{"id":"66877dae-1eda-4842-b3de-4cc6bfacddc2","arxiv_id":"1908.03435","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A weighted product of muzzle velocity, effective range, projectile mass, and rate of fire yields a fitted superexponential trend for small arms from 1200 CE to 2015 CE.","lead":"The authors fit a composite Figure of Regularity to small arms from 1200 CE to today, and report that it grows faster than exponentially across bows, muskets, and rifles. A generalist might read it as a test of whether long-range technology forecasting can work without a single agreed performance measure.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Model selection contradiction: Model E (piecewise exponential) beats the chosen quadratic Model B on BIC and R2 in Table 2, so Eq. 5 is not the best-supported 800-year law by the paper's own criteria.","rationale":"The reader's stated weakest assumption is historical data accuracy for medieval and early modern weapons. That is a genuine threat to the fitted exponents and to the apparent smoothness of the trend, and it deserves a sensitivity analysis. However, the more load-bearing problem is internal: the paper's own model-comparison table contradicts its choice of the quadratic law. Section 2 establishes BIC as the model-selection criterion, and Table 2 shows Model E has lower BIC and higher R2 than Model B. The paper then switches to MAPE to justify Model B without explaining why BIC should be overruled. Since the headline result is exactly the Model B parameterization in Eq. 5, this is not a peripheral issue; it is the evidence for the central claim. In addition, the printed definition of Model E is internally inconsistent, which means the comparison as reported cannot be independently audited. A cross-validation comparison would settle whether the data support a single smooth superexponential law or a piecewise regime-change model. Because the final verdict remains conditional rather than accept or reject, and the reader already recommended conditional acceptance with revisions, my stress-test does not move the verdict.","tokens_in":12212,"tokens_out":11850,"duration_ms":118405,"concrete_test":"Run a 10-fold or leave-one-out cross-validation that re-estimates both the FoR exponents and the temporal parameters inside each training fold for Models B and E, using the same fitting routine as Section 4.4, and compare out-of-sample MAPE and log-likelihood on the held-out folds. If Model E has lower out-of-sample error than Model B, the quadratic 800-year law should be withdrawn in favor of a regime-change description. If Model B wins out-of-sample, the model-selection objection is resolved and the conditional acceptance can stand.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The centerpiece of the paper, Eq. 5, is the parameterization of Model B. Section 5 states Model B is best 'in terms of parsimony, BIC, R2, and MAPE.' Table 2, however, reports Model E with BIC -334 vs -326 and R2 0.961 vs 0.955; Model B wins only on MAPE (0.112 vs 0.153). Since BIC is introduced in Section 2 as the model-selection statistic, choosing B requires an unstated decision to privilege MAPE over BIC. The paper also mis-specifies Model E: both branches of the piecewise formula are printed identically (exp(theta1+theta2 t_i) on both sides), with a constraint theta4>theta1 that is undefined as written, so the fitted form behind Table 2 is not transparent. If Model E is correctly specified and its better BIC/R2 are real, the data favor a discontinuity around 1832, not a single quadratic superexponential law from 1200 CE to the present. This is the paper's own evidence undermining the headline claim. A secondary consequence is that the FoR exponents (alpha_i) are re-estimated for each temporal model (Table 2 rows), so the 'law' is selected jointly with the exponents; without a proper out-of-sample comparison between B and E, the specific Eq. 5 cannot be taken as a discovered regularity rather than one of several equally flexible descriptions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a composite measure called a Figure of Regularity (FoR) for technologies lacking a standard measure of performance, defined as a multiplicative power-law of four small-arms attributes: muzzle velocity, effective range, projectile mass, and rate of fire. Using a dataset of over 120 weapons from 1180 to 2015 CE, the authors jointly fit the FoR exponents and one of five temporal dynamics models, and select a quadratic-exponential model (Model B) as best. This yields Eq. 4, FoR = (1.1e-6) V^2.0 D^2.35 M^0.61 R^0.39, and Eq. 5, log(FoR) = 1.0 + 8.27e-6 (t-1200)^2, claimed to describe an 800-year superexponential regularity. The paper then uses Eq. 5 to forecast log(FoR) growth to 2050 and compares it to a sum of increments from plausible R&D improvements in muzzle velocity, range, and projectile mass.","tokens_in":12524,"tokens_out":3350,"duration_ms":31618,"significance":"If the claimed regularity held, it would be a remarkable empirical finding: a single weighted product of four weapon attributes following a smooth accelerating curve for over eight centuries, and a methodological template for studying technologies without an accepted measure of performance. The paper is clearly written and transparent about its data and fitting procedure, and the compilation of historical small-arms data in the companion technical report is a useful contribution. However, the statistical support for the specific 800-year law is substantially weaker than the presentation suggests, because the model-selection evidence in the paper's own Table 2 points to a piecewise model, the model comparison is made on in-sample fits, and the forecast feasibility