# AMRAL × Lebesgue Universal Covering — Round 00 ## Comprehensive Finite-Closure Methodology, Research Handoff Protocol, and Semi-Autonomous AI Attack Program **Document ID:** AMRAL-LUC-FC-R00 **Version:** v0.2 **Date:** 2026-09-18 **Research status:** Round 00 / Methodology and research protocol / not a declaration of completed proof **Research mode:** Human-Directed + Semi-Autonomous AI Mathematical Research **Topic:** Lebesgue Universal Covering Problem / Finite Closure Program **Research initiator and methodology source:** Neo.K **Primary methodology source:** Relational Constraint–Handoff Methodology (RCHM); integrating Neo.K's existing prior research on the Lebesgue universal covering problem, skew/tension fields, $S^1$ indexing, finite certificates, and the AMRAL research process. **AI collaborating researcher and primary executor:** Aletheia / ChatGPT, GPT-5.6 Sol **Other future collaborators:** local-end AI, other independent AI, formalization tools, numerical verifiers, and human reviewers; each actual use is recorded separately as it occurs, with no conclusion assumed in advance. --- # 0. Document Positioning This document is round zero of the new attack line on the Lebesgue universal covering problem. It does not attempt, in Round 00, to claim that: 1. the Lebesgue universal covering problem has been proved; 2. a new rigorous lower or upper bound has been obtained; 3. the optimal covering body has been proved to possess some specific symmetry; 4. finitely many witnesses have been proved sufficient to determine the global optimum; 5. any certificate that still requires large-scale computation has been completed; 6. the limit problem of a finite hierarchy has been automatically upgraded into a finite-closure theorem. The purpose of Round 00 is to fix the canonical interface for future research, so that subsequent Round 01, 02, 03, $\ldots$ can accumulate, rather than starting over each round. There are six core tasks: 1. rewrite the Lebesgue problem as an RCHM-tractable relational handoff problem; 2. reclassify past research into proof interfaces, search operators, candidate generators, and verifiers; 3. define the strict working meaning of "finite closure"; 4. establish the four-stage closure chain Existence $\to$ Descent $\to$ Saturation $\to$ Global Closure; 5. establish a per-round reproducible format for research records, computation debt, counterexamples, corrections, and handoff; 6. specify the first exact gate at which real attack work begins, starting from Round 01. Therefore, this document is not a "research summary," but the parent protocol for all subsequent research. --- # 1. The Problem and the Current External Boundary Let $\mathcal K_1$ denote the family of planar compact sets of diameter at most one. The Lebesgue universal covering problem asks for the planar convex set $U$ of minimum area such that every $K\in\mathcal K_1$ can be embedded into $U$ after some planar rigid motion. Written as: $$ a_{\mathrm{Leb}} = \inf_U \left\{ \operatorname{Area}(U): \forall K\in\mathcal K_1, \ \exists g\in E(2), \ gK\subseteq U \right\}. $$ As of 2026-09-18, this program adopts the following external benchmarks as its research starting point, not as new results of this research: - The published upper bound given by Gibbs: $$ a_{\mathrm{Leb}} \le 0.8440935944. $$ - The rigorous lower bound proposed by Mishra on 2026-08-31: $$ a_{\mathrm{Leb}} \ge 0.8344. $$ Its proof uses constant-width test sets such as circles, the Reuleaux triangle, and the Reuleaux pentagon, compresses the placement problem into a finite-dimensional search, and completes an exhaustive subdivision via a large, verifiable certificate. - The Reuleaux-type variational hierarchy proposed by Zeng on 2026-09-01: $$ \Lambda_1 \le \Lambda_2 \le \cdots \le a_{\mathrm{Leb}}, $$ and: $$ a_{\mathrm{Leb}} = \lim_{M\to\infty}\Lambda_M, $$ and claiming that there exists a constant $C$ such that: $$ 0 \le a_{\mathrm{Leb}}-\Lambda_M \le CM^{-2}. $$ Hence the currently available published numerical bracket is: $$ 0.8344 \le a_{\mathrm{Leb}} \le 0.8440935944. $$ The current gap is: $$ 0.8440935944-0.8344 = 0.0096935944. $$ But this program does not treat "continuing to narrow this decimal gap" as its sole objective. The more central objective is to investigate: $$ \boxed{ \text{Can this infinite-dimensional geometric optimization be rigorously converted into finitely many certifiable branch types?} } $$ --- # 2. Principles for Inheriting Past Research This program does not start from zero, nor does it treat the entire body of past research as an already-proved premise. All prior results must be reclassified as: $$ \text{REUSE}, \quad \text{REBUILD}, \quad \text{TEST}, \quad \text{REJECT}, \quad \text{DEFER}. $$ where: - REUSE: a definition or a sufficiently reliable structure that can serve directly as a shared interface; - REBUILD: the idea is retained, but the proof or scope must be rebuilt; - TEST: a high-value conjecture that can only be treated as a falsifiable hypothesis; - REJECT: the old argument is circular, out of scope, or invalid, and is no longer relied upon; - DEFER: possibly valuable, but not currently the minimal residual gate. Round 00 repositions the main prior lines as follows. --- # 3. Prior Line A: The One-Dimensional Metric Kernel and Information-Faithful Representation Past research proposed using a one-dimensional index $I=[0,1]$ together with a distance kernel: $$ D_K(s,t) = \|q_K(s)-q_K(t)\| $$ to build an information-faithful representation. Its value is retained as: 1. showing that "reducing to a one-dimensional index" need not mean losing all two-dimensional geometric information; 2. providing a cross-representation consistency check; 3. allowing a fallback to the distance-kernel layer for auditing when a lighter representation such as the support function shows suspected information distortion; 4. being usable to define representation fidelity, rather than serving directly as the primary computational dimension-reduction tool. This program does not treat the full distance kernel as the primary source of finitization, because: $$ D_K:I\times I\to\mathbb R $$ is still a high-information two-variable object. Hence this line's role in the new program is: $$ \boxed{ \text{faithfulness auditor / fallback representation} } $$ rather than the primary finite compiler. **Round 00 classification: REUSE + REBUILD.