{"id":"8f9b9d3a-ac7f-4945-9a9d-89ceef478079","arxiv_id":"2501.08961","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"OEFPIL, an errors-in-variables fitting algorithm, calibrates the nonlinear SThM voltage-conductivity curve and estimates unknown sample conductivities and uncertainties in a single automated step, with results comparable to Monte Carlo.","lead":"This paper applies an existing statistical fitting algorithm, OEFPIL, to calibrate a scanning thermal microscope and to estimate unknown thermal conductivities together with their uncertainties in a single automated step. A generalist reader might care because the method replaces a slow Bayesian calibration with a fast deterministic one, making nanoscale heat measurements more practical.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed robust uncertainty evaluation is not supported: all fits have very low p-values, and Table 2 shows OEFPIL and Monte Carlo disagree dramatically for kC (dataset 1) and kE (dataset 2).","rationale":"The paper's intended contribution is a deterministic, single-step calibration that yields robust uncertainties. The strongest evidence for this would be a validated uncertainty budget, and that evidence is missing. The authors themselves flag very low p-values in every fit, which means the model or the input uncertainties are wrong. Without resolving this, any reported uncertainty is conditional on an unverified assumption. The discrepancy in Table 2 is the most direct contradiction: OEFPIL's linearized uncertainty for kC and kE differs from Monte Carlo by orders of magnitude. The paper dismisses kC as asymmetry, but the MC distribution for kC has a standard deviation of 1134.5, indicating the linearized covariance is not a meaningful summary. The MC comparison for dataset 2 is also compromised by non-convergence, yet the conclusion claims agreement. This is an internal inconsistency in the validation argument, not merely a disagreement with external consensus. The reader's weakest assumption about Eq. (2) is the likely root cause; our concern is that the paper provides no check of that assumption and instead reports p-values that actively contradict it. The proposed test with an augmented model would distinguish misspecification from underestimated uncertainties. In the meantime, the central claim should be qualified: the method appears useful in the 0.1–10 W/m/K range, but its uncertainty estimates are not robust near the asymptote. This is consistent with the reader's CONDITIONAL verdict, so no verdict change is needed.","tokens_in":11315,"tokens_out":14078,"duration_ms":146984,"concrete_test":"Reanalyze dataset 3 with an augmented model Y = a k/(b+k) + c + d k and test the significance of the extra term (F-test or ΔAIC). If d is significant at the 5% level, or if any estimated k_A...k_E shifts by more than its reported uncertainty, Eq. (2) is misspecified and the reported uncertainties are not reliable. This directly probes the paper's self-reported low p-values and would settle whether the calibration curve or the uncertainty budget is the cause.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of robust single-step uncertainty evaluation is not established by the evidence in the paper. In Section 3 the authors report that \"in all cases, the fit exhibit very low p-values,\" and they attribute this to \"either a problem with the assumed calibration curve expression or significantly underestimated uncertainties,\" stating that the source is unclear. Since the calibration curve Y = ak/(b+k) + c (Eq. 2) is the sole link between measured voltage and conductivity, any misspecification propagates into every reported k and uncertainty. The paper's own Monte Carlo validation is inconsistent with the claimed accuracy: for dataset 1, kC is 9.05 ± 4.39 by OEFPIL but 27.3 ± 1134.5 by MC; for dataset 2, kE is 0.2174 ± 0.0060 by OEFPIL but 0.236 ± 0.327 by MC, roughly a factor of 54 in uncertainty. The MC runs also failed to converge for many generated datasets, especially dataset 2. The paper's conclusion that agreement with MC is \"a few percent for values and around ten percent for uncertainties\" is therefore not supported by Table 2. Consequently, the headline claim of robust uncertainty evaluation is overstated; at minimum it holds only in a narrow mid-range of conductivities, and the uncertainty estimates near the asymptote are not reliable.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies the OEFPIL errors-in-variables fitting algorithm to the calibration of a scanning thermal microscope. The authors define an intermediate measurand Y as the voltage difference with respect to an SiO2 reference, model the calibration curve as Y = a k/(b+k) + c, and fit the curve parameters and the unknown thermal conductivities of five samples simultaneously. They demonstrate the procedure on two independent measurement campaigns, a manually corrected version of dataset 1, and a Mandel-Paule pooled dataset, and they compare the OEFPIL results with Monte Carlo simulations. The central claim is that OEFPIL provides robust and single-step evaluation of local thermal conductivities and their uncertainties without Bayesian priors.","tokens_in":11597,"tokens_out":6914,"duration_ms":73760,"significance":"If the central claim were fully supported, the paper would make a useful methodological contribution to SThM calibration by replacing iterative or Bayesian procedures with a deterministic errors-in-variables fit and by automating the extraction of calibration data from images. The manuscript has several genuine strengths: the automated measurement protocol, the explicit treatment of Y and k as uncertain quantities, the