{"id":"97454f54-bce8-4aa6-b606-c2f8dac7b179","arxiv_id":"2607.24483","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.5,"correctness_risk":"low","formal_verification":"partial","parameter_count":0,"one_line_summary":"The shortest guaranteed escape path in the golden gnomon has exact length C=1.282676… and is a symmetric seven-piece segment–arc–tangent path.","lead":"The paper gives the first exact shortest escape path for Bellman’s lost-in-a-forest problem on the golden gnomon (36°–108°–36° triangle): a seven-piece curve of length C≈1.282676. It matters as a rare closed-form optimum below the 45° base-angle barrier, with a new calibration method and Lean-checked certificates.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"Same soft spot as the reader: Prop. 5.1 is the sole non-algebraic bridge. If its planar order exclusions miss one shortest competitor, the ledger need not stay in the unit disk.","rationale":"This matches the reader’s weakest_assumption rather than adding a new objection. The exact constant, candidate length, eighteen support windows, quartic isolation, calibration identities, Abel/ledger algebra, and several metric cores are kernel-checked with pinned dependencies; there are also independent Newton/integer-relation and exact rational-witness cross-checks, which are real evidence even though not load-bearing. The remaining risk is localized: not a consensus dispute, not circularity in defining C, and not a visible algebraic inconsistency. It is the possibility that the prose planar surgery/order classification admits a shortest non-anchored competitor. That would be decisive if present, but the paper narrows it to finite, checkable configurations with explicit rational margins (κ=1.29 versus 1.2925184/1.2944 in (A.14)) and states precisely which parts are unformalized. Given that transparency and the amount of independent verification around the bridge, I would not move the verdict from ACCEPT; I would treat full formalization of the named Appendix A lemmas as the assurance milestone rather than as a current refutation.","tokens_in":29190,"tokens_out":5476,"duration_ms":191298,"concrete_test":"Make Appendix A.2–A.4 one independent machine-checked target: from planar axioms, with Theorem A.6 either reproved for polygonal arcs or taken as an explicit hypothesis, prove Lemmas A.4, A.5, A.8, A.9 and A.10. In particular, enumerate all weak temporal orders consistent with F≺M≺T and at least one of B,D outside (F,T), verify the list reduces exactly to (A.13), and certify each exceptional four-chord lower bound >129/100 under (A.12). If the enumeration finds another order, a bound ≤129/100, or a feasible alternative (L), Prop. 5.1 fails; otherwise this concern is closed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The equality mechanism is solid only for competitors that admit an anchored marking: Prop. 4.5 gives C≤Iµ(η)≤len η precisely after Lemma 4.4 keeps every suffix state |R(t)|≤1. The load-bearing step is therefore Prop. 5.1: every standard polygonal minimizer with len η<C must be forced into the anchored order. That reduction rests on a chain of prose planar facts rather than on the Lean-checked algebra: Theorem A.6 supplies the Λ-gap; Lemma A.7 converts it to F≺M≺T with h<C/(1+√2)<s; Lemma A.5 must exclude alternative (L) and prove the (1+√2)h surgery; Lemma A.8 must have the fourteen exceptional orders in (A.13) be exhaustive and each four-chord bound exceed κ=129/100; only then do Z1,Z2 become interior anchors and Lemma A.9/A.10 confine out-of-order contacts to the two migrations tolerated in (4.10). A single missed case—a feasible (L) configuration, a non-exhaustive temporal order, or an exceptional order whose supported sum is ≤129/100—would allow a contact on a decreasing fan phase to enter reversed temporal order. Then a ledger state could leave the unit disk, Iµ(η)≤len η would not follow, and a len<C escape path would evade the contradiction. I do not see an exhibited counterexample; the margins in (A.14) are positive and several metric cores are formalized. But this is where the theorem would break if it breaks.","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The paper determines exactly the value of Bellman's lost-in-a-forest problem for the golden gnomon G (equal sides 1, apex angle 108°): the shortest curve guaranteed to reach ∂G from unknown position and heading is a symmetric seven-piece line/arc path Γ of length C = 2sc((b−a)/c + λ) ≈ 1.282676025459, where (a,b,λ) are built from a uniquely isolated root of an explicit quartic Q (Eq. B.1). The upper bound is a certified escape calculation: a support-fan identification (Lemma 3.5, via a one-turn support principle) reduces the escape criterion (3.1) to eighteen support windows checked by exact rational interval arithmetic, with the contact system's determinant bounded below by 3 (2.5). The lower bound