{"id":"cf4cf796-832b-4d8c-8def-6b7744a2a997","arxiv_id":"2607.19218","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Heavy-tailed noise in diffusion can be exactly calibrated via a tempered-stable volatility chain, but is provably a nuisance whenever the denoiser sees it and the chain is drawn independently of the data.","lead":"A new diffusion model for financial returns drives heavy-tailed noise with a persistent 'volatility clock' — an AR(1) chain of tempered-stable steps — with closed-form formulas for fat tails and volatility clustering. The paper then proves that when the generator is told the clock, the noise law provably stops mattering; control returns only when the clock is inferred from the data by a variational encoder.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Invariance theorem rests on x0⊥A; v2 breaks this premise and restores control, so the 'nuisance' claim is narrower than the abstract states.","rationale":"The reader identified the same weakest assumption: x0⊥A in v1 training. This is indeed the most load-bearing premise of Proposition 5, because the theorem's conclusion q(x0|A)=q(x0) collapses without it. The paper is transparent about this: Section 3.1 states the independence, Section 3.5 introduces v2 precisely to break it, and Section 4.6 shows the consequence (shock ratio 3.00 vs 0.88). Thus the concern is not a hidden flaw but a boundary on the theorem's applicability. The abstract, however, omits the qualifier and overstates the invariance as a general property of conditioning. This justifies the reader's CONDITIONAL verdict: the paper's central claims are correct under their stated conditions, but the headline presentation overgeneralizes. A dependent-A ablation is the cleanest way to test the theorem's boundary experimentally; the paper's v2 already provides strong evidence in that direction. The verdict remains CONDITIONAL because the core mathematics and the majority of experiments are sound, but the scope limitation and the small-scale empirical basis call for conditions.","tokens_in":13910,"tokens_out":14687,"duration_ms":153746,"concrete_test":"Train a conditioned model with dependent A: after drawing x0 from the GARCH-t data, set A_i from the path's true conditional variances (or a correlated noise), rescaled to unit mean; keep architecture, schedule, and budget identical to Section 4.3. Then perform the Section 4.4 ×8 shock intervention. If the envelope ratio at the shock peak moves substantially away from ~1 toward the naive 2.83, the independence premise is exactly what produces the invariance, confirming the concern. Alternatively, re-run v2's intervention with a held-out seed and verify the shock ratio exceeds 2.5 (the paper reports 3.00±0.14), which would again show that breaking independence restores control.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 5 (Section 3.4) proves p(x0|A)=q(x0|A)=q(x0) using q(x0|A)=q(x0), which holds only because training draws x0 and A independently (Section 3.1, Eq. 9). This is a stated assumption, but it is the load-bearing one: the theorem does not show that conditioning on A is generally nuisance-invariant; it shows that the law of an independent nuisance is irrelevant in the exact-denoiser limit. The paper's own v2 (Section 3.5) demonstrates that breaking this premise by amortized inference makes the same ×8 intervention move the envelope by 3.00±0.14 instead of 0.88±0.03. Consequently, the abstract's 'when the denoiser is conditioned on the mixing variables, their law is a nuisance' is incomplete: the qualifier 'trained with A independent of x0' is essential. Any user who conditions on an observation informative about x0 (e.g., a volatility estimate derived from the path) reopens the mechanism channel. The proof is correct and the experiments are consistent, but the claim's scope is narrower than the headline.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the Denoising Subordinated Probabilistic Model (DSPM), a DDPM whose per-coordinate noise variance is a stationary AR(1) chain driven by tempered-stable increments. Conditionally on the chain, the standard DDPM posterior, ELBO, and ancestral sampling hold unchanged (Propositions 1--2). The authors prove closed-form identities for excess kurtosis and squared-noise autocorrelation (Proposition 3) and give an analytic calibration inverse with a feasibility bound (Corollary 1); DDPM, i.i.d.