{"id":"4a623381-a671-464f-a1a1-25b3b80b6d46","arxiv_id":"2607.26999","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"A reflection involution splits conditioned friction into antisymmetric boundary physics and symmetric differencing bias, producing a two-layer arrow of time and unimposed Cheshire-cat property currents measurable by two-state gEDMD.","lead":"Pre- and post-selected quantum trajectories yield a measurable split of friction into boundary physics and estimator artifact, so the arrow of time has two layers at fixed resolution. The same tool makes a Cheshire-cat property current emerge from port-only conditioning and scale to a million-dimensional spin chain.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Exact-derivative reversal at the fluctuation layer is verified only for involution-preserving ensembles at N=8; bright classes used at scale break that protection.","rationale":"The reader correctly flags the real-symmetric/independent-seed hypotheses and the q-dependence of N=20 depths. Those are real but secondary for the qualitative claim: sign of mode-layer reversal is q-stable (Table IV), and residuals ρ_A/ρ_S are reported. The more load-bearing gap is the untested interaction of those hypotheses with the exact-derivative fluctuation-layer control that licenses “immunity belongs to the inference, not the ensemble.” Algebraic sample identities at N=20 (Sec. V A) protect only the independent class; bright-class mode reversal is empirical and modulation-driven (Fig. 8 control), not theorem-guaranteed at the dictionary layer. Until the concrete N=8 exact-derivative check on a bright class is done—or an N=20 dictionary exact run—the central two-layer slogan overreaches its verified domain. That keeps the verdict CONDITIONAL (not REJECT): Cheshire-cat generator response, unique γ_A/γ_S split under the stated hypotheses, and mode-layer reversal remain on solid ground. Code still unshipped is a separate reproducibility debit already noted by the reader. No change of verdict category; confidence in the unrestricted “both layers reverse” half should be lower than the reader’s HIGH on the full strongest_claim.","tokens_in":30491,"tokens_out":770,"duration_ms":70291,"concrete_test":"At N=8, rerun the 29-element dictionary regression with the exact derivative (Eq. 2) on the future-consistent class and on φ_T ∝ e^{εΛ}Π_K|ψ(T)⟩ at K=D (same M and windows as Fig. 5d/Fig. 9). If half-window γ_exact fails to change sign, or |γ_A| stays far below the independent-class ±0.274, the “immunity is inference” claim does not extend to the bright classes used at D=2^{20} and the abstract’s two-layer wording needs scope restriction.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The headline claim that fluctuation-layer immunity is a property of the inference (not the ensemble) rests on exact-derivative dictionary-layer friction reversing about the midpoint. That control is reported only for the independent class at N=8 (Sec. IV D: antisymmetric to 9.6e-13, half-window averages ±0.274). Appendix C and Sec. VI E make plain that the involution—and thus guaranteed antisymmetry of γ_exact—requires independent, conjugation-invariant seeds and real-symmetric H. The scalable N=20 results (Table II, Fig. 11) instead use bright classes built from |ψ(T)⟩ that break measure preservation. Sec. IV E already flags the critical combination: “the dictionary layer of a class that breaks the involution” is “the one place where conditioning leaves no trace of reversal at all.” No exact-derivative dictionary-layer run is given for any involution-breaking class at either size (Sec. VI E, fifth limitation). Consequently the portable half of the two-layer story—both layers reverse once the scheme component is removed—has not been established for the ensembles that actually remove the 2^{-N/2} obstruction. Mode-layer reversal under modulation remains well supported; the inference-only reading of fluctuation-layer immunity does not automatically carry over.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.5","summary":"The manuscript introduces two-state gEDMD, transferring generator extraction to complex weak-value trajectories via the exact identity dA_w/dt = i⟨[H,A]⟩_w. Two main results follow. First, a reflection involution on pre-/post-selected ensembles forces every window-fitted friction to split uniquely as γ_fwd = γ_A + γ_S (antisymmetric boundary-condition part plus symmetric differencing artifact), both measurable as (γ_fwd ± γ_bwd)/2; at fixed inference resolution coherent modes reverse about the midpoint while the fluctuation layer does not, because γ_S is ~34× larger there, yet both reverse under the exact derivative. Second, in a port-conditioned lattice interferometer the Cheshire-cat structure emerges unimposed as separate continuity equations for