{"id":"ecc0448b-58d5-40eb-b727-4f0d6bd63ecb","arxiv_id":"2505.03758","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"ForEXy converts EXFOR nuclear data files to a JSON format and back, and its covariance tools feed a new 237Np(n,f) evaluation that is higher than JENDL-5.","lead":"This paper presents ForEXy, a set of open-source Python tools that convert the EXFOR nuclear reaction database into a readable JSON format and back, plus utilities for managing and checking the database. The authors use these tools to produce a preliminary new evaluation of the neutron fission cross section of neptunium-237, which comes out systematically higher than the JENDL-5 library.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'systematic increase over JENDL-5' rests on a post-hoc shape-ratio rescaling of the Paradela dataset, and the paper provides no sensitivity test showing the increase survives without that adjustment.","rationale":"The paper's software contribution is concrete and supported: the codes are publicly released, the J4 files are distributed, and round-trip examples are shown. My concern is limited to the evaluation claim, which the reader also flagged as the weaker part. The reader's weakest assumption mentions both hand-assigned HED correlation coefficients and the Paradela shape-ratio rescaling; I agree that the evaluation inputs are not shipped, but the sharper, internally testable issue is the post-hoc fitting of the Paradela normalisation. Because the paper explicitly says this treatment was adopted after seeing that the dataset was discrepant, and does not show how the final 'systematically higher than JENDL-5' conclusion changes if that treatment is removed, the headline nuclear-data result is not yet firmly supported. This is exactly the kind of sensitivity check that should be a condition on the evaluation claim. I would keep the reader's CONDITIONAL verdict rather than moving to accept or reject: the software contribution stands, and the evaluation is labelled preliminary, but the paper should either quantify the posterior difference with uncertainties or demonstrate that the increase is robust to the shape-ratio treatment.","tokens_in":28837,"tokens_out":5141,"duration_ms":49943,"concrete_test":"Re-run the SOK fit with the same y and V but (i) keep Paradela 51015.002.1 as an absolute ratio without the fitted 0.96 normalisation and (ii) exclude it entirely; compute the posterior x1 and covariance X1 from Eqs. (8)-(9) on the 237Np(n,f) grid and compare x1 with JENDL-5 at each energy. If the posterior remains above JENDL-5 by more than the relevant diagonal uncertainty in both runs, the systematic-increase claim survives; if not, the increase is an artifact of the shape-ratio treatment and the summary statement should be revised or explicitly qualified as preliminary.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Sec. 3.2 reports that Paradela et al.'s 237Np(n,f)/235U(n,f) ratio was found to be systematically higher during trial fitting, so in the final fit it was treated as a 'shape ratio' in arbitrary units and its overall normalisation (0.96) was fit as an additional parameter. This is a post-hoc modification of the data model for the dataset that is most discrepant, and the paper gives no justification beyond the discrepancy itself. Because the central summary claim is that the posterior is systematically higher than JENDL-5, this claim may be an artifact of the rescaling: changing the normalisation changes y, and through Eq. (8) the posterior x1, and its effect is neither quantified nor separated from the prior/posterior comparison in Fig. 14. Fig. 15 itself shows that the scaled Paradela ratios remain above the Shcherbakov ratios, which JENDL-5 follows, making the treatment's contribution to the upward shift especially opaque. The conclusion also lacks any uncertainty on the posterior-minus-JENDL-5 difference, so 'systematically higher' is stated without a statistical threshold. The supporting SACS comparison is likewise not propagated through the fitted normalisation uncertainty.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces ForEXy, a Python package of seventeen utilities for the EXFOR experimental nuclear reaction database, built around a new JSON representation (J4) intended to be reversible with the EXFOR format. Core tools convert EXFOR files to J4 and back (X4TOJ4, J4TOX4), extract pointed datasets (POIPOI, EXTMUL), construct covariances from partial uncertainties (MAKCOV), manage dictionaries, and support bibliography/reference handling. The second part applies these codes to construct covariance matrices for 237Np(n,f) data in EXFOR and to perform a simultaneous SOK least-squares evaluation from 100 keV to 200 MeV, reporting a preliminary posterior that is systematically higher than JENDL-5. The paper also validates a rewritten Python version of the SOX input-generation code against the previous Perl version, showing sub-0.1‰ changes in evaluated cross sections.","tokens_in":29128,"tokens_out":5745,"duration_ms":49774,"significance":"If the reversibility claims hold, ForEXy is a genuinely