** --- # 4. Prior Line B: Uncovered Tension and Universal Tension Past research defined: $$ N_p(K,U;g) = \left( \int_K \operatorname{dist}(gx,U)^p \,d\nu_K(x) \right)^{1/p}, $$ and: $$ N_p(K,U) = \inf_{g\in E(2)}N_p(K,U;g), $$ and then defined: $$ \mathfrak N_p(U) = \sup_{K\in\mathcal K_1}N_p(K,U). $$ This line is retained, but its division of labor is reassigned. It is suited to: - hard-case ranking; - candidate covering-body reduction; - adversarial search; - numerical continuation; - local active-region discovery; - quantification of failing configurations. But this program does not allow "very small tension" to be substituted for exact containment. The final proof layer still requires: $$ gK\subseteq U. $$ Hence: $$ \boxed{ \text{tension is a search and diagnosis functional; exact zero needs an exact gate.} } $$ **Round 00 classification: REUSE.** --- # 5. Prior Line C: Reassigning the Division of Labor Across the Six Parallel Verification Lines Past six-way parallel Lebesgue research produced six classes of structure. Round 00 reassigns them as follows. ## 5.1 Continuity / Lipschitz If the support function, angle, or placement direction admits provable Lipschitz control, then the continuous domain can be bracketed by an interval cover or a finite grid. New role: $$ \boxed{ \text{certified discretization interface} } $$ rather than "the sampling is dense enough, so it presumably holds." ## 5.2 Support function The support function is upgraded to this program's primary shared interface. New role: $$ \boxed{ \text{primary handoff interface} } $$ ## 5.3 Constant-width reduction The constant-width family is the core witness domain currently also in use by the latest external research. New role: $$ \boxed{ \text{target-domain reduction candidate} } $$ But Round 01 must re-fix its precise theorem, premises, and scope, rather than directly citing the informal conclusions of earlier drafts. ## 5.4 Finite sampling bounds New role: $$ \boxed{ \text{continuum-to-interval certificate compiler} } $$ ## 5.5 Predictive $SE(2)$ search New role: $$ \boxed{ \text{search accelerator only} } $$ It can save computation, but cannot serve as a proof that "no other configuration exists." ## 5.6 Removable-region / skew field New role: $$ \boxed{ \text{candidate cover shrinker + active constraint detector} } $$ It is used to generate the next candidate; it does not directly bear the burden of the universality proof. **Round 00 classification: all six lines are retained, but proof responsibility is recut.** --- # 6. Prior Line D: $S^1$, $H_2$, $C_{10}$, and Group Orbits The direction space naturally has: $$ S^1 $$ structure, so $S^1$ is no longer merely an analogy, but the primary direction-parameter domain. On the other hand, the previously proposed: $$ H_2, \qquad C_{10}, \qquad \text{fivefold / ten-direction patterns} $$ are not treated as a priori truths. They are downgraded to: $$ \boxed{ \text{symmetry detector / branch hypothesis generator} } $$ Only when active-contact data repeatedly shows the same orbit structure, and this can be retained by an exact argument, are they upgraded to a theorem candidate. If numerical or symbolic audit shows that the optimal branch does not possess that symmetry, it is removed immediately; the hypothesis is not protected. **Round 00 classification: $S^1$ REUSE; $H_2/C_{10}$ TEST.** --- # 7. RCHM's Scope Mapping onto the Lebesgue Problem RCHM requires that every notion of "finite," "infinite," and "closed" first have its scope fixed. This program defines: $$ \Sigma_{\mathrm{LUC}} = (\mathcal X,\mathcal T,\mathcal N,\mathcal P,\mathcal O), $$ where, provisionally: $$ \mathcal X = \mathbb R^2, $$ $$ \mathcal T = \text{Euclidean topology + group action topology}, $$ $$ \mathcal N = \text{diameter / support norm / area objective}, $$ $$ \mathcal P = \text{admissible shapes, placements, directions and cover parameters}, $$ $$ \mathcal O = \text{support function, slack field, active contacts and certificates}. $$ This scope may be refined in later rounds, but it may not be silently swapped out. Any round that changes $\Sigma_{\mathrm{LUC}}$ must record this explicitly in the Change Log. --- # 8. The Original Infinite Domains and Sources of Escape The Lebesgue problem is not a single "infinity." There are at least five distinct sources. ## 8.1 The Target Shape Domain The diameter-one target sets, or their constant-width completions, form an infinite family. Its escape is not ordinary coordinate unboundedness, but rather: $$ \boxed{ \text{shape-complexity escape} } $$ for example, a new curvature distribution, a new Fourier mode, a new non-finite-arc structure, or new extremal contact geometry. ## 8.2 The Continuous Direction Domain $$ \theta\in S^1. $$ Its form of escape is: $$ \boxed{ \text{unsampled-direction escape} } $$ ## 8.3 The Rigid-Motion Placement Domain $$ g\in E(2). $$ Its degrees of freedom include rotation and translation. Its risk is: $$ \boxed{ \text{configuration escape} } $$ ## 8.4 The Candidate Universal-Cover Boundary Domain $U$ itself may also be an infinite-dimensional family of convex bodies. Its risk is: $$ \boxed{ \text{boundary-mode escape} } $$ ## 8.5 Contact combinatorics Even with a fixed finite-parameter representation, the combinatorial type of the active constraints can still explode. Its risk is: $$ \boxed{ \text{branch-combinatorics escape} } $$ Hence, whenever this program speaks of finite closure, it must state item by item which kind of escape has been eliminated, rather than merely saying "the dimension has already been reduced." --- # 9. Primary Shared Interface: The Support Function For a compact convex set $K\subset\mathbb R^2$, define the support function: $$ h_K(u) = \sup_{x\in K}x\cdot u, \qquad u\in S^1. $$ Below, the angle notation $u_\theta$ denotes a unit direction, and we write: $$ h_K(\theta) = h_K(u_\theta). $$ If $g=(R_\phi,t)\in E(2)$, then: $$ h_{gK}(\theta) = h_K(\theta-\phi)+t\cdot u_\theta. $$ For convex closed sets, if: $$ h_{gK}(\theta) \le h_U(\theta) \qquad \forall\theta\in S^1, $$ then we obtain the natural support-function interface for containment. We therefore define the slack field: $$ \delta_{K,U,g}(\theta) = h_U(\theta)-h_{gK}(\theta). $$ A legal containment state requires: $$ \delta_{K,U,g}(\theta) \ge 0 \qquad \forall\theta\in S^1. $$ An active direction is defined as: $$ \delta_{K,U,g}(\theta)=0. $$ This brings the past notions of tension, skew, direction indexing, and group action into the same interface for the first time. --- # 10. The RCHM Handoff Domain Let: $$ A = \text{admissible target bodies}, $$ $$ B = E(2), $$ $$ C = C(S^1) $$ or a suitable subsequent subspace of support functions. And let: $$ U = \text{candidate universal cover}. $$ The handoff between the target and its placement can be written as: $$ (K,g) \mapsto h_{gK}. $$ The interface for the candidate cover is: $$ U \mapsto h_U. $$ The legality condition is: $$ \Gamma(K,g;U)=1 $$ if and only if: $$ h_{gK}(\theta) \le h_U(\theta) \qquad \forall\theta\in S^1. $$ Lebesgue universality can then be rewritten as: $$ \forall K\in A, \quad \exists g\in B, \quad \Gamma(K,g;U)=1. $$ The core of this program is not to quietly drop this quantifier, but to progressively compress the escape structure inside it into finitely many verifiable branch types. --- # 11. The Working Definition of Finite Closure This program explicitly distinguishes: $$ \text{finite points} $$ from: $$ \text{finite branch types}. $$ We do not require that only finitely many geometric points remain in the end. The actual definition of finite closure is: there exist finitely many branch domains: $$ D_1,\ldots,D_N, $$ such that all legal extremal states fall within: $$ D_{\mathrm{ext}} = \bigcup_{i=1}^N D_i, $$ and each $D_i$ satisfies at least: 1. finite-dimensional parameterization; 2. compactness, or a provable compactification; 3. fixed branch combinatorics; 4. boundary and singular cases explicitly included; 5. amenability to interval arithmetic, convex optimization, real algebraic methods, SMT, formal proof, or other exact verifiers; 6. every legal extremal state has at least one branch witness. Hence: $$ \boxed{ \text{finite closure} \neq \text{finite number of configurations}. } $$ but rather: $$ \boxed{ \text{finite number of saturated, certifiable branch types}. } $$ If each $D_i$ can further be shown to be semialgebraic, piecewise-analytic, or to have even stronger structure, this is recorded as Strong Finite Closure. --- # 12. The Four-Stage Closure Chain Round 00 adopts the following main chain: $$ \boxed{ \text{Existence} \longrightarrow \text{Descent} \longrightarrow \text{Saturation} \longrightarrow \text{Global Closure}. } $$ This is a general research prototype extracted from the recent RCHM-Hodge experiments, now formally transplanted onto the Lebesgue line. --- # 13. Gate E: Existence The Existence gate asks only: > Has a concrete structure capable of bearing the extremum or the obstruction been found? For example: - some Reuleaux witness family; - some finite-arc extremizer; - some contact graph; - some Fourier-mode profile; - some symmetric orbit; - some lower-bound configuration; - some candidate universal-cover boundary. Passing Existence means only: $$ \exists x\in D_{\mathrm{candidate}}. $$ It does not mean: $$ D_{\mathrm{candidate}} = D_{\mathrm{global}}. $$ No round may write candidate existence as global necessity. --- # 14. Gate D: Descent The Descent gate asks: > Can a conclusion obtained in a finite model, a discrete model, a truncated model, or an intermediate representation be legally handed back to the original continuum problem? Typical form: $$ \text{finite model statement} \stackrel{?}{\Longrightarrow} \text{original geometric statement}. $$ Descent obstructions that may arise on the Lebesgue line include: 1. a finite angular grid failing to cover the directions between grid points; 2. Fourier truncation losing high-frequency modes; 3. the Reuleaux finite-arc hierarchy giving only an approximation rather than the exact extremizer; 4. a numerical optimizer finding only a local optimum; 5. the support-function parameterization failing to handle the translation gauge; 6. the contact equations ignoring degenerate/merged contacts; 7. a finite-precision certificate failing to include outward rounding; 8. symmetry reduction excluding a non-symmetric extremizer. A successful Descent must have an explicit proof or certificate; it may not be replaced by "this should be fine analytically." --- # 15. Gate S: Saturation Saturation is the most important new core gate of this program. Suppose a set of legal witnesses or branches is already known: $$ W = \{D_1,\ldots,D_N\}. $$ Define the legal extremal domain: $$ D_{\mathrm{legal}}. $$ Saturation asks: $$ \boxed{ \operatorname{Im}(W) \stackrel{?