particle and polarization, with a local field rotating only the polarization phase at exactly 2B and appearing as a rigid imaginary generator shift. The involution identity and two-layer structure are verified from D=2^8 to 2^20, with self-averaging removing the 2^{-N/2} brightness obstruction among classes sharing a boundary modulation.","tokens_in":30754,"tokens_out":1533,"duration_ms":33325,"significance":"If the claims hold, the paper supplies an operational, measurable criterion (|γ_A| ≷ γ_S) for when a time-symmetric description exhibits a reversing arrow, separating boundary-condition physics from estimator artifacts. The Appendix C lemma/theorem chain is clean under stated hypotheses; sample-level identities at 10^{-14}–10^{-13} up to D=2^{20}, machine-precision interferometer responses (exact 2B precession, particle density invariant to ~10^{-16}), and the self-averaging route around the overlap obstruction are genuine strengths. The portable methodological half—decompose measured friction before interpreting it as physics—extends beyond the specific models to smoothed estimation and classical fluctuation-path settings. Code availability and labelled numerical checks further raise the evidentiary standard.","major_comments":[{"comment":"The headline claim that fluctuation-layer immunity is a property of the inference (not the ensemble) rests on exact-derivative dictionary-layer friction reversing about the midpoint (Sec. IV D: antisymmetric to 9.6×10^{-13}, half-window averages ±0.274). That control is reported only for the independent, involution-preserving class at N=8. Appendix C and Sec. VI E state that guaranteed antisymmetry of γ_exact requires independent conjugation-invariant seeds and real-symmetric H. The scalable N=20 results (Table II, Fig. 11) use bright classes built from |ψ(T)⟩ that break measure preservation. Sec. IV E already flags that “the dictionary layer of a class that breaks the involution” is “the one place where conditioning leaves no trace of reversal at all,” yet no exact-derivative dictionary-layer run is given for any involution-breaking class at either size (fifth limitation, Sec. VI E). Ei","section":"Sec. IV D, Sec. VI E, Table II, Abstract"},{"comment":"At N=20 the quoted mode-layer reversal depths depend strongly on the post-hoc ill-conditioned-window threshold q. Appendix B, Table IV shows the second-half γ_exact for mask X f=1/16 moving from −0.030 at q=0 to −0.198 at the adopted q=0.15 (and to −0.605 at q=0.30); depths in Table II are therefore convention-dependent summaries even though the sign is stable. The paper already reports threshold-insensitive residuals ρ_A/ρ_S, which is the right diagnostic. The main-text and Table II emphasis should be shifted onto those residuals (and onto P(flip)), with half-window depths clearly labelled as convention-dependent, so that magnitude comparisons across classes and sizes are not over-read.","section":"Appendix B, Table IV, Table II, Sec. V B"}],"minor_comments":[{"comment":"The factor-of-34 comparison of γ_S between layers (Sec. IV D) is quoted at a single Δt_cg=0.04; Appendix B shows γ_S ∝ Δt_cg, so the numerical factor should be stated together with the resolution, as is done for the no-flip inequality elsewhere.","section":"Sec. IV D, Appendix B"},{"comment":"Fig. 5(c)–(d) and Fig. 11 would benefit from an explicit panel or inset of |γ_A|/γ_S versus window center (or versus Δt_cg) so the dominance claim is visible without reading off dashed levels from the text.","section":"Fig. 5, Fig. 11"},{"comment":"The companion paper [1] is cited as under review (arXiv:2605.05604). For stand-alone readability, a one-paragraph restatement of what causal gEDMD returns (positive forward friction, vanishing time-symmetric friction) would help readers who lack [1].","section":"Sec. I"},{"comment":"Notation: A_w, (A)_w and ⟨A⟩_w are declared interchangeable, but the text also uses overlines for |g|^2-weighted ensemble means; a short notation table in Appendix A would reduce load.","section":"Sec. I, Sec. II D"},{"comment":"In Sec. III B point 2, P_x is introduced in the same sentence as a warning not to confuse it with the overlap g; breaking that sentence and defining P_x in a displayed equation would help.","section":"Sec. III B"},{"comment":"Table III lists C(Δt_cg) inconsistently across columns (40 / 20 / 10) while physical Δt_cg is 0.4 / 0.04 / 0.04; stating Δt_cg in time units in the table header would avoid misreading.","section":"Table III"}],"recommendation":"major_revision","confidential_remarks":"The skeptic’s concern about exact-derivative fluctuation-layer reversal being verified only for involution-preserving ensembles is load-bearing for the abstract’s strongest phrasing and is why I chose major_revision rather than minor_revision. The rest of the manuscript—theorem, interferometer, mode-layer scaling, self-averaging—is in good shape and would support a strong accept once that claim is either closed with a bright-class exact-derivative control or scoped down. Fit to a quant-ph theory journal is appropriate; the dual Cheshire-cat / many-body structure is unusual but methodologically unified."