Design notes — Part XI — v3.0, Phase 12: the carry-overs (steps 76–77: Anod finished — the deterministic seed, the strain buffer, the Chirikov grid, the spark reading; the Boehm layout, the split surface, preset-next — moved directly after Kessho on 2026-10-14, `DECISIONS_10 #13`; formerly `DECISIONS_11.md`)

Opened 2026-10-14 at step 76; merged into DECISIONS.md as Part XI at the phase’s fold (2026-10-18, the author’s word). The spec is PROJECT_SPEC.md (v5) with TO_PROJECT_SPEC.md §10.10, §14.7 and §15.4 as the phase’s draft, corrected per step; the roadmap ROADMAP_6.md (Phase 12); the author’s delta _work/specs/ROADMAP_6_delta_phase_order_and_spark.md (the spark reading, §B). DECISIONS_11 #n means an entry here; DECISIONS_10 #n Part X (Phase 11). The core is OPEN for feature work through step 77 and frozen from step 78 to step 92 (the phase order of DECISIONS_10 #13); the phase invariant stands — tests/fixtures/field_512_metal.bin bitwise on Metal, the §10.1 switching gate across the media.

1. Step 76 — Anod finished: the deterministic seed, the strain buffer, the Chirikov grid and the spark reading (2026-10-14; libsumi 1.11.0, additive; evidence docs/evidence/step76/). The seed (DECISIONS_7 #25 answered): sumi_set_seed / sumi_get_seed — the mapper’s strike-phase generator is one 32-bit word, the core’s only draw. Measured first: the word alone left Anod divergent run to run; the rest was the mapper’s gesture CLOCKS — the press-fed torsion sweep’s phase and its sub-floor remainder, the stir’s pending rotation and sense, the running spark and burst episodes, the smoothed controls that never quite reach their targets — so the set rebases them too (the state a recording cannot carry, at rest). The replay’s header gains seed <n> (schema 1, additive: older readers skip it, older files have none); the player sets it after the session, before its first frame; the desktop’s recorder and the bench’s write it. Gate 1 on the bench (--seed-test): one gesture script — eight strikes, a press at two spots for a second — twice from seed 7 reads bitwise in Sumi, Anod AND Kessho (2 097 152 bytes each, 0 differ); the red: twice without the reseed, Anod differs in 1.07 million bytes (the step-59 finding reproduced on purpose; Sumi never cared); the word read back after eight strikes and set back reruns bitwise. Gate 2: the replay gate on the bench’s recording (--record-demo, 480 frames, now carrying seed) GREEN on the desktop tier, the 60 Hz re-bucketing diverging as required (the mobile tier is the tablets’ sub-steps’). The strain buffer (SPEC §10.4’s carried question, answered): a side RGBA16F target at the field’s size, ping-pong, stepped once a frame after the deformations by a third program of composite.glsl (strain) through the stencil kessho_strain — now a @block with three readers (Kessho’s Layer A, the spark, the strain pass; the plasma keeps its own inline copy of the same arithmetic, bitwise 1.10.0 by the Anod golden): x the texel’s σ, y the FAST activity (the strain added per frame, summed over 50 ms), z the SLOW activity (the same over spark_afterglow seconds), w the EXPOSURE (σ’s EMA over the same). Reset by the dip and at creation; an export of a foreign field reads instantaneous σ (SUMI_COMPOSITE_NO_STRAIN). The field’s payload and both fixtures are untouched; the plasma never reads it. What “the strain buffer” had to mean: σ of the accumulated map never decays — a channel that “cools” can only be the strain’s RATE integrated, so the buffer holds activity, not σ; the afterglow is its slow channel (gate 8: a press feed’s spot reads 11.30 at the press end and 4.11 one time constant later, the ratio 0.3634 against 1/e = 0.3679), and Kessho’s Layer A reads the exposure (a time constant on the fringes; the Kessho bench gates were not rerun against it in this step — the roadmap’s gate 4 is open until the Kessho test is run with a settle of three time constants, [ITERATE]). The spark reading (the delta’s