useful open-source infrastructure for the EXFOR community, and the worked round-trip examples, pointer-extraction demonstrations, and SOX validation are concrete evidence for the software part. The independent SACS comparison (1361 mb vs 1347 mb) gives some directional support for the posterior shift. However, the evaluation's central claim currently rests on a post-hoc rescaling of the most discrepant dataset and lacks uncertainty quantification on the posterior-minus-JENDL-5 difference, so the significance of the evaluation section is conditional; the software contribution is more solid than the evaluation claim.","major_comments":[{"comment":"The paper states that during trial fitting Paradela et al.'s 237Np(n,f)/235U(n,f) data were found to be systematically higher, so in the final fit the dataset was treated as a 'shape ratio' in arbitrary units with an overall normalization fitted to 0.96. This is a post-hoc modification of the data model for the dataset most discrepant from the prior, and no justification beyond the discrepancy itself is provided. Because y enters the posterior through Eq. (8), the fitted normalization directly affects x1, and the paper does not quantify how the posterior or the posterior-minus-JENDL-5 difference would change if the rescaling were omitted or if the normalization were allowed to vary within its uncertainty. A sensitivity test (e.g., fitting with and without the rescaling, or profiling over the normalization) is needed before the 'systematically higher' claim can be supported.","section":"Sec. 3.2, Fig. 15"},{"comment":"The conclusion that the newly evaluated 237Np(n,f) cross section is systematically higher than JENDL-5 is stated without any uncertainty on the posterior or on the difference from JENDL-5. Figure 14 shows prior and posterior curves but no posterior uncertainty band, and the SACS comparison (1361 mb vs 1347 mb) is quoted without the Mannhart SACS uncertainty, so the reader cannot judge whether the increase is statistically significant. The SACS benchmark is also not propagated through the uncertainty of the fitted Paradela normalization, which limits its use as independent support. Please provide a quantitative statement of the difference with its uncertainty and indicate the energy range in which the difference is significant.","section":"Sec. 3.2 and Summary"},{"comment":"The design requirements state 'It is possible to reproduce the original EXFOR file from the JSON file' and 'It supports all logical structures of the EXFOR format.' The paper demonstrates reversibility with one entry (Fig. 3) and pointer extraction with one subentry (Figs. 4-7), but no systematic round-trip test over a large sample or the full library is reported. Since these two requirements are central to the software claim and distinguish J4 from simpler JSON representations, the authors should either provide a large-scale validation (e.g., round-tripping all entries of EXFOR-2024 and reporting the failure/success rate) or explicitly narrow the claim to the structures exercised in the paper.","section":"Sec. 2.2"},{"comment":"The covariance matrix V used in the SOK update is built from partial uncertainties with correlation coefficients C^k_ij supplied by the evaluator through HED files. The paper correctly emphasizes that V is the most important input and cites previous work showing sensitivity of fits to V (Ref. [51], Fig. 2), but no sensitivity analysis of the 237Np posterior to alternative HED assignments is given. Because the central evaluation claim depends on V, the paper should include at least a limited sensitivity study (e.g., treating a systematic uncertainty as fully correlated versus uncorrelated), or present the evaluation strictly as an illustration of the ForEXy workflow rather than as a new cross-section result.","section":"Sec. 2.4 and Sec. 3.1"}],"minor_comments":[{"comment":"The text twice refers to the 'EXFOR Maser File'; this should be 'Master File'.","section":"Sec. 2.1"},{"comment":"Several names contain stray spaces, e.g., 'V orobyev' in Sec. 3.2 and 'V .A. Kushnir' in the discussion of Fig. 3; these should be cleaned up.","section":"Throughout"},{"comment":"The prior and posterior curves are difficult to distinguish; using distinct line styles or adding a posterior uncertainty band would make the comparison clearer.","section":"Fig. 14"},{"comment":"SOX is described as rewritten in Python, but it is not listed among the seventeen codes in Table 1; please clarify whether SOX is part of ForEXy or a separate program.","section":"Table 1 and Sec. 3.1"},{"comment":"The sentence 'It means we treated their absolute ratio in the EXFOR library as a ratio in arbitrary units...' is a run-on and would be clearer if split into two sentences.","section":"Sec. 3.2"}],"recommendation":"major_revision","confidential_remarks":"The software part of the paper is a solid contribution and likely publishable, but the evaluation part needs to be either strengthened with sensitivity and uncertainty analysis or explicitly reframed as a workflow demonstration rather than a new evaluation result. The summary's 'systematically higher' claim is currently stronger than the evidence in the manuscript. I would also ask the editor to consider whether the explicitly preliminary status of the evaluation is appropriate for an archival journal article, or whether the authors should be asked to tighten the claims to match the presented analysis."