1. Start with the actual question
You may begin with partial mathematics, an anomaly, an intuition, a historical dispute, AI-generated code, a diagram, or a perceptual idea. No preliminary credential gate is required.
2. Build the smallest real version
Turn the question into something that can run or be encountered: a calculation, proof, program, simulation, apparatus, dataset comparison, or virtual world.
3. Run the real path
Observe what it actually does. Keep the output, the failure, and enough source or instructions for another person to inspect or repeat it. That is evidence debt in its minimal form.
4. When stuck, change the representation
Do not assume that more search inside the same language is the only answer. Change the objects, coordinates, diagram, interface, scale, or sensory relation. Then build that alternative too.
5. Compare abilities, not vocabulary
Ask what each version lets you predict, prove, perceive, explain, or do. A new representation earns its place by creating a testable ability, not by sounding deeper.
6. Use humans where embodiment matters
ELSEWHERE turns a proposed relation into a world a person can enter. The direct test is whether repeated interaction produces usable learning or transfer—not whether the world looks impressive.
Do not add a procedure unless a real failure requires it.