Skill Resolve

Resolve

Choose the evidence method that can actually settle the question.

Resolve selects among a derivation, literature investigation, or disposable probe. The choice follows the question, existing evidence, assurance need, and cost.

For multi-step scholarly questions, available Consensus Threads receive one complete brief. Provider Pro is the default; Deep requires an explicit user request. Capacity checks, timed polling and unchanged document preservation precede independent primary-source and reception scrutiny. Missing tools retain labeled fallbacks.

For research-gap questions, its literature method builds a coverage map from the retrieved papers, labels sparse cells as potential gaps, and checks them with a validation search before proposing a focused question. An empty cell does not establish that no research exists.

Use it when: A live correctness, research, or empirical question cannot be settled by an ordinary bounded read or check.

Use something simpler when: The matter is pure preference, the direct check already answers it, or the decision is already made and the research would only decorate it.

Illustrative request

Can this queue design lose an acknowledged item during restart? Use the cheapest sufficient method to answer that specific question.

What a useful result looks like: A supported answer, counterexample, or named limit, with the relevant derivation, claim-evidence matrix, or probe record. A probe declares its question and disposal boundary before execution.

Boundary: A toy probe does not prove production behavior. A citation does not by itself verify a scientific claim. The resulting evidence informs the decision owner; it does not issue an implementation approval.

Recon maps territory; Gauntlet can adjudicate a consequential decision using the resulting evidence.

Read the canonical v7.3.0 method · All methods