Candidate layout factory

[TGALLOWAY1]

Core idea

Instead of “the greedy solver,” think of it as a candidate factory with:
1. multiple seeded starting layouts
2. multiple local update policies
3. a diversity penalty / novelty reward
4. an explanation layer that says why each result is interesting

[TGALLOWAY1]

1. Natural Pose Anchor

Goal: maximize immediate comfort and intuitive finger dominance

Seed logic
• Rank sounds by weighted importance:
• frequency
• section coverage
• density contribution
• accent/backbone role
• Place top sounds on the “golden pads” under index/middle fingers in the natural hand pose.
• Keep left-hand dominant sounds in left zone, right-hand dominant sounds in right zone.

What it tends to discover
• Very playable “obvious” layouts
• Good starting point for finger drumming
• High learnability

Risk
• Can become too conservative
• May miss clever cross-grid structures

Good explanation text
• “This layout prioritizes dominant fingers and low home-pose drift.”
• “Most frequently used backbone sounds are placed in high-control pads.”
• “Low stretch and low zone violation are the main reasons it scores well.”

[TGALLOWAY1]

2. Grouped / Motif Cluster

Goal: keep musically related sounds physically close

Seed logic
• Build a graph where edge weights reflect:
• same group membership
• co-occurrence
• frequent transitions
• phrase-level reuse
• Cluster that graph, then assign each cluster to compact local regions on the grid.
• Within each cluster, order pads by internal transition frequency.

What it tends to discover
• Layouts with strong memorability
• Better handling of recurring phrases
• Easier local improvisation within a cluster

Risk
• Can create local comfort but global hand congestion
• Might overload one zone if a group is too dense

Good explanation text
• “This layout keeps drum/lead/fill families spatially compact.”
• “Frequent within-group transitions are short.”
• “It scores well on learnability and motif consistency, but may trade off some hand balance.”

[TGALLOWAY1]

Coordination Pattern Seed

Goal: optimize for rhythmic interaction patterns, not just static comfort

This is the most interesting “outside the box” one.

Seed logic
Infer recurring coordination motifs such as:
• left-right alternation
• two-hand simultaneity
• downward sweeps
• rolls
• call-and-response phrases
• repeated anchor + accent patterns

Then assign sounds so the physical motion pattern matches the temporal motif.

Examples:
• alternating notes placed across hands
• downbeat accents placed under coordinated downward strikes
• recurring two-note figures placed as mirrored pairs
• rolls mapped along a direction that matches the hand motion

What it tends to discover
• layouts that feel surprisingly musical
• better flow at tempo
• more “performed” and less “placed”

Risk
• can look less obvious statically
• may be harder to explain unless diagnostics are good

Good explanation text
• “This layout reduces burden by converting rapid alternations into cross-hand patterns.”
• “Recurring accent pairs align with coordinated downward motions.”
• “It improves transition efficiency more than static pose comfort.”

[TGALLOWAY1]

Exploration / Randomized Novelty Seed

Goal: intentionally find layouts a human would not normally try

Seed logic
• Start from random or semi-random placement
• But constrain to basic feasibility:
• no absurd initial spread
• preserve core hand zones loosely
• maybe pin a few backbone sounds
• Run stochastic hill-climbing / greedy improvement with randomness in move choice

What it tends to discover
• surprising layouts
• unconventional but effective asymmetries
• new musical possibilities

Risk
• explanation quality matters a lot
• can produce weird but not meaningful diversity unless filtered

Good explanation text
• “This candidate began from a novelty-biased seed and survived because it reduced transition burden in dense passages.”
• “Although less conventional, it improves hand alternation and reduces repeated-finger overload.”
• “Its value is in a different tradeoff profile, not familiarity.”

[TGALLOWAY1]

Update-policy options

Here are some good variants.

A. Strict Greedy Descent

Always take the best immediate move.

Use for:
• natural-pose candidate
• easy interpretability

Weakness:
• local minima

B. Soft Greedy / Boltzmann Move Selection

Prefer better moves, but not always the absolute best.

Use for:
• exploration candidate
• avoiding repetitive convergence

C. Tabu Greedy

Remember recently reversed swaps/moves and ban them temporarily.

Use for:
• preventing oscillation
• cleaner search trajectories

D. Alternating Objective Passes

One pass optimizes:
• hand balance
Next pass optimizes:
• transition burden
Next pass optimizes:
• clustering / learnability

Use for:
• discovering structured tradeoffs
• avoiding single-score tunnel vision

E. Constraint-first then score

First eliminate feasibility disasters, then optimize ergonomic score.

Use for:
• making sure random candidates don’t waste time

[TGALLOWAY1]

https://chatgpt.com/s/t_69be0c3f3e68819195165360c04d0e36