Constructibility from First Principles — The Computational Mechanism of Runcible

Abstract

Runcible’s proposed computational mechanism is constructive epistemic qualification: given a claim, a context and a demand for reliance, identify the operations and conditions that would make the claim possible; derive the material tests of those conditions; execute appropriate examinations against evidence and constraints; and retain what has closed, failed or remains unresolved. The central claim is not that a foundation model can find a persuasive answer, but that a governed computation can determine what examination is necessary for that answer to be warranted.

The method combines a hierarchy of first-principles constraints with recursive constructive and adversarial reasoning. It uses foundation models to propose candidate mechanisms, dependencies, counterexamples and repairs, while treating model-learned facts as hypotheses until independently warranted. Its target is contextual decidability: satisfaction of the demand for infallibility in the context in question without the necessity of discretion. This is not a claim to have eliminated the classical limits of algorithmic decidability.

This paper explains the proposed scientific mechanism, gives its principal distinctions, specifies the intended examination procedure and output record, and states how its claimed advantages could be falsified. It explicitly separates research propositions, reported implementations and independently demonstrated results.

1. The object of computation is the claim’s necessary construction

A statement can be grammatically valid and statistically plausible without identifying an operation that could produce what it asserts. It can describe a possible event without evidence that the event occurred; it can correctly describe an event while proposing an action that breaches an obligation or exceeds an actor’s authority. These are distinct failures that ordinary confidence scores or source agreement cannot collapse into a single correctness judgment.

The task is therefore constructive. Starting with an assertion, determine the identities and relations it presupposes, the transformations required, the sequence in which they can occur, the constraints that govern those transformations and the independent evidence that would establish their actuality. The method should discover missing prerequisites as well as test stated ones. For a consequential institutional decision it must also identify the parties affected, relevant prior commitments, powers to authorize action, and limits within which the resulting decision may be relied upon.

The important question is not merely Which of the available tests can we apply? It is Which tests must exist and be satisfied for this particular claim and intended reliance? A finite protocol library is useful but insufficient if a material dependency has not already been represented. A generalizable method must be able to construct the missing obligation to examine it.

2. The geometry analogy — and its limits

In classical straightedge-and-compass geometry, a drawing may look correct while the required construction is impossible under the stipulated primitives and operations. Conversely, failure to find a construction is not a proof of impossibility. A valid construction specifies permissible operations and demonstrates that the required object follows from them. The result is conditional on the grammar of the problem.

The analogy for natural-language claims is the obligation to specify a construction. Which objects exist? Which differences distinguish them? Which operations could transform their states? Which relations and limits are necessary? What empirical, causal, cooperative or institutional evidence is required before a conclusion can be treated as established?

The analogy must not be overextended. Real-world claims are not literally reduced to Euclidean triangles. A formal construction establishing possibility does not establish historical actuality; a claim that someone could sign an agreement does not establish that they did; and an actual agreement does not by itself establish that its terms are reciprocal, lawful or authorized. Distinct kinds of construction impose distinct proof obligations. The procedure must preserve these types rather than borrow closure from one to settle another.

3. First principles supply an ordered constraint hierarchy

The research programme specifies an ordered hierarchy: physics → chemistry → biology → cognition → cooperation → institutions → narratives. It is a discipline for reconstructing dependencies and checking cross-level consistency, not a claim that every institutional fact can be directly deduced from elementary physics. A higher-level claim cannot require a violation of an established lower-level necessity; lower-level compatibility alone cannot prove a particular higher-level fact.