is a balanced support calibration: the escape inequality is integrated against a folded source measure µ balanced by the normal relation 2c n₀+n₁+n₂ = 0; normal-cone aggregation compresses this to a finite zero-sum vector family, and Abel summation bounds its total by path length provided the suffix ledger stays in the unit disk (Lemmas 4.1–4.4). The geometric bridge, Proposition 5.1, forces any shorter standard polygonal minimizer into the \"anchored\" temporal order the ledger tolerates, using an imported Λ-configuration theorem, a two-gap surgery, a rational supported-sum exclusion of fourteen exceptional orders, and cyclic bitonicity. Corollaries give the sharp homothetic worm cover C⁻¹G and the transcendence of C. Two finite certificate families, the ledger algebra, and several","tokens_in":29557,"tokens_out":6097,"duration_ms":204273,"significance":"If correct, this is the first proved exact optimum for an isosceles triangle with base angle below 45°, a regime where the extremal curve has moving circular contacts and the tetral-arc classification of Movshovich–Wetzel does not apply; it converts Gibbs's numerical \"Tunnel\" prediction into a theorem with an exact, transcendental constant. The balanced-support-calibration method is the main conceptual contribution and is explicitly presented as angle-independent (Remark 4.6), with the golden-specific work cleanly isolated. The manuscript ships strong verification assets: Lean 4 developments for both finite certificate families with exact rational interval arithmetic, an axiom audit (no sorry, no native_decide, dependencies confined to Mathlib's classical trio), independent Python re-expansion of the supported sums, a quartic-free numerical cross-check, and a precise statement (§B.3) of which steps are machine-checked and which remain prose. The result is falsifiable (explicit constant, explicit curve) and the equality mechanism (tightness exactly on supp ν, Remark 5.2) is transparent. This is a substantial and well-documented advance on a classical problem.","major_comments":[{"comment":"The entire lower bound funnels through the exclusion of alternative (L) in Lemma A.5 (via Lemma A.4 → A.7 → Prop. 5.1): if a shorter minimizer admitted an (L)-configuration, the (1+√2)h surgery fails and the anchored marking need not exist. This step is pure prose: 'monotone turning of the counterclockwise convex boundary' forces direction angles into [π,2π], giving height monotonicity along α_{C1,L1}, contradicting (A.4). The deduction is plausible and short, but it is the one load-bearing planar argument with neither a formalized core nor a displayed criterion. I ask that the monotone-turning fact used here be stated as a displayed lemma with proof (or a precise citation), so that the only non-machine-checked bridge in the proof is inspectable at the same standard as the rest of Appendix A.","section":"Appendix A.2, Lemma A.5 (exclusion of alternative (L))"},{"comment":"The exclusion of the fourteen exceptional temporal orders rests on the enumeration in (A.13a): twenty orders compatible with F≺M≺T, six interior, fourteen exceptional collapsing to eight symmetry classes, each with a supported-sum bound in (A.14) exceeding κ = 129/100 (margins are exact rationals, e.g. 100978/78125 = 1.2925184 > κ). The per-order estimates are independently re-expanded by verify_rational_supports.py and OuterTetral.lean retains certificates for all fourteen orders — but it is unclear whether the exhaustiveness of the enumeration itself (the claim that these eight classes cover all cases, including coincident labels under the outer-side convention) is machine-checked or by hand. Since a single missed order would admit a decreasing-fan contact and break the ledger, please state explicitly how exhaustiveness is established.","section":"Appendix A.3, Lemma A.8, Eqs. (A.13)-(A.14)"}],"minor_comments":[{"comment":"The symbol 'buθ' for the physical normal appears without explanation (presumably a bold/hatted u); define or unify the notation.","section":"Prop. 3.6, Eq. (3.5)"},{"comment":"The 3×3 matrix M is hard to parse; consider displaying its columns factored by c_a (already cleared) or aligning the entries typographically. The bound det M > 3 is kernel-checked, so this is purely presentational.","section":"Eq. (2.4)"},{"comment":"The queue of conditions '−0.121 < p < −0.119' etc. is described as 'coarse boxes' but their provenance (from the root isolation (B.2)–(B.4) and (B.5)) is only implicit; a one-line forward reference would help.","section":"Lemma 2.2, Eq. (2.8)"},{"comment":"The claim h ≤ h_max = 147/250 uses h < s < 147/250 from (A.9); displaying sin(π/5) < 147/250 among the certified bounds of (A.12) would make the chain self-contained at the point of use.","section":"Appendix A.3, (A.11)"},{"comment":"The liberal use of LLMs is commendably disclosed in the front matter; consider