use of the Mandel-Paule estimator for between-location variability, and the candid discussion of convergence failures and the limited useful range of the calibration curve. However, as it stands, the evidence does not establish the robustness claim: all fits show very low p-values, the Monte Carlo comparison is internally inconsistent in several entries, and for some samples no meaningful conductivity could be derived. These issues are load-bearing because the calibration curve is the only link between the measured voltage difference and the thermal conductivity.","major_comments":[{"comment":"The authors report that 'in all cases, the fit exhibit very low p-values' and attribute this to 'either a problem with the assumed calibration curve expression or significantly underestimated uncertainties.' Since Eq. (2) is the sole functional link between Y and the thermal conductivity, a misspecified curve or an underestimated uncertainty propagates into every reported k* and its uncertainty. The paper does not resolve this ambiguity, so the central claim of robust single-step uncertainty evaluation is not established. Please provide quantitative goodness-of-fit statistics, residual diagnostics, or an independent validation of Eq. (2), and either remove or qualify the robustness claim accordingly.","section":"Section 3, paragraph after Table 1"},{"comment":"The Monte Carlo comparison does not support the Conclusion's statement that agreement is 'a few percent for values and around ten percent for uncertainties.' For dataset 1, kC is 9.05 ± 4.39 from OEFPIL versus 27.3 ± 1134.5 from Monte Carlo; for dataset 2, kE is 0.2174 ± 0.0060 versus 0.236 ± 0.327 and b is 0.0327 ± 0.0109 versus −0.0056 ± 0.4182. The text also notes that the algorithm did not converge for all generated datasets, especially dataset 2. Moreover, because the Monte Carlo runs sample from the same model and the same data as the OEFPIL fit, the comparison tests internal numerical consistency rather than model validity. The validation claim should be restricted to the cases where agreement is actually demonstrated and should be described as a consistency check, not as an independent validation.","section":"Section 3, Table 2 and Conclusion"},{"comment":"The claimed scope of the method is broader than what the data demonstrate. For sample C, no meaningful value could be deduced in datasets 2 and 4; for sample D, no meaningful value could be deduced in datasets 1 and 2; and one sample had to be removed from dataset 2 because results were inconsistent with the assumed model. The text further states that the method 'will not give useful results for thermal conductivities above approx. 10 W m−1 K−1.' The abstract and title promise single-step evaluation of local thermal conductivities without this restriction. Please state the valid conductivity range as part of the central claim and report the success/failure rate for all five unknowns, not only the well-behaved ones.","section":"Section 3, Tables 1 and 2"},{"comment":"The provenance of dataset 3 is contradictory. The text says 'manual postprocessing performed on dataset 1 resulting in dataset 3,' while the Table 1 caption states '3 is a manual correction of 2.' Since dataset 3 is used in Table 2 and Figure 5, this inconsistency must be corrected and the affected results re-examined.","section":"Section 3 and Table 1 caption"}],"minor_comments":[{"comment":"There are typographical errors that should be corrected, including 'aquired' in Section 2.2, 'conducitivites' in Section 3, and 'Reserch' in Reference [14].","section":"Throughout"},{"comment":"The sentence 'The results are given in Table 2 typical probability distributions...' is grammatically incomplete and should be rewritten.","section":"Section 3, before Table 2"},{"comment":"No raw data or processing code are provided. Depositing the data and the OEFPIL/Monte Carlo scripts would allow independent verification of the p-values, the Mandel-Paule estimates, and the Monte Carlo convergence behavior.","section":"Data availability"},{"comment":"The phrase 'very low p-values' is vague; actual p-values or equivalent goodness-of-fit statistics should be reported so that the reader can judge the magnitude of the model-data discrepancy.","section":"Section 3, p-value discussion"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is honest about several limitations, but the abstract overstates the robustness of the method. The contradictory description of dataset 3 must be resolved before further review. I would not recommend rejection if the authors provide the requested goodness-of-fit analysis, restrict the validation and scope claims to what the data support, and clarify the Monte Carlo comparison."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this is basically a case study in applying the existing OEFPIL algorithm to SThM calibration, with a useful equivalence result. The paper is honest about its problems, but the headline 'robust and single-step evaluation of uncertainties' goes beyond what the data show.\n\nWhat's actually new: the application itself, the automated image-based processing, and the demonstration that one-step joint fitting gives the same results as the two-step curve-then-predict procedure. The experimental detail is good, and the authors clearly state the limitations of the high-conductivity asymptote. They also admit that all fits have very low p-values and that the source of the model or uncertainty mismatch is unclear. That candor counts for something.