{"id":"e13ad8a0-3829-4152-8cdc-e3fdbe8b5b6f","arxiv_id":"2607.11551","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Trajectory libraries define behavioral spaces that enable exact future prediction from past windows under continuation maps, spectral inclusion of visible eigenvalues, and immersion conditions.","lead":"Libraries of stored output trajectories can predict the future of a new trajectory from its past without identifying the generating model. The theory gives exact conditions for linear systems, noise robustness, compositional generalization via interconnections, and extension to some nonlinear systems via immersion.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The paper’s strongest claim (exact prediction from mixed libraries under richness and r ≥ s_mix, and exact out-of-library prediction under visible spectral inclusion) is supported by self-contained linear-algebra arguments that do not rely on undisclosed parameters or circular fitting. The reader’s weakest-assumption diagnosis matches the places where the exact statements would fail if the hypotheses were dropped. Because those hypotheses are classical, explicitly stated, and necessary rather than hidden, they do not constitute a correctness risk that would move the verdict. The recommended verification step simply re-checks the foundational continuation map; success leaves the ACCEPT recommendation intact.","tokens_in":37725,"tokens_out":390,"duration_ms":4680,"concrete_test":"Independently re-derive the continuation identity Hf = L⋆ Hp of Lemma 3(i) from Fact 1 and the factorization HY = OT X alone (without invoking the latent-variable argument of part (ii)); confirm that the same L⋆ acts correctly on every admissible past when rank(X) = n. If the identity fails under those hypotheses the foundational mechanism collapses; otherwise the strongest claim stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims rest on classical linear-algebraic facts (continuation maps from the observability index, image factorization of template matrices under data richness, and spectral inclusion for scalar-output diagonalizable systems). These are stated with explicit hypotheses (r ≥ s_mix, rank Xi = ni, diagonalizability for the spectral results) and proved from standard systems theory without circularity or hidden free parameters. The reader correctly flags data richness and the past-length requirement as the weakest assumptions; they are classical, necessary for the exact statements, and clearly scoped. No internal inconsistency or load-bearing gap that would overturn Theorem 1 / Proposition 2 / Theorem 5 was found.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper develops a theory of trajectory prediction from libraries of stored output templates, without identifying the generating state-space model. Libraries of finite-length output windows from one or more LTI systems define behavioral spaces; exact prediction is characterized via continuation maps (Lemma 3), mixed-library factorization (Lemma 4, Theorem 1), and, for out-of-library systems, spectral inclusion of output-visible eigenvalues (Proposition 2 / Theorem 5). Robustness to noisy observations and libraries is quantified by exact error identities (Theorems 3–4); interconnection constraints (feedback, series, parallel) generate new libraries with emergent modes (Lemmas 5–10, Theorem 6, Proposition 6); and the framework extends to nonlinear systems whose outputs are contained in or immersed into finite-dimensional linear behaviors (Proposition 7). Numerical examples illustrate noisy prediction, sparse modal recovery, feedback-generated modes, and nonlinear immersion.","tokens_in":37874,"tokens_out":724,"duration_ms":7312,"significance":"If the results hold, the paper supplies a clean algebraic foundation for template-based prediction machines that generalize beyond stored trajectories and, via spectral inclusion or interconnection, beyond the generating systems themselves. The core contributions rest on standard linear-algebra and systems-theory facts (observability index, data richness, Vandermonde structure) with explicit hypotheses and exact error identities; the interconnection construction and the immersion route to nonlinear trajectories are particularly distinctive. Strengths include parameter-free exact statements under clearly scoped assumptions, explicit equivalence between spectral inclusion, output inclusion and linear immersion (Lemmas 11–12, Corollary 2), and concrete numerical illustrations. The work sits naturally at the intersection of behavioral data-driven control and sequence completion, and opens well-defined directions (library design, inputs, online recursion).","major_comments":[],"minor_comments":[{"comment":"The manuscript is long and dense; a short roadmap or table of main theorems (Theorem 1, Proposition 2, Theorem 6, Proposition 7) early in Section II would help readers navigate the progression from single-system to multi-system, out-of-library, interconnection and nonlinear cases.","section":null},{"comment":"Notation for the continuation map switches among L*, Lmix and L; a single consistent symbol (with subscripts only when needed) would reduce cognitive load, especially in Sections V–VII.","section":null},{"comment":"Figure 1 (noisy-data histogram) and Figure 3 (sparse modal prediction) would benefit from explicit axis units and a brief caption statement of the SNR / r / T values used, so that the plots are self-contained.","section":null},{"comment":"The restriction of the spectral results (Propositions 2–4, 6–7) to scalar-output diagonalizable systems is clearly stated, but a short remark on the multi-output or non-diagonalizable obstacles (and whether they are merely technical) would help readers assess the scope.","section":null},{"comment":"A few typographical slips remain (e.g., “visible the spectral inclusion” in the proof of Proposition 2; occasional missing spaces around math). A final copy-edit pass would polish the presentation.","section":null}],"recommendation":"accept","confidential_remarks":"The paper is technically solid and well within the scope of a strong systems/control journal. The reader’s and skeptic’s assessments align with mine: the load-bearing claims are classical and correctly proved under explicit hypotheses. I see no reason to request major technical revisions; the minor presentation points can be handled in production or a light revision cycle."