B.2): params.anod_glow_mode 0 plasma / 1 spark, spark_afterglow (0.05..5 s, default 0.6), spark_branch (κ, 0..1), SUMI_GLOW_PLASMA / _SPARK. Ridges, not sheets: a texel survives if σ is within a tenth of its neighbours’ maximum across ê_ψ⊥ at the stencil’s half-width (the tolerance keeps a ridge continuous across the half-float’s dips; without it the ridges were one-texel dust); the transfer 1 − exp(−(σ/1.5·anod_glow)⁴); the LEADER = ridge × the fast activity saturating at 0.01 (above the strain rate’s half-float noise — at 0.03 the leader flickered per texel and the skeleton fragmented), toward white-blue, the bloom’s only source; the AFTERGLOW = ridge × the slow activity (reference 0.3, gain 0.55) in the charge’s own colour; the flicker 15 % at 7 Hz with the speckle’s value as the phase; charge only, as the plasma. Branches, as built: the delta’s “threshold against σ + κ·grain” cannot touch a hot ridge — on a swirled filament σ is several, far above the reference, and the transfer saturates whatever the dither (measured twice: 11.67 endpoints per 100 texels of skeleton at κ 0 and 11.67 at κ 0.5 with an additive dither; 11.62 and 11.95 with a multiplicative one of ±1.5σ) — and a texel-scale grain only punches holes a two-texel ridge absorbs. So κ GATES the ridge against a grain COARSER than a ridge (nine texels a cycle at 1080p, screen-locked): cut wherever the grain exceeds 1 − 1.2κ — never at κ 0, three texels in ten at κ 0.5 — a chain of segments with dead ends, the dendrite’s spurs; lighting, not geometry. What the gates measured (--spark-reading-test, the swirl scene at 1080p: twelve large strikes torn by their sparks, 180 frames of a strong Lamb–Oseen swirl with a tine across; the press scene for the afterglow): (5) the plasma prints bitwise with the spark’s knobs moved (0 bytes), the field under the spark bitwise the plasma run’s (0 bytes) while the print differs in 6 167 — a reader; the Anod golden tests/fixtures/composite_512_metal_anod.rgba, captured from the tree BEFORE the step and gated by composite_gate.py --medium 1, bitwise; (6) thinness: the lit ridges’ mean width 2.49 texels (area 6 354 over a skeleton of 2 548; Zhang–Suen); the red, the suppression off: 6.3 texels — the roadmap’s 12 was a soak scene’s guess, this scene’s strained charge is thinner, the bar is 6 and twice the ridges; (7) sparsity: the spark lights 0.31 % of the canvas above the bright threshold, the plasma 0.59 % — the roadmap’s ”≈ 30 %” was the same guess: this scene’s charge covers 3 % of the canvas; the bar is 5 % and 0.6× the plasma; (8) the afterglow above; (9) the branches: 11.62 endpoints per 100 texels at κ 0 (a one-texel ridge has ends of its own — “no spurs” is relative), 16.23 at κ 0.5 (1.40×), the red (the branch grain off at κ 0.5) 11.82; (10) the glass under a 120 BPM roll scroll with the flicker’s clock running and the grid off prints bitwise (0 bytes); (11) the A/B renders at the author’s defaults — spark_ab/plasma.png, spark.png, spark_branches.png — the author’s viewing, the Anod default per medium stays plasma until it. The Chirikov grid (DECISIONS_5 #89): --chirikov-grid <dir> — one pass of the operator at K 0.3 / 0.6 / 1.0 × periods 2 / 4 / 8 × ε 0.15 / 0.3 / 0.6 on a sheet with two drops, 27 renders and an index (docs/evidence/step76/chirikov_grid/); the author’s choice becomes the defaults by their word (today K 1.0, periods 2, ε 0.5). The frozen paths: the field and the Sumi composite fixtures bitwise, the Anod golden bitwise, ctest 14/14, the web gate green. Not in this step: the Lichtenberg operator (open decision 21: 76b or 4.0, the author’s); the tablets’ recorders’ seed and their Glow-reading rows (their sub-steps); the Kessho gates against the exposure (above). Next: step 77, the Boehm layout, the split surface and preset-next — the last core-open step. Amended by #2 (2026-10-15): the branch trick is retired and spark_branch dropped before 1.11.0 ships; the strike’s shape follows glow_mode from 76b.