-stable, and shared-scale Student-t noise are recovered as boundary cases (Proposition 4). Proposition 5 states that, with exact conditional reverse kernels, the generated distribution is invariant to the law of the mixing chain if the denoiser is conditioned on it. Experiments with GARCH-t data confirm that conditioned models do not transmit the mixing law, blind models do, and a variational encoder (v2) that couples A to x0 restores control. The paper ships reproducible scripts for all numbers.","tokens_in":14139,"tokens_out":15676,"duration_ms":153821,"significance":"The DSPM construction is elegant and the closed-form calibration is a genuine practical contribution. The paper is unusually honest: code is provided, every headline number is seed-averaged where it matters, and the limitations of the blind ablation and the v2 fidelity/control trade-off are stated explicitly. The nuisance-invariance theorem is a useful delimiting result for the heavy-tailed diffusion literature. Its scope, however, is narrower than the abstract says: it applies to a latent A that is independent of the data at training time, not to any conditioning on A. The v2 experiment is a well-designed controlled demonstration of exactly this boundary.","major_comments":[{"comment":"Proposition 5's conclusion p(x0|A)=q(x0|A)=q(x0) uses q(x0|A)=q(x0), which is valid only because training draws x0 and A independently (stated in §3.1). The abstract's sentence 'when the denoiser is conditioned on the mixing variables, their law is a nuisance' omits this premise and is false as written: v2 (§3.5, Table 4) conditions the denoiser on A, yet interventions move the envelope by 3.00±0.14 exactly because the encoder breaks x0⊥A. Please add the independence qualifier to the abstract and to the formal statement of Proposition 5, and rephrase the headline claim as 'an independently drawn mixing chain is a nuisance in the exact-denoiser limit.'","section":"Abstract; §3.4, Eq. (9)"},{"comment":"The blind-model quantitative claim ('blind models transmit the mechanism exactly as calibrated') rests on a single training run and on a self-described 'crude' pairing of a marginal denoiser with A-scaled updates. This is acceptable as an ablation but not as the 'exactly as calibrated' headline. Please either supply multiple seeds for the blind rows or weaken the wording in the abstract and Section 4.5.","section":"§4.5, Table 3"}],"minor_comments":[{"comment":"The independence hypothesis x0⊥A should be part of the formal statement of Proposition 5, not only a sentence in the preceding paragraph. As written, readers may quote the proposition without the premise.","section":"§3.4"},{"comment":"The claim that restoring the log-determinant term is 'necessary and sufficient' for the encoder to train is stronger than the evidence: the paper shows two failure modes and one success. Suggest phrasing it as 'necessary in our experiments' or providing a formal argument.","section":"Remark 2, Eq. (10)"},{"comment":"The single-run blind ablation should either report seed variability or be clearly marked in the abstract, since the abstract presents the 0.151 vs 0.161 agreement as a headline result.","section":"§4.5"},{"comment":"References to 'Anonymous, 2026' and 'Baule, 2026' are working papers without full author lists. Unless the submission is double-blind, these should be updated to complete archival citations before publication.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is fundamentally sound: the core algebra (Propositions 1, 3, Corollary 1) is correct, and the experiments are honest and reproducible. The main risk is that the unqualified nuisance-invariance claim could be read as a proof that conditioning on any volatility estimate derived from the path is useless, which is not what the theorem says; the required revision is mostly reframing plus a small amount of additional empirical evidence for the blind ablation. I would not reject."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Candid take: this is a solid paper with two genuinely new pieces — the DSPM construction (a DDPM with a correlated tempered-stable AR(1) volatility chain, giving closed-form kurtosis and squared-noise ACF and an exactly invertible calibration) and Proposition 5, the nuisance-invariance theorem for conditioned denoisers. I checked the core algebra; Proposition 3's identities, Corollary 1's inverse map, and the feasibility bound are correct. The experiments are honestly run: pooled estimators, three seeds for the main comparisons, the seed-variance retraction in §4.3(b), and a documented failure mode when the simplified loss is used without the SV term. The blind-model result — calibrated lag-one ACF 0.151 vs 0.161 with geometric decay — is a clean demonstration that the conditioning bit, not the noise law, controls transmission.