# Chapter 3 — Philosophical Framework --- ## 3.1 Introduction Ontology asks what exists and how it is represented. Bunge and Wand and Weber provide a source for assessing representations in information systems [CITE:Bunge_ontology] [CITE:Wand_Weber_ontology]. Epistemology concerns knowledge and its grounds; provenance models provide one operational vocabulary [CITE:W3C_PROV2013]. Praxeology concerns action and practical reasoning, for which systems engineering offers methods [CITE:INCOSE2023]. Axiology concerns values, purposes and consequences, including professional obligations [CITE:APEGA_RWO2021]. These are related questions, not a sequence of mandatory forms. **Normative commitments** state values or obligations. **Design hypotheses** propose means of serving them. **Empirical findings** report what examination establishes. A coherent design still needs evidence of its performance. --- ## 3.2 The Four Pillars ### 3.2.1 Ontology — What Exists in the System The ontological commitment is to preserve distinctions that matter to the work. A requirement, an assumption, a physical model, a proposed operation, a result and an acceptance are different things. Confusing them can authorize an action that nobody intended or make a result appear better supported than it is. Three levels must remain distinct: | Level | Example | What examination must establish | |---|---|---| | Work or domain entity | A support, an intended deliverable, a required input | What it is, how it relates to the purpose, and which constraints apply | | Representation | A model object, ScopeOfWork, dependency declaration or graph view | What it denotes, what it omits, and whether it remains faithful | | Implementation convenience | A lock, cache, index or temporary file | What operation it supports and whether its failure affects the work | Domain representations should express the work faithfully [CITE:Bunge_ontology] [CITE:Wand_Weber_ontology]. Locks and caches serve implementation needs; their presence does not give them authority over domain meaning. The filesystem is a shared working medium, not the ontology itself. In current repository practice, a deliverable folder brings together commitments, design and declared needs. Tools derive relationships from those sources. The folder can misrepresent the work; the graph can omit a dependency. A query must report its basis and unresolved inputs rather than treat successful parsing as proof of suitability. Product engines may own domain state under their applicable contracts, so not every authoritative fact is a Markdown file. The tree answers what belongs within a larger undertaking. The dependency graph answers what one contribution needs from another. A network can expose other relevant connections, such as a common source or decision. These are views of the work and need not be separately authored stores. Attention is the activity through which participants relate these views to the undertaking. A missing edge cannot establish independence: shared writes, resources and interfaces may still connect apparently separate tasks. The distinction between descriptive and constitutive records also requires care. A stress result describes a state of affairs and can be wrong. An approval record may be part of an institutional act if the applicable process gives it that role. Its formal presence alone does not establish that the actor had authority, adequately examined the subject or engaged with its consequences. A recorded decision and a well-founded decision remain different claims. ### 3.2.2 Epistemology — What Can Be Known, and How Fluent output is not self-certifying. A numerical value and plausible citation can look credible while being unsupported. Retrieval and other techniques can improve grounding, but the presence of a source does not establish that it supports every inference made from it [CITE:Ji2023]. The framework therefore asks that material grounds and their limits be inspectable. A source can support a definition without supporting a numerical formula extracted from it. A test can establish specified behaviour without establishing fitness for every use. Agreement among agents can show consistency with a shared premise while leaving that premise unexamined. #### The Ontology of the Epistemology The framework uses six concepts to expose these distinctions. They are tools for inquiry, not an exhaustive state space of knowledge. | Concept | Meaning | Limit | |---|---|---| | Claim | An assertion about a subject | A clear assertion may still be false | | Warrant | Relevant grounds: a source, observation, calculation, test or argument | Confidence and fluency alone do not establish adequacy | | Status | A scoped classification useful to examination | A label is not its supporting evidence | | Gap | Missing information or grounds whose consequence matters | Recording a gap does not resolve it | | Conflict | Incompatible claims requiring examination of sources, scope or meaning | Some conflicts can be resolved by inquiry; others require a choice | | Ruling | A decision by the relevant authority | Authority can settle a choice, not manufacture factual support | FACT, ASSUMPTION, PROPOSAL and TBD are useful conventions where a consumer needs them. They must not imply that a cited statement is universally true or that an adopted assumption has become an observation. Nor does every routine implementation proposal require a new owner decision. Its authority depends on the assignment, commitments and reserved choices. Provenance, explicit uncertainty, conflict surfacing and examination can make unsupported claims easier to find. A source-field validator detects a missing field; assessing the source's faithfulness, relevance and sufficiency requires inquiry. An instruction against