2. The first spark build viewed, and the branch trick retired (2026-10-15; the author’s ROADMAP_6_delta_76b_leader_stroke.md, folded with corrections). The author was not satisfied, and the delta says why: on the strikes the three readings — plasma, spark, spark with branches — have the same silhouette. The reason is in the operator, not the reader: the Anod note-on’s shear episode is a stack of triangle waves, whose slope is piecewise constant, so σ across the blob is a set of PLATEAUS; non-maximum suppression thins only a peaked profile (and #1’s tolerance of a tenth keeps a plateau whole on purpose), so every texel of the blob survives. #1’s ridge numbers were true but came from the swirl scene, where differential rotation grades σ; a strike alone has no line in it, and a reader cannot invent one. The line has to be written: step 76b, the leader stroke (the Lichtenberg operator of open decision 21, now inverted from “recorded, not built” to “the fix”). Retired: the branch trick — measured in #1 as unable to make endpoints on a hot ridge, it cannot make them on a plateau either; params.spark_branch is dropped from the header, the ABI history (1.11.0 lists two fields), the preset table, the INI, the settings window and the bench, and its dev mode with it. Kept, unchanged: the seed, the strain buffer, the Chirikov grid, the reading’s NMS, transfer, leader, afterglow and flicker — they were right, with nothing thin to act on. The fold’s corrections to the delta: 76b on the Mac (the Linux box is out); the roadmap’s decisions are 21 (inverted) and 22–24 (the trick retired, walk vs DBM, the drift direction), plus 25 the §10.2 principle — the delta’s 20–23 were off by one; “1.6.0” in its ABI note is 1.11.0; the walk’s lengths in canvas heights, not texels (ℓ ≈ 0.012, the window’s half-width 0.003 — thirteen and three texels at 1080p); the plateau red in an operational form (the suppression keeps ≥ 80 % of a blob’s charge — a stack of three has several flat levels, so the 5 % cross-section is the figure, not the gate); “exactly 1 per stroke at p_b = 0” read as the single leader’s two ends; and the stroke as an EPISODE like the spark’s — the tips advancing over frames, one batched pass a frame per stroke — so the pass budget holds by construction and the buffer’s leader rides the tip. The first build’s captures are archived under docs/evidence/step76/spark_first_build/. The amended step-76 gates: (6) and (9) move to 76b; (12) the plateau and (13) the torsion-scene thinning are added, measured at 76b’s start.

3. Step 76b — the leader stroke, as shipped (2026-10-15; the delta of #2, ROADMAP_6_delta_76b_leader_stroke.md; libsumi stays 1.11.0, additive: sumi_add_leader, SUMI_DEFORM_LEADER = 17; evidence docs/evidence/step76b/). The operator is the delta’s: under glow_mode 1 the Anod note-on is a stroke along a jagged tree from the key instead of the blob episode (which mode 0 keeps bitwise — the Anod golden holds, the field under plasma holds); the walk ℓ = 0.012 canvas heights, ±30° about the strike’s pitch axis, p_b = 0.15 at ±45° with half the parent’s count, depth 3, N = 8 + 32·v (v the strike’s value: r = r_min + span·√v, so the radius the mapper hands the strike gives v back as ((r − r_min)/span)² — no second velocity path); the generator a 32-bit LCG on the strike’s draw from the seed, the same draw the blob’s phase takes, so the replay gate of step 76 holds with strokes (gate 2: the press scene twice from seed 7 bitwise, 16.6 MB, 0 bytes differ). Each segment a slip along its own direction in a C² quintic window across (h = 0.003) and along; one pass per stroke per frame composes the frame’s sub-steps last-first (the pre-image of a composition), up to eight entries. Resolved in the building, none of it in the delta:

   • The channel is charged. The first build’s stroke walked into dark water: the reading lights charge only and a strike’s charge is a small disc, so the whole tree was a crack through the far-field grid, lit nowhere (spark_first_build-style, archived in the step’s leader_ab/ as the plateau renders’ sibling). The leader IS the ionized path, so the pass writes the strike’s own band (phase base + the radial fraction across, as a drop’s; the strike’s aux with it) within h of the segment, round caps at the joints; the slip stays. The mapper notes the strike’s band at the strike (sumi_voice_mapper_note_charge; the engine’s tap notes its drop’s); the gesture route’s sumi_add_leader takes a drop count and a band of its own. Under the soak the charge is OFF (sumi_debug_set_leader_charge): the class gate reads the bare shear, the charge a deposit excluded by nature as every strike’s drop is (#13 of Part V’s precedent).
