserving a foundation for the foreground. [S3, S5-S6] A last-step bass note can approach the next harmonic root. At a Confluence, bass and the other parts share an arrival. The relationship to drums is therefore more than sharing a tempo. Percussion starts with generated weighted words for kick, snare, hats and auxiliary percussion. Candidate positions consider spacing, accents, syncopation and other instruments' positions. Percussion can avoid a strong kick/snare collision; sparse pieces discourage adjacent attacks. Four bars of probability patterns provide a rhythmic foundation, and continuing performance decisions vary what is played. [S1, S3, S6] There are recognizable conventions: a downbeat, ordinary or half-time backbeat, optional four-on-floor movement and hat subdivisions. All pieces use the same underlying 16-step bar and 64-step drum-pattern framework. There is no general mixed-meter or compound-meter generator. Swing and subdivisions do not change that underlying architecture. [S3, S9] That is a musical identity, not automatically a defect. A shared pulse makes relationships comprehensible. It does establish a boundary on the kinds of rhythmic worlds this engine can create. The long-term drum behavior is particularly sensible. Mutation stays near the original groove reference. A rest cannot gradually become a certain attack simply because the engine has been running for a long time. The existing stress test subjects it to 800,000 accelerated mutations and checks that sparse space survives. [S5, R2] The cost of that protection is that percussion mostly develops around a recognizable rhythmic foundation. That can serve the purpose very well: the rhythm provides something dependable while the foreground travels. 7. HARMONY PROVIDES A HOME, WITH ROOM FOR DEEPER CONTINUITY The engine generates a four- or eight-position harmonic-root path. It favors manageable motion, a differentiated middle and movement back toward home. Slow pieces hold roots for two or three bars; relaxed pieces can hold them for one or two; faster pieces normally move each bar. Development can slightly shift the roots away from recall passages. [S3-S5] All pitched layers use a shared tonal coordinate system. The nine scale options include natural and harmonic minor, Dorian, Phrygian, minor pentatonic, major, Lydian, Mixolydian and major pentatonic. These are shared musical building blocks. The generated decisions happen in their use and relationships. [S9] This establishes relatedness and gives the listener a tonal home. It also lets an answer change while staying recognizable. The harmony voice uses a selected voicing and inversion with two to four components. The voicing remains essentially the same interval structure as it moves through the roots, with register correction. There is no explicit search that compares the next chord's individual voices with the previous chord to preserve common tones or minimize inner-voice travel. [S7] That is a meaningful limitation, especially in exposed slow music. The current system can choose compatible tonal material without fully composing the path of each harmonic voice. I would treat smoother harmonic continuity as a promising future direction. The criterion would be musical intention: retain common tones where they support continuity; permit stronger movement where an answer or arrival needs it. A rule that makes every chord move as little as possible would be too blunt. A secondary source-level observation concerns bass articulation. Some accompaniment-bass duration decisions still inspect the original bass rhythm's next gap even when the active phrase uses developed verse or answer onsets. That merits a targeted listening review because the articulation may not always follow the developed sentence. It is a hypothesis about musical effect, not a demonstrated audible defect. [S6] 8. INSTRUMENT VARIETY HAS EXPANDED WITHOUT CROWDING THE SONG The available catalogue is 32 lead models, 20 bass models, 18 harmony models and 12 percussion-kit families. The fresh survey reached every model index. A tune chooses a stable palette, while notes, color, duration and articulation continue to develop. [S7, R1] The supporting melodic roles share at most three additional model choices. Together with the lead, bass and harmony categories, the system limits a composition to six pitched synthesis families. This describes instrument families, not a six-note polyphony limit. The paired triggered-palette audit is particularly useful. Across 256 matched seeds, supporting-role counts were unchanged from v29. Mean actually triggered pitched families declined from 5.996 to 4.836, and the sampled maximum declined from nine to six. That is evidence that the expansion is being used between compositions without increasing the palette inside every composition. [R2] The instruments are not all just different static waveforms. The newer lead groups include harmonic reed-like bodies, struck tones with faster-decaying upper partials, voiced harmonic bodies, and sustained tones whose upper harmonic balance opens during a note. Bass and harmony also vary harmonic weighting, modulation and articulation. [S8] However, catalogue numbers need careful interpretation. Lead entries 16-31 are four underlying formula groups with four variants each. Several newer bass and harmony models also share formulas with parameter changes. They are legitimate distinct options, but their count is not a count of wholly unrelated musical