\n\nThe soft spots are real, though. No code or raw data are provided, and the Monte Carlo validation shows serious disagreements for exactly the samples you'd worry about: kC in dataset 1 and kE in dataset 2 differ hugely in uncertainty between OEFPIL and MC, and many MC runs didn't converge. The paper says agreement is 'a few percent for values and around ten percent for uncertainties,' which is not supported by Table 2. The SiO2-relative definition of Y also fixes one calibration point at zero, which is a bit circular, but the LNE reference values give some external grounding. Two of the five unknowns couldn't be assigned values at all. These are all things a careful reader can find, but the paper buries them in section 3 and acknowledges them, so it's not deceptive—just optimistic.\n\nThe model itself, Y = a k/(b+k) + c, is the single load-bearing assumption and it's not independently validated. Given that the paper flags the low p-values, that's more than a minor caveat. Still, the central method is sound for the mid-range conductivities where the slope is steep, and the equivalence result seems correct under the stated conditions.\n\nWho is this for? SThM users and metrologists working with nonlinear calibration. It won't matter much outside that niche. It deserves a serious referee, but the review should require the authors to release code and data, address the uncertainty underestimation, and soften the robustness claim to match the evidence.\n\nRecommendation: accept it for peer review, with major revision conditional on data release and an honest re-framing.","headline":"A useful, honest case study in applying an existing algorithm to SThM calibration, but the robustness claim goes beyond the evidence and needs revision.","tokens_in":12146,"tokens_out":2144,"would_cite":false,"duration_ms":22249,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"OEFPIL, a deterministic errors-in-variables fitting algorithm, calibrates a scanning thermal microscope and estimates unknown thermal conductivities with uncertainties in a single automated step, giving the same results as the…","keywords":["scanning thermal microscopy","SThM calibration","OEFPIL","errors-in-variables regression","thermal conductivity measurement","uncertainty evaluation","automated measurement","calibration curve fitting"],"falsifier":"Take a set of samples whose thermal conductivities are known independently by a traceable method (for example laser flash), spanning the steep part of the curve from roughly $0.1$ to $10$ W/(m K), apply the OEFPIL calibration, and check whether the reported 68 percent uncertainty intervals contain the independent values; if coverage is systematically below 68 percent, or if the fit p-values remain very low when more and better data are used, the assumed curve or uncertainty model is wrong.","tokens_in":11135,"feed_emoji":"🌡️","tokens_out":10604,"duration_ms":101082,"temperature":0.7,"pith_summary":"Scanning thermal microscopy can map heat conduction at the nanoscale, but calibrating the probe is hard because the response is nonlinear and only a few reference materials are available. The paper claims that the OEFPIL algorithm, a deterministic errors-in-variables fitting method, can take the raw reference-subtracted voltage data, fit the calibration curve, and estimate the unknown thermal conductivities of new samples in one step, complete with their uncertainties. If this works, SThM users no longer need Bayesian priors or separate manual calibration and prediction stages, and the whole procedure can be automated for routine laboratory use. The authors demonstrate it on two independent measurement sets with six reference and five unknown samples, and report that the one-step and two-step fitting routes give identical estimates of the curve parameters and the unknown conductivities.","feed_headline":"One-step fit yields thermal conductivities and uncertainties","feed_subtitle":"A deterministic fitting algorithm automates nanoscale heat-probe calibration in one step.","key_machinery":"The machinery is the OEFPIL algorithm for nonlinear errors-in-variables regression, which separates directly measured quantities (the reference conductivities and the voltage-difference measurands $Y$) from indirect measurands (the curve parameters and the unknown conductivities) and iterates a linearized solution until convergence, returning estimates along with a covariance matrix. The model is constrained by equations (3) and (4), which force the calibration and unknown samples to satisfy the same curve $Y = a k/(b+k)+c$. Repeated image measurements are combined into one value and uncertainty per sample, and the fitting step absorbs both reference-sample errors and measurement errors without prior distributions.","core_discovery":"The central claim is that the nonlinear calibration model $Y = a k/(b+k)+c$, fitted jointly to the reference samples and the unknown samples by OEFPIL, yields a consistent estimate of both the curve parameters $(a,b,c)$ and the thermal conductivities of the unknown samples together with their covariance matrix, so calibration and uncertainty propagation are solved at once. The paper also claims that in this model the simultaneous fit and the two-step procedure (fit the curve first, predict conductivities later) give identical estimated values and uncertainties, because the estimators of the curve parameters are uncorrelated with the estimators of the unknown-sample $Y$ values. This is supported by Monte Carlo comparison, with agreement to a few percent for values and around ten percent for uncertainties for most samples.","pith_inferences":["Beyond the paper: if the block-diagonal separability that makes the one-step and two-step estimates identical holds for this whole class of errors-in-variables models, OEFPIL could replace multi-stage or