{"id":"99a48cd7-c0f8-4d4e-8fcb-e36360adb796","arxiv_id":"2411.17014","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper proposes a temperature-parameterized entropy model for parking occupancy and a dynamic-programming policy (TIPP) that outperforms two simple policies in a simulated garage.","lead":"A parking-policy paper borrows the canonical ensemble from statistical mechanics, fitting a single 'temperature' to predict how full each floor of a garage is, then uses dynamic programming to send cars to the best floor. The paper reports that its Temperature-Informed Parking Policy beats two simple heuristic policies in simulation, but the central occupancy equation is internally inconsistent and the evaluation is circular.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. 1 is not a valid probability for positive energies (it gives p>1), and its low-temperature limit contradicts the text; the entropy-model foundation is internally inconsistent, so the central claim collapses.","rationale":"The reader's weakest assumption identifies the thermal-equilibrium premise as the fragile point, and that is indeed a legitimate concern for real garages with rush-hour surges. However, the more fundamental, decisive flaw is that Eq. 1 cannot represent a probability for the positive energies the paper explicitly uses: it yields values in (1,2), and its low-temperature limit is 2 rather than 0. This internal contradiction invalidates the prediction model regardless of whether a garage is in equilibrium. The DP control recursion (Eq. 5) is mathematically coherent in isolation, but it consumes probabilities p_i that the entropy model cannot provide. The simulation experiments are also circular because the garage occupancy is generated from the same entropy model TIPP assumes, so the reported \"outperformance\" only demonstrates self-consistency, not real-world predictive or policy value. The reported MSE of 0.173 on binary outcomes further undermines the claimed \"excellent fit.\" These issues are not cosmetic; they strike at the central claim. Since the reader's REJECT verdict already reflects this level of concern, no adjustment to the verdict is needed.","tokens_in":6802,"tokens_out":3408,"duration_ms":33795,"concrete_test":"Evaluate Eq. 1 at E=1, k_B T=1: the computed value is 2e\\/(1+e) ≈ 1.462, which exceeds 1, immediately showing it is not a probability. Then take the low-temperature limit for fixed positive E: lim_{T→0} p = 2, not 0, contradicting the paper's claim that low temperatures give low occupancy probabilities. This arithmetic check settles whether Eq. 1 is internally valid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing premise of the prediction component is Eq. 1 in Section III: p(i)=2e^{E/kT}\\/(1+e^{E/kT}). For any positive E(i), which the paper assumes throughout (energy proportional to squared distance in Section III, and E(i)=(i/N)^2 in Section V), the argument x=e^{E/kT} exceeds 1, so p=2x\\/(1+x)>1, approaching 2 as T→0. Thus the formula outputs probabilities greater than 1 exactly in the regime the text describes as \"low temperatures... states will have a low probability of being occupied.\" This is not a disagreement with external consensus; it is an internal mathematical contradiction. Because the entropy model is the sole source of the occupancy probabilities p_i used in the dynamic-programming recursion (Eq. 5), the policy's inputs are not well-defined even under the thermal-equilibrium assumption. The empirical validation is likewise problematic: the pooled MSE of 0.173 on 0-1 fill-rate data corresponds to an RMS error of roughly 0.42, yet the paper calls the fit \"excellent.\" A corrected formula might be salvageable, but as written the central claim that a single temperature predicts occupancy is unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a single-parameter 'temperature' model for the occupancy distribution of spots in a multi-story parking garage, with the occupancy probability of a spot of energy E claimed to be given by Eq. (1). Based on these probabilities, it formulates a dynamic-programming policy (TIPP) that chooses which lower floor to visit next, and compares TIPP with Benchmark, Inverse, and Optimal policies in a simulated 10-level garage with 30 spots per level at temperature 0.5. The paper also reports an in-sample MSE of 0.173 for the entropy-model fit on ten real parking lots, and concludes that the model is accurate, sample-efficient, and suitable for real-time parking assignment.","tokens_in":118,"tokens_out":6783,"duration_ms":92036,"significance":"The dynamic-programming component is a genuine and clearly stated contribution: given reliable occupancy probabilities, the recursion in Eqs. (5)-(6) defines a sensible, implementable policy, and the simulation protocol is transparent. The paper also has the virtue of explicitly acknowledging that TIPP does not beat the optimal policy on average. However, the central model equation is mathematically invalid as printed, the reported fit is in-sample and of poor quality, and the simulation validates the policy only under the same model that generated the data. These issues are load-bearing for the paper's main claims, so the manuscript cannot be accepted in its current form.","major_comments":[{"comment":"Equation (1) is not a valid probability