Resources / Start here

Runcible, in Plain Language

From a plausible answer to a decision an organization can responsibly act on.

Revised 10 November 2026

In this article
  1. Start with a decision people recognize
  2. Four questions before reliance
  3. Why we call it “The Other Half of AI”
  4. Oversing connects the answer to the organization
  5. The organization keeps what it learns

An organization can obtain an AI answer in seconds. Deciding whether to rely on it can still require a lawyer, a domain expert, an operational manager, and someone willing to accept responsibility. Runcible and Oversing address that gap between an answer and an accountable result.

Runcible examines what the organization can responsibly conclude. Oversing governs what the organization may do about it.

Start with a decision people recognize

A manager wants to pool sales commissions to encourage teamwork. An AI can quickly draft the new policy. But a polished policy leaves the consequential questions unanswered: Which commissions have already been earned? What did the contracts promise? Who can change future terms? Does pooling actually improve the behavior the manager wants?

Those questions belong to different kinds of examination. Calculating the new payment amounts does not establish a right to change them. The manager’s approval does not establish the claimed effect on teamwork. A desirable objective does not settle who bears the cost.

Runcible breaks the proposal into those claims and dependencies. It examines the relevant evidence and constraints, identifies what fails or remains unknown, and develops alternatives that preserve the legitimate objective. A prospective arrangement with the necessary agreement and approval may be a candidate repair. It must be examined in its own right.

Four questions before reliance

  • Possibility: Can the proposed operation be constructed with the available people, resources, sequence, and conditions?
  • Ethics and morality: Whose interests, obligations, costs, and risks are affected? Does the proposal depend on deception, an uncompensated imposition, or a concealed transfer?
  • Truth and testifiability: What can be testified to from the evidence? Which claims survive examination, and which remain unsupported?
  • Warrantability and liability: What reliance can a responsible party undertake, under which limits, with what remedy if the conclusion fails?

A favorable result in one category cannot silently answer the others. A feasible action may still violate an obligation. An accurate fact may still be confidential. A permitted action may still rest on a false premise.

Why we call it “The Other Half of AI”

Generative models supply candidate language, interpretations, plans, and explanations. Runcible supplies constructive falsification: it makes a claim explicit enough to expose the conditions that could defeat it, then examines those conditions.

Construction is part of the test. Naming the objects, operations, resources, and dependencies can reveal an impossibility or a missing prerequisite before a proposal reaches execution. Surviving the required challenges supports a bounded conclusion; it does not turn an unanswered question into a fact.

This also separates truth from alignment. A company’s goals, a jurisdiction’s rules, or an individual’s preferences can affect what may be done or communicated. They do not change whether a factual claim corresponds to the evidence. The system keeps those judgments distinct.

Oversing connects the answer to the organization

Oversing is the enterprise operating platform: a shared system for strategy, people, customers, commitments, work, resources, accounting, permissions, and institutional memory. A question can therefore be examined against the organization’s actual roles, agreements, budgets, tasks, and history.

The architecture separates three responsibilities:

  1. Generation and examination: The generating model and Runcible produce and test candidates.
  2. Decision and authorization: Oversing applies the organization’s procedures and authority limits, with human validation or an explicitly bounded delegation.
  3. Execution: Oversing issues the authorized task to a separate executing model or tool. The generating model cannot enlarge its own authority by rewriting a recommendation.

In the compensation example, a qualified prospective policy is still a proposal. The authorized people must approve it; the relevant agreements and effective dates must be recorded; the payroll change must occur within the approved mandate.

The organization keeps what it learns

The Decidability Record preserves the claim, the examination, evidence, failed conditions, possible repairs, and remaining questions. Oversing connects that record to the decision, action, and observed consequences.

That gives the next person more than a previous answer. They can see why it held, what would reopen it, and what changed after execution. Reviewed cases can improve protocols, RDL, and compiler behavior over time. The immediate improvement path is better examination and cheaper, more controlled use of existing foundation models.

The practical objective is an organization that can adapt continuously: discover a problem, understand its grounds, authorize a repair, carry it through, and retain the result.