also noting in §B.3 whether any Lean proof scripts were machine-assisted in drafting, to complete the provenance picture. Reference [11] is appropriately flagged as unrefereed; the anticipation credit to Gibbs is fair.","section":"Front matter / §B.3"},{"comment":"Remark 1.2's 4.22% comparison to ζ sin β checks out numerically. Suggest adding one sentence after Corollary B.1 noting that a, b, λ are algebraic while C is transcendental — a striking and easily missed feature of the constant.","section":"Remark 1.2 / Cor. B.1"}],"recommendation":"minor_revision","confidential_remarks":"The novelty claim (first proved exact optimum for an isosceles triangle with base angle below 45°) appears accurate against the cited literature [15, 16, 20, 21]; the concurrent Wichiramala–Panraksa preprint concerns a nonisosceles triangle and is properly distinguished. The LLM-use disclosure is unusually candid and, given the axiom-audited Lean development with an explicit prose/formal boundary in §B.3, does not diminish confidence in the results — if anything it sets a good standard. My two major comments are requests for documentation of steps I believe to be correct, not identified errors; the rational margins in (A.14) are positive and exactly verifiable. Well within the journal's scope; I recommend handling as a minor revision."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is the first proved exact escape length for an isosceles triangle with base angle below 45°. They give an explicit seven-piece optimizer Γ, a constant C from one isolated quartic root, transcendence of C, and the sharp homothet cover statement. That is real progress past Gibbs’s numerical Tunnel paths and the known [45°,60°] range.\n\nWhat is new methodologically is the balanced support calibration. One source measure folds through the three normals into a zero-sum vector family; normal-cone aggregation plus Abel summation bounds the support integral by path length once every ledger suffix stays in the unit disk. The candidate saturates the measure on eighteen exact windows, so upper and lower bounds meet cleanly. The algebraic certificates (quartic isolation, contact matrix det M>3, window margins, length identity, ledger junctions) and the reusable discrete ledger facts are Lean-checked with pinned Mathlib deps and no sorry. That is serious evidence, not decoration.\n\nThe soft spot is exactly where the stress-test puts it: Proposition 5.1. Every shortest polygonal competitor with length <C must be forced into an anchored marking so the ledger applies. That step uses the imported Λ-configuration theorem, two-gap surgery with exclusion of alternative (L), the fourteen exceptional-order bounds in A.8, and cyclic bitonicity. Several metric cores are formalized; the remaining planar combinatorics are classical and the rational margins clear κ. I do not see a missed case on the page, and the equality mechanism itself is not circular—C is not fitted. Still, if a non-anchored minimizer slipped the surgeries, the unit-disk bound would not fire. That is a localized prose bridge, not a hole in the algebra.\n\nCitations look right (Zalgaller, Besicovitch/Movshovich, Wetzel survey, Gibbs). No free parameters, no invented physics.\n\nThis is for people who work on worm problems, escape paths, or calibration methods in convex geometry. Worth a careful referee. I would accept it for peer review and would cite the result and the ledger idea.","headline":"Exact E(G)=C for the golden gnomon via a new calibration/ledger method, with Lean-checked algebra and one localized planar rigidity bridge.","tokens_in":30697,"tokens_out":519,"would_cite":true,"duration_ms":10379,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["52A40","52C15","49Q10"],"pacs":[],"model":"grok-4.5","headline":"The shortest guaranteed escape path from the golden gnomon has exact length C ≈ 1.282676, attained by a seven-piece curve.","keywords":["Bellman's lost-in-a-forest problem","escape path","golden gnomon","support function","calibration","convex geometry","worm problem","homothetic cover"],"falsifier":"Exhibit a polygonal escape path for G whose length is strictly less than C, or a unit arc that cannot be placed inside any positive homothet of G smaller than scale 1/C; alternatively, produce a length-minimal polygonal escape path whose support contacts refuse the anchored ledger order after the two-gap surgery.","tokens_in":30533,"feed_emoji":"📐","tokens_out":1065,"duration_ms":16442,"temperature":0.7,"pith_summary":"Bellman’s lost-in-a-forest problem asks for the shortest curve that is guaranteed to reach the boundary of a known shape no matter where you start and which way you face. This paper solves it exactly for the golden gnomon—an isosceles triangle with equal sides 1 and apex 108°. The shortest such escape path is a symmetric seven-piece route of straight segments, circular