First-Principles Rebuild

Own the separation of facts, assumptions, goals, and constraints, followed by a fresh derivation. Do not use “first principles” to ignore evidence, standards, or safety limits.

Route

Use when recurring patches suggest a wrong frame or path dependence, historical choices block current options, or the user explicitly asks to derive from zero rather than inherit industry convention.

Skip direct facts, fixed procedures, deterministic bugs, and local repairs whose cost is proportionate and sufficient. During an incident, restore safety first; revisit architecture only after containment or when explicitly asked.

Procedure

1. Define the desired result without naming the current solution.
2. Make four lists: confirmed unavoidable facts; habitual but unverified assumptions; the real objective; and resources or hard constraints. Demote unsupported “facts” to assumptions.
3. Identify which parts of the current approach merely preserve historical structure and which alleged constraints are changeable.
4. Derive at least one path using only confirmed facts, the objective, and hard constraints; “everyone does it” is not a premise.
5. State the new path’s necessary conditions and the variable most likely to invalidate it.
6. Design the smallest reversible check that distinguishes the old path from the new one, with an observation and stop or pivot rule.

Handoffs and Boundaries

• Use $fact-claim-audit for disputed inputs.
• Use $cross-domain-mechanism-transfer only when the bottom structure is clear and direct domain routes are exhausted.
• Use $bidirectional-steelman for a material choice between the rebuilt path and credible alternatives.

Completion Gate

Facts, assumptions, goals, and constraints are not mixed. The new path is traceable to them, its key premise is testable, and the first step has a measurable signal and a stop or pivot condition.

Source Status

Distilled from the user-supplied “第一性原理” method card. The card proposes a reconstruction workflow but supplies no comparative result evidence.