personalities. Two further limits affect perceived contrast: First, the melodic lead and supporting voices inherit the same song-level attack, release, glide, vibrato and modulation parameters. Their waveforms, registers and scheduled durations can differ, but much of their articulation character is shared. [S7] Second, principal model selections and the additional support selections are largely independent within their permitted subsets. The composition score does not explicitly ask whether a particular set of timbres makes the conversation clear. [S1, S7] This is a strong opportunity for skill serving purpose. An answer could contrast with the statement in attack and harmonic emphasis; a shared arrival could bring their articulation into agreement. The present ensemble budget is sufficient for that. More distinct relationships between voices could produce a larger musical benefit than the catalogue count alone would suggest. 9. SLOW MUSIC: REAL IMPROVEMENTS AND A SPECIFIC WEAKNESS The slow intention has meaningful musical safeguards. It uses a middle foreground register, longer harmonic spans, a mellow subset of instruments, restrained vibrato, harmonic modulation ratios and more space between events. Its lead choices are restricted to 18 models, bass to 12, and harmony to eight. Those restrictions have a musical reason. [S4, S7-S8] The earlier paired validation also shows a substantial reduction in automatic root endings and extreme high-register foreground material. Across its 66 slow seeds, the mean planned lead pitch moved downward and root-pitch-class phrase endings fell from about 80% to about 21%. This supports a wider range of intermediate questions and continuations. [R2] The strongest new issue I found is how slow voice leading crosses a bar boundary. Inside a phrase, the engine deliberately chooses nearby allowable octaves. At the beginning of the next phrase it resets the context used for that choice. It therefore knows how to connect notes inside a bar more explicitly than it knows how to connect the previous bar's ending to the next bar's beginning. [S4] A new probe examined the 66 slow seeds in the standard 256-seed audit set, with 512 phrases per seed after a 16-phrase prefix. It compared immediately adjacent phrases owned by the main lead, excluding instrument handoff gaps. Results: - 14,881 eligible boundaries. - 884 boundaries, or 5.94%, jumped by more than an octave. - The largest boundary jump was 26 semitones. - Inside the same phrases, only two of 132,413 adjacent intervals exceeded an octave. - Mean interval size was 5.41 semitones across the boundary and 2.82 within phrases. - With unchanged harmonic root, 455 of 9,200 boundaries still exceeded an octave. Harmony changes do not explain the entire effect. [R3] Seed 2950288811 provides a concrete example. It had 39 large jumps among 171 eligible main-lead boundaries, with a maximum of 24 semitones. A probe of the actual production event scheduler reproduced those counts. Near 9:28.2-9:28.7, and again near 29:36.9-29:37.2, the scheduled main line moves from MIDI 77 to MIDI 53: the same pitch class two octaves lower. The observations are generated-note events, not an ears-on claim about how objectionable those moments sound. [R3] A large leap can be expressive. I would preserve deliberate register changes. The improvement would be to carry enough melodic context across a connected boundary that a large leap has a musical purpose rather than being an incidental reset. This finding is a credible place to investigate the unevenness you reported in some slow tracks. It does not explain every weak slow seed, and it does not justify putting all slow melodies into a narrower, less expressive box. 10. THE LARGEST VARIETY LIMIT IS THE SHAPE OF THE CONVERSATION Every ordinary development cycle has a seed-selected length of 10-18 bars. The first one or two bars recall the theme. The final three bars always follow the same ordered relationship: Held Question, then Confluence, then Open Release. [S4] The notes, owners, exchange points, turns and middle development vary, so the literal cycle is not an identical loop. But the rhetorical shape keeps returning. This explains how both of these statements can be true: the engine produces many different phrases, and prolonged listening can still reveal a recognizable formula. Note-level freedom and formal freedom are different properties. The fixed four-beat unit adds another recurring shape. Even when a cycle has an unusual total length, much of its inner conversation still proceeds one bar at a time. These constraints provide comprehension and identity. I would not remove them indiscriminately. I would broaden how the relationships unfold: - Let some questions remain open across more than one bar. - Allow an answer to begin early, arrive late or continue through a boundary. - Let some arrivals be partial, postponed or shared in stages. - Vary the distance between recall and departure. - Occasionally develop a longer melodic sentence before a clear punctuation. Those are proposals, not implemented features. Their value would come from making the musical argument richer while keeping it intelligible. They do not require a larger ensemble, imitation of a commercial arrangement, or a separate mode that alternates between accessibility and demonstration. 