}{=} D_{\mathrm{legal}}. } $$ That is: > Does there exist any legal but not-yet-witnessed escape state that the current branch system fails to witness? Typical saturation failures: - a hidden high-frequency shape mode; - a non-Reuleaux constant-width body; - a new contact topology; - contact merger / bifurcation; - a symmetry-breaking branch; - a boundary case; - a non-generic branch; - a placement configuration not covered by the existing enumeration; - a hole in the interval cover; - the candidate cover's boundary still admitting a local reduction. Only after Saturation passes can one say that "the current finite branch system is complete." --- # 16. Gate G: Global Closure Global Closure requires the lower side and the upper side to close simultaneously. The ideal form is that there exist $A_\ast$ and a candidate universal cover $U_\ast$ such that: $$ \operatorname{Area}(U) \ge A_\ast $$ holds for all universal covers $U$, and: $$ \operatorname{Area}(U_\ast) \le A_\ast. $$ Hence: $$ \operatorname{Area}(U_\ast) = A_\ast = a_{\mathrm{Leb}}. $$ Written as the closure identity: $$ \boxed{ \text{Exact Lower Certificate} + \text{Exact Universal Upper Certificate} + \text{Saturation} \Longrightarrow \text{Global Closure}. } $$ If the lower and upper bounds only narrow without meeting, this still counts as progress; closure is not claimed. --- # 17. A New Version of the Finite Closure Conjecture Round 00 does not adopt the overly strong version, "finitely many witness points must necessarily determine the global optimum." It adopts instead: ## Conjecture FC-LUC There exist finitely many certifiable branch types: $$ D_1,\ldots,D_N $$ such that every legal extremal state that determines the Lebesgue optimum falls within: $$ D_{\mathrm{ext}} = \bigcup_{i=1}^{N}D_i, $$ and each $D_i$ is finite-dimensional, compact or admits a controlled compactification, has fixed branch combinatorics, and can be verified by an exact certificate. This conjecture allows continuous parameters to remain within a branch. Hence it is weaker than: $$ \text{there exist finitely many points that fully determine the optimum} $$ but stronger than the bare statement: $$ \Lambda_M\to a_{\mathrm{Leb}}. $$ One of this program's long-term goals is to determine whether FC-LUC is true. If FC-LUC is refuted by a counterexample, the research can still fall back to a controlled hierarchy; this is not treated as a failure of the program. --- # 18. The Freedom Ledger Every round must maintain: $$ \mathcal F_n = \{ \text{degrees of freedom and escape types not yet eliminated} \}. $$ The initial ledger provisionally records: $$ \mathcal F_0 = \{ F_{\mathrm{shape}}, F_{\mathrm{direction}}, F_{E(2)}, F_{\mathrm{boundary}}, F_{\mathrm{contact}}, F_{\mathrm{witness}}, F_{\mathrm{certificate}} \}. $$ where: - $F_{\mathrm{shape}}$: the target shape family; - $F_{\mathrm{direction}}$: the $S^1$ continuum; - $F_{E(2)}$: rotation / translation; - $F_{\mathrm{boundary}}$: candidate-cover boundary modes; - $F_{\mathrm{contact}}$: active-contact topology; - $F_{\mathrm{witness}}$: extremal-witness completeness; - $F_{\mathrm{certificate}}$: the numerical-to-exact gap. Every round must record: $$ \Delta\mathcal F_n = \text{the degrees of freedom genuinely eliminated or compressed this round}. $$ and update: $$ \mathcal F_{n+1} = \mathcal F_n\setminus\Delta\mathcal F_n $$ or, when necessary, add newly discovered obstructions. An increase in the number of documents may not be substituted for a reduction in degrees of freedom. --- # 19. Proof Graph, First Version Round 00 establishes the following dependency graph. ## P0 — Canonical problem statement Fix the Lebesgue constant, the admissible objects, convexity, and the rigid-motion convention. ## P1 — Completion / constant-width reduction Rigorously rebuild: which targets can be reduced to constant-width bodies, and what the scope and premises are. ## P2 — Support-function compiler Establish: $$ (K,g,U) \mapsto \delta_{K,U,g}(\theta) $$ as an exact interface. ## P3 — Gauge reduction Eliminate unnecessary translation / rotation redundancy, and establish a compact placement-parameter domain or an equivalent normalization. ## P4 — Certified continuum control Use Lipschitz control, interval covers, analytic bounds, or other tools to turn the $S^1$ direction continuum into a certifiable finite cover. ## P5 — Finite hierarchy reconstruction Independently reconstruct the existing Reuleaux / finite-arc hierarchy, explicitly marking which parts are literature theorems and which are lemmas reproved by this research. ## P6 — Active-contact branch compiler Extract contact topology, orbits, and degenerate branches from candidate extrema. ## P7 — Descent closure Prove that the finite representations of P4–P6 contain no representation artifact. ## P8 — Saturation Prove that all legal extremal states are witnessed by the branch system, or find a counterexample and extend the branch system. ## P9 — Exact lower certificate Compress the finite branches into a rigorous lower-bound certificate. ## P10 — Exact upper certificate Give a genuine universal-containment certificate for the candidate $U_\ast$. ## P11 — Equality closure If: $$ P9=P10, $$ and P8 is already closed, then global closure is complete. At present, only P0's problem background and part of the P1-P5 material have prior research / external research to inherit from; the graph as a whole is not regarded as complete. --- # 20. The Method Library No Longer Competes Internally, but Becomes an Operator Library This program does not require