invention expresses a norm and needs examination of the resulting behaviour. The representations serving these purposes remain revisable; [Appendix E](appendix_e_development_history.md) records their development. #### Grounds, Examination and Reliance Three relationships support the inquiry: | Question | Relevant relation | |---|---| | What supports this claim? | Claim to evidence, with scope and limits | | What examination has occurred? | A subject and revision to a review method, examiner and findings | | What reliance has been accepted? | An accountable person to identified content, purpose and conditions | These relationships can change independently. A claim can acquire contrary evidence after review. A technically supported result may have no accepted use. A person can accept reliance under stated uncertainty while that uncertainty remains unresolved. Authentication changes normative standing; it adds no experimental observation. Review is inquiry into adequacy for a purpose. A content hash can identify what was examined or accepted where that binding is required. Assessing the meaning, quality and sufficiency of examination requires its grounds and findings. The person's committed engagement remains distinct from the review record. ### 3.2.3 Praxeology — How Work Is Done The praxeological question is how intention becomes a usable result through investigation, production, examination, repair and integration. Responsibility must persist across these activities. It need not be divided among agents at every boundary. The current roles describe contributions. Agent 0 holds purpose and continuity with the owner. HELPS_HUMANS contributes design. WORKING_ITEMS owns an implementation undertaking and integration. TASK executes a bounded assignment and does not delegate. Except for Agent 0's continuing responsibilities, the undertaking determines which roles are useful. One session may carry the work without delegation. Responsibility, staffing and permissions are different. A role identifies a contribution; an assignment identifies the present objective and limits; a host supplies actual tools and enforces some permissions. The role name alone cannot grant tools or impose a universal read-only rule. Written boundaries and mechanical containment must be described according to what they actually enforce. One coordinator owns a shared result. Independent work need not wait behind an unresolved branch. Shared writes, interfaces, resource limits and review capacity can nevertheless justify sequencing. The graph identifies declared relationships; participants must examine whether apparent independence is real. Owners continue through ordinary repairs. A finding changes the relevant action, not necessarily the staffing. Existing evidence remains useful unless a change, contradiction or uncovered concern invalidates it. An uncertain outcome requires inspection before an action is repeated. A separate approval is needed only where authority or an applicable condition requires it. This practice preserves explicit holds and owner-reserved commitments. It removes the inference that accountable work must have a gate at every phase or a separate record at every handoff. The method must support progress, correction and recovery within authority, with costs proportionate to the undertaking. ### 3.2.4 Axiology — What the System Values Professional responsibility and public welfare remain central commitments. Agent capability does not authorize an agent to certify, seal or issue professional work for reliance. Applicable professional obligations remain with the person who undertakes them. The axiology also includes useful attention, maintainability, user agency and timely delivery. These are not costs external to accountability. A register can bury a consequential decision. A repeated approval can train a person to respond mechanically. Unnecessary checking can displace examination of a more consequential uncertainty. Removing such a control can improve both throughput and responsible engagement. A control is justified when its contribution to understanding, reliability or appropriate authority warrants its attention, delay and complexity. That is not permission to disregard a required condition. It is a basis for choosing and revising the mechanisms that meet the condition. Silent numerical error warrants a different examination from a reversible wording change. Axiological inquiry distinguishes applicable obligations and constraints, uncertain premises, negotiable preferences and the costs of control. An obligation must be identified in its actual scope; a preference for faster work cannot simply outweigh it. An uncertain load assumption concerns the grounds of an evaluation, rather than a value to trade away. Preferences among viable alternatives can concern maintainability, access, cost or user agency. Checks and approvals themselves consume time, attention and implementation effort. Those costs belong in the assessment of how a condition is met. These considerations need no universal scalar value function or mandated sequence. They can conflict, and their interpretation can change through use. Agents can expose consequences and compare practical alternatives. The person judges which purposes, constraints and tradeoffs they can stand behind, including when the question or available alternatives need revision. Values can be expressed in requirements, defaults, interfaces and prohibitions. Those expressions can guide action without becoming the person's commitment. The framework cannot mechanize the person's answerability by encoding a value function. It can make tradeoffs visible, preserve their consequences and help the person respond to them. --- ## 3.3 The Fractal Property The *fractal property* names