for any positive energy E(i). Since E(i)>0 throughout the paper (E is proportional to squared distance in Section III and E(i)=(i/N)^2 in Section V), the argument x=exp(E/k_B T) exceeds 1, and p(i)=2x/(1+x) lies strictly between 1 and 2, approaching 2 as T approaches 0. This contradicts the text's claim that 'at low temperatures, states will have a low probability of being occupied.' Because Eq. (1) is the only source of the probabilities p_i used in the dynamic-programming recursion Eq. (5), the policy's inputs are not well-defined. A corrected Boltzmann factor of the form p ∝ exp(-E/k_B T) would restore probabilities in [0,1], but then all fits, figures, and simulation temperatures would need to be redone.","section":"Section III, Eq. (1)"},{"comment":"The validation reported for the entropy model is not predictive. The temperature for each lot is obtained by minimizing the MSE of the same entropy model on that lot using gradient descent, and the energy-distance relation E ∝ d^2 was itself selected as the best fit on the same data. Reporting the pooled in-sample MSE of 0.173 as 'excellent' is unsupported: on 0-1 fill-rate data, this corresponds to a root-mean-square error of roughly 0.42. No held-out data, cross-validation, or uncertainty estimates are provided, so the claim that the model is an accurate representation of parking-spot occupancy is not established.","section":"Section III, Figure 2"},{"comment":"The simulation cannot validate the entropy-based prediction. The garage occupancy is generated from Eq. (1) at a temperature of 0.5, and TIPP's prediction component uses exactly the same model with a fitted temperature. Thus the experiments show only that, conditional on the entropy model being correct, the dynamic-programming policy is useful; they provide no evidence that Eq. (1) predicts real parking occupancy. This circularity, combined with the invalid Eq. (1), means the central claim that a single temperature parameter can predict parking-lot occupancy is unsupported.","section":"Section V, simulation setup"}],"minor_comments":[{"comment":"The time parameters t1, t2, and t3 are introduced in Section II but no numerical values or units are given in Section V; the y-axis of Figure 5 is therefore in unspecified units, and the simulation is not reproducible as described.","section":"Section V"},{"comment":"The symbol T is used for both temperature and total parking time (Eq. (2)), and the state space is written as S = {i, T}, which conflates a parameter with a state variable.","section":"Sections II and IV"},{"comment":"The model is called an 'entropy model,' but Eq. (1) is a canonical-ensemble occupation probability; no entropy is computed or modeled, so the terminology is misleading.","section":"Section III"},{"comment":"The sample-efficiency result is presented as a single curve without error bars or multiple lots, so the claim that 'very few observations are necessary' is not statistically supported.","section":"Section III, Figure 3"},{"comment":"Reference [4] is malformed, and several references lack complete bibliographic details, which should be corrected.","section":"References"}],"recommendation":"reject","confidential_remarks":"The paper's core defect is the invalid probability formula and the circular simulation; these are not routine presentation issues. A corrected model would require new derivation, refitting, and a simulator that is not generated by the model under test, which I view as beyond a normal revision. I see no reason to suspect misconduct, but the central claims are not supported as submitted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know before reading this. The occupancy model at the heart of the paper is not a probability distribution as printed, and the simulation validates the model against data generated by the same model. Those two flaws are load-bearing.\n\nWhat is genuinely new here is the framing: treating the occupancy distribution of a multi-story garage as a canonical ensemble with a single temperature parameter, and using dynamic programming to choose which floor to direct a car to given that distribution. The DP formulation in Section IV is a reasonable way to set up the floor-choice problem, and factoring in walking time for the passenger is a nice touch. That part mostly holds together on its own.\n\nThe problem is Section III. Equation (1) writes p(i) = 2 exp(E(i)/kT) / (1 + exp(E(i)/kT)). For any positive energy, which the paper assumes everywhere, this value is between 1 and 2, not a probability. And the low-temperature limit the formula gives is p -> 2, while the text explicitly says 'At low temperatures, states will have a low probability of being occupied.' That is not a disagreement with an external baseline; it is an internal contradiction in the central equation. Since the DP policy takes these p_i as inputs, the policy is not well-defined even on the paper's own assumptions.\n\nThe empirical support is weaker than the prose suggests. The reported fit is in-sample: the temperature is chosen by gradient descent on the same lots used to report the MSE of 0.173. For a 0-1 fill rate that corresponds to an RMS error of about 0.42, which is not 'excellent.' And the simulated garage in Section V is generated from the same entropy model with T=0.5, so TIPP is essentially testing the model against itself. The comparison also omits the stronger baselines cited in the introduction, like the RL approaches.