choosing a single "uniquely correct method" up front. Each method is responsible only for a specific residual domain. ## 20.1 RCHM Responsible for: - scope; - interface; - legal handoff; - escape analysis; - freedom ledger; - witness completeness; - descent; - saturation; - closure responsibility. ## 20.2 RIITG When the current proof graph breaks at some edge, work backward from the result to the intermediate theorem it requires. ## 20.3 RAB When a numerical / structural pattern appears stably, work backward to the minimal necessary axiom, condition, or lemma, then subject it to counterexample testing. ## 20.4 KCPE Perform knowledge-conditioned quasi-enumeration over the combinatorial / contact domain that has already become a finite branch. ## 20.5 Skew / tension field Perform ranking, sensitivity analysis, local reduction, and hard-case discovery on the residual slack. ## 20.6 Fourier / basis methods Responsible for the controlled parameterization and mode audit of the constant-width support-function family. ## 20.7 Group action / orbit methods Responsible for quotient redundancy, the symmetry detector, and active-orbit compression. ## 20.8 Interval arithmetic / exact optimization Responsible for the final leg from numerical candidate to rigorous certificate. Hence: $$ \boxed{ \text{it is not a matter of picking one method to solve the whole problem, but of letting the residual obstruction choose the next operator.} } $$ --- # 21. The Fixed Format for Each Round of Research Starting from Round 01, every round's canonical Markdown source must contain at least the following fields. ## 21.1 Metadata - Round number; - date; - version; - executor; - external literature versions used; - code / data / certificates used; - the handoff ID from the previous round. ## 21.2 Previous state List the previous round's: $$ \mathcal F_n, $$ and the one or few remaining gates. ## 21.3 Objective Specify only the exact gate to be attacked in this round. It is forbidden for a single round to expand without limit into ten new directions with no gate priority order. ## 21.4 Definitions and normalization All new notation, scales, gauges, and scope changes must be stated explicitly. ## 21.5 Derivation Fully retain: - why the derivation moves from the previous round to this step; - which theorem was used; - which step is the researcher's own derivation; - which step is only a hypothesis; - which step depends on computation; - which step might fail. ## 21.6 Claim ledger Every important result may use only one of the following states: - PROVED-ANALYTIC; - VERIFIED-COMPUTATIONAL; - VERIFIED-THEOREM-APPLICATION; - DERIVED-CANDIDATE; - NUMERICAL-EVIDENCE; - CONDITIONAL; - COMPUTE-DEFERRED; - OPEN; - REJECTED; - CORRECTED. ## 21.7 Counterexample / adversarial section Every round must ask at least: 1. which hidden branch could escape? 2. if some hypothesis is wrong, what does the minimal counterexample look like? 3. which numerical pattern might be only a local optimum? 4. which symmetry might be only a search bias? ## 21.8 Freedom ledger update Explicitly record: $$ \Delta\mathcal F_n. $$ ## 21.9 Computation debt List the large-scale computations not yet run. ## 21.10 Reproducibility package Record: - exact equations; - parameter ranges; - pseudocode or code; - expected output; - acceptance criterion; - certificate format. ## 21.11 Handoff Leave only: - remaining escape; - unwitnessed legal states; - global closure obstruction; - the minimal gate for the next round. --- # 22. The Deferred-Computation Protocol This program adopts: $$ \boxed{ \text{progress first, computation debt explicit, no fake verification.} } $$ If the computational load is too large, this round may continue theoretical progress, but must mark the computation status as: $$ \text{COMPUTE-DEFERRED}. $$ In that case, the following must be left behind: 1. the exact mathematical object to be computed; 2. variables and parameter ranges; 3. the algorithm or pseudocode; 4. the required numerical precision; 5. the expected invariant; 6. the pass / fail criterion; 7. the certificate schema; 8. any known complexity estimate; 9. which proof-graph node the result should be written back to once the local computation is complete. It may not be written as: $$ \text{a large computation should pass, so it is treated as passed}. $$ It may only be written as: $$ \text{theoretical reduction complete; certificate pending}. $$ --- # 23. The Multi-AI Autonomous Checking and Correction Protocol This program treats error correction as part of the research process, not as a record of failure. It permits subsequent involvement by: - local-end models; - other cloud models; - formal provers; - independent verifiers; - human reviewers; to re-derive or stress-test any round. But the following must be observed. ## 23.1 No silent overwrite If an old conclusion is wrong, the history is not deleted; it is marked in the new round as: $$ \text{CORRECTED} $$ or: $$ \text{REJECTED}. $$ ## 23.2 Separation of provenance The following must be distinguished: - the methodology / research direction provided by Neo.K; - the AI's autonomous derivation; - external literature theorems; - local-end computation; - re-verification results from other AI. ## 23.3 Counterexamples take priority If a second AI can find a counterexample, the counterexample is preserved with priority; it is not overridden by a majority vote among AIs. ## 23.4 Re-derivation outranks tonal agreement An independent AI is not required to "agree with the previous AI"; it is required instead to reconstruct as much as possible from the original definitions. This makes: $$ \boxed{ \text{multi-AI verification} \approx \text{independent derivation + adversarial audit} } $$ rather than a consensus vote. --- # 24. Defining Roles in Semi-Autonomous AI Mathematical Research This series' research mode is explicitly recorded as: $$ \boxed{ \text{Human-directed, semi-autonomous AI mathematical research}. } $$ The division of roles is as follows. ## Neo.K Responsible for: - research problem selection; - RCHM and other proprietary methodology sources; - research direction and priority; - decisions to publish, withhold, or hand off results; - providing new cross-domain methods or counterfactual hypotheses when necessary; - arranging large-scale local-end computation and verification by other AI. ## Aletheia / GPT-5.6 Sol Responsible for: - active research execution each round; - literature reconstruction; - proof-graph maintenance; - theorem and lemma derivation; - counterexample search; - RCHM handoff; - formalizing computational problems; - certificate specification; - self-audit; - producing the canonical Markdown source. Hence this series is neither "the AI merely helps with formatting" nor "the AI alone decides the research goals." --- # 25. Canonical Source Specification The canonical source of each round's formal output is UTF-8 Markdown. The mathematical source uses only: - inline: `$...$`; - display: `$$...$$`. The formal source may not depend on a chat-rendering display. It is forbidden to first convert LaTeX formulas into Unicode math characters and then treat that as the canonical source. `unicode_escape`-style round trips are forbidden. Before formal commit, at least the following must be verified: 1. UTF-8 decodability; 2. display-delimiter pairing; 3. inline-delimiter pairing; 4. no mixing with other math delimiters; 5. key formulas can be reconstructed directly from the source; 6. the file contains no secrets, API keys, or private data. --- # 26. Per-Round Output, Cloud Crystallization, and the Dual-Conversation Collaboration Principle This program adopts "one canonical MD per round." There is no requirement of "output every day"; the output trigger is **the completion of each research round**. Multiple rounds may be completed on the same day, and equally a single round may take more than one day to complete. Upon completing each round: 1. produce the complete UTF-8 Markdown canonical source; 2. if there is code / JSON / a certificate, additionally package it as a ZIP; 3. the chat display shows only this round's summary, status, progress, and the next gate; 4. the complete reasoning, sources, rationale for progress, failed branches, counterexamples, computation debt, and reproducibility information are kept in the MD; 5. the user is not required to copy formal formulas from the chat display; 6. the round's canonical MD is synced to a dedicated cloud research folder; 7. an append-only `State_Crystal_Round_NN` is produced at the same time, recording the latest gate, closed gates, open questions, computation debt, dependency versions, and the entry point for the next round. Hence the long-term research adopts: $$ \boxed{ \text{chat = discussion / rendering view}, \qquad \text{MD = canonical research record}, \qquad \text{cloud crystal = cross-conversation handoff state}. } $$ ## 26.1 The Cloud-Sync Crystallization Method The cloud research folder uses an append-only structure; it does not use a single overwritable `CURRENT` file as the sole source of truth. Every round leaves behind at least: $$ \texttt{Round\_NN\_...md} $$ and: $$ \texttt{State\_Crystal\_Round\_NN.md}. $$ The Round file preserves the complete research; the State Crystal preserves only the minimal sufficient state needed to continue across conversations. The latest state is determined by the largest Round number. This lets the research simultaneously satisfy: - append-only; - traceability; - replayability; - independence from chat context; - not losing research history due to a single mutable pointer becoming corrupted; - a new AI conversation being able to continue directly from the latest crystal. ## 26.2 Asymmetric Dual-Conversation Collaboration This research permits two deliberately desynchronized AI research lines. ### Main Line A: Frontier / Canonical Line This conversation maintains the latest research round, and is responsible for: - advancing the latest exact gate; - producing the canonical Round MD; - maintaining the proof graph and the constraint ledger; - updating the cloud State Crystal; - determining the current canonical state. ### Verification Line B: Lagged Verification / Expansion Line A separate, independent AI conversation deliberately lags the main line by several rounds, and is responsible for: - secondary verification; - independent re-derivation; - counterexample search; - filling gaps; - alternative representations; - branch-line expansion; - adversarial audit of existing gates. The verification line does not directly overwrite the canonical history. Its results are first marked as: $$ \text{AUDIT}, \quad \text{CORRECTION}, \quad \text{EXTENSION}, \quad \text{COUNTEREXAMPLE}, \quad \text{INDEPENDENT-REPRODUCTION}. $$ and are then absorbed, rejected, or reclassified by the main line in a subsequent Round. The two lines therefore form: $$ \boxed{ \text{Frontier progression} \;\parallel\; \text{Lagged independent verification} } $$ rather than two AIs synchronously duplicating the same line of reasoning. ## 26.3 Corrections, Catching Escapes, and Multi-AI Checking This research explicitly permits later correction. If an error in a previous Round is discovered, the history is not deleted; instead a correction edge is created in the new Round: $$ R_i \xrightarrow{\text{audit / counterexample}} R_j. $$ If a large computation cannot be completed