\n\nThe soft spots are real but most are admitted in the paper. The abstract oversells two things. First, the v2 'coupling the chain to the data' is actually a log-normal Clark clock, not the tempered-stable chain of Definition 1; the substitution is stated in §3.5, but the abstract presents it as coupling that chain. Second, the headline numbers (pooled correlation 0.76, shock ratio 3.07) are the optimistic versions; the honest within-path correlation is 0.567 and the seeded shock ratio is 3.00±0.14. The nuisance-invariance theorem is narrower than the abstract suggests: it holds only under independent training draws of x0 and A and in the exact-denoiser limit. The stress-test note is correct on this — condition the sampler on anything informative about x0 and you reopen the mechanism channel. The paper's own v2 demonstrates exactly that, so this is a scope qualifier, not a flaw in the proof.\n\nThe single 'Anonymous 2026' citation grounding the novelty of Proposition 5 is unverifiable; that's an overlap risk the editors should check.\n\nWho it's for: anyone working on heavy-tailed diffusion mechanics or financial generative models. It deserves a serious referee. The strongest referee angle is the abstract/body mismatch and the precise scope of the invariance claim.","headline":"Correct math and an honest experiment program, but the abstract overstates the coupling claim and the invariance theorem's scope — still worth refereeing.","tokens_in":14801,"tokens_out":2698,"would_cite":true,"duration_ms":28114,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"In exact conditioned diffusion models, the choice of heavy-tailed noise law cannot shape the generated distribution; only coupling the noise to the data restores control.","keywords":["denoising diffusion","heavy-tailed noise","volatility clustering","tempered-stable subordinator","variance mixture","stochastic volatility","nuisance parameter","financial time series generation"],"falsifier":"Train a conditioned denoiser with independent A, let it converge, and sample under a designed ×8 volatility path; Proposition 5 predicts the amplitude envelope ratio stays near 1 (measured 0.88±0.03), against the naive √A prediction of 2.83 — a measured ratio far from 1 would falsify the invariance.","tokens_in":13572,"feed_emoji":"📈","tokens_out":14339,"duration_ms":135971,"temperature":0.7,"pith_summary":"This paper tries to establish two opposite things at once. Constructively, it defines a diffusion model whose per-coordinate noise variance follows a stationary AR(1) chain driven by tempered-stable jumps, so the noise itself carries volatility clustering; it derives closed-form formulas for kurtosis and squared-noise autocorrelation and an exactly identified, analytically invertible calibration, with Gaussian, i.i.d. stable, and Student-t noise as boundary cases. Delimitingly, it proves that if the denoiser is conditioned on the mixing chain and the chain is drawn independently of the data, the exact reverse process terminates at the data distribution regardless of the chain's law, so interventions on volatility do nothing. Controlled experiments confirm this invariance and show that a blind denoiser transmits the calibrated clustering exactly. The paper then breaks the independence with a variational encoder and recovers control over shocks, at the cost of unconditional kurtosis falling below the Gaussian baseline; the paper is explicit that fidelity and control are demonstrated separately, not jointly.","feed_headline":"When denoisers see the noise, its law stops mattering","feed_subtitle":"A new proof and experiments show only coupling volatility to data restores control over generated returns.","key_machinery":"The central object is the variance chain A: a stationary AR(1) recursion A_i = φ A_{i-1} + η_i, with tempered-stable increments η_i normalized so E A_i = 1. Because every forward covariance is a multiple of the same diagonal matrix D_A, the conditional Gaussian posterior and denoising objective factorize, so the denoising diffusion machinery survives verbatim once the denoiser is fed √A and whitened inputs. The key identities are K = 3(1−α)/(θ(1+φ)) and ρ_r2(h) = v φ^h/(3v+2) with v = Var(A_i), giving an exactly identified calibration with analytic inverse and feasibility bound ρ_r2(1) < K/(3(K+2)). The load-bearing theorem is Proposition 5: under training-time independence of x0 and A, exac","core_discovery":"The paper's central claim is a dichotomy. Constructively, the proposed Denoising Subordinated