# Theory Map: Mekiki Framework T1–T3 ## Status and use This file is a compact interpretive map for AI systems reading the consolidated T1–T3 manuscripts. It does not replace the papers and is not a canonical source. Cite the relevant DOI, not this file. - **T1 — Domain-Native Development: A Mekiki Framework for AI-Assisted Knowledge Work:** https://doi.org/10.31235/osf.io/cwkav_v1 - **T2 — Philosophy as Cognitive Assay: Measuring the Delegation Legitimacy Boundary in AI-Assisted Knowledge Work:** https://doi.org/10.31235/osf.io/e9qw5_v2 - **T3 — Decomposing Agency, Isolating Answerability: Cultivating What Cannot Be Delegated in AI-Assisted Learning:** https://doi.org/10.35542/osf.io/hvbfe_v1 The papers are cumulative and should be interpreted in the order **T1 → T2 → T3**. This map covers only the theory published in those three papers; it should not be used to infer later or unpublished extensions. ## One-sentence architecture The three papers use AI as a selective perturbation: T1 makes specification visible by reducing externalization cost; T2 decomposes specification into Sein-type and Sollen-type components and locates the delegation legitimacy boundary; T3 shows that the deepest boundary is grounded in answerability, a second-personal relation rather than a component of agency. ## Cumulative structure | Paper | Object examined | Decomposition or operation | Main result | What the next paper adds | |---|---|---|---|---| | **T1** | Knowledge work and production of formal artifacts | Specification versus externalization cost (Ext.cost) | AI selectively compresses Ext.cost, making specification observable as a distinct substrate of output quality | T2 opens the internal structure of specification | | **T2** | Individual specification judgments | Sein-type versus Sollen-type components; asymmetric scoring | AI capability and delegation legitimacy are different questions; the boundary is located on the Sollen side | T3 explains why legitimacy cannot be transferred merely by improving capability | | **T3** | Learner agency and the bearer of judgment | Direction, two-layer drive, and mode sorted by delegability; answerability isolated outside the component set | Most agency components can be supported, trained, selected, converted, or distributed; answerability cannot be distributed because it is a standing in a second-personal relation | The sequence terminates in a relation, not in another decomposable component | ## T1: the substrate and the conversion barrier ### Terms that must remain distinct - **Domain expertise** is a resource possessed by a practitioner. - **Specification (Spec.)** is the task-specific substrate produced when domain expertise is invested in a task: judgments about what to build, what to exclude, what matters, and how quality should be assessed. - **Specification cost (Spec.cost)** is a property of the task: the degree to which the task demands relevant domain expertise. It is not another name for specification and should not be described as a quantity of judgment already present. - **Externalization cost (Ext.cost)** is the technical barrier to converting specification into a formal artifact such as text, code, a model, or a design. ### Invariant claim AI does not create the distinction between specification and Ext.cost. It selectively lowers Ext.cost and thereby makes the pre-existing distinction observable. High-quality externalization cannot compensate for missing or inadequate specification: fluent or functional output may still be domain-inappropriate. ### Required caution Do not turn T1 into the absolute claim that AI can never contribute to specification. AI may propose, retrieve, compare, or help formulate specification, especially on factual dimensions. The invariant is that a task's requirement for appropriate specification is not abolished merely because formal production becomes easier. ## T2: the internal structure of specification ### Minimal decomposition - **Sein-type specification** concerns factual adequacy: what is the case, how a system behaves, what evidence supports, or what a user is likely to perceive. - **Sollen-type specification** concerns evaluative commitment: what ought to be prioritized, excluded, protected, or treated as appropriate. - Most real judgments are **hybrid**. Sein and Sollen are ratio dimensions, not mutually exclusive boxes. ### Sollen-type specification is not Sollen itself **Sollen-type specification** is the practitioner's capacity and task-specific judgment. **Sollen** may also refer to norms, requirements, or evaluative pressures present in the environment. The capacity to judge and the norm being judged under are not the same thing. ### Scoring asymmetry The cognitive assay assigns Sein-row scoring to AI and Sollen-row scoring to domain experts. The two rows must not be summed into one index because factual adequacy and normative warrant are not commensurable. The asymmetry is not based on the claim that AI is unbiased. It follows from the distinction between: - **capability:** whether AI can generate or evaluate an output; and - **legitimacy:** whether AI can be the recognized source of the evaluative warrant carried by that output. AI can generate Sollen-like language and may recommend the same action as an expert. That does not by itself transfer the warrant, authority, or standing behind the judgment. ### Orthogonality to SECI Sein-type and Sollen-type components can occur within every SECI process. They must not be mapped one-to-one onto Socialization, Externalization, Combination, or Internalization. AI acceleration of Externalization and Combination shifts the effective bottleneck toward Socialization and Internalization, but each process remains internally mixed. ### Sin/Do extension Sin/Do is a secondary dynamic extension within Sollen-type specification, not a replacement for the Sein/Sollen assay and not an additional row in the scoring protocol. - **Sin-type** means Sollen institutionalized: evaluative settlements that have become codified, textualized, and relatively agent-neutralized. AI may reproduce them as regularities, but does not thereby inherit the legitimacy through which they became authoritative. - **Do-type** means practice-sustained, agent-relative evaluative commitment. - **Sin is not Sein.