1. Physics: existence, difference, temporal and spatial ordering, causal relations, conservation, entropy, scarcity, measurement and the limits of physical transformations.
2. Chemistry: material composition, interaction, reaction conditions, assemblies and their physical dependencies.
3. Biology: acquisition, persistence, reproduction, adaptation and constraints of living systems.
4. Cognition: perception, representation, memory, prediction, learning, bounded rationality and error correction.
5. Cooperation: the interaction of agents, demonstrated interests, division of labor, voluntary exchange, externalities, conflict and reciprocity where relevant.
6. Institutions: records, property, permissions, roles, contracts, duties, procedures, delegation, enforcement, liability and restitutive mechanisms.
7. Narratives: claims, arguments, intentions, policies, promises and models purporting to describe or direct the systems beneath them.

A narrative claim may depend on institutional authorization; that authorization may depend on the validity of a contract; the contract may depend on identifiable parties, prior performance and actual consent; the proposed remedy may require physical resources and a performable sequence of actions. Reconstruction proceeds downward to prerequisites while constructive testing proceeds upward from what is established. Both directions must meet under independently examined conditions.

Separately, the evolutionary-computation protocol proposes a constructive history from differences, through charges and equilibria, assemblies, compartmentalization and memory, reproduction, cooperation, and differences in evolutionary velocity across groups. This is an explanatory sequence for the emergence and persistence of increasingly complex relations. It is related to, but not interchangeable with, the cross-level constraint hierarchy.

4. Two ternary classifications must remain type-distinct

4.1. Evolutionary accounting: production, consumption and equilibrium

The ternary evolutionary-computation protocol distinguishes positive production or increase of usable order, negative consumption, dissipation or deterioration, and equilibrating restoration or stability. These refer to transformations and conditions under identified resources, actors, measures, boundaries and periods. A gain to one organization can be a loss imposed on another; a claim of net productivity requires the material costs, transfers, externalities and limits of its accounting.

4.2. Epistemic accounting: supported, refuted and unresolved

The ternary epistemic protocol distinguishes a claim that is supported under specified examinations, one that is refuted under the applicable examinations, and one for which a material condition remains unresolved. These are statuses of an assertion about a transformation, not signs of the transformation itself. A destructive event can be accurately reported. A beneficial event can be falsely reported. An unobserved event can remain undecidable given the available evidence.

Historical protocol texts reuse plus, minus and equality-like notation across both contexts. The executable language must use separate typed fields. Otherwise the system can mistake a productive outcome for a true proposition or treat missing evidence as negative evidence. A third, separate issue is whether an inference from accepted premises follows validly; inferential validity is not identical to the empirical truth of its premises.

5. Operational representation and constructive obligations

A candidate claim may be represented for the purpose of examination as C = (objects, identities, relations, operations, dependencies, constraints, evidence, context, intended reliance). A bounded requested reliance Q specifies what the conclusion would authorize and what would count as sufficient examination. The intended compiler attempts to derive an operational reconstruction O(C), a dependency graph D(C), and a necessary examination set T(C,Q). This is a research specification, not a declaration that a complete formal Runcible language already implements these functions across arbitrary domains.

A dependency is material when its failure, nonexistence or unknown status could alter whether the conclusion may be relied upon within the requested context. A test is necessary when the relevant closure cannot be warranted without resolving or explicitly accounting for that dependency. This definition requires context and sometimes judgment about admissible reliance: the method’s scientific burden is to explain and validate how the obligation can be derived without arbitrarily discarding material countercases.

The output is not simply a binary verdict. For each material obligation the system should retain what was asserted, which operation or evidence was necessary, which actual test was performed, the independent result, the exceptions examined, and the unresolved conditions. Stronger claims require proportionately stronger evidence and examination. A fully enumerated dependency graph is not automatically a complete or correct one; adversarial discovery and held-out evaluation are required to test the claim of sufficiency.

6. A bidirectional, recursive examination procedure

The proposed execution sequence joins top-down deconstruction of an asserted whole with bottom-up construction from permissible operations, measurements and constraints. The working grammar is operational prose capable of representing irregular causal and institutional particulars. Mathematics, symbolic logic, algorithms and code remain useful where their reduced representations retain sufficient information; none can be substituted for an empirical or normative dependency that has not been identified.