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What should you know: the thing worth remembering here is ForEXy, not the neptunium evaluation. The paper's real contribution is a Python package of 17 utilities built around a reversible JSON representation of EXFOR. Previous JSON converters were one-way; the J4 round-trip (X4TOJ4/J4TOX4) is a genuinely new capability, and the package ships code, data, and worked examples. That is formal, reproducible work and it earns credit.\n\nThe J4 design is described with sensible examples — the decay-data keys, the pointer cancellation, the dictionary handling. The authors also demonstrate a reproducibility check: converting an EXFOR entry to J4 and back, with the differences called out explicitly (trailing zeros, counters). That is the right level of honesty for infrastructure work.\n\nThe 237Np(n,f) evaluation is a different matter. The reader's conditional verdict is fair. The soft spot the stress-test found is real: Paradela's ratio was seen to be systematically higher, then entered the fit as a 'shape ratio' with a fitted 0.96 normalization. The paper offers no sensitivity test showing the posterior shift survives without that rescaling, and the 'systematically higher' conclusion is stated without an uncertainty on the difference. So the headline nuclear-data claim is not yet supported. To the authors' credit, they label the evaluation preliminary, they show the SACS comparison, and they cite an independent benchmark that points in the same direction — but the 0.96 fit remains a post-hoc adjustment on the most discrepant dataset, and that needs a robustness check before the claim is taken seriously.\n\nThe evaluation inputs (HED files, the exact covariance assignments) are not shipped in the paper, so a reader cannot reproduce the posterior without reconstructing the configuration. That is a moderate problem for the evaluation section, not for the software section.\n\nWho gets value: anyone working with EXFOR as a data source — evaluators, compilers, people training ML on nuclear data. This is the kind of paper that deserves a serious referee. I would send it to review, expecting the reviewer to focus on the evaluation methodology and request a sensitivity analysis. If the authors can show the upward shift is not an artifact of the 0.96 rescaling, the paper is a double contribution; if not, the software half stands on its own.\n\nRecommendation: accept the paper for review, with a clear request for a sensitivity test around the shape-ratio treatment.","headline":"ForEXy is the real contribution here; the 237Np evaluation's 'systematically higher' claim is preliminary and rests on a fitted 0.96 rescaling that needs a sensitivity check.","tokens_in":29645,"tokens_out":2498,"would_cite":true,"duration_ms":22722,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["29.87.+g","25.85.Ec","29.85.Fj"],"model":"deepseek-v4-flash","headline":"The paper claims the EXFOR library can be losslessly converted to a JSON format (J4) and back again, and that a J4-built covariance pipeline puts the 237Np(n,f) evaluation systematically above JENDL-5.","keywords":["EXFOR library","J4 JSON format","ForEXy utility codes","EXFOR-J4 conversion","covariance estimation","neptunium-237 fission","SOK least-squares evaluation","JENDL-5"],"falsifier":"Rerun the SOK fit with the Paradela 237Np/235U ratio restored to its published face value (no 0.96 factor) and with the HED correlation coefficients set to their extremes (all zero, then all one): if the posterior 237Np(n,f) curve no longer sits systematically above JENDL-5, the evaluation claim is an artifact of the shape-ratio rescaling and hand-assigned covariances. Alternatively, once the Vorobyev et al. ratio dataset is compiled into EXFOR, repeat the fit and check whether the systematic rise over JENDL-5 survives.","tokens_in":28658,"feed_emoji":"⚛️","tokens_out":13050,"duration_ms":101958,"temperature":0.7,"pith_summary":"This paper tries to establish that the 60-year-old EXFOR experimental nuclear reaction data library can be represented, without loss, in a modern JSON format. It introduces J4, a JSON schema in which every EXFOR code string is decomposed into keys defined by the EXFOR dictionary, and two converters, X4TOJ4 and J4TOX4, that round-trip an EXFOR file through J4, together with a suite of seventeen Python utilities for maintaining the library, cancelling multiple-reaction pointers, estimating covariances, managing bibliographies and dictionaries, and spell-checking. As a demonstration, the authors apply the covariance builder MAKCOV and the SOK least-squares evaluator to the 237Np(n,f) reaction between 100 keV and 200 MeV, obtaining a preliminary evaluation systematically higher than the JENDL-5 library value. The paper matters