\n\nI want to be fair: the dynamic-programming recursion itself is mostly sensible. If the authors replaced Eq. 1 with the standard Fermi-Dirac form 1/(1+exp(E/kT)), or otherwise fixed the normalization, the temperature story might be salvageable. But as written the central claim collapses. I would not cite this paper, and I would not send it to review in its current form. The right next step is to fix the formula, validate out-of-sample on real garage data, and run non-circular simulations or a small field test.","headline":"A fresh single-parameter parking model that unfortunately rests on an invalid probability formula and a circular simulation, so the central claim doesn't hold as written.","tokens_in":7587,"tokens_out":3301,"would_cite":false,"duration_ms":28605,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single fitted temperature predicts how parking spots are distributed through a garage, and a dynamic-programming policy using that prediction cuts expected parking time.","keywords":["entropy model","parking occupancy prediction","statistical mechanics","dynamic programming","Temperature-Informed Parking Policy","multi-story parking garage","canonical ensemble","temperature parameter"],"falsifier":"Record the occupied or empty status of every spot in a real multi-story garage along with each spot's distance to the main entrance, then fit the single-temperature entropy curve to a time window that includes a rush-hour surge; if the best-fit curve has error far above the 0.173 MSE reported here, or if the fitted temperature changes so quickly that one value cannot describe the garage at that moment, the central claim is contradicted.","tokens_in":6544,"feed_emoji":"🚗","tokens_out":8071,"duration_ms":65447,"temperature":0.7,"pith_summary":"The paper tries to show that the distribution of occupied spots across the floors of a multi-story garage can be captured by fitting one number, a 'temperature,' to an entropy model borrowed from statistical mechanics. If that works, a garage operator can estimate which floors still have spaces from a handful of observations rather than instrumenting every spot. On top of this predictor, the paper builds a dynamic program that tells each incoming car which floor to try next so that the sum of scanning, driving, and walking-back time is minimized. In a simulated 10-floor garage, the resulting Temperature-Informed Parking Policy (TIPP) beats a top-down Benchmark policy and an Inverse policy, and approaches the performance of a policy that sees every free spot.","feed_headline":"One temperature predicts where parking spots open in a garage","feed_subtitle":"An entropy model plus dynamic programming guides cars to the right floor and beats simple policies in simulation.","key_machinery":"The key object is the single-parameter entropy model, in which each spot's energy is proportional to the square of its normalized distance to the point of interest and the temperature $T$ encodes how busy the garage is; the occupancy probability formula from the canonical ensemble turns that energy and temperature into a predicted fill rate for every spot. The second piece is the dynamic-programming recursion for $f(i)$ and the resulting floor-selection rule $u(i)$, which converts the predicted distribution into an explicit parking decision for a car at floor $i$. Together they let a system update its temperature estimate from a small sample of observed spots, re-run the recursion, and direct each car to the floor that minimizes expected time to park and return to the entrance.","core_discovery":"The central claim is that a parking garage behaves like a canonical ensemble: the probability that spot $i$ is occupied is $p(i) = \\frac{2e^{E(i)/k_B T}}{1+e^{E(i)/k_B T}}$, with $E(i)$ taken as the squared normalized distance from the spot to the single point of interest and with temperature $T$ as the only fitted parameter. Fitting $T$ by minimizing mean squared error on ten real parking lots gives an aggregate MSE of $0.173$, and about ten observations from a 105-spot lot already produce a good estimate of the whole distribution. The paper then treats floor choice as a Markov decision process, defines $f(i)$ as the minimum expected total time starting from floor $i$, and computes $f(i)=p_i(t_1+it_2)+(1-p_i)(t_1+\\min_{j>i}((j-i)t_3+f(j)))$, with control $u(i)=\\operatorname{arg\\,min}_{j>i}[(j-i)t_3+f(j)]$. Feeding the entropy-model probabilities into this recursion yields TIPP, which in simulation outperforms Benchmark and Inverse policies at medium and high occupancy.","pith_inferences":["If the equilibrium assumption is the fragile link, a natural stress test is to apply the model to a garage with known rush-hour surges; one would expect the single-temperature fit to degrade exactly when inflow exceeds outflow, suggesting a two-temperature or time-varying extension.","The dynamic-programming layer is agnostic to how the $p_i$ are obtained, so any better occupancy predictor could be swapped in without changing the control recursion.","The sample-efficiency result suggests an active-sampling rule: instead of observing fixed spots, choose the few spots whose occupancy is most informative about $T$; the model makes