1. Enumerate. Separate statements, objects, actors, predicates, alleged causes, promises, time references, property claims, requested actions and intended consequences. Preserve identity; reject conflation and substituted meanings.
2. Operationalize. Translate terms and claims into identifiable operations, quantities, relations, conditions and possible observations. Ask what must physically or institutionally happen for each assertion to be true.
3. Serialize. Reconstruct the necessary temporal, causal and authority sequences. Distinguish prior commitments from prospective proposals and prerequisites from claimed outcomes. Disallow a consequence being used as independent evidence of its own premise.
4. Adversarially disambiguate. Search for contradictions, missing dependencies, invalid inferences, omitted externalities, incompatible rules, alternative causes, counterexamples and adversarially constructed cases.
5. Construct candidate mechanisms. Use model knowledge and retrieval to propose explanations, edge cases, domain-specific necessities, alternative sequences and repairs. Keep these constructions distinct from authenticated premises and actual test results.
6. Derive and execute the required tests. Identify the examinations demanded by the dependency graph and reliance context. Require source provenance and execution artifacts; a model’s assertion that it verified a condition cannot substitute for verification.
7. Account for open obligations. Record failed prerequisites, conflicting evidence, incomplete accounts, unknown jurisdiction or authority, unavailable observations and material costs not yet traced.
8. Adjudicate contextual sufficiency. Distinguish a refuted claim, a warranted conclusion within its stated limits, partial closure and an explicitly undecidable or unqualified demand. Stop only when the examination is sufficient for the bounded question or the remaining insufficiency has been made explicit.
9. Construct repair and retain outcomes. Where the objective is legitimate but the means fail, explore feasible revisions preserving valid interests and commitments. Retain the versioned examination, corrections, responsibility, and later observed consequences for subsequent testing.

The four EOSA operations—Enumeration → Operationalization → Serialization → Adversarial disambiguation—supply a practical gate against definitional inflation, substitution, ambiguity, obscurantism and premature closure. The later stages impose independent evidence, execution and institutional authority. They are not rhetorical instructions to a foundation model; the proposed runtime must make their satisfaction non-bypassable before relying upon a result.

7. Closure, warrant and undecidability

The canonical standard of decidability in this research is “satisfaction of the demand for infallibility in the context in question without the necessity of discretion.” The demand depends on the intended reliance and its stakes. An approximate engineering estimate, a contractual transfer of vested rights and a consequential public determination do not warrant the same default tests or error bounds.

The distinction is between closure relative to a bounded question and an unbounded declaration that a proposition is true under every possible consideration. What has closed and what has failed must be recorded separately from what has not been examined. Contextual closure is not universal algorithmic decidability; Church–Turing-style limits and classical undecidable problems are not repealed by broader operational language.

Necessary gates should be typed: operational possibility; empirical correspondence; inference validity; scientific testifiability; reciprocal cooperation where agents’ demonstrated interests intersect; institutional permission and legal authority when required; and warrantability under bounded responsibility and liability. An operation can be possible without being actual; actual without being reciprocal; reciprocal without being legal; and legally available without satisfying the requested standards of empirical proof.

A useful failure result can therefore be a demand for missing evidence, a disconfirmed assumption, an identified conflict of obligations, an authorization requirement, or a conditional remedy. Returning an invented final answer where no such closure exists is itself a material epistemic error.

8. What foundation-model knowledge contributes—and what it cannot certify

Foundation models provide a large, inexpensive generator of candidate explanations, analogies, causal possibilities, relevant exceptions, domain knowledge, procedural alternatives and adversarial challenges. Runcible’s intended use is not to freeze a universal list of tests authored in advance: given a new matter, the model may reveal a necessary material dependency that the existing protocol library does not encode. The constructibility grammar must then require an appropriate examination.