because 237Np(n,f) is a candidate reference standard in the MeV region, and because a reversible machine-readable mirror of EXFOR could lower the barrier for new tools built around the world's only comprehensive archive of experimental nuclear reaction data.","feed_headline":"Round-trips the 60-year-old EXFOR library through JSON","feed_subtitle":"Lossless J4 conversion powers a neptunium-237 fission evaluation that lands systematically above JENDL-5.","key_machinery":"The load-bearing mechanism is the J4 schema together with its two converters. J4 is a JSON representation in which every EXFOR code string is expanded into keys whose names and value types are defined by the EXFOR/CINDA Dictionary, which is what makes the library machine-readable without losing any logical structure; X4TOJ4 and J4TOX4 implement the round trip, with SEQADD re-inserting the line counters that J4 deliberately zeroes. The second mechanism is MAKCOV, which builds an experimental covariance matrix from partial uncertainties (ERR-S, ERR-1, ERR-2, ...) using the identity $cov(y_i,y_j) = \\sum_k C^k_{ij}\\,\\Delta^k y_i\\,\\Delta^k y_j$, where the correlation coefficients $C^k_{ij}$ are read from a hand-written HED file. The third is the SOK evaluator, which performs the least-squares update of a prior estimate to a posterior estimate, $x_1 = x_0 + X_0C^T(CX_0C^T+V)^{-1}(y-Cx_0)$, on logarithms of cross sections and ratios, so that the hand-built matrix $V$ directly determines the posterior.","core_discovery":"The central claim has two parts, and both are stated by the authors as accomplished facts. The first is that the J4 JSON representation carries the full logical structure of the EXFOR format: a code string such as DECAY-DATA is decomposed into keyed subfields (nuclide, half-life, radiation type, energy, intensity) defined by the EXFOR/CINDA dictionary, so that a program can read any value directly, and the conversion is reversible, with X4TOJ4 followed by J4TOX4 reproducing the original EXFOR file apart from trailing-zero formatting (preserved by the -s option) and line-counter integers that SEQADD restores. The second is scientific: using the J4-based chain to build experimental covariance matrices for the fast-neutron fission of neptunium-237 and feeding them to the SOK simultaneous least-squares evaluator produces a preliminary 237Np(n,f) cross section, evaluated between 100 keV and 200 MeV, that is systematically higher than the corresponding JENDL-5 cross section. The authors note the direction is supported by the 252Cf spontaneous-fission spectrum-averaged cross section (Mannhart's 1361 mb against JENDL-5's 1347 mb), and they explicitly label the result preliminary, pending inclusion of the newly published Vorobyev et al. ratio dataset once it enters EXFOR.","pith_inferences":["A formal round-trip test over the entire EXFOR-2024 Master File, rather than the single-entry examples shown, would settle the losslessness claim in full generality; 'supports all logical structures' is asserted for the design, and the paper demonstrates equivalence on individual entries.","The hand-written HED correlation coefficients are the main lever on the result: recomputing the 237Np posterior with HED correlations at their extremes (all zero versus all one) would bound how much of the systematic rise over JENDL-5 is experimental data and how much is evaluator judgment.","The fitted 0.96 factor on the Paradela ratio is the largest single normalization judgment in the fit; the authors' own plan to add the Vorobyev et al. dataset gives a direct test, because if that addition moves the factor or erases the rise, the evaluation would need revision.","If J4 becomes the interchange layer for the library, it would extend earlier JSON-database efforts by adding a reversible JSON Dictionary (DICA2J and DICJ2A) and pointer-cancellation tools, which those efforts did not address."],"forward_implications":["J4 files distributed alongside the ForEXy codes let anyone build EXFOR-processing software without learning the 80-column EXFOR layout, because the JSON carries every logical structure of the original file.","The switch from the Perl SOX to the Python/J4 SOX is production-safe: the experimental data vector $y$ is reproduced exactly, and differences in the covariance matrix $V$ change evaluated fission cross sections by less than 0.1‰, with the largest effect in 240Pu.","MAKCOV turns covariance construction from EXFOR partial uncertainties into a scriptable, documented step, and doubles as a total-uncertainty calculator through the quadrature sum of the partial uncertainties.","The preliminary SOK evaluation places the 237Np(n,f) cross section systematically above JENDL-5 across 100 keV to 200 MeV, a direction consistent with the 252Cf spectrum-averaged cross section comparison.","The POIPOI and EXTMUL pointer-cancellation tools extract a single dataset from a multiple-reaction-formalism subentry, removing a known hurdle for programmatic users of EXFOR."],"supporting_citations":[{"why":"Defines the EXFOR library whose full logical structure J4 must carry and round-trip; the