Only a goal
Describe the outcome. Main maps the missing professional and real-use lenses instead of expecting you to name every stakeholder.
Agent-led planning quality control
Better plans before specs and code.
The right perspectives. No more. When you have only a goal or partial domain knowledge and want an AI agent to drive design, Complex Enough improves the Plan before assumptions propagate into specs and code. You remain the boss; Main acts as the meeting manager.
1 · Where it fits
You need a sound Plan—not every answer upfront.
Complex Enough is for agent-led product and software planning where missing a user consequence, authority boundary, source of truth, or handoff could distort everything built afterward.
Partial domain knowledge
Your perspective stays visible and reviewable while independent lenses challenge gaps you may not know to ask about.
Consequential handoffs
Use it when state, authority, evidence, or responsibility crosses users, teams, systems, or a physical handoff.
Keep simple, single-actor, locally reversible work in an ordinary session. Selective routing is part of the product.
2 · Difference from direct design
A quality gate before one perspective becomes the Plan.
A direct session lets one agent fill missing details from one working perspective. Complex Enough first asks whether independent lenses could materially change the result.
Direct session
One path fills the gaps
Goal or partial knowledge → one agent’s assumptions → Plan → Spec → Implementation.
Complex Enough
Perspectives challenge the gaps
Goal or partial knowledge → selective routing → user-reviewed independent lenses → evidence-based synthesis → confirmed Plan.
The stable technical identifier remains orchestrate-multi-perspective-panel. Complex Enough preserves the public 1.x contracts while making meeting formation visible and adjustable.
01
Route selectively
Main checks consequences, handoffs, authority, evidence, reversibility, and stale-state harm before creating meeting state.
02
Review the slate
You see complete role definitions and can accept, edit, merge, split, remove, or import positioning text before anything runs.
03
Synthesize evidence
Independent perspectives are adjudicated by evidence and authority—not votes—and return one auditable result.
3 · Planned extension
Follow the discussion. Learn why the decision holds.
The current skills-only release already returns structured public evidence and decisions. A planned GUI will make that deliberation easier for the human boss to follow before the final Plan is locked.
Understand the domain
See public claims, evidence, conflicts, consequences, and concise decision reasons—not just the final answer.
Intervene before lock
Ask Main for clarification or request an added or split perspective for a newly confirmed next round.
Keep reasoning boundaries
This is human learning, not model training or permanent memory. Hidden chain-of-thought and raw private transcripts remain excluded.
The GUI is roadmap work and is not included in the current skills-only 1.1.0 release.
Directional evidence
Catch ambiguity before it travels downstream.
Plans become inputs to later specs and implementation decisions. Complex Enough is designed to expose user consequences, authority boundaries, handoffs, and recovery gaps before those assumptions harden.
5.0%
Complete six-task result
Relative mean planning-score uplift with all six tasks retained, including a deliberately simple negative-applicability case.
12.5%
Separate compact benchmark
Relative mean planning-score uplift in an independent focused three-task compact-panel comparison.
120/120
Current contract gate
Behavioral assertions passed by three fresh blind public-output graders across 26 isolated cases.
Inside the six-task evaluation, higher meeting-value cases with multi-party state, authority, or physical handoffs averaged a +0.403/5 score delta, versus +0.146/5 for the two lower meeting-value positive cases—about 2.8x the observed delta magnitude. This is descriptive scenario segmentation, not a claim that quality became 2.8 times higher.
These are directional planning results from simulated evaluators in one model family—not guarantees of production outcomes. Downstream amplification has not yet been measured directly. Read the evidence boundaries and calculations.
Clear boundaries
Decision support, not artificial consensus.
Generated perspectives are analytical lenses, not real stakeholders or user research. Public output excludes hidden reasoning, private scratch work, raw internal messages, and panelist transcripts.