the recurrence of the four questions at different scales. A numerical result, a deliverable, an application and the framework itself can each be examined for their entities, grounds, course of action and values. The term describes this recurrence rather than a mathematical fractal. A ScopeOfWork or Design document can answer several of these questions, while tools derive other useful views. The choice of representation follows the consumer's need. [Appendix E](appendix_e_development_history.md) records the particular document arrangements explored in developing this approach. | Scale | Example of a useful distinction | |---|---| | Numerical result | A passing example does not establish accuracy outside its tested range | | Deliverable | Implementation completion does not establish owner acceptance | | Application | An available operation does not establish authority to apply it | | Framework | A coherent mechanism does not establish that its benefit exceeds its cost | Repeating the questions can expose a neglected relationship. Repeating the forms without that purpose can create administrative work. Recurrence is a design resource, not proof of completeness or a requirement for four documents. --- ## 3.4 Mutually Corrective Perspectives Epistemology receives particular emphasis because plausible agent output can conceal weak grounds. Adequacy nevertheless depends on all four perspectives. Evidence matters in relation to an intended use; distinctions matter through what they enable people to understand and do. The four perspectives constrain one another: - Purpose and consequences determine which uncertainty is worth investigating. - Ontological distinctions identify the subject of a claim or operation. - Evidence can expose a flawed distinction, assumption or intended use. - Practice determines whether inquiry, correction and acceptance can affect the actual result. - Experience of control costs can require a different method of examination. Epistemology thus has a particular emphasis in this study, while adequacy depends on all four. Inspectable warrant is valuable because of what people need to do and stand behind. It must be connected to a subject and a workable course of action. No pillar alone establishes suitability for professional practice. Productivity and accountable reliance can support each other. Redundant records can impede both; useful checks can improve both. The design question is which arrangement enables useful work with adequate evidence and appropriate authority, not how much administrative structure demonstrates concern for accountability. --- ## 3.5 The Pillars as the Ontology of Professional Accountability The four pillars offer a compact evaluative framework. They are not an exhaustive ontology of everything a professional can perceive or a proof that every accountable system must adopt Chirality's vocabulary. | Question | A failure it can expose | |---|---| | What kind of thing is this? | Treating a proposed model change as an applied result | | What supports the claim? | Relying on a passing check beyond the claim it tested | | How does this become usable work? | Leaving completed contributions without integration | | What matters and to whom? | Improving a local metric while reducing user control or safety | Consider an agent proposing a support change in SWBPIPE. The physical model, proposal, applied operation, analysis result and acceptance must remain distinct. The result depends on identified inputs, methods and assumptions. The operation needs an actual route through validation and application, including refusal of stale inputs and recovery where supported. The person retains applicable engineering and reliance decisions. These questions guide the design without selecting a unique API, file format or process topology. Systems engineering supplies methods that can serve those questions [CITE:INCOSE2023]. Their selection is a design hypothesis to examine in use. A failed recovery can reveal that the outcome model was incomplete. A new risk can require a different test. A more capable agent can make an old supervision step unnecessary. These revisions can improve coherence because the framework is mutually corrective, rather than a one-way derivation from philosophy to code. --- ## 3.6 The Chirality of Knowledge The project name identifies one explanatory claim: there is a permanent accountability gap between externalizable information and accountable knowing. The metaphor is deliberately bounded. It does not supply a geometry of knowledge, divide knowledge into a fixed number of parts, or make every duality in the architecture chiral. ### 3.6.1 Information and the Knower Information can be recorded, transmitted, copied, cited, compared, and organized. Knowledge is a situated achievement of a knower. Without a knower, information is not known as knowledge. The thesis does not impose the usual factive condition that only true belief can count as knowledge: a person may know wrongly, incompletely, or provisionally and later revise what they know. Professional evidence and review matter partly because knowing can be mistaken. This position separates two questions that are easily conflated: 1. **What does a person know from the information?** This is situated in the knower and may differ with context, purpose, experience, and time. 2. **What reliance has an accountable actor accepted?