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a clean, self-contained theory paper that takes the behavioral / fundamental-lemma line and systematically turns multi-system output libraries into prediction machines. The punchline is Theorem 1 plus the out-of-library spectral results: under data richness and past length at least the aggregate observability index, any least-squares fit of the past against the mixed library recovers the future exactly for every trajectory from any system in the library; exact out-of-library prediction holds when the new system’s output-visible spectrum sits inside the library spectrum (Proposition 2 / Theorem 5). That is new as a package, even though the single-system continuation map is classical.\n\nWhat they do well is the architecture. They give the mixed-library factorization, the exact noise-error identities (Theorems 3–4), the sparse modal version that replaces the global r ≥ s_mix requirement by a spark condition depending only on the number of active modes, and the construction of feedback / series / parallel libraries from atomic data via the compatibility kernel Θ, which genuinely produces emergent modes without new experiments. The immersion extension to nonlinear systems with finite-dimensional observation spaces is the natural next step and is cleanly stated. Proofs stay inside standard linear algebra and systems theory; the numerical examples are only illustrative and do not pretend otherwise. Citations sit where they should (Willems, Markovsky–Rapisarda, Koopman immersion literature).\n\nSoft spots are real but proportionate. Data richness and r ≥ s_mix (or the spark condition) are load-bearing; without them feasibility and uniqueness fail. The sharp spectral statements need scalar output and diagonalizability. The nonlinear claim is limited to systems that already immerse into a finite-dimensional linear behavior. None of these are hidden, and none overturn the central claims under the stated hypotheses. The paper is autonomous-first; inputs are left for later.\n\nThis is for people who work on data-driven simulation, behavioral control, or dictionary-style sequence completion for dynamical systems. It deserves a serious referee. I would bring it to reading group and I would cite the mixed-library and interconnection constructions. Send it to peer review.","headline":"Solid theory paper that turns multi-system trajectory libraries into exact predictors, with clean spectral and interconnection extensions; classical assumptions, no load-bearing holes.","tokens_in":38446,"tokens_out":526,"would_cite":true,"duration_ms":7612,"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":"Stored output trajectories form behavioral spaces that predict futures without identifying the generating system.","keywords":["trajectory prediction","template libraries","behavioral spaces","continuation maps","spectral inclusion","data-driven prediction","immersion","interconnected libraries"],"falsifier":"Build a library from several diagonalizable linear systems whose visible eigenvalues are known; generate a new scalar-output trajectory whose spectrum is not contained in that set and check whether the least-squares predictor still recovers the future exactly for past length equal to the aggregate observability index.","tokens_in":38631,"feed_emoji":"📈","tokens_out":545,"duration_ms":4870,"temperature":0.7,"pith_summary":"The paper shows that a library of finite-length output windows, drawn from one or several linear systems, can be used as a prediction machine: given only the past of a new trajectory, a linear fit against the library past block recovers the future exactly whenever the past is long enough and the data are rich. Exactness holds even when the trajectory comes from a different system whose visible eigenvalues already appear in the library, and it continues to hold after the library is enlarged by feedback, series or parallel interconnection constraints that create new modes. The same mechanism extends to nonlinear systems whose outputs lie in a finite-dimensional linear observation space. The practical payoff is that prediction no longer requires first recovering a state-space model; the stored templates themselves act as the generative model.","feed_headline":"Templates alone predict futures without system ID","feed_subtitle":"Stored output windows form behavioral spaces that recover exact futures when spectra match","key_machinery":"The continuation map of the aggregate (or interconnection-generated) library: the linear relation Hf = Lmix Hp that turns any feasible encoding of the past into an exact future decoding, independent of which particular coefficient vector is chosen.","core_discovery":"Libraries of stored output trajectories define behavioral spaces that serve as exact prediction machines: under data richness and a past window at least as long as the aggregate observability index, any coefficient vector that matches an observed past against the library yields the correct future for every trajectory generated by any system represented in the library, and for many systems outside it whose output-visible spectrum is covered by the library.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Templates predict futures without identifying the system","Stored trajectory libraries form exact prediction machines","Behavioral spaces from templates recover futures via past matching","Output templates forecast trajectories under spectral coverage","Libraries of templates yield futures beyond generating systems"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The stored trajectories must be rich enough to span every mode of every template system, and the observed past must be at least as long as the observability index of the whole library.","fun_headline_variants_meta":{"raw":{"variants":["Templates predict futures without identifying the system","Stored trajectory libraries form exact prediction machines","Behavioral spaces from templates recover futures via past matching","Output templates forecast trajectories under spectral coverage","Libraries of templates yield futures beyond generating systems"]},"model":"grok-4.5","effort":"low","cost_usd":0.00723,"raw_usage":{"total_tokens":1729,"prompt_tokens":690,"num_sources_used":0,"completion_tokens":66,"cost_in_usd_ticks":72300000,"prompt_tokens_details":{"text_tokens":690,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":973,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":690,"tokens_out":66,"duration_ms":9686,"temperature":1.0,"reasoning_tokens":973,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T04:48:18.418040+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Build a library from several diagonalizable linear systems whose visible eigenvalues are known; generate a new scalar-output trajectory whose spectrum is not contained in that set and check whether the least-squares predictor still recovers the future exactly for past length equal to the aggregate observability index.","supporting_citations":[],"review_version":1}