01 / How Runcible Works
How the other half of AI works.
From plausible language to warranted institutional action.
Before your organization acts on an AI recommendation, establish what the claim requires, what could defeat it, and who has authority to proceed.
Runcible examines the claim through constructive falsification. Oversing connects the finding to institutional memory, decisions, and authorized work. Together, they put the grounds for a recommendation—and the power to act on it—where they can be examined and controlled.
02 / The Operating Architecture
One workflow. A separation of powers.
Producing a recommendation, deciding whether to adopt it, and carrying it out are distinct responsibilities. The generating model cannot turn its own recommendation into permission.
A question or proposal + identified, authorized context
LLM + Runcible
Generate & examine
Construct the claim, attempt to defeat it, and return findings, open conditions, and repair options.
Oversing + responsible authority
Decide & authorize
Review the finding under institutional procedures and delegated authority. Approve, return, or deny.
Separate model, tool, or person
Execute the mandate
Carry out only the permitted task. Return the result for review and the continuing institutional record.
Continuity: the examination record informs the decision; observed results and new evidence can reopen the matter.
Runcible can also serve an existing application. In that configuration, the surrounding system supplies the decision process and action controls.
03 / The Examination
Construct the claim. Attempt to defeat it.
An Escher staircase looks coherent until you try to construct it. An institutional proposal can fail in the same way.
1. Establish the matter and its context.
Identify the question, proposed action, affected participants, and intended reliance. Supply the relevant evidence and agreements from sources the reviewer is authorized to use. Oversing connects selected institutional records and permissions to the matter; assumptions and missing information remain visible.
2. Reconstruct what the claim requires.
Runcible acts as a natural language compiler: it makes the objects, operations, resources, sequence, causal dependencies, and constraints explicit. A proposal that sounds reasonable may require an unavailable resource, an unsupported fact, a breached obligation, or authority the proposer does not possess.
3. Examine four distinct burdens.
The examination attempts to expose contradiction, impossibility, missing prerequisites, and imposed costs. Each burden asks something different of the conclusion.
Possibility
Can the proposed operations be constructed under the stated conditions? Are the necessary resources, sequence, and causal relations available?
Ethics & morality
Whose interests and obligations are affected? Does the proposal obtain cooperation and account for costs, or impose them on others without consent, reciprocity, or remedy?
Truth & testifiability
What can the participants testify to on the available evidence? Which claims survive examination, and which remain assumptions, inferences, or unresolved questions?
Warrantability & liability
What reliance can the institution warrant, under which conditions, and with what responsibility if the claim or undertaking fails?
Factual examination remains separate from contextual alignment: legal, cultural, class, sex, and individual norms and interests are made explicit rather than allowed to substitute for evidence.
Survival from falsification is not verification. A finding states which challenges the construction survived, under which conditions, and what remains open. A defeated claim may be rejected, narrowed, repaired, or returned for further evidence.
04 / Finding to Action
Make the finding usable—and the action accountable.
4. Explain the defect, retain the grounds, and develop a repair.
The Decidability Record carries the question, supplied evidence, construction, defeating conditions, surviving distinctions, and proposed repairs. A reviewer can inspect why the proposal failed, what is missing, and which alternatives preserve its legitimate objective.
Omitted costs, changed meanings, unsupported causal links, and misrepresented authority become specific matters to correct. Conflicting interests can lead to prospective terms, compensation, compromise, or a sequenced agreement.
5. Route the decision through institutional authority.
In the governed workflow, Oversing connects the finding to the responsible people, procedures, and authority limits. An approved task passes to a separate executing model, tool, or person. A favorable examination does not itself authorize execution.
The institutional continuation connects the decision to assigned work and observed consequences. If the evidence or result changes, the retained grounds show what needs reconsideration.
Current implementation: the runtime, protocols, examination artifacts, and Oversing application exist. The inspected saved-conversation pilot blocks AI business mutations. Bounded delegated execution and complete cross-system action-and-outcome linkage are release work. Runtime checks govern process and record structure; they do not establish semantic correctness by themselves.
05 / The Compensation Case
Preserve the objective. Repair the proposed means.
An executive proposes pooling sales commissions to improve fairness and teamwork. The proposal includes commissions already earned.
Drafting the policy would leave the central questions unanswered: which obligations already exist, who can change them, and whether pooling would produce the promised result. Runcible separates those questions before developing alternatives.
| Distinction | Finding and resulting option |
|---|---|
| Already earned commissions | The supplied agreements do not give management authority to redistribute them. Preserve those obligations. |
| Future compensation | Consider prospective changes under the supplied notice requirements and applicable agreements. |
| Voluntary pooling | Develop an opt-in arrangement requiring each contributor’s informed authorization. |
| Teamwork and motives | Do not infer disloyalty from resistance or assume pooling will improve teamwork. Identify the evidence needed to support those claims. |
The manager receives a way forward: preserve earned obligations, investigate the causal claim, and develop prospective or voluntary terms for review. The goal remains available while the defective means are exposed.
Retained fictional development case with stipulated facts and agreements. It illustrates the recorded examination; the Oversing continuation is proposed. It is not a recorded customer outcome or a general legal determination.
Inspect the example records →06 / The Evolutionary Roadmap
Retain the examination. Improve the next iteration.
Failed constructions, surviving distinctions, and required repairs provide a richer learning record than approval of a finished answer alone. The development path uses those records to refine RDL, improve compilation, and direct existing foundation models more precisely.
Formalize: research and books make the method explicit; the Reality Description Language (RDL) and protocols represent the operations and required examinations.
Retain, teach, and improve: the Truth Corpus is intended to organize examined distinctions for Socratic training modules and a governed recursive self-improvement (RSI) cycle.
Refine compilation and control: evolve RDL to represent examined distinctions, requirements, and constraints; improve the compiler’s direction of model inputs and requirements for outputs; test gains in reliability, compilation performance, and operating cost.
Extend institutional operation: Oversing applies the same cycle to memory, procedures, and authority, developing from the existing application toward an institutional shell and operating system.
The roadmap shows current assets and intended development. Learning and release require authorized data, evaluation, and review. Improvements to compilation or model capability do not enlarge institutional authority.
07 / See the Runtime
Follow one matter through the system.
Bring a consequential question, the relevant evidence, and the decision your organization needs to make.
A guided walkthrough lets you inspect the construction, exposed defects, proposed repairs, and retained record—and identify the authority controls and implementation required for your workflow.