╔═══════════════════════╗
  ║                       ║
  ║   S O N A T A         ║
  ║                       ║
  ║   28 keys.            ║
  ║   Form fully          ║
  ║   realized.           ║
  ╚═══════════════════════╝

Design Concept & Approach

Sonata is the third in a family. Cadenza (36) is the daily driver. Coda (22) is the minimalist concept. Sonata sits between, asking: can a 28-key board offer the structural cleanliness of Coda (zero lateral stretch, no inner column) with the linguistic completeness of Cadenza (full alphabet on base, full HRM)?

The answer through v1.x was: yes structurally, but at a measurable SFB cost vs Miryoku-on-36. v1.0 oversold this. v1.1 corrected the marketing (1.79% honest SFB, "sibling, not successor"). v2.0 closes most of the SFB gap through a full alpha redesign that keeps the Colemak-DH home row but optimally reassigns every off-home letter.

The v2.0 process explicitly considered an alternative home row (the "C2" I↔O swap) and rejected it after full layout placement showed C1 dominates. It also explicitly considered a more aggressive variant (C1-bestV: U on L_pinky_bot, ~0.97% SFB) and chose the safer C1-balance for the released v2.0 — leaving bestV available as a v2.1 if the residual SU/US SFB on L_mid proves uncomfortable in daily use.

Physical Layout

4 finger columns × 3 rows × 2 sides + 2 thumb keys × 2 sides = 28 keys. No inner column. Column-stagger geometry. Custom 28-key PCB target. Below is the base layer — the same structural shape is reused on every layer that follows.

Design Principles

Letter Placement (v2.0 — preserved through v3.0)

The Colemak-DH home row is preserved unchanged. All 16 off-home letters were reassigned to minimize SFB while honoring three structural constraints: D cannot share a column with N (ND ~1.5% German), H cannot share a column with E (HE ~3% English), C cannot share a column with S (SC ~0.30%). U was the trickiest: it cannot go on R_pinky (recreates v1.1's IU disaster) or R_mid (EU is German same-finger). U landed on L_mid_bot — paying ~0.30% SU/US SFB as the cheapest U-on-left option.

Ergonomic Metrics

SFB by Column — Honest Breakdown

The total SFB number alone hides which columns actually carry the cost. Here's the honest column-by-column breakdown for v2.0 (alpha grid unchanged in v3.0, so SFB metrics carry over):

The two remaining hotspots are L_mid (G·S·U) at ~0.35% — driven by SU/US, the cost of placing U on the left side without using the pinky — and R_mid (Y·E·F) at ~0.38% — the structural cost of E being so reactive that any non-trivial column-mate creates moderate SFBs. The R_ring fix (V·I·U → P·I·B) is the single biggest win and the structural justification for the v2.0 redesign.

Layer Reference

Sonata exposes nine layers (L0 base + eight modal layers). Every layer is rendered below as the full 28-key board. The gold-glowing key on each board is the one you press and hold to activate that layer; bilateral triggers light up both keys.

Comparison vs Other Layouts

The Family — Cadenza · Cadence · Sonata · Coda

Sonata is one of four related layouts in a single family. Each shares a Colemak-DH base and Tap Dance Home Row Mods; the differences come from key count, hardware target, and how much of the layer system can survive the reduction. The musical naming describes how each variant relates to the others: Cadenza is the demanding solo passage (the origin), Cadence the same piece adapted for a smaller hardware, Sonata the complete musical form (synthesis of the family ideas), and Coda the concluding distillation.