on the main line for the time being, a $$ \texttt{COMPUTE-DEFERRED} $$ entry may first be created, fully recording the input, algorithm, expected output, acceptance criterion, and the gate it affects, to be computed later by a local-end or other AI / computing environment. While the computation remains incomplete, a numerical candidate may not be written as a proved theorem; but the whole line of research is not required to stop. The research may continue to advance every structural gate that does not depend on that numerical certificate. --- # 27. The Adaptive Attack Route from Round 01 Onward The subsequent rounds are not locked to a fixed number of documents, but a priority order is established first. ## Round 01 — Canonical Support-Function Compiler The first formal attack gate: $$ \boxed{ \text{the rigorous reconstruction of P0 + P1 + P2 and the fixing of their scope}. } $$ Objectives: 1. rigorously fix the target domain of the Lebesgue problem; 2. rebuild the legal chain of convexification / completion / constant-width reduction; 3. establish the support-function equivalence of containment; 4. explicitly handle the translation / rotation action; 5. establish the slack field; 6. establish the first version of the freedom ledger; 7. find the first genuinely unclosed gate. Round 01 gives priority to analytic derivation, and in principle requires no large-scale computation. ## Round 02 — Hierarchy Reconstruction and Descent Audit Independently reconstruct the latest finite-arc / Reuleaux hierarchy. The point is not to trust the external paper, but to break it down into: - theorem dependency; - finite-dimensional parameterization; - convergence rate; - descent gap; - saturation gap. ## Round 03 — Certificate Anatomy Dissect the current latest certified lower-bound work: - witness family; - placement dimension; - box subdivision; - interval / floating-point error control; - independent verifier; - branch-explosion structure. The purpose is not to reproduce the number, but to extract a reusable certificate compiler. ## Round 04+ — Active Branch Search Begin using: - RCHM; - skew / tension; - Fourier; - group orbit; - KCPE; - adversarial search; to search for new branch types, hidden escapes, or saturation counterexamples. From this point on, the round number is determined by the residual gate, and the total number of rounds is not fixed in advance. --- # 28. Acceptable Forms of Progress in the First Phase "Raising the bound" is not the only thing that counts as progress. Any of the following counts as substantive progress: 1. proving that a class of escape is impossible; 2. proving the descent of a reduction; 3. finding a hidden gap in a finite hierarchy; 4. finding a new active-contact branch; 5. proving that some symmetry hypothesis must be abandoned; 6. compressing an infinite family into finite branch types; 7. turning a numerical conjecture into an exact certificate problem; 8. giving an executable verifier specification; 9. finding a counterexample that overturns an old hypothesis; 10. compressing a global obstruction into a single exact gate. Hence what this program tracks is: $$ \boxed{ \text{residual freedom reduction} } $$ rather than: $$ \boxed{ \text{paper count}. } $$ --- # 29. Stop / Pivot Rules To prevent autonomous research from degenerating into a large volume of ineffective iteration, the following rules are added. ## 29.1 Three-Round No-Freedom-Reduction Rule If, for three consecutive rounds: $$ \Delta\mathcal F_n=\varnothing $$ or only numerical samples are added without narrowing the global quantifier, then that line is paused. ## 29.2 Branch-Explosion Rule If the number of finite branches grows rapidly without a provable pruning invariant, pure enumeration is stopped and the work returns to symbolic reduction. ## 29.3 Symmetry-Falsification Rule If active data continues to produce non-$H_2/C_{10}$ branches, no further resources are invested in protecting the fivefold / tenfold hypothesis. ## 29.4 Numerical-Plateau Rule If the bound improves by only a tiny amount over multiple rounds with no closure structure, priority shifts to attacking descent or saturation instead. ## 29.5 Certificate-Debt Rule If COMPUTE-DEFERRED items accumulate excessively, generating new numerical conjectures is paused until the certificate debt is cleared. --- # 30. Round 00 Initial Risk Table ## Risk A — The Constant-Width Reduction Being Oversimplified Mitigation: Round 01 rebuilds from the literature's theorem and its full premises, rather than using an informal version. ## Risk B — Support-Function Gauge Confusion Mitigation: explicitly separate translation, rotation, and width normalization. ## Risk C — A Finite Hierarchy Being Mistaken for Finite Closure Mitigation: enforce the Descent + Saturation gates. ## Risk D — A Numerical Certificate Verifying Only the Searcher's Own Branch Mitigation: audit the certificate domain and branch completeness separately. ## Risk E — A Symmetry Hypothesis Causing Search Bias Mitigation: retain an asymmetric adversarial branch. ## Risk F — The AI's Autonomous Derivation Misciting the Literature Mitigation: mark external theorems with source and date; independently rebuild major dependencies. ## Risk G — A Large Computation Unable to Complete in Real Time Mitigation: COMPUTE-DEFERRED; leave behind the complete computation specification first. ## Risk H — Corrections Causing Inconsistency Between Documents Over Time Mitigation: no silent overwrite; use CORRECTED / REJECTED and a handoff version. --- # 31. The Substantive Output of Round 00 This round does not claim a new Lebesgue bound. What this round genuinely adds is the following research structure. ## 31.1 New