   • The along-ramp is the class. Across the band a slip is a pure shear (det J = 1 for any profile — the tine’s argument); along it the window is on the coordinate being slipped, so the ramps compress by a·w′_max/ramp, w′_max = 1.875 for the quintic. The first build’s ramp of h made det J ≈ 0.06 at the ramps’ steepest; the ramp is 2ℓ = 0.024 and the slip a = h·(0.6 + 1.4 v) ≤ 2h = ramp/4, so det J ≥ 0.53 and ONE sub-step at every velocity (K = ⌈4a/ramp − 10⁻⁴⌉ = 1; the formula stays general). The soak: det min 0.763, mean 1.00000 everywhere after one sub-step; 500 pairs 0.15 texel of pre-image drift, the mass within 0.9 %; the stream (a stroke every 12 frames at half energy, 6000 frames) grows the mass +20.6 %, 3.4·10⁻⁵ a pass — the far ramps’ bilinear magnification, within the class’s rate bar (5·10⁻⁵; the burst’s and the wake’s bar) — and the erosion reads below the glide tine’s control. [ITERATE] a gentler ramp (3ℓ) if the growth matters in play; the strokes are brief events there.
   • The advance is time-based. 120 sub-steps a second as a fractional credit on dt (one a frame at the scripted 1/120 s; capped at eight so a stall never bursts), every alive branch together, the frame’s entries in time order in the one pass; the budget refused → nothing committed, the stroke waits a frame (the first build advanced the counter and lost the segment). The storm of fifteen strokes at 127: 15 alive at once, at most 30 passes a frame (the fifteen strokes and their fifteen charges on the strike’s frame; the budget 64), the last walked out at frame 59.
   • The channel’s gain 0.55 → 0.4 (SPARK_AFTER_GAIN, a step-76 [ITERATE]): the delta’s “the leader ≥ 3× the channel at the tip” read 2.6 at 0.55 and reads 3.03 at 0.4 — measured on the charged texels with the bloom off (the composite’s own terms: the tip the brightest, the channel the crest’s 90th percentile of the texels whose fast activity has gone), because the lit mask’s threshold clips the channel’s distribution from below and the first measure could not move. Plasma is untouched by the constant.
   • The seed test’s red control, found and fixed in the bench: a run cut short mid-press carried its held press into the next run — the press’s own state is not the seed’s to rebase — and the two “identical” runs differed by 3.6 MB. Every leader-test run now ends its press (sumi_gesture_press_end); the determinism itself was never in doubt. The measures, corrected against the delta’s figures (the roadmap’s gate text is the author’s to amend): (3)‘s red — “the blob keeps ≥ 80 % of its charge lit” — is the suppression’s share alone, NMS on against NMS off on the same reading: 32 % on one blob, 48 % on the press scene’s eight (a thinned line would keep a few percent: two widths over a diameter), the silhouette 100 % within two texels of the plasma’s; the plasma-relative figure (45 %) conflates the spark’s steeper transfer with the suppression; raw σ over the blobs’ interior reads (max − mean)/max = 0.53 — a stack of three is terraces, not one plateau, which the fold’s note anticipated. (4) — ”≥ 4 endpoints per 100 texels of ridge” — is read on the CHARGE’s skeleton (the channel the stroke writes, thinned, spurs pruned four texels): the lit mask’s skeleton counts the flicker’s dashes (41.8 ends a stroke at p_b 0), and the strain’s band overshoots every kink by the along-ramp (6.8); on the charge, six strokes at p_b 0.15 end in 3.83 tips each (0.58 per 100 texels — the “4 per 100” assumed the dashes), at p_b 0 in 1.50 — the tip, and the root where the charge disc’s skeleton leaves one — and the blob episodes’ discs give 0.09 endpoints per 100 texels of charge; the bar is ≥ 3.5 a stroke and ≥ 1.5× the single leaders’. The amended step-76 gates: (12) the plateau — 47.9 % kept with the suppression on, 100 % within two texels — and (13) the torsion scene — the rings 1.91 texels wide under the reading, 3.9 with the suppression off — are green (--spark-reading-test 7/7, the sparsity gate now the plasma reading of the SAME field: 1.06 % against 1.95 %). Gate 6, the reach chart: 22 / 31 / 47 / 51 frames and 2.4 / 5.6 / 10.4 / 13.4 thousand strained texels at 32 / 64 / 96 / 127. Thinness: 1.70 texels on one stroke (3.6 k lit texels over a 2.1 k skeleton). --leader-test 6/6, --soak leader green, ctest 14/14, the field fixture within its tier, the Sumi composite and the Anod golden bitwise, the web gate green. Kept open for the author: the viewing (leader_ab/: plasma strike / leader / leader at p_b 0 on the press scene at frames 40 and 100, stroke_one.png at the defaults, blob_under_spark.png the plateau figure, storm_f30.png) and the Anod default per glow_mode; DBM against the walk (decision 23 of the roadmap: the walk is kept unless the viewing says too regular); the drift direction (24: the pitch axis shipped); the constants ℓ, h, p_b, the rate, the gain — [ITERATE]; the tablets’ Glow-reading rows (their own sub-steps) carry the stroke with them, nothing to add there. The --leader-pb <p> flag sets this run’s branch probability for the live window and replays (the test sets its own). Amended at the author’s viewing (2026-10-18):
   • The stroke is struck BESIDE the shear episode, not instead of it. The first ship replaced the Anod note-on’s blob episode by the stroke under glow_mode 1, and the bolt lost its impact — the blast and the streamers are the strike’s weight, the leader grows out of them. The mapper’s switch now adds the stroke and falls through to the episode; the strike’s draw order is the stroke’s first, the blob’s phase after (the determinism gate holds: bitwise twice from one seed). Every render under leader_ab/ is re-cut this way.