Development timeline

Sonata is not a simple descendant of any single layout. v3.0 is a synthesis point — it consolidates lessons from a parallel improvement of Cadenza and the build of Cadence. The seven-step chronology:

 ①  Cadenza v1.0          36 keys · Corne Choc · the origin
                           │
 ②  Sonata v1.0            28 keys · started — first reduction attempt
                           │
 ③  Coda v0.1              22 keys · started in parallel — distillation
                           │
 ④  Sonata v2.0            alpha-grid redesign (Sonata's own SFB work)
                           │
 ⑤  Cadenza (improved)     refinements to the 36-key origin
                           │
 ⑥  Cadence v1.5.0         34 keys · Ferris Sweep
                           │             port of the improved Cadenza
                           │
 ⑦  Sonata v3.0    ◄───────┴──── synthesis of:
                                    · learnings from ⑤ Cadenza improvements
                                    · learnings from ⑥ building Cadence
                                  applied to Sonata's layer-access architecture:
                                    · combos eliminated → bilateral letter pairs
                                    · all access keys + HRMs as Tap Dance
                                    · L_SYM rebuild by frequency

Coda's chronology is parallel to early Sonata work — it started as a separate "what's the absolute minimum" exploration and has stayed at concept stage (v0.1). Cadence was built specifically to bring the improved Cadenza muscle memory onto Ferris Sweep hardware — same design philosophy, two thumb keys less.

Side-by-side

Cadence's SFB and pinky load not stated explicitly in its own documentation; values shown inherit from Cadenza by self-description ("same design philosophy, same muscle memory, two thumb keys less"). Cadenza/Coda metrics are taken from Sonata's own comparison table where the originals don't provide a direct number.

What's preserved across the family

What's distinctive about each

Which one to pick

Honest Verdict

Verdict: Sonata v3.0 is a credible sibling on the same Pareto frontier; it no longer trails on SFB. The remaining gaps (~2 pp pinky, ~4.5 pp home row) are intrinsic to the 28-key constraint. Pinky comfort + proven design → Miryoku on a Sweep. Lowest lateral stretch + German EU/SCH fidelity + cross-hand Meta+Alt + dedicated Backspace → Sonata v3.0.

Future Variant — the C1-bestV alpha redesign

Both v2.0 and v3.0 use the conservative C1-balance; the more aggressive C1-bestV (see process notes in sonata-spec.json under v2_0_proposals_explored). The bestV variant trades:

bestV candidate layout (future variant):

 ┌───┬───┬───┬───┐          ┌───┬───┬───┬───┐
 │ Z │ W │ F │ M │          │ L │ K │ P │ J │
 ├───┼───┼───┼───┤          ├───┼───┼───┼───┤
 │ A │ R │ S │ T │          │ N │ E │ I │ O │
 ├───┼───┼───┼───┤          ├───┼───┼───┼───┤
 │ U │ C │ G │ D │          │ H │ Y │ B │ V │
 └───┴───┴───┴───┘          └───┴───┴───┴───┘

The decision rule for migrating: if after 3–4 weeks of daily use the residual SU/US SFB on L_mid genuinely interferes with typing flow, the bestV reshuffle is a small surgical change (six keys move) and the SFB win is meaningful. If the SU/US never bothers you, the current placement is the better long-term home because U-on-pinky is the riskier ergonomic bet.

QMK Implementation

config.h essentials

#define TAP_DANCE_ENTRIES 34
#define TAPPING_TERM 200
#define PERMISSIVE_HOLD
#define HOLD_ON_OTHER_KEY_PRESS
#define COMBO_TERM 40
#define COMBO_ONLY_FROM_LAYER

Layer enum

enum sonata_layers {
    L_BASE = 0,
    L_OVF,    // Tab-hold (single)
    L_SYM,    // Ent-hold (single)
    L_NUM,    // Bsp-hold (single)
    L_NAV,    // Spc-hold (single)
    L_MOUSE,  // V-hold OR K-hold (bilateral)
    L_FN,     // W-hold OR P-hold (bilateral)
    L_CLI,    // U-hold OR F-hold (bilateral)
    L_WM      // C-hold OR B-hold (bilateral)
};

Tap-dance breakdown

OS layout

US International. Umlauts via "+a/o/u (deadkey method); ß via RAlt+S; € via RAlt+5. Literal " via M9 macro (sends "+Space) to bypass the deadkey when needed.

MIT License · 2026 · one7two99 · github.com/one7two99/sonata
Based on Colemak-DH by stevep99 — home row preserved, column-mates redesigned in v2.0; layer-access architecture redesigned in v3.0.
Family: Cadenza (36) · Sonata (28) · Coda (22)