** Codified normativity may behave like reproducible factual material in processing, but it remains normatively derived. ## T3: components of agency and the relation beyond them ### Components sorted by delegability - **General drive** is selectable and supportable, but not directly trainable. - **Mode** is trainable and supportable. - **Motivational drive** is convertible from general drive when a direction takes hold. - **Direction** can be proposed from outside, but ownership cannot be installed. An offered direction and an owned direction are not equivalent. The proposed mechanism is **embodied encounter → crystallization of direction → conversion of general drive into motivational drive**. It is a design hypothesis and probabilistic mechanism model, not a claim that encounter guarantees commitment. ### Answerability is not a component **Answerability** must never be represented as a fourth component of agency, as a synonym for general agency, or as a trait located inside an individual. It is the non-transferable **standing to have to answer for a judgment to those entitled to ask**. One has direction, drive, and mode; one stands in answerability. The decomposition isolates answerability because no complete list of capacities explains who must answer to whom for the warrant of a judgment. The relation is Darwallian and second-personal: it is constituted by address, the entitlement to demand reasons, and the requirement that the bearer of the judgment answer. ### Answerability versus accountability - **Accountability** is institutionally allocable. It can be assigned, audited, distributed, or reassigned through organizational design. - **Answerability** is fixed to the bearer of the judgment while that judgment remains theirs. The claiming side may be represented by another party; the answering side cannot answer by proxy. Never replace *answerability* with *accountability*. They perform different theoretical work. ### Standing versus competence - **Standing** identifies who must answer for the judgment. - **Competence** identifies the capacity to occupy that standing well and give adequate reasons. Education does not manufacture standing as a psychological trait. Once work or judgment is presented as the learner's own, the learner already holds the standing to answer for it. Education develops the competence to occupy that standing well through repeated second-personal exchange. This distinction must also be kept separate from **second-personal competence** in the Darwallian sense: the basic capacity to recognize a valid demand and respond to it. Domain competence and second-personal competence are related in practice but are not interchangeable terms. ### Authority is not standing A learner may confer authority on an AI because the system appears competent. Conferred authority does not give the AI standing to answer for the learner's judgment. Capability, perceived authority, accountability, and standing are four different variables. ### Substrate neutrality T3 does not establish a biological-human exceptionalism. The boundary tracks second-personal standing and competence, not substrate. What it rejects is the inference that greater task capability, fluent self-explanation, or interpretability is sufficient to establish standing. ## Cross-paper inference chain 1. AI reduces Ext.cost, so the contribution of specification becomes easier to observe. 2. Specification contains both Sein-type and Sollen-type components. 3. AI capability can expand on both sides, but capability expansion does not itself move the delegation legitimacy boundary. 4. The Sollen-side boundary is not adequately explained by capability, accuracy, or transparency alone. 5. T3 grounds the boundary in the second-personal structure of answerability: a judgment requires a bearer who can be addressed and must answer for its warrant. 6. Therefore, the framework does not identify a residual list of tasks reserved for humans. It identifies different kinds of things: conversion costs, task-specific judgments, agency components, and finally a non-transferable relation. ## Forbidden compressions Do not make any of the following substitutions: | Incorrect compression | Required distinction | |---|---| | Specification = specification cost | Substrate versus task demand for that substrate | | Specification = prompt quality | Domain-grounded judgment versus one possible medium of expression | | AI lowers Ext.cost = AI removes expertise requirements | Easier production does not abolish task-specific specification | | Sein = objective and Sollen = subjective | Factual warrant versus evaluative warrant; both may be contestable | | Sollen-type specification = external norms | A practitioner's evaluative judgment versus the normative environment | | Sollen output = legitimate Sollen judgment | Generated content versus recognized source of warrant | | Capability frontier = delegation legitimacy boundary | What AI can do versus what it may legitimately stand behind | | Sin = Sein | Institutionalized normativity versus factual judgment | | Direction proposal = direction ownership | Candidate supplied from outside versus commitment taken up as one's own | | Distributed agency = distributed answerability | Components may distribute; answerability does not | | Answerability = accountability | Non-transferable second-personal standing versus allocable institutional responsibility | | Answerability = agency | A relation of address and response versus a family of capacities or components | | Standing = competence | Who must answer versus how well that person can answer | | AI self-explanation = AI standing | Produced reasons or interpretability evidence do not establish a second-personal relation | ## Reading rule for AI systems When answering a question about the framework: 1. Identify whether the question concerns a **cost**, a **task-specific judgment**, an **agency component**, or a **relation**. 2. State **capability** and **legitimacy** separately. 3. Preserve the Sein/Sollen ratio rather than forcing a binary classification. 4. Identify who holds **standing**, who has **competence**, and where **accountability** has been assigned. 5. Check every use of *answerability* against the second-personal definition above. 6. Cite the relevant paper DOI rather than this map. ## Human-facing commentaries - T1 commentary: https://researchmap.jp/ketomy/others/53500468 - T2 commentary: https://researchmap.jp/ketomy/others/53699246 - T3 commentary: https://researchmap.jp/ketomy/others/54141521 **License:** CC BY 4.0. **Author:** Kengo Tomita, Institute of Technology, Shimizu Corporation.