reference the new format is measured against.","marker":"[1]"},{"why":"Prior EXFOR-to-JSON converter and MongoDB tool; the earlier approach that J4 is designed to supersede with full reversibility.","marker":"[13]"},{"why":"The ForEXy manual specifying the J4 and HED file syntax on which X4TOJ4, J4TOX4 and MAKCOV depend.","marker":"[17]"},{"why":"Documents the SOK least-squares formalism whose prior-to-posterior update equations the 237Np evaluation runs.","marker":"[48]"},{"why":"The EXFOR-based simultaneous uranium and plutonium fission evaluation framework and original Perl SOX whose output the Python SOX must reproduce.","marker":"[51]"},{"why":"Provides the neutron data standards used as reference reactions in the simultaneous evaluation framework.","marker":"[56]"},{"why":"Shcherbakov et al.'s 237Np/235U ratio dataset, the measurement the posterior follows closely and the key post-2002 data absent from JENDL-4.0.","marker":"[61]"},{"why":"Paradela et al.'s n_TOF 237Np/235U ratio dataset, entered in the fit as a shape ratio with fitted factor 0.96; removing it would remove the largest single normalization judgment in the evaluation.","marker":"[62]"},{"why":"Mannhart's 252Cf spectrum-averaged cross section recommendation used as the consistency check supporting the systematic increase over JENDL-5.","marker":"[63]"},{"why":"Vorobyev et al.'s newly published 237Np(n,f)/235U(n,f) ratio dataset, which the authors state will be used to check whether the systematic rise persists.","marker":"[64]"}],"fun_headline_variants":["EXFOR data roundtrips through JSON to refine Np-237 fission","J4 JSON conversion fuels Np-237 fission evaluation","New ForEXy codes round-trip EXFOR data via JSON for fission eval","Np-237 fission evaluation from JSON-based EXFOR runs above JENDL-5","EXFOR-to-JSON codes yield higher Np-237 fission cross-section than JENDL"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The evaluation result rests on the covariance matrix $V$ that the evaluator builds from EXFOR partial uncertainties using correlation coefficients assigned by hand in HED files, plus the decision to rescale the Paradela ratio dataset by a fitted factor of 0.96; if those assignments or that rescaling misrepresent the data, the systematic rise over JENDL-5 is not supported.","fun_headline_variants_meta":{"raw":{"variants":["EXFOR data roundtrips through JSON to refine Np-237 fission","J4 JSON conversion fuels Np-237 fission evaluation","New ForEXy codes round-trip EXFOR data via JSON for fission eval","Np-237 fission evaluation from JSON-based EXFOR runs above JENDL-5","EXFOR-to-JSON codes yield higher Np-237 fission cross-section than JENDL"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000801,"raw_usage":{"total_tokens":3540,"prompt_tokens":983,"completion_tokens":2557,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":599,"completion_tokens_details":{"reasoning_tokens":2452}},"tokens_in":599,"tokens_out":2557,"duration_ms":15974,"temperature":1.0,"reasoning_tokens":2452,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:48:31.602334+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the SOK fit with the Paradela 237Np/235U ratio restored to its published face value (no 0.96 factor) and with the HED correlation coefficients set to their extremes (all zero, then all one): if the posterior 237Np(n,f) curve no longer sits systematically above JENDL-5, the evaluation claim is an artifact of the shape-ratio rescaling and hand-assigned covariances. Alternatively, once the Vorobyev et al. ratio dataset is compiled into EXFOR, repeat the fit and check whether the systematic rise over JENDL-5 survives.","supporting_citations":[{"cited_title":"Otuka, ForEXy: Utility codes for EXFOR Library, Technical Report IAEA-NDS-244 Rev","cited_arxiv_id":null,"evidence_quote":"The ForEXy manual specifying the J4 and HED file syntax on which X4TOJ4, J4TOX4 and MAKCOV depend."},{"cited_title":"Kawano, H","cited_arxiv_id":null,"evidence_quote":"Documents the SOK least-squares formalism whose prior-to-posterior update equations the 237Np evaluation runs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the neutron data standards used as reference reactions in the simultaneous evaluation framework."},{"cited_title":"Paradela, L","cited_arxiv_id":null,"evidence_quote":"Paradela et al.'s n_TOF 237Np/235U ratio dataset, entered in the fit as a shape ratio with fitted factor 0.96; removing it would remove the largest single normalization judgment in the evaluation."},{"cited_title":"Mannhart, Response of activation reactions in the neutron field of californium-252 spontaneous fission, Techni- cal Report STI/DOC/010/452, International Atomic Energy Agency, 2006","cited_arxiv_id":null,"evidence_quote":"Mannhart's 252Cf spectrum-averaged cross section recommendation used as the consistency check supporting the systematic increase over JENDL-5."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Vorobyev et al.'s newly published 237Np(n,f)/235U(n,f) ratio dataset, which the authors state will be used to check whether the systematic rise persists."}],"review_version":1}