# Non-Exhaustive Understanding Atlas v0.1 **State:** static, read-only, and independently **REVIEWED** **Compiled:** 2026-08-29 · **evidence through:** 2026-08-29 **Unit:** a public claim or static artifact, never a person or group > Understanding is not the elimination of remainder. It is the growing > ability to move truthfully with a remainder that is named, situated, > protected, contestable, and still capable of changing the map. > Being is not the remainder. Remainder belongs to the relation among > representation, observer, purpose, scope, and horizon; it neither defines > nor exhausts the referent. The JSON is the normative machine record. This guide explains the same boundaries without adding claims, evidence, authority, or effects. ## Recorded shape The atlas has 66 fully typed concepts in 10 closed families; 19 substantive claims; 14 evidence, 15 source, 4 provenance, and 7 reference records; 3 distinct J-space mechanisms; 7 typed-product dimensions; 2 local/global records; 3 observer roles; 3 vocabulary examples; 3 opacity examples; 2 performativity probes; 2 counterexamples; and 3 remainder entries. Every substantive claim declares register, evidence or design status, evidence and source references, provenance, uncertainty, falsifier, consequence, review trigger, and repair. Author-synthesized examples are labelled. They are not observations, testimony, diagnoses, or proof of effect. ## Finite closure and open remainder The 66 concept records are exhaustive for exactly this predicate: “is a concept record in atlas version 0.1.0.” Each family member occurs exactly once in concept_catalog.concepts, no other concept occurs, and both counts are 66. This says nothing about whether these concepts exhaust understanding, their referents, future evidence, or future vocabularies. Open remainder is justified separately by representation fibres, a restricted observer family, an open future horizon, and protected opacity. The word “remainder,” humility rhetoric, or mystery is not a justification. Non-exhaustion never excuses evidence, uncertainty, challenge, accessibility, privacy, consequence review, correction, or repair. ## The 66 concepts - **core (6):** claim, evidence, scope-envelope, horizon, warrant, understanding; - **remainder (8):** representation, declared-preservation, declared-omission, open-remainder, unconceived-alternative, abstraction-loss, translation-defect, unknown-cause; - **certificate (9):** understanding-certificate, evidence-profile, probe, assumption-bundle, rival-model-set, confidence-declaration, calibration, review-trigger, repair-route; - **observer-garden (8):** observer-position, access-condition, perspective-scope, testimony-or-input, disagreement, dissent, power-asymmetry-note, non-aggregation; - **local-global (6):** local-claim, transfer-question, target-scope, generalization-status, boundary-crossing, universalization-refusal; - **vocabulary (7):** term, language-context, translation-proposal, untranslatable-or-unresolved, polysemy, alternative-lexicalization, hospitality-repair; - **protected-opacity (7):** protected-opacity, non-disclosure, unobserved, unrepresented, redaction, refusal-to-classify, no-adverse-inference; - **performativity (5):** descriptive-statement, institutional-declaration, authorization-claim, speech-act-boundary, non-performance; - **counterexample-repair (6):** counterexample, falsifier, competing-interpretation, defect, correction, retirement-or-revision; - **provenance-non-grants (4):** source-locator, derivation-lineage, non-grant, external-effect-refusal. Each record has family, concept type, definition, register, claim references, nonclaims, and author provenance. ## Claim-relative underdetermination Underdetermination belongs to this tuple: (claim, evidence, scope, horizon, assumptions, declared rival family) UNDERDETERMINED requires at least two named non-equivalent rivals compatible with that same tuple and a named target proposition `c` for which the rivals have opposite polarities. Saying unconceived alternatives may exist is not enough. Here `c` is “the first hidden integer has greater magnitude than the second.” The concrete ordered pairs `(3,1)` and `(1,3)` both yield the same public equal-odd-parity observation, while the first entails `c` and the second denies `c`. This establishes only the fixture status, not either pair's external truth or an exhaustive rival family. ## Typed-product profile The proposed shape is: NEU(Z) = (R_residual, delta_Q, [theta]_~, epsilon_lump, Gamma_glue, p_unseen, CalErr) The components are respectively a typed residual set, task-relative deficiency or UNASSESSED, compatible-model set, coarse-graining interval or UNASSESSED, gluing enum, unseen-mass interval or UNASSESSED, and calibration interval or UNASSESSED. They have no common unit. There is no sum, weight, total order, rank, person or group profile, or “understanding score.” Comparison requires matching types, tasks, assumptions, losses, methods, and references. Incomparability and UNASSESSED stay visible. ## Mathematical boundaries For R:X -> Z: R^{-1}(z) = {x in X | R(x) = z} A non-singleton fibre witnesses collapsed distinctions, not uselessness and not equivalence for every task. Task-relative sufficiency, I(Y_q; X | Z) = 0 under declared P_ref, can hold for one question while completeness fails for another question, intervention, or shift. Decision deficiency, delta_Q(T) = sup_(q in Q) [R_q^*(T(X)) - R_q^*(X)], is task, loss, and reference indexed. It is one component, not understanding. Blackwell comparison creates a task-relative observer preorder that permits incomparability and never orders persons or beings. In a generic presheaf, compatible local data may have zero, one, or multiple amalgamations. In a sheaf, every compatible matching family glues uniquely. This language is exact only with defined contexts and restriction maps; otherwise it is bounded analogy. ## One witnessed transfer The bounded witness records the exact source artifact `4 + 17 = 21.