that question well-posed.","With multiple entrances or points of interest, the paper's proposed additive energy extension is directly testable in the same simulation framework."],"forward_implications":["A garage operator could predict the full occupancy distribution from as few as ten observed spots, with the fitted temperature carrying all the information.","Because the model needs only distance-to-entrance and a small sample, it offers a low-cost alternative to sensor-heavy full-information parking systems and avoids sharing vehicle-level data.","Under TIPP, the recommended floor shifts as the chosen floor fills up; the paper interprets the resulting fluctuations as the policy correcting its own temperature estimate.","At low occupancy the simple Benchmark policy is competitive, but TIPP's advantage grows as the garage becomes congested, exactly when parking decisions matter most."],"supporting_citations":[{"why":"Supplies the canonical ensemble formula and the thermal-equilibrium assumption on which the entropy model is built.","marker":"[9]"},{"why":"Motivates the problem by quantifying the cost of searching for parking in the US, UK, and Germany.","marker":"[1]"},{"why":"Provides a graph-based parking assignment baseline that reduces vehicle cruising time and frames the assignment problem.","marker":"[2]"},{"why":"Gives a heuristic parking assignment procedure that the paper cites as prior work with a time restriction.","marker":"[3]"},{"why":"Represents earlier fuzzy-logic prediction of parking availability that the entropy model is compared against conceptually.","marker":"[4]"},{"why":"Supplies a multi-agent deep reinforcement learning baseline for online parking assignment with partially connected vehicles.","marker":"[6]"},{"why":"Shows a deep reinforcement learning approach to autonomous valet parking, representing the connectivity-heavy methods TIPP avoids.","marker":"[7]"}],"fun_headline_variants":["Know your parking floor with one temperature: entropy model","A single temperature predicts garage floor occupancy","Entropy and dynamic programming steer you to the empty floor","Temperature-informed parking policy finds spots faster","One parameter to park them all: physics-based guidance"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a parking garage is in thermal equilibrium, meaning the average inflow of cars equals the average outflow, so that a single temperature can determine the entire occupancy distribution; if real garages spend long periods in rush-hour imbalance, this premise fails and the entropy model loses its foundation.","fun_headline_variants_meta":{"raw":{"variants":["Know your parking floor with one temperature: entropy model","A single temperature predicts garage floor occupancy","Entropy and dynamic programming steer you to the empty floor","Temperature-informed parking policy finds spots faster","One parameter to park them all: physics-based guidance"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000933,"raw_usage":{"total_tokens":3997,"prompt_tokens":952,"completion_tokens":3045,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":2974}},"tokens_in":568,"tokens_out":3045,"duration_ms":22356,"temperature":1.0,"reasoning_tokens":2974,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:37:23.932464+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record the occupied or empty status of every spot in a real multi-story garage along with each spot's distance to the main entrance, then fit the single-temperature entropy curve to a time window that includes a rush-hour surge; if the best-fit curve has error far above the 0.173 MSE reported here, or if the fitted temperature changes so quickly that one value cannot describe the garage at that moment, the central claim is contradicted.","supporting_citations":[{"cited_title":"Huang, Statistical Mechanics","cited_arxiv_id":null,"evidence_quote":"Supplies the canonical ensemble formula and the thermal-equilibrium assumption on which the entropy model is built."},{"cited_title":"The impact of parking pain in the us, uk and germany car,","cited_arxiv_id":null,"evidence_quote":"Motivates the problem by quantifying the cost of searching for parking in the US, UK, and Germany."},{"cited_title":"A new heuristic for solving the parking assignment problem,","cited_arxiv_id":null,"evidence_quote":"Gives a heuristic parking assignment procedure that the paper cites as prior work with a time restriction."},{"cited_title":"Development of fuzzy logic forecast models for location-based parking finding services,","cited_arxiv_id":null,"evidence_quote":"Represents earlier fuzzy-logic prediction of parking availability that the entropy model is compared against conceptually."},{"cited_title":"Online parking assignment in an environment of partially connected vehicles: A multi- agent deep reinforcement learning approach,","cited_arxiv_id":null,"evidence_quote":"Supplies a multi-agent deep reinforcement learning baseline for online parking assignment with partially connected vehicles."},{"cited_title":"Deep reinforcement learning-based long-range autonomous valet park- ing for smart cities,","cited_arxiv_id":null,"evidence_quote":"Shows a deep reinforcement learning approach to autonomous valet parking, representing the connectivity-heavy methods TIPP avoids."}],"review_version":1}