   • The habit and the tine profile are every medium’s. The drop’s habit (drop_habit, habit_roundness, habit_blades) and the tine’s profile (tine_profile) shape the drop and the tine under Sumi and Anod as under Kessho, and sat under Kessho’s Mineral panel; the desktop’s settings window gives them a Drop & tine shape section of their own, always present (the author: “the habit affects all the mediums”); a mineral preset still pairs a habit and a profile with its optics, and editing either still marks the mineral Edited. The settings reference moves the two rows with it.
   • A step-76 bug, found by gate 5 and fixed: the strain pass compared a float σ against its own half-rounded previous value (the buffer is RGBA16F), a residual of up to half an ulp every frame on a static field — positive where the rounding went down — so a strained texel kept a permanent activity floor: a stroke’s channel decayed to 0.44 instead of 1/e, and a texel of σ ≳ 16 would have read as a leader forever. The pass now rounds σ to half (packHalf2x16, the value stored is the value compared); the press scene’s afterglow gate of step 76 reads 0.358 after the fix (0.364 before — its feed spot sat at a lower σ), and the fixtures are untouched (the plasma and Sumi composites never read the buffer; Kessho’s exposure channel is the same EMA of the same σ).
   • The measures, once more: with the strike’s blast under every stroke, the branch count and the afterglow are read on STROKES ALONE through the gesture route (sumi_add_leader: six on the tap grid’s points at the tap’s energy; one at full energy) — the torn charge would sit in every count; thinness is read beyond the strike’s blob and streamers (2.5 r about the key); the reach is the channel’s charge beyond the same disc; the leader’s contrast is read at frame 30 with the tip walking, the channel the texels whose leader is under a tenth; the storm’s cost is the pass TOTAL over 320 frames against the same storm under plasma — the shear episodes’ steps land where the budget lets them, so a per-frame peak compares two different frames. The figures are the evidence’s table.
   • The viewing’s verdict and the default (the author, 2026-10-18): “this looks good” — the walk is kept (DBM stays a 4.0 item unless asked for; roadmap decision 23), the pitch axis stays the drift (24), and plasma stays Anod’s default glow_mode: the spark reading with its stroke is the player’s choice in the settings, the presets and the INI. The roadmap’s gate 8 is folded.

4. Step 77 — the Boehm flute, the split surface, preset-next (2026-10-18; libsumi 1.12.0, additive — THE LAST CORE-OPEN STEP of the arc, the core frozen from step 78 to step 92; evidence docs/evidence/step77/). The roadmap’s one line left the instrument to design; the ambiguities it resolved, in order:

   • Which woodwind, and what the cells are. The Boehm FLUTE (the system’s name is the flute’s; the clarinet’s register key gives a twelfth and a different chart). Its cells are REGISTERS, as the brass’s are partials: three — the fundamental, the octave, the second octave, +0 / +12 / +24 — on the brass geometry unchanged (the column’s three rows are taller than the touch width, so the radius is the width’s half, 0.125; the arc’s three sit at π, π/2, 0 on the same ring; the one trumpet_arc flag, named “Registers on an arc” under the flute). The idealised instrument, as the brass: no intonation, the flute’s irregular third octave replaced by a clean +24.