Probabilistic Model (DSPM) puts a stationary AR(1) tempered-stable chain inside the noise variance; conditioned on the chain, the denoising diffusion probabilistic model's posterior and objective are unchanged, and squared-noise autocorrelation is closed-form, giving exact calibration. Delimitingly, Proposition 5 proves that if the denoiser is conditioned on the chain and the chain is independent of the data, the exact reverse chain terminates at the data distribution — the chain's law is a nuisance and interventions do nothing. Experiments confirm both halves: conditioned models reproduce clusterin","pith_inferences":["Testable extension: run the four-way conditioned comparison at larger scale with a strict measure of the exact-denoiser limit; because Proposition 5 is scale-free, any systematic between-noise difference in dependence structure would expose a finite-T or receptive-field effect the paper leaves open.","The feasibility bound ρ_r2(1) < K/(3(K+2)) (always below 1/3) is a cheap pre-training screen: a dataset with lag-one squared autocorrelation above the bound at matched kurtosis cannot be described by a single AR(1) mixing chain, pushing toward multi-factor (supOU) volatility before any neural training.","For financial scenario generation, the paper's dichotomy means conditioned heavy-tailed diffusion is inert for volatility stress-testing unless the latent is coupled to the path; the decisive test would be path-dependent pricing under generated measures, which the paper leaves for future work.","The paper states plainly that v2 runs on a log-normal volatility clock, not the tempered-stable chain of the constructive section; a model carrying the tempered-stable prior via Monte-Carlo KL, or with a supOU multi-factor chain, is the natural step toward joint fidelity and control."],"forward_implications":["Conditioned models reproduce the data's volatility clustering regardless of whether the noise is Gaussian, i.i.d. stable, shared Student-t scale, or a persistent tempered-stable chain; all measured lag-one squared-return autocorrelations land within 0.017 of the data's 0.159.","In the exact-denoiser limit, scenario design through the mixing chain is inert: a ×8 volatility shock changes the generated amplitude envelope by a factor 0.88±0.03, where √A-scaling would predict 2.83.","Blind denoisers transmit the mechanism as calibrated: blind DSPM samples show lag-one squared-return autocorrelation 0.151 against the calibrated 0.161, with the predicted geometric decay, while the i.i.d. variant's clustering collapses to 0.008.","Coupling the chain to data through a variational encoder restores quantitative control: the same ×8 shock scales the envelope by 3.00±0.14 and the calm regime by 0.52±0.01, bracketing naive √A predictions.","The simplified denoising loss cannot train the volatility encoder on its own because the whitened residual is nearly scale-free in A; the exact stochastic-volatility likelihood's log-determinant term is required."],"fun_headline_variants":["Denoiser sees volatility, volatility law becomes nuisance","Independent volatility clock is inconsequential in diffusion","Diffusion noise law irrelevant when denoiser is conditioned","Coupling volatility to data restores diffusion control"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The invariance theorem presupposes that in training the data and the volatility chain are drawn independently, and that reverse kernels are exact; couple the two — as the paper's own coupled model does — and the theorem no longer applies.","fun_headline_variants_meta":{"raw":{"variants":["Denoiser sees volatility, volatility law becomes nuisance","Independent volatility clock is inconsequential in diffusion","Diffusion noise law irrelevant when denoiser is conditioned","Coupling volatility to data restores diffusion control"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000241,"raw_usage":{"total_tokens":1415,"prompt_tokens":859,"completion_tokens":556,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":495}},"tokens_in":603,"tokens_out":556,"duration_ms":6294,"temperature":1.0,"reasoning_tokens":495,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T13:05:12.632063+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train a conditioned denoiser with independent A, let it converge, and sample under a designed ×8 volatility path; Proposition 5 predicts the amplitude envelope ratio stays near 1 (measured 0.88±0.03), against the naive √A prediction of 2.83 — a measured ratio far from 1 would falsify the invariance.","supporting_citations":[],"review_version":1}