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The thing worth knowing is the operational split. Once you have the reflection involution on independent pre/post ensembles (real-symmetric H, conjugation-invariant seeds), every window-fitted friction decomposes uniquely into antisymmetric boundary physics and symmetric scheme bias, both readable from the same forward/backward pair. That is not just a rewrite of the companion causal-friction story; it is a measurable criterion for when a conditioned arrow reverses. The Cheshire half is also clean: port-only conditioning, separate continuity equations, 2B phase response, rigid imaginary generator shift, particle density frozen to machine precision. No ontology fight required.\n\nAppendix C is short and honest about its hypotheses. Sample-level identities hold at 10^{-14}–10^{-13} out to D=2^{20}. Self-averaging among bright classes that share the boundary modulation is a practical result: it actually removes the 2^{-N/2} obstruction. The factor-of-34 scheme disparity between mode and dictionary layers at fixed Δt_cg is the right way to state the two-layer claim—as degree, not as a metaphysical immunity of fluctuations.\n\nSoft spots, in proportion. The stress-test lands on a real gap, not a quibble: exact-derivative dictionary-layer reversal is shown for the independent class at N=8; the scalable bright classes break measure preservation, and the paper itself flags that combination as the place where conditioning can leave no reversal trace. So the slogan “immunity belongs to the inference, not the ensemble” is fully secured only where the involution holds. Mode-layer reversal under modulation is solid across classes and sizes. Second, half-window depths at N=20 move hard with the ill-conditioned-window threshold q even though the sign is stable—magnitudes are convention-dependent summaries. Code is promised, not shipped. Complex H is left as a sketch.\n\nNone of that inverts the central qualitative picture. This is for people who extract effective dynamics from trajectories under two-time boundary conditions—quantum smoothing, weak-value transport, classical fluctuation paths. It deserves a serious referee. I would bring it to reading group and engage; I would not treat the fluctuation-layer exact-derivative slogan as settled for the bright ensembles until that control is run.","headline":"Clean theorem-plus-numerics paper: the γ_A/γ_S split is real and useful; the portable “both layers reverse under exact derivative” claim is only fully checked for the involution-preserving ensemble at N=8.","tokens_in":31490,"tokens_out":576,"would_cite":true,"duration_ms":16699,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Conditioning on past and future splits measured friction into a reversible boundary arrow and an irreversible inference artifact, so the arrow of time has two layers.","keywords":["weak values","two-state vector formalism","dynamic mode decomposition","arrow of time","quantum Cheshire cat","pre- and post-selection","property currents","quantum smoothing"],"falsifier":"On a reflection-paired ensemble with independent conditioning, check whether the half-window mode friction reverses while the dictionary-layer forward friction stays positive at the stated coarse-graining interval, and whether both reverse when the exact derivative replaces finite differences; failure of the sample-level identity O'_w(t) = ε_O O_w(T−t) to machine precision would kill the theorem.","tokens_in":31259,"feed_emoji":"⏳","tokens_out":1075,"duration_ms":22538,"temperature":0.7,"pith_summary":"An observer who knows both the start and the end of a quantum run assigns intermediate properties that a purely causal observer never sees. This paper shows those assignments obey an exact reflection symmetry and can be read off trajectory data with the same generator-extraction tools used for ordinary dynamics. Every measured friction splits uniquely into an antisymmetric part fixed by the boundary conditions and a symmetric part injected by the finite-difference estimator; both parts come from the same forward and backward fits. At fixed resolution, coherent modes reverse their arrow at the midpoint while the fluctuation layer does not, because the estimator bias is far larger there—yet with the exact derivative both layers reverse, so the second layer’s immunity is a property of the description, not of the ensemble. The same machinery makes a Cheshire-cat separation of particle and polarization