   • The keys and their CCs. Nine keys in buttons bits 0..8: the left hand’s three fingers reuse the valves’ CC 110–112 (the same bits, so the brass code reads buttons & 7 unchanged), the thumb, the G♯ lever, R1, R2, R3 and the D♯ key on CC 114–119 — the six free undefined controllers above the slide’s 113; CC 120+ are channel-mode messages and the bench holds 102–109. The foot joint’s C♯ and C rollers are dropped (two more keys would have needed scattered numbers): the chart’s bottom is D4, the flute without its foot — one full chromatic octave on the fundamental.
   • The chart as a total function. The vent rule (the highest open hole sounds), the holes top to bottom with the Boehm mechanism’s two linkages (the A key and the F key both close the B♭ hole: the “1 and 1” B♭ comes for free), the G♯ and D♯ keys closed-standing (pressed = open), and ONE cross fingering kept because the standard chart has it (R3 alone under an open F hole: F♯ — a real acoustic effect, not a linkage). Every one of the 512 states sounds a note in D4..C♯5; the twelve standard fingerings are the goldens (G_BOEHM), transcribed independently of the rule, with the linkages and the cross fingering asserted by name and the D♯ key held as a flutist holds it (silent below the vent). sumi_layout_boehm_note is exported so a key panel can label the note.
   • The split surface’s home for a wire note. A touch knows its half; a note from the wire does not. Resolved by a SPLIT NOTE (split_note, C4): the left half plays the notes below it, the right the notes from it — a keyboard split — and the probe REFUSES a cell whose note belongs to the other half, so a touch’s note and the engine’s drop agree. The alternative, two MPE zones (lower = left, upper = right), is the Seaboard’s split and the right long-term answer, but the normalizer models one zone; recorded as a 4.0 item. Each half is its layout at half width with the aspect halved (canvas-height units carry over, so radii and steps need no scaling; a layout’s normalized x halves back into the full frame — the axis too, found by the test: the first cut forgot it). The public queries wrap unsplit bodies (probe_one, position_one, cells_one, delta_one) and the bodies call each other through those, never through the public names — a public call applied the split twice (the grids’ cells quartered their x; the test caught it). The cells are both halves’ unfiltered (the API carries no notes); the shells grey. Rolls scroll the whole sheet and FIFTHS has no cells: a non-Play layout on either side switches the split off rather than drawing nonsense.
   • Preset-next is hostmpe’s _preset_set / _count / _next in the quick-switch’s shape over opaque preset ids (the shell’s indices or hashes, 32 at most); the pads are the tablets’ sub-steps, as the roadmap says of the panels — nothing on the desktop, which has no strip. hostmpe’s HOSTMPE_QUICK_MAX grows to 14; the strip’s announce restates fifteen messages (the nine keys then the slide), its reset lifts every key.
   • The shells. The desktop: the layout combo of fourteen, the split rows (the checkbox, the right half’s combo, the split note with its name), the INI’s split_*, the preset’s three fields; the web: the Boehm in the layout list, its nine key toggles, the overlay’s “keys” line (the split is not on the web’s GUI — its param shim stops at step 66’s ids; the overlay would draw a split a preset carried); the iPad’s and the Tab’s pickers name the layout (playable) so their surfaces play the registers with every key up (C♯5 on the fundamental) — their key panels and their split surfaces are their own sub-steps. The chart row for CC 114–119 is conditional until a tablet’s panel logs them. Gates: normalizer_tests 46 820 checks (the chart’s twelve goldens, the linkages, the cross fingering, the 512 states in range, three register cells on the column and arc goldens at two aspects, every fingering × every register through the probe and back through the placement, the key CCs decoded on the master and refused on a member; the split: the placement by the note, the probe by the touch and refused across the split, the cells of both halves, the axis halved, a roll switching it off), hostmpe_tests 3 175 (the keys, the announce of fifteen, the reset, preset-next), abi_c_compile 33 symbols at 1.12.0 (the enum, the CC constants, the chart, the probe under two fingerings), ctest, the field fixture, the Sumi and Anod composites, the web gate — all on the final build. Carried: the layouts’ microtonal mapping (§14.6), the two-zone split, the tablets’ panels and surfaces, the chart row’s log. The phase close — the commit, the fold of this file as Part XI, v3.0.0-alpha.2 — is the author’s.