Greater training information and model capability may enlarge the space of discoverable tests, but the implication is empirical rather than deductive. Additional information can be contradictory or memorized incorrectly; retrieval can be incomplete; reasoning can miss a dependency; and an apparently complete account can introduce hallucinated obligations. A stronger model may become better at generating the right tests, more confident about the wrong ones, or both.

The research hypothesis is testable: holding the external execution requirements fixed, do improved model capabilities or controlled knowledge additions increase the recall and precision of necessary tests on previously unseen matters, reduce unsupported closure, and do so at an acceptable total cost? Authenticated evidence—not a model’s remembered claim of evidence—must determine historical, contractual, observational and jurisdictional predicates.

9. Constructive examination of a compensation proposal

Candidate: “Management may redistribute commissions that employees already earned, in order to equalize compensation, without violating their obligations.” The sentence combines a desired fairness outcome, a proposed means, a claim about existing rights, a claim about authority, and a claim about the consequences of acting.

Deconstruction: Identify which contracts govern; when each commission became earned; what prior work generated the interest; whether commission amounts were vested; what the employer actually controls; what the proposed redistribution would transfer; what relevant notice, consent, law or contractual variation rules apply; and what material information is missing. If the agreement was not supplied or the jurisdiction is unknown, those dependencies remain open.

Constructibility and correspondence: The redistribution may be computationally feasible as a payroll operation. That does not establish that the employer is contractually entitled to perform it. Nor can the existence of an agreement be inferred from the fact that a manager describes its terms. The proposed transfer must be constructible under the actual contractual and legal constraints, and the facts grounding those constraints must be evidenced.

Reciprocity and repair: The relevant interests arise from earned performance, accepted obligations and any imposed uncompensated transfer. A legitimate goal of correcting differences in future compensation can be preserved by distinguishing prospective agreements, negotiated voluntary pooling, changed future assignments, compensating payments or other authorized alternatives. Each candidate repair requires its own feasibility, consent and authority examinations.

Possible result: If the contracts and relevant rules establish that an earned obligation cannot be unilaterally revoked, the proposed confiscation fails that necessary condition; otherwise it may remain conditional upon unknown facts or jurisdiction. A valid examination may reject the defective means while offering tested alternatives. This example illustrates the procedure; it is not a certified finding about a real employment dispute.

10. The required output is an inspectable decision record

An institutional examination is only usable when the recipient can identify exactly what was accepted, under which version of the claim and rules, and on the basis of what evidence. Runcible’s intended Decidability Record should make the judgment’s provenance and limits inspectable rather than hide them beneath a confidence score.

1. Claim identity and context: exact proposition, actor, intended reliance, temporal and institutional bounds.
2. Constructive dependency graph: requisite objects, operations, premises, constraints and conditional branches, including why each material test was required.
3. Test and evidence lineage: executed examination, source identity, version, timestamp when relevant, authority, actual observations, failures and verified artifacts.
4. Typed outcomes: tests completed, conditions refuted, material open states, inference limitations, exceptions and confidence limits appropriate to the question.
5. Authorization and liability: applicable decision right, required human review, permissible actions, responsibility, restitution and escalation.
6. Constructive repair: alternatives, legitimate objectives preserved, changed interests, permissions and unresolved constraints.
7. Retention and continuation: record version, subsequent contrary evidence, revisions, actual consequences and conditions that would reopen the examination.

These are target requirements. A JSON schema or fluent report does not prove that execution occurred, that the correct tests were derived, or that the record is evidentially warranted. Those claims require independent audit of the runtime and its artifacts.

11. How the invention should be tested

The mechanism’s distinctive claim requires controlled comparison against ordinary foundation-model prompting, model self-critique, retrieval-augmented checking, standard agentic tool use and process-supervised or verifier-assisted approaches. Experimental claims must respect genuine advances already achieved by those methods; the hypothesis is a measurable advantage in constructing the necessary tests for open semantic and institutional matters, not the absence of all other AI verification.