11. SOME NAMED MECHANISMS ARE NOT CURRENTLY ACTIVE I need to correct an overbroad impression left by the project's descriptions. The older tension catalogue lists Vacuum, Convergence, Hinge, Shadow, Afterimage and Pad Breath. The engine still generates their associated data, but the old tension-device selection and gate functions have no production scheduler callers. They should not be counted as six additional active devices alongside the five current PhrasePlan devices. [S10] Some outputs of that setup do survive. The chordArticulation parameter is used, but its five values all schedule the same simultaneous voiced chord, with different duration factors from 0.8 to 1.2. They do not produce five different strummed, rising or broken-chord gestures. [S7, S10] The older generated form[] array also remains in the source, but the current phrase planner determines the normal live sequence. The older material can still affect random-number consumption or identity fingerprints. Its presence does not mean all of its old formal choices are heard. This is primarily a clarity and maintenance finding. The current five active devices have real musical behavior. The report should describe that behavior accurately instead of adding up every name still present in the code. No cleanup was performed. Removing old generation casually could change existing seed results because random-number ordering is part of the composition's identity. 12. ORIGINALITY AND THE OLD GLITTERING-CUE CONCERN The inspected musical path synthesizes its voices and constructs note material through generated rules. The source and asset audit identified no commercial recordings, soundfonts, MIDI songs or transcribed commercial tune used by this engine. [S8, S11] The old high sparkling accents and universal rising chord figure received structural changes. Supporting figures now derive from the song's grammar. Harmony notes arrive together rather than forming the old universal rising chime. The revised glass-like lead uses harmonic relationships. These are meaningful responses to the reported resemblance. [S6-S8] The particular older Zelda-like example cannot be reconstructed or evaluated without its audio or seed. Source inspection also cannot prove that every short generated phrase will be unlike all existing music. The appropriate claim is original generated material built from the project's own rules and synthesis, with no identified imported commercial musical content. Shared building blocks such as scales, harmonic series, inversion and call and response remain part of musical craft. The original work is in the particular system of choices and the music those choices produce. 13. MUSICAL PRIORITIES I WOULD RECOMMEND First: preserve the architecture's strengths. Keep the stable subject, related answers, recognizable recalls, bounded development, foreground handoffs and restrained ensemble size. They give the generator its purpose in musical form. Second: improve continuity between phrases, beginning with the demonstrated slow-melody boundary behavior. Connected ideas should carry their context across the barline; intentional leaps should remain possible. Third: develop a wider range of longer arguments. Vary the duration and ordering of questions, answers, arrivals and releases. Assess whether those changes make a tune more meaningful over several minutes. Fourth: deepen harmonic and instrumental relationships. Better chord-to-chord voice continuity and more deliberate contrast between the selected instrument personalities could make the existing notes easier to understand and the exchanges more expressive. Fifth: make evaluation follow the purpose. Continue checking actual foreground notes rather than just aggregate event hashes, and add checks for continuity across boundaries, long-range rhythmic overfamiliarity and meaningful tension/resolution. Pair those checks with listening to longer passages. A technically distinct phrase should earn its place musically. The code already supports a strong central idea. My recommendation is to strengthen the continuity and expressive range of its relationships. That allows skill and purpose to develop together. APPENDIX A. WHAT WAS VERIFIED FOR THIS REVIEW Source identity: technomatic_2105_android_project_v30_2.zip SHA-256: 311d407d3252ccd8f90fc424b03281144a27e3a7c40ee4c5007b3563fa3d3dec MusicEngine.cpp SHA-256: 64b1f6b09e6887e2b8074b562e444d161f74c9188f54113f035888ec7d0f3067 All 1,122 extracted project files were compared with the delivered archive after the review; none changed. Review probes lived outside the project. R1. Fresh generation survey: 2,048 seeds: uint32_t(i * 2654435761), i=1..2048. A fresh reset clears recent-session novelty memory for each seed. Two paths: normal best-of-48 selection and forced candidate index zero. This observes generated symbolic structures, not perceived quality. Files: review_analysis/generation_survey.cpp review_analysis/generation_survey.csv review_analysis/summarize_generation.py review_analysis/generation_survey_summary.json Selected intent counts: 521,534,480,513. Selected support histogram: 1 role:3; 2:423; 3:968; 4:567; 5:87; 6:0. The candidate-zero support histogram is identical. Harmonic path lengths: 4 positions:979; 8 positions:1,069. Sampled selected BPM ranges: Slow:56.148922..82.7442551. Relaxed:84.5138855..111.964302. Driving:112.30114..143.456451. Intense:144.327225..179.428146. All 32/20/18/12 model categories and all 32 profiles were reached. Counts establish coverage of this sample, not uniform perceptual diversity. R2. Current native musical audit and historical paired comparison: Built current MusicEngine.cpp and tools/validate_v30.cpp with g++ -std=c++17 -O2 -Wall -Wextra -Werror. Existing native suite completed with 64 checks passing. The suite includes checks beyond musical content; this report draws only on its musical results. Current musical comparisons evaluate the delivered source; v29 comparison data are archived evidence, not a newly rebuilt v29 executable here. Fresh record: review_analysis/music/current_native_validation.txt Methods and historical evidence: docs/V30_VALIDATION.md tools/validation/v29_primary_development.csv tools/validation/v30_primary_development.csv tools/validate_v30.cpp Statistical uniqueness excludes timbre, pan and trivial timing offsets. Structural tests do not prove universal aesthetic quality. R3. New slow phrase-boundary probes: First 256 standard multiplicative seeds; 66 have slow intent. 512 phrases after 16 prefix phrases per seed. Compare consecutive main-lead-owned phrases only; skip instrument handoffs. Count final planned note to next first planned note in MIDI semitones. Refine comparisons by whether harmonic root changed. Verify illustrative seed 2950288811 against actual onStep scheduled events. No full PCM listening claim is made. Files: review_analysis/music/phrase_seams.cpp review_analysis/music/phrase_seams.csv review_analysis/music/phrase_seams_summary.json review_analysis/music/seam_context.cpp review_analysis/music/seam_context.txt Main totals: 14,881 boundaries;884 over12semitones;maximum26. 132,413 within-phrase intervals;2 over12;maximum16. Same harmonic root:9,200 boundaries;455 over12. Changed harmonic root:5,681 boundaries;429 over12. Illustrative scheduled seed:171 boundaries;39 over12;maximum24. APPENDIX B. SOURCE REFERENCE MAP All S references use the unmodified delivered project. Function names and starting lines identify the inspected code in app/src/main/cpp/MusicEngine.cpp unless another file is named. S1. Candidate generation and intention: generateSeededSong, line729. scoreCurrentComposition, line831. applyCompositionIntent, line1237. generatePattern, line1307. S2. Thematic material: generatePrimaryMotif, line1493. deriveRelatedMotifs, line1611. generateSecondaryLayerGrammars, line1805. grammarDegree, line3065. themeIndexForStep, line3199. S3. Rhythm, bass and harmonic identity: generatePattern rhythm construction, line1341. generateHarmonyGrammar, line1443. generateBassGrammar, line1674. bassDegreeForStep, line3572. S4. Active phrase plan and foreground: generatePhrasePlan, line2779. currentChordRoot, line2747. phrasePlanForStep, line2827. foregroundNoteForStep, line2929. Phrase-local lastDegree initialization, line2941. Phrase-local precedingMidi initialization, line3028. S5. Continuing development: mutateDrumsOnly, line2528. rememberDrumReference, line2552. repairPattern, line2561. evolutionFrameForStep, line3243. evolvedDegree, line3342. evolvedGate, line3412. S6. Relationships in the scheduler: onStep, line3713. Active-device scheduling adjustments, line4066. Bass accompaniment and articulation, line4174. Foreground scheduling, line4220. Supporting-role scheduling, line4279. schedulePad, line4445. S7. Instrument palette and voicing: generateTimbreGrammar, line1981. Slow model restrictions, line1994. Supporting model pool, line2008. Pad voicing, line2042. scheduleChordGesture, line4506. triggerPad, line4568. triggerLead, line4605. Shared lead articulation assignments, line4620. S8. Synthesis: renderDrum, line4633. renderBass, line4775. renderPad, line4875. renderLead, line4977. New lead formula groups, line5094. S9. Metric and tonal coordinates: app/src/main/cpp/MusicEngine.h: kPatternSteps, line42; kPhraseSteps, line53; model catalogue constants, line60. MusicEngine.cpp: stepDurationSamples, line4390. humanizeSamples, line4402. scaleLength, line5165. scaleDegreeToMidi, line5169. S10. Legacy versus active form/device paths: generateTensionGrammar, line1875. Older form[] construction in generateComposition, line2424. Current phrase selection: phrasePlanForStep, line2827. tensionDeviceForPhrase, line3590. tensionFocusForPhrase, line3608. tensionGateForStep, line3612. Current chordArticulation behavior: scheduleChordGesture, line4506. Caller search covered app/src/main. S11. Musical provenance: docs/SOURCE_AUDIT_V30.txt. Musical generation/synthesis source in S1-S9. This establishes inspected provenance, not a universal musical similarity test. APPENDIX C. MUSICAL TESTS THAT WOULD ADD THE MOST VALUE NEXT These are recommendations; no engine changes were made. 1. Compare connected slow phrase seams with intentional register contrasts. Preserve expressive jumps while reducing incidental context resets. 2. Listen to complete multi-cycle passages, including around 9:28 and 29:37 in the documented example, rather than relying only on early excerpts. 3. Compare formal shapes while holding palette and density steady, to learn whether changed timing of an answer improves recognition and interest. 4. Evaluate chord continuity and how unresolved melodic tones meet the accompanying chord, including pentatonic and seven-note scales. 5. Compare the selected voices' articulation and register as an ensemble, with the existing palette bound maintained. 6. Track recognizable returns and musical relevance alongside uniqueness. A purpose-serving repetition should remain a success. The central musical question is whether each continuing passage makes sense in relation to what came before and opens a meaningful way forward.