** This is recorded by a scoped act bound to identified content. Evidence and review may discipline what a person knows. Authentication answers the second question by evidencing the actor's attributable acceptance of reliance; it neither creates knowledge nor establishes reality as it ultimately is. ### 3.6.2 The Accountability Gap and Operational `Gap` The permanent **accountability gap** is the non-identity between information and accountable knowing. No artifact, provenance chain, semantic model, or approval record is identical with a person's knowing. More information may change what a person knows; it does not remove the need for a knower or make the resulting knowing universal. The operational primitive `Gap` has a narrower meaning. It records that a warrant has not been found for a claim. An operational `Gap` is remediable: a source may be located or a claim revised. A ruling may authorize action under uncertainty without supplying the missing warrant. The accountability gap is not missing evidence and cannot be closed by another citation. Keeping the terms separate prevents a permanent feature of knowing from being mistaken for a workflow defect. ### 3.6.3 Configurational Multiplicity Identical information may occasion different knowledge in different knowers, or in one knower under different contexts, purposes, or times. This thesis calls that openness **configurational multiplicity**. The phrase does not claim that every interpretation is equally good or that evidence is irrelevant. It says that information underdetermines the situated knowing it may occasion. Chirality's schemas, Knowledge Types, semantic lenses, and knowledge graphs are therefore scaffolding rather than exhaustive categorizations. They make important questions and relationships inspectable, support comparison, and focus professional review. They do not define a closed state space containing everything any knower can perceive in the information. A semantic model can organize a work product for a stated purpose; it does not legislate the limits of knowledge. Authentication stabilizes one accountable relation within this multiplicity. An identifiable actor binds acceptance to identified content or SHA, scope, and purpose. That act gives the information accountable-reliance status within the stated relation. Another knower may know something different from the same information, and the authenticating actor may later revise their own knowing through a new attributable act. ### 3.6.4 Philosophical Precedents and Limits The framework draws resources from six philosophical precedents without claiming exact correspondence or derivation. **Niels Bohr's complementarity** [CITE:Bohr1958] provides a precedent for resisting the demand that every adequate account collapse into one description. The thesis borrows only that restraint. It does not import a physical theory into epistemology. **Michael Polanyi's personal and tacit knowing** [CITE:Polanyi1958] [CITE:Polanyi1966] and his later work with Harry Prosch [CITE:Polanyi1975] support the priority of the knower, the personal contribution to knowing, and the fiduciary character of professional commitment. Polanyi is a principal resource for understanding why knowing cannot be exhausted by explicit information. The framework does not claim to formalize Polanyi exactly. **Brian Cantwell Smith's registration and answerability** [CITE:Smith1996], together with his distinction between computational reckoning and judgment [CITE:Smith2019], provides a principal resource for the accountability gap. Registration is an achievement of a situated subject, and judgment involves answerability not supplied by formal calculation alone. The framework uses this distinction without treating Smith's metaphysics as an architectural specification. **Reckoning can be extensive.** In this framework, agents can investigate, infer, compare, propose, assess, plan, check and act within granted authority. Those activities can involve sophisticated reasoning and adaptation. Calling them reckoning does not reduce them to trivial calculation or imply a need for rigid procedures. **Judgment is situated and committed.** It is the person's world-involving engagement with what the work means, what matters and what they will stand behind. It shapes purpose, framing, interpretation and action throughout the undertaking. Final acceptance is one expression of judgment; an approval record neither exhausts it nor proves sufficient engagement. **Responsibility is distinct from capability.** The present professional allocation places accountable judgment and duty of care with the person. Greater agent capability can change delegation, supervision and verification without itself transferring that responsibility. This is not a metaphysical proof of what every future machine could or could not become. It also prescribes no read-only Agent 0, mandatory manager layer or repeated approval of ordinary work. For clarity, this thesis uses *assessment*, *evaluation*, *interpretation* and *selection* for agent activities when *judgment* would confuse this distinction. Ordinary operational uses of the word elsewhere do not confer professional authority. **Wilfrid Sellars' space of reasons** [CITE:Sellars1956] clarifies that justification is not reducible to causal description. In Chirality this supports the practical distinction between producing an output and giving inspectable grounds for reliance. It does not place machines and humans in mutually exclusive metaphysical realms. **Robert Brandom's inferentialism** [CITE:Brandom1994] illuminates the normative statuses undertaken when a person makes and defends a claim. It helps explain why attributed commitment matters, but the project's operational `Claim`, `Warrant`, and `Ruling` primitives are governance constructs rather than translations of Brandom's system. **Nishida Kitaro's account of maintained tension** [CITE:Nishida1945] provides a precedent for relations whose difference need not be erased through synthesis. The framework takes from Nishida the permission to leave a constitutive gap open, not a claim that his contradictory self-identity and this