sampling-based calibration in other nonlinear measurement problems with independent direct measurements.","Beyond the paper: a targeted test would be to run the same protocol on samples with independently certified conductivities spanning the steep part of the curve and check whether the reported uncertainty intervals actually cover the certified values.","Beyond the paper: the very low p-values reported for all fits indicate that either the assumed calibration-curve shape is incomplete or the quoted uncertainties are too small; identifying that missing uncertainty source is the natural next step before the method's coverage claims are taken as settled.","Beyond the paper: the visible differences between the two datasets hint that a between-dataset uncertainty component is needed; a multi-day repetition campaign would show whether the within-dataset uncertainties are stable."],"forward_implications":["SThM calibration can be run as an automated, unsupervised measurement: the same script acquires images, computes the reference-subtracted voltages, and feeds them to OEFPIL.","Users get the thermal conductivity of an unknown sample together with a standard uncertainty in a single deterministic computation, with no priors and no MCMC sampling.","The one-step/two-step equivalence means a calibration curve can be estimated once and reused later for prediction without losing consistency, as long as the uncertainty matrix of the direct measurements remains block-diagonal.","In the demonstrated setup, the useful range of the method is roughly $0.1$ to $10$ W/(m K); above that the curve is too flat to resolve conductivity.","Manual post-processing of images affected by contamination or probe jumps reduces the uncertainty of the estimated conductivities by 15 to 60 percent, so automated acquisition still benefits from careful data screening."],"supporting_citations":[{"why":"supplies the calibration model and measurement protocol that the new OEFPIL procedure is designed to simplify.","marker":"[11]"},{"why":"introduces the distinction between direct and indirect measurands on which the errors-in-variables formulation rests.","marker":"[12]"},{"why":"provides the existing calibration samples that give traceability to the new, smaller sample set.","marker":"[13]"},{"why":"supplies the consensus-value estimator used to combine repeated image measurements into one value per sample.","marker":"[14]"},{"why":"provides the OEFPIL iterated-linearization method and notation used for the joint fit.","marker":"[17]"}],"fun_headline_variants":["Single-step fit automates nanoscale heat-probe calibration","One-step algorithm calibrates SThM and quantifies uncertainty","OEFPIL fit yields thermal values and errors in one go","Automated SThM calibration with simultaneous uncertainty","Nonlinear fit solves SThM calibration in one step"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the three-parameter curve $Y = a k/(b+k)+c$ correctly describes the true relation between the voltage difference and thermal conductivity for every calibration and unknown sample; if that shape is wrong, every estimated conductivity and its uncertainty inherits the error.","fun_headline_variants_meta":{"raw":{"variants":["Single-step fit automates nanoscale heat-probe calibration","One-step algorithm calibrates SThM and quantifies uncertainty","OEFPIL fit yields thermal values and errors in one go","Automated SThM calibration with simultaneous uncertainty","Nonlinear fit solves SThM calibration in one step"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000152,"raw_usage":{"total_tokens":1124,"prompt_tokens":788,"completion_tokens":336,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":404,"completion_tokens_details":{"reasoning_tokens":254}},"tokens_in":404,"tokens_out":336,"duration_ms":3889,"temperature":1.0,"reasoning_tokens":254,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:12:59.146133+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a set of samples whose thermal conductivities are known independently by a traceable method (for example laser flash), spanning the steep part of the curve from roughly $0.1$ to $10$ W/(m K), apply the OEFPIL calibration, and check whether the reported 68 percent uncertainty intervals contain the independent values; if coverage is systematically below 68 percent, or if the fit p-values remain very low when more and better data are used, the assumed curve or uncertainty model is wrong.","supporting_citations":[{"cited_title":"Quantitative measurement of thermal conductivity by sthm technique: Measurements, calibration protocols and uncertainty evalua- tion","cited_arxiv_id":null,"evidence_quote":"supplies the calibration model and measurement protocol that the new OEFPIL procedure is designed to simplify."},{"cited_title":"Elsevier, North Holland, 1988","cited_arxiv_id":null,"evidence_quote":"introduces the distinction between direct and indirect measurands on which the errors-in-variables formulation rests."},{"cited_title":"Quantiheat project: Main progresses","cited_arxiv_id":null,"evidence_quote":"provides the existing calibration samples that give traceability to the new, smaller sample set."},{"cited_title":"Paule and John Mandel","cited_arxiv_id":null,"evidence_quote":"supplies the consensus-value estimator used to combine repeated image measurements into one value per sample."},{"cited_title":"Estimation of function parameters through iterated linearization for nonlinear errors-in-variable regression with corre- lated variables","cited_arxiv_id":null,"evidence_quote":"provides the OEFPIL iterated-linearization method and notation used for the joint fit."}],"review_version":1}