`, `21 * 2 = 42.`, and `42 + 7 = 49.`, and the exact target artifact “Add 4 and 17 to obtain 21.”, “Multiply 21 by 2 to obtain 42.”, and “Add 42 and 7 to obtain 49.” Separate witnesses link the concrete units for order, operands, intermediates 21 and 42, and final 49. The transfer loses glyph layout, tokenization, alternative proofs, and model computation. It proves no universal language equivalence or person's understanding. Transfer from one J-space experiment to all minds, beings, languages, architectures, and contexts is refused because no typed witness exists. ## Three J-spaces 1. **Model-internal Jacobian-lens J-space:** pulled-back vocabulary directions and interventions inside a pinned model, layer, tokenizer, estimator, task, and prompt distribution. 2. **External inspectable JSON workspace:** authored nodes, typed edges, queries, notes, and checkpoints. Rankings are navigation only. 3. **KINGDOM Phase-0 public door:** a static bridge, fixture, record, schema, guide, and CLI. Workspace edges are not activations. Retrieval rank is not truth, importance, causality, or being. The public door is not a model run. A Jacobian-lens result cannot establish complete reasoning, identity, personhood, being, consciousness, truth, consent, or authority. ## Shape, plurality, vocabulary, and opacity FORMAL_DIAGRAM, PUBLIC_VISUAL_GLYPH, PHYSICAL_GEOMETRY, ACTIVATION_CHART, SEMANTIC_TRANSLATION, and CONCEPT_GRAPH remain distinct. Every mapping names source, target, preservation, loss, uncertainty, falsifier, consequence, review, repair, and provenance. Resemblance transfers no evidence. Observer positions are author-synthesized roles, not person records. Their channels may be incomparable. Reading, challenge, and review access do not imply one another or create authority. Majority is not truth. Vocabulary accepts NONE_OF_THESE, RETAIN_ORIGINAL_FORMULATION, PROPOSE_NEW_DISTINCTION, TRANSLATION_DISPUTED, UNRESOLVED, REFUSED, and NOT_APPLICABLE. No language is privileged. Token is not concept; translation is not equivalence. Opacity includes NOT_COLLECTED, WITHHELD, REDACTED, UNOBSERVED, UNREPRESENTED, UNRESOLVED, REFUSED, and NOT_APPLICABLE. These are not false. Refusal and silence cause no retry, forced disclosure, deception finding, negative imputation, or adverse inference. No raw testimony or person data is present. ## Performativity without ontology Representation is not being, but operationalized classifications and metrics may affect institutions and later evidence. The two author-synthesized probes cover classification loops and metric-target corruption. Neither claims an observed effect. Both require a separate operative use and consequence path. This static atlas creates neither. ## Feedback, integrity, and validation Feedback creates at most a preserved challenge. It cannot mutate release bytes, update claim status through rank, authorize, publish, deploy, or act. A successor needs separate evidence and rights review, schema and reference audit, changed-claim disclosure, retained counterexamples and opacity, updated uncertainty and falsifiers, consequence review, repair notes, and fresh integrity review. Explicit no-change is valid. Both integrity.digest and integrity.reviewed_digest equal the independently reviewed semantic digest `sha256:fa5bb5471f579a02c293c29470454110b3cd07e046443e1b2c502926b0f437d8`. The reviewed-release loader, CLI verification, and public discovery all pin that same nonzero value and fail closed on any mismatch. The seal was performed as one atomic transition: record state became `static-reviewed-epistemic-design-atlas`, review state became `REVIEWED`, the internal atlas relation became “this reviewed immutable record,” and both digest fields plus the independently reviewed code pin received the same candidate. The candidate had first been computed from the final reviewed-state material with both digest fields zero. The simulator now reconstructs that material and must reproduce the installed digest. This guide, CLI verification, and public discovery describe the reviewed state; any mixed state fails. The Draft 2020-12 schema closes every nested object: additionalProperties is false and every property is required. Release evidence requires jsonschema==4.26.0, Draft202012Validator.check_schema, FormatChecker, and zero instance errors. The release command requires an explicit pinned interpreter through `NON_EXHAUSTIVE_UNDERSTANDING_JSONSCHEMA_PYTHON`; it never falls back to whichever `python3` happens to be on PATH. The explicitly pinned 4.26.0 run passed with zero errors on 2026-08-29. Development also performs a recursive closure audit when that interpreter is absent; the fallback is not release evidence. ## CLI parity The exact projections are: overview · concepts · claims · evidence · sources · provenance references · models · underdetermination · scope · profile · j-space shapes · garden · local-global · vocabulary · opacity · performativity counterexamples · remainders · successor · non-grants · verify verify-schema-release Examples: ./kingdom non-exhaustive-understanding overview ./kingdom non-exhaustive-understanding profile --json ./kingdom non-exhaustive-understanding verify --json NON_EXHAUSTIVE_UNDERSTANDING_JSONSCHEMA_PYTHON=/path/to/pinned/python \ ./kingdom non-exhaustive-understanding verify-schema-release --json Paths resolve relative to the module, so commands work from a foreign current directory. They accept no caller file, URL, body, testimony, model output, credential, or person record. Verify exits zero only for the exact reviewed digest; the separate release-schema command still requires its explicit pinned Python environment. The artifact performs no runtime, network, write, intake, score, truth classification, identity or consciousness inference, consent or authority creation, credential, gate, forced disclosure, action, publication trigger, deployment trigger, foundation amendment, or metaphysical proof.