accountability account share one structure. Polanyi and Smith carry the principal philosophical weight: Polanyi for the knower's personal and fiduciary involvement, Smith for situated registration, judgment, and answerability. The other precedents sharpen particular features. None warrants a claim of geometric refinement or exact correspondence. ### 3.6.5 What the Chirality Metaphor Contributes The metaphor gives the project a compact name for an orientation that cannot be delegated away: information may be externalized, while knowing and accountable reliance remain situated in persons. The architecture can expose the informational substrate and record an accountable relation across the gap. It cannot replace the knower with the record. Polanyi's account of tacit integration gives this limit a positive structure [CITE:Polanyi1975]. In the from–to structure of knowing, a person attends *from* subsidiary particulars *to* a focal whole, and particulars scrutinized in themselves lose the joint meaning they subtend. The governed record is designed to occupy the subsidiary position: the practitioner attends from claims, grounds, models and their relationships to the engineering reality that remains focal. On this reading, explicit representation can support integrated understanding. A failure occurs when recording displaces attention to the work — a failure mode this thesis names directly (§6.9.4, §9.2.6). Several disciplines serve the intended orientation: generated output must not claim acceptance in its own voice (§6.8.3), and examination responds to the consequence and detectability of error (§5.7). A record is useful through what it lets participants examine and do, not merely because it exists. This contribution is narrower than an architectural prescription. `Claim` and `Warrant`, meaning and commitment, and the four accountability pillars remain important distinctions, but they are not independent chiral structures. The primary chirality is only the accountability gap. The value of the metaphor is explanatory economy, not formal precision. ### 3.6.6 Why the Gap Is Not a Synthesis Problem The contrast with Hegelian synthesis is practical rather than a claim about all dialectical philosophy. In professional AI governance, it would be a mistake to treat better output as progressively eliminating the accountable professional. Capability can reduce effort and error while leaving the allocation of authority unchanged. Chirality therefore optimizes for visibility and attributable reliance across the gap, not for erasing it. This position does not deny that human and machine activity can be deeply integrated. It denies only that integration, accuracy, or fluency by itself transfers duty of care or makes authentication automatic. The gap persists because information and accountable knowing are not the same kind of status. ### 3.6.7 Engineering Regulation and Accountable Reliance The APEGA practice standard *Relying on the Work of Others and Outsourcing* [CITE:APEGA_RWO2021] is evidence for the practical importance of this distinction, not proof of a metaphysical thesis. Its requirements for direct supervision and control, thorough review, and authentication place obligations on the professional who relies on work prepared by others. They do not make the producer's capability a substitute for the professional's attributable act. Authentication, as used here, binds an accountable actor to identified content, scope and purpose. It confers accountable-reliance status; it does not turn information into knowledge, guarantee correctness or determine what another knower must know. Evidence, examination and accountable commitment remain distinct. Whether a particular process satisfies a professional standard requires an assessment of the actual work and applicable obligations. Other frameworks address supporting parts of the architecture. Bunge and Wand and Weber [CITE:Bunge_ontology] [CITE:Wand_Weber_ontology] inform the ontological pillar; W3C PROV [CITE:W3C_PROV2013] supplies a provenance vocabulary; Toulmin [CITE:Toulmin1958] is a precedent for the operational distinction between claims and warrants; and the INCOSE handbook [CITE:INCOSE2023] supplies the systems-engineering disciplines discussed in Chapter 7. These relationships support the architecture without multiplying the primary chirality. --- ### 3.6.8 Perspectives and iterative Feedback The [shared introduction](INTRODUCTION.md#one-undertaking-three-perspectives) relates the normative, operative and evaluative emphases of the three theses. Philosophically, that relation keeps a commitment distinct from its proposed means and from evidence of its adequacy. A finding about an application can change what participants understand or value, as well as how they act. The four questions make those revisions intelligible while preserving the person's answerability for the purpose and reliance at issue. ## 3.7 Summary The four pillars provide enduring, mutually corrective questions about work. They direct attention to relevant distinctions, grounds, courses of action and values. They do not uniquely derive a document kit, gate sequence, staffing arrangement or runtime. The reckoning/judgment distinction preserves both sophisticated agent capability and the person's situated commitment. Agents can carry substantial authorized work. The person remains responsible for the professional reliance they accept, and that engagement begins before the final approval. The primary chirality remains the non-identity between externalizable information and accountable knowing. Evidence and records can support that relation without performing it. The recurring four questions apply to Chirality itself: its mechanisms require examination for adequacy, cost and limits of transfer.