Load-bearing assumption test

Contract

Field	Bound contract
Trigger	User says "tell me why this is wrong", or "poke holes in this" about a plan, design, or claim
Authority	Read-only. No file, VCS, credential, paid, published, deployed, or remote mutation.
Side effect	Chat output only: a single root objection and first-nail recommendation as { root, first_nail }; no files written
Done	Root is genuinely load-bearing and the first nail is the cheapest experiment beating the plan or null when none exists

Inputs

The plan, design, or claim under attack, supplied by the user in the conversation. No file or remote input is required; the skill reads only what the user presents.

Procedure

1. Pin the load-bearing assumption: identify the single thing that must hold for the whole plan to stand. Done when: one assumption is named whose failure collapses the plan.
2. Attack on whatever axes apply: a fact that may be false, confabulation, analogy mistaken for isomorphism, a future-tense suture, or the sharpest one, a principle cited but its opposite implemented. For empirical plans, also leakage and statistical power or family-wise error. Done when: every applicable axis has been tested against the pinned assumption.
3. Collapse to one root: the single objection whose failure makes the rest moot. Not a list. Done when: one root objection is selected, not a register of objections.
4. Find the first nail: the cheapest falsification available before the expensive program runs. Done when: one experiment is named that is cheaper than the plan and would prove whether root matters, or null is returned when no such experiment exists.
5. Return { root, first_nail }. Nothing else. Done when: the pair is returned and no additional objections or scaffolding are appended.

Failure and recovery

• No load-bearing assumption found: state that the plan has no single point of failure and return the sharpest objection found, labeled as non-root. Do not fabricate a root.
• No cheaper first nail exists: return the root with first_nail set to null and state that no falsification is cheaper than running the plan. Do not invent an experiment.
• Never widen to a risk register or a list of objections; that is a different contract. If the user asks for more than one objection, stop and name the scope mismatch.
• Never swallow the absence of a root by pretending the done predicate holds.

Output

A single { root, first_nail } pair in chat: root is one load-bearing objection whose failure makes the rest moot; first_nail is the cheapest experiment that would prove whether root matters, or null when no falsification is cheaper than the plan.