Reasoning Foundation Router

Select methods; do not replace the active domain skill. The domain owner keeps facts, calculations, implementation, safety standards, and the final deliverable.

Cheap Route Check

A task is method-complex when any condition holds:

• one missing user fact can change the route;
• two or more credible choices or causal explanations carry meaningful cost, risk, lock-in, or opportunity cost;
• stakeholders, objectives, or time horizons conflict;
• hidden assumptions, path dependence, or a conclusion-flipping variable matter;
• the task must synthesize evidence across time, comparators, disciplines, or distant domains.

Length is not complexity. A long format conversion is simple; a one-line migration choice may be complex. Skip the package for direct facts, translation, formatting, fixed instructions, and deterministic implementation or bug fixes settled by inspection and tests.

Choose One Primary Owner

Choose by the requested final deliverable, not by which method sounds most sophisticated:

1. $socratic-problem-framing: the problem is not decision-ready and unavailable user input can change the route.
2. $fact-claim-audit: truth, credibility, causality, or evidence quality is the core deliverable.
3. $dual-layer-explanation: the core deliverable is learning one concept at beginner and professional levels.
4. $reverse-engineer-example: the user supplied an inspectable example and wants reusable mechanisms.
5. $longitudinal-horizontal-analysis: the core deliverable combines evolution, comparators, and conditional outlook.
6. $first-principles-rebuild: path dependence or patchwork must be replaced by a derivation from facts, goals, and constraints.
7. $cross-domain-mechanism-transfer: the bottom structure is known and the core deliverable is borrowing mechanisms from distant fields.
8. $expert-perspective-panel: the outcome depends on three genuinely complementary professional method lenses.
9. $bidirectional-steelman: the core deliverable is a strongest-case contest around one disputable choice or recommendation.

When more than one matches, write the final-deliverable owner as primary. Example: “verify cases, then decide whether to migrate” uses $bidirectional-steelman as primary and $fact-claim-audit as an evidence layer; “only verify the cases” uses $fact-claim-audit alone.

Compose Only on Dependency

Default to one method. Use at most three only when one layer produces a required input for the next:

1. Framing or evidence: $socratic-problem-framing and/or $fact-claim-audit.
2. Construction or research: one of the other domain-shaped methods.
3. Contest check: $bidirectional-steelman, only when its own trigger is met.

Valid examples include framing → first principles, fact audit → longitudinal-horizontal analysis, first principles → cross-domain transfer, and fact audit → first principles → steelman.

Do not run the whole package. Do not combine $expert-perspective-panel with $bidirectional-steelman merely to create more voices. Add steelman after a panel only when one material, conclusion-flipping choice remains.

Hard Boundaries

• Multi-perspective reasoning is not multi-agent execution. Never spawn agents just because $expert-perspective-panel uses three lenses.
• Multiple agents do not prove a claim; use evidence standards and the relevant domain owner.
• Film, security, medical, legal, financial, data, and other specialized routes keep their existing mandatory owners and gates.
• Questions should not stall reversible work. Ask one decisive question unless the user explicitly enters the bounded Socratic sequence.
• In high-impact contestable decisions, include $bidirectional-steelman even when the user presents only one side.

Delivery and Receipt

Keep routing backstage. Deliver the requested conclusion or artifact, not a tour of methods. Show a compact route summary only if the user asks which methods were used or a missing input blocks the work.

Before finalizing, verify: exactly one primary owner; no more than three methods; every helper has an explicit input-output dependency; external claims follow current evidence rules; and the result ends in a usable explanation, judgment, artifact, experiment, or one decisive question.

Package Contents

• $socratic-problem-framing
• $dual-layer-explanation
• $reverse-engineer-example
• $longitudinal-horizontal-analysis
• $fact-claim-audit
• $expert-perspective-panel
• $first-principles-rebuild
• $cross-domain-mechanism-transfer
• $bidirectional-steelman

See source-notes.md for provenance and evidence status, and ecosystem-integration.md for neighboring-skill ownership.