Working Definition Formal definition: $$ \boxed{ \text{finite closure} = \text{finite saturated certifiable branch types}, } $$ rather than "finitely many points." ## 31.2 New Main Closure Chain $$ \boxed{ \text{Existence} \to \text{Descent} \to \text{Saturation} \to \text{Global Closure}. } $$ ## 31.3 New Residual-Domain Viewpoint The various mathematical methods no longer compete for ownership of the whole problem; instead, the residual obstruction chooses the operator. ## 31.4 New Research Governance Formally fixed: - semi-autonomous AI execution; - computation debt; - multi-AI adversarial correction; - claim-state ledger; - one-round-one-MD; - canonical UTF-8 source. ## 31.5 New First Gate Round 01 does not chase the bound directly, and instead first establishes: $$ \boxed{ \text{the canonical support-function + constant-width handoff compiler}. } $$ --- # 32. Round 00 Freedom Ledger Initial state: $$ \mathcal F_0 = \{ F_{\mathrm{shape}}, F_{\mathrm{direction}}, F_{E(2)}, F_{\mathrm{boundary}}, F_{\mathrm{contact}}, F_{\mathrm{witness}}, F_{\mathrm{certificate}} \}. $$ This round does not claim to have eliminated any of these mathematical degrees of freedom. What this round completes is a meta-level closure: $$ \Delta\mathcal F_{\mathrm{method}} = \{ \text{ambiguous responsibility}, \text{method overlap}, \text{untracked computation debt}, \text{silent correction}, \text{finite-hierarchy / finite-closure conflation} \}. $$ So: $$ \boxed{ \text{Mathematical closure: OPEN} } $$ but: $$ \boxed{ \text{Methodological Round 00: CLOSED} } $$ --- # 33. Round 01 Handoff Contract The next round accepts only the following tasks. ## Input 1. Lebesgue canonical problem statement; 2. convexification / completion / constant-width literature chain; 3. support-function representation; 4. $E(2)$ action; 5. Round 00 freedom ledger. ## Required output 1. exact definitions; 2. proved containment lemmas; 3. constant-width reduction status; 4. gauge normalization; 5. first exact escape ledger; 6. first theorem dependency graph; 7. unresolved gates; 8. Round 02 handoff. ## Forbidden shortcut Round 01 may not, just because the latest literature uses constant-width bodies, directly replace: $$ \text{all targets} $$ with: $$ \text{finite Reuleaux family}. $$ The intermediate reduction must exist explicitly. --- # 34. Final Research Philosophy This program does not presuppose that the Lebesgue universal covering problem will necessarily be completely solved within a finite number of rounds. But it adopts a strategy stronger than "keep trying new configurations": $$ \boxed{ \text{every round must ask where else there is still room to escape from.} } $$ If some round succeeds in sealing off an escape, it is removed from the ledger. If a new escape is discovered, it is added to the ledger. If a finite model cannot descend, the finite result is not passed off as a continuum theorem. If the witness family has not achieved saturation, a large volume of certificates is not passed off as global completeness. If the computation has not yet been run, a reproducible computation debt is left behind. If another AI finds a loophole, the loophole is upgraded into the next gate. Hence the core of the entire Lebesgue Finite Closure Program is not: $$ \text{the AI guessing the answer in a single shot}, $$ but rather: $$ \boxed{ \text{AI and humans together compress the global degrees of freedom layer by layer, until only a verifiable closure problem remains.} } $$ This is also the formal research protocol for this series from Round 01 onward. --- # 35. Status Summary **Round 00: CLOSED** **Global Lebesgue problem: OPEN** **New bound claimed: NO** **Finite closure proved: NO** **Finite closure framework defined: YES** **RCHM integration complete at methodology level: YES** **Computation required in Round 00: NO** **Next gate: Round 01 — Canonical Support-Function + Constant-Width Handoff Compiler** --- # References / Current Baseline 1. Philip Gibbs, *An Upper Bound for Lebesgue's Covering Problem*, arXiv:1810.10089, 2018. https://arxiv.org/abs/1810.10089 2. Ujjwal Mishra, *Curves of constant width and Lebesgue's covering problem*, arXiv:2608.30538, 2026-08-31. https://arxiv.org/abs/2608.30538 3. Shuai Zeng, *An exact hierarchy for Lebesgue's universal covering constant and a certified 0.834 lower bound*, arXiv:2609.01284, 2026-09-01. https://arxiv.org/abs/2609.01284 4. Wacharin Wichiramala and Chatchawan Panraksa, *Wetzel's 30-60-90 Triangle Covers Unit Arcs*, arXiv:2606.14625, 2026. https://arxiv.org/abs/2606.14625 5. AMRAL Research Lab, PROGRAM-UNIVERSAL-COVERING current public context, accessed 2026-09-18. https://amral.evemisslab.com/ 6. Neo.K, *Relational Constraint–Handoff Methodology (RCHM) v0.1*, internal research source, 2026-09. 7. Neo.K, *From a One-Dimensional Metric Line to Universal-Covering Tension: A One-Dimensionalization Proposition Conjecture, Information-Faithfulness Condition, and Local-Reduction Methodology for the Lebesgue Universal Covering Problem*, internal research source, 2026-07-07. 8. Neo.K, *From Annular Fibers to Planar Covering Tension: $S^1$ Indexing, the $H_2$ Fivefold–Tenfold Structure, Group Action, and a 120-Cell-Analogy Proposition in the Two-Dimensional Lebesgue Universal Covering Problem*, internal research source, 2026-07-07. 9. Neo.K, *EML-LUC-2026-SIXWAY: Six-Way Parallel Verification of the Lebesgue Universal Covering Problem v0.1*, internal research source, 2026-07. --- # Canonical Source Declaration This file is the UTF-8 canonical Markdown source for Round 00. Any rendering, summary, or formula display in the chat interface is not a substitute for this round's official original manuscript.