Design space exploration

Design Space Exploration

Expand the solution space deliberately before choosing. Read references/design-space-methods.md when decomposition, parallel generation, coverage, or source detail matters.

Workflow

1. State the desired outcome, users, system boundary, invariants, material constraints, and forbidden outcomes. Separate requirements from inherited implementation assumptions.
2. Reframe the problem from at least two credible perspectives. Do not generate a large option list for the wrong question.
3. Decompose the space into independent or weakly coupled dimensions such as ownership, state, interaction, deployment, economics, reversibility, and operating model. Add domain-specific dimensions rather than forcing this example taxonomy.
4. Generate materially different option families. Include the status quo, smallest viable change, orthogonal designs, a high-upside frontier design, and hybrids only where their mechanisms are coherent.
5. When parallel agents are available, give them different lenses or regions of the space. Independent constraints, analogies, failure models, or design philosophies produce more diversity than repeating one open prompt.
6. Normalize and cluster the results by mechanism. Merge cosmetic variants, preserve genuinely different causal structures, and expose combinations that violate constraints.
7. Map explored and unvisited regions, key assumptions, option dependencies, irreversible commitments, and the cheapest probe that could retire each important uncertainty.
8. Stop when another generation pass is unlikely to reveal a materially new family relative to its cost. Hand the portfolio to the appropriate decision process without silently choosing a winner.

Output

Produce a Design Space Portfolio:

• Challenge and boundary — outcome, users, invariants, constraints, and rejected framings.
• Exploration dimensions — the axes used to structure coverage.
• Option families — mechanism, distinctive value, assumptions, failure modes, reversibility, and dependencies.
• Coverage map — explored, combined, invalid, and intentionally unvisited regions.
• Discriminating probes — cheap experiments, prototypes, research, or evidence that would reduce uncertainty.
• Decision handoff — choices still alive and the criteria or authority needed for selection.

Boundaries

• Use critical-analysis to compare explanations about what is true.
• Use structured-deliberation when independent participants must exchange positions, objections, and dissent.
• Use decision-quality-standard to select and record the final material option.
• A list of minor feature variations is not a design space. Preserve options only when they differ in a decision-relevant mechanism, tradeoff, or future.
• Do not score options during initial generation unless a hard invariant makes one invalid; early ranking collapses diversity.
• Do not claim complete coverage. Record the bounded search and remaining regions honestly.