emerge from port-only conditioning as a pair of conservation laws, and the whole structure holds from 256 to a million dimensions.","feed_headline":"Two layers of time’s arrow split by how you measure","feed_subtitle":"Boundary conditions reverse coherent modes; the estimator bias does not—until you use the exact derivative.","key_machinery":"Two-state gEDMD plus the reflection involution: because weak values obey dA_w/dt = i⟨[H,A]⟩_w exactly, generator regression applies unchanged and supplies an exact-derivative baseline; the involution then forces the unique split γ_fwd = γ_A + γ_S and makes the physical and inferential arrows operationally separable.","core_discovery":"A reflection involution on the pre-/post-selected ensemble forces every window-fitted friction to decompose uniquely as γ_fwd = γ_A + γ_S, with γ_A antisymmetric about the midpoint and carrying the boundary-condition physics of coherent modes, and γ_S symmetric and generated by the differencing scheme; both are obtained from the same data as (γ_fwd ± γ_bwd)/2. At fixed inference resolution the arrow of time therefore has two layers: modes reverse, fluctuations do not, because γ_S is roughly 34 times larger at the fluctuation layer—yet with the exact derivative both reverse. Independently, port-only conditioning on a lattice interferometer produces unimposed Cheshire-cat property currents, wi","pith_inferences":["Experiments that report friction or dissipation under pre- and post-selection should publish both forward and backward estimators so readers can separate γ_A from γ_S.","The factor-of-two phase response to a local field is a clean weak-measurement signature that could be sought in existing Cheshire-cat neutron or photon setups.","If the sketched time-reversal extension for complex Hamiltonians holds, the two-layer split would apply to systems with magnetic fluxes and driven interferometers.","Self-averaging at large dimension suggests that boundary modulation design matters more than fine subspace structure for scalable two-state inference."],"forward_implications":["Whether a conditioned observer sees a reversing arrow is decided by the measurable inequality |γ_A| ≷ γ_S at that layer and resolution, not by interpretation.","The brightest post-selection class sharing the same boundary modulation can be used at large N without changing the physics, removing the exponential overlap obstruction.","Cheshire-cat separation is a dynamical pair of conservation laws readable from trajectories, not only a single-instant assignment.","Apparent friction that scales with the differencing interval and vanishes under the exact derivative is an estimator artifact, not a property of the ensemble.","The same decomposition applies wherever effective dynamics is inferred from trajectories with both-end boundary conditions, including classical fluctuation paths."],"fun_headline_variants":["Two-layer time arrow: modes reverse, fluctuations do not","Reflection splits friction into γ_A boundary and γ_S bias","Exact derivative flips both layers of the quantum time arrow","Port-only conditioning yields unimposed Cheshire-cat currents","Coherent modes reverse at midpoint; estimator layer stays put"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The exact symmetry proof needs a real-symmetric Hamiltonian and independently drawn, conjugation-invariant boundary seeds; a complex Hamiltonian or a post-selection built from the evolved state breaks the argument as written.","fun_headline_variants_meta":{"raw":{"variants":["Two-layer time arrow: modes reverse, fluctuations do not","Reflection splits friction into γ_A boundary and γ_S bias","Exact derivative flips both layers of the quantum time arrow","Port-only conditioning yields unimposed Cheshire-cat currents","Coherent modes reverse at midpoint; estimator layer stays put"]},"model":"grok-4.5","effort":"low","cost_usd":0.005441,"raw_usage":{"total_tokens":1668,"prompt_tokens":1033,"num_sources_used":0,"completion_tokens":85,"cost_in_usd_ticks":54408000,"prompt_tokens_details":{"text_tokens":1033,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":550,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":1033,"tokens_out":85,"duration_ms":9659,"temperature":1.0,"reasoning_tokens":550,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T14:26:40.976555+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On a reflection-paired ensemble with independent conditioning, check whether the half-window mode friction reverses while the dictionary-layer forward friction stays positive at the stated coarse-graining interval, and whether both reverse when the exact derivative replaces finite differences; failure of the sample-level identity O'_w(t) = ε_O O_w(T−t) to machine precision would kill the theorem.","supporting_citations":[],"review_version":1}