Decision Space Expansion

Decision-Space Expansion — expand the option space before you decide

Core methodology (established 2026-08): before making an important decision, use a heterogeneous multi-model committee to widen the option space, structure the divergence, then fold real constraints back in — and let a human make the final call. Decide inside a space you intentionally expanded, not inside your own prior cognition.

The complete loop has three stages:

Stage 1 — Diverge (widen the options)

• Feed the same brief to models from different families (Anthropic / OpenAI / DeepSeek / Google / local OSS), answered in isolation (invisible to each other — prevents anchoring).
• Assign a stance to each model: architecture / cost / security / product → plans naturally diverge.
• Any heterogeneous model pool works: provider aggregators, multiple CLIs (Claude Code, Codex, Gemini CLI, opencode), local Ollama/vLLM endpoints, or a self-hosted gateway. What matters is model-family diversity, not count.
• Mandatory question patterns:
  • "Why didn't you choose X?" — forces out excluded options + reasons (highest information density)
  • "Give one non-mainstream option with a real justification." — kicks the model off its training distribution
  • "Where is this plan most vulnerable?" — cross-examination a single model never performs on itself

Stage 2 — Structure (sort the options)

Classify the proposals:

• Consensus zone → adopt directly (the parts with standard answers, e.g. a state machine + WebSocket for a queueing system).
• True divergence zone → worth human arbitration.
• Pseudo-divergence zone → noise, discard (e.g. PHP vs Go for a CRUD system is not a material difference).

Stage 3 — Converge (decide with real constraints)

• Fold back real constraints: budget, team skills, timeline, market/region, compliance — let the AI weigh options inside the already-expanded space.
• Optional: pyDecision (AHP / TOPSIS / PROMETHEUS) turns options into comparable score tables.
• Final review is always human. Models fail confidently.

Core beliefs

• Decision power = size of the option space, not the ability to pick. Your cognitive surface is your decision ceiling.
• Model disagreement = training-data geography/ecosystem bias + post-hoc rationalization, not deep reasoning. The most valuable divergence is the geography/ecosystem blind spot (e.g. China-region models raise ICP filing / WeChat mini-programs / Aliyun; Western models default to AWS).
• Same-model-multi-role gives role diversity only, not cognitive diversity — heterogeneous models produce real divergence.
• The committee expands your known unknowns; it cannot reach unknown unknowns — and being forced to understand the divergence while converging is the cognitive growth.
• Best posture: ask AI when you don't know what you don't know, not when you already know what you want.

Pitfalls

• Don't treat the committee as an "auto-optimal-answer" machine — open fusion tools (MoA / LLM-Blender) merge answers by default and delete the human arbitration step, which is the opposite direction.
• Pseudo-divergence wastes time: models fabricate complete rationales for any assigned stance (preference first, reasons after).
• When requirements don't constrain the choice, models output their own familiar default — not an optimum.
• Models silently assume hidden constraints (concurrency, budget, team); humans must backfill the real ones.

Output template

A decision report should include: option skeletons → consensus / true-divergence / pseudo-divergence zones → trade-offs for each true divergence → recommendation + rejection reasons → a one-liner on the point the human must decide.