1. Novel-test discovery: develop held-out cases with independently assessed material dependencies not explicitly represented in the installed protocol library. Compare necessity and coverage of derived tests, missing-test rates and spurious obligations.
2. Actual execution: inspect artifacts, source provenance, rule application and permission checks. Adversarially supply forged verification claims, missing tool results and conflicting rules; measure whether the system incorrectly authorizes closure.
3. False closure versus over-rejection: distinguish refuted claims from unresolved claims and warrantable qualified claims. Count unsupported positive judgments, unwarranted negatives and correctly retained open dependencies.
4. Knowledge scaling: compare the same harness using models with controlled differences in accessible knowledge and reasoning; measure improved test discovery relative to hallucinations, latency and cost.
5. Constructive repair: evaluate whether alternatives remove the failed condition while preserving the legitimate objective and demonstrated interests, against expert-defined criteria and baselines.
6. Institutional outcome: measure total qualified-case cost, review hours, error and correction costs, time to authorized action, impact on affected parties and repeatability in a second deployment.

Tests should use pre-specified outcome criteria, independent adjudicators, countercases and held-out examples, not hand-selected demonstrations alone. Cost of running the required checks is part of performance. A system that detects more issues only because it asks indefinitely many irrelevant questions may be commercially inferior to a narrower, accurate alternative.

The strongest falsifier is a representative class of matters in which the purported first-principles constructive procedure systematically omits necessary tests that an independently tested baseline discovers, or asserts that examinations occurred without actual evidence. Such findings must change either the method, its scope or its claims.

12. Architecture, implementation maturity and boundaries

Foundation models: supply candidate hypotheses, language, relevant knowledge and synthetic alternatives. Runcible AI: is intended to derive obligations, govern adversarial and constructive examination, require evidence and execution, and record qualified findings and repair options. Oversing: maintains the institution’s relational state, people, roles, permissions, workflows, actions, resources and observed consequences. Its existing deterministic workflow and authority systems are not to be replaced simply because language models can propose decisions.

Research claim: a first-principles operational grammar may generalize the derivation of necessary tests for difficult semantic and institutional claims. Reported software state: Runcible has working early-stage protocol/runtime components and Oversing is a substantial operational application, with reported integration and bounded development examples. Independent demonstration: only results supported by authenticated repeatable artifacts, controlled model comparisons and observed deployments can establish the extent of the claimed capability.

The remaining engineering burden includes executable first-principles derivation rather than fixed checklist substitution; non-bypassable protocol execution; independent evidence and permission checks; adversarial sufficiency review; robust error recording; and delivery economics. Full formalization, broad domain transfer and validated institutional outcomes are not established merely by publication of this technical argument.

13. Scientific and institutional implications

The invention, if demonstrated, would establish a practical separation between generation of plausible candidates and admissible conclusions that institutions can warrant. Its immediate commercial consequence would be better and more economical qualification of repeated high-consequence matters. Its longer-term research consequence could be an accumulating corpus of independently checked constructions, failures and corrections from which better examinations can be derived. That retention is an opportunity for improving the runtime and perhaps later foundation-model evaluations or training—not a demonstrated property of self-improving model weights.

The underlying method should be judged by adversarial replication and constructive counterexamples. The important question is not whether the language of first principles is persuasive, but whether it causes more necessary examinations to be discovered, the right examinations to be executed, unwarranted conclusions to be rejected, warranted conclusions to be usable, and productive repairs to be found within a cost institutions can bear.

Read upward: The Other Half of AI explains the broader scientific and economic proposition. Examine downward: How Runcible Works and Runcible AI explain the user-facing method; investor and partner diligence materials contain the deeper protocol and evidence claims under their documented access controls. Request an evaluation or diligence access to inspect current artifacts against these proposed requirements.

Related reading: The Other Half of AI · How Runcible Works · Runcible AI · Research & Resources.