shoulders, and tangents, with length C determined by one isolated root of an explicit quartic; C is transcendental and equals about 1.282676. Equivalently, the triangle scaled by 1/C is the smallest positive copy of itself that can cover every curve of length 1. The authors obtain the matching lower bound by a balanced support calibration: one weighted family of escape inequalities, saturated by the candidate, is aggregated into a finite zero-sum vector family whose running balance (the ledger) stays inside the unit disk and therefore cannot exceed path length. Local surgery and cyclic order force any shorter polygonal competitor into the ledger’s safe temporal order. This is the first proved exact optimum for an isosceles triangle with base angle below 45°.","feed_headline":"Shortest escape from the golden gnomon is exactly C≈1.282676","feed_subtitle":"A seven-piece path solves Bellman’s problem for the 108° isosceles triangle; the constant is transcendental","key_machinery":"Balanced support calibration: a positive source measure on the triangle’s three normals folds into a balanced vector measure µ whose integral against support functions is at least C for every escape hull and, after normal-cone aggregation and Abel summation, is at most path length whenever the running suffix (the ledger) stays in the unit disk; rigidity forces shortest polygonal counterexamples into the ledger’s allowed order.","core_discovery":"For the golden gnomon G, the escape length equals the exact constant C = 2sc((b−a)/c + λ) built from the unique root p in [−1/8, −1/9] of an explicit quartic; the minimum is attained by the symmetric seven-piece path Γ of segments, circular shoulders of radius s, and tangents. Equivalently, C⁻¹G is the smallest positive homothet of G that contains a congruent copy of every unit-length rectifiable arc.","pith_inferences":["The same ledger-plus-rigidity pattern should decide other triangles inside the numerical “Tunnel” regime (roughly 27°–42° base angle) once the corresponding quartic or algebraic system is written down.","Because the escape threshold forces the shoulder radius to equal the triangle’s altitude factor s, curvature of optimal contacts is dictated by the normal dependence rather than chosen by hand.","A computer-assisted search that returns a shorter simple polygonal escape path would immediately falsify either the surgery lemmas or the unit-disk ledger bound."],"forward_implications":["E(G) equals the explicit transcendental constant C ≈ 1.282676025459, so the classical scaled-Zalgaller benchmark is not optimal for this triangle.","C⁻¹G is a sharp homothetic worm cover: it contains a copy of every unit arc, and no smaller positive homothet of G does.","The same calibration engine applies to any triangle once a sharp source measure and a rigidity argument placing competitors in ledger order are supplied.","Parameters of the optimum reduce to one isolated quartic root; Lean checks the finite algebraic certificates and ledger identities.","This is the first proved exact escape length for an isosceles triangle with base angle below 45°."],"fun_headline_variants":["Exact golden-gnomon escape length is C=1.282676…","Seven-piece path solves Bellman escape for 108° gnomon","First exact optimum for isosceles base angles under 45°","Symmetric Γ attains C from one quartic root for gnomon G","C⁻¹G is smallest homothet covering all unit arcs"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"Every length-minimal polygonal escape path shorter than C can be forced, by two-gap surgery and cyclic boundary order, into exactly the contact sequence the unit-disk ledger tolerates.","fun_headline_variants_meta":{"raw":{"variants":["Exact golden-gnomon escape length is C=1.282676…","Seven-piece path solves Bellman escape for 108° gnomon","First exact optimum for isosceles base angles under 45°","Symmetric Γ attains C from one quartic root for gnomon G","C⁻¹G is smallest homothet covering all unit arcs"]},"model":"grok-4.5","effort":"low","cost_usd":0.004569,"raw_usage":{"total_tokens":1420,"prompt_tokens":879,"num_sources_used":0,"completion_tokens":81,"cost_in_usd_ticks":45688000,"prompt_tokens_details":{"text_tokens":879,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":460,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":879,"tokens_out":81,"duration_ms":7667,"temperature":1.0,"reasoning_tokens":460,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T13:36:15.411098+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Exhibit a polygonal escape path for G whose length is strictly less than C, or a unit arc that cannot be placed inside any positive homothet of G smaller than scale 1/C; alternatively, produce a length-minimal polygonal escape path whose support contacts refuse the anchored ledger order after the two-gap surgery.","supporting_citations":[],"review_version":1}