{"id":"305d9ea4-0bf8-43aa-b0f0-f61ef87b118c","arxiv_id":"1908.00468","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Data informativity gives exact conditions for when measured data, rich or not, suffice for certifying controllability, designing stabilizing or deadbeat feedback, or solving LQR from data.","lead":"This paper defines when measured data are informative enough to answer control questions directly, without first building a model. It shows that for some tasks, like state-feedback stabilization, the data need not be rich enough for system identification, while for linear quadratic regulation they must be.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the proofs are internally consistent; the explicit noise-free LTI assumption is a scope limitation, not a flaw.","rationale":"The reader's verdict is ACCEPT with moderate confidence, and I agree that the noiseless, known-dimension LTI assumption is the most fragile point of the paper. However, I do not regard this as a load-bearing technical objection: the paper states the assumption explicitly, and the central claim is a set of necessary and sufficient conditions within that exact-data model class. I examined the main proof steps where a hidden flaw would most likely live: the data-driven Hautus test (Theorem 8), the common-feedback-gain argument (Lemma 15 and Theorem 16), the LMI characterization (Theorem 17), and the LQR necessity theorem (Theorem 26). In each case the algebra is internally consistent. The only omitted proof is the stabilizability half of Theorem 8, but the controllability proof's construction extends directly by restricting λ to the closed exterior of the unit disk, so the omission does not create a correctness risk. The finite-horizon LQR question and the input/output necessary-and-sufficient conditions are explicitly left open by the authors and do not undermine the stated central results. A useful independent check would be a small randomized re-verification of the stabilizability construction, but I found no concrete reason to expect it to fail. Therefore the appropriate outcome is to leave the reader's ACCEPT verdict unchanged.","tokens_in":25587,"tokens_out":21382,"duration_ms":221848,"concrete_test":"Implement the omitted stabilizability direction of Theorem 8: for random small data (X−, U−, X+) generated by LTI systems, find all λ with |λ| ≥ 1 where rank(X+ − λX−) < n; when such a λ exists, use the paper's complex-λ construction with the real and imaginary parts of a left null vector to build a real consistent (Ā, B̄) and check via the PBH test that (Ā, B̄) is not stabilizable. If this construction ever fails, the 'only if' direction of Theorem 8 for stabilizability would be invalid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the main theorems (Theorems 8, 16, 17, 26, 29, 34, and Corollary 36) and their proofs. The affine structure of the consistent-set Σ_i/s is used correctly, and the rank conditions follow from it. The complex-eigenvalue construction in Theorem 8 is valid, and the same argument with |λ| ≥ 1 covers the omitted stabilizability direction, so the omission is a presentation choice rather than a gap. Lemma 15's nilpotency argument is sound, and the dichotomy in Theorem 26 (identification versus the K = 0 pathological case) follows from the homogeneous-system argument. No internal inconsistency or missing step that would change the central claim was found. The weakest point remains the stated assumption that the data are generated exactly by a noiseless LTI system of known dimension; under noise the consistent set defined by (6) or (42) can be empty or misleading, but the paper explicitly flags this and all theorems are conditional on it. This is a scope limitation rather than a correctness risk.","agreement_with_reader":"partial"},"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-14T15:55:03.466709+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}