check is internal to the fitted equations.","major_comments":[{"comment":"The claim that Model B is best 'in terms of parsimony, BIC, R2, and MAPE' is contradicted by Table 2: Model E has a lower BIC (-334 vs -326) and a higher R2 (0.961 vs 0.955) than Model B, which wins only on MAPE (0.112 vs 0.153). Since Section 2 introduces BIC as the model-selection statistic, the choice of Model B requires an explicit justification for prioritizing MAPE, which the paper does not provide. If Model E is correctly specified, the data favor a piecewise exponential with a break near 1832, not a single quadratic superexponential law from 1200 CE to the present.","section":"Section 5 and Table 2"},{"comment":"Model E is mis-specified as printed: both branches of the piecewise formula are identical, exp(theta1 + theta2 t_i), and the stated constraint theta4 > theta1 references a parameter theta4 that does not appear in the formula. As written, this is not a two-regime model and cannot reproduce the fitted values in Table 2. The authors must provide the correct piecewise specification and refit the model before the comparison between Models B and E can be evaluated.","section":"Section 4.3, Model E"},{"comment":"The FoR exponents alpha_i are re-estimated for each temporal model on the same dataset, and the temporal model is selected on that same dataset. The reported R2, BIC, and MAPE therefore measure in-sample fit of a model class that was selected to fit, not evidence of an independent regularity. To support the claim of discovering an 800-year law, the paper needs an out-of-sample or holdout analysis, for example fitting on pre-1800 data and testing on post-1800 data, or a comparison against a null model that randomizes the attribute values while preserving the time series.","section":"Sections 4.2-4.4 and Table 2"},{"comment":"The forecast and feasibility check are internal to the fitted model. The forecast log(FoR)=6.97 in 2050 uses Eq. 5 directly, and the feasibility check sums increments computed from the same fitted FoR exponents (mislabeled as Eq. 5 in the text). The close agreement between the required increment (0.80) and the sum of feasible increments (0.78) is therefore an arithmetic consequence of using the same fitted equations, not independent evidence of forecast feasibility.","section":"Section 6"},{"comment":"The central regularity rests on historical attribute values, especially maximum effective range and rate of fire for medieval and early modern weapons, which are rough historical judgments rather than precisely measured quantities. No sensitivity analysis is provided to show how the fitted exponents in Eq. 4 or the temporal curve in Eq. 5 would change under plausible perturbations of these values. Without such an analysis, the stability of the claimed 800-year regularity is not demonstrated.","section":"Section 4.1 and Table 1"}],"minor_comments":[{"comment":"The symbol k is used both as the number of free parameters and as the product index in Eq. 1; please use different notation to avoid confusion.","section":"Section 2, Eq. 2"},{"comment":"The text says 'Table 2 shows examples of the data points' but the relevant table is Table 1; please correct the cross-reference.","section":"Section 2, Step 1"},{"comment":"The verbal description says 'the fraction is a quadratic function of time' but the model is cubic in time; please align the wording with the equation.","section":"Section 4.3, Model C"},{"comment":"The MAPE is described as an average over three conditions with different fitting and testing periods; this is a form of out-of-sample evaluation and deserves a more prominent and precise description, including the number of data points in each period.","section":"Section 4.4, MAPE description"},{"comment":"The text refers to 'Eq. 5' when discussing the FoR model with velocity, range, and mass terms, but those terms appear in Eq. 4; please correct the equation references throughout the section.","section":"Section 6"},{"comment":"The claim that this is 'the longest exponential or superexponential trend reported for any technology' would benefit from a more careful comparison with prior literature, which includes multi-century trends in other domains; as stated, the claim is too broad for the evidence presented.","section":"Section 7"}],"recommendation":"major_revision","confidential_remarks":"The paper's central claim is not supported by its own model-selection table, and the piecewise model that actually wins on BIC and R2 is mis-specified in the text. The authors should be asked to correct the specification, refit all models, and provide a proper out-of-sample or robustness analysis before the claim of an 800-year law is taken seriously. The topic is appropriate for the journal and the underlying data are valuable, so a major revision rather than rejection seems warranted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something genuinely new: it extends Alexander-Nelson regression to several temporal growth models and applies the idea to an 800-year dataset of small arms, finding a composite index that fits a smooth superexponential curve remarkably well. The writing is clear, the domain interpretation of the exponents is sensible, and the out-of-sample MAPE checks are a nice touch. I believe the authors are serious and the empirical regularity is worth discussing.\n\nThe problem is that the central claim is undercut by the paper's own model-selection table. Table 2 shows Model E (piecewise exponential) with BIC -334 vs -326 and R2 0.961 vs 0.955, yet the text says Model B is best in terms of BIC and R2. That is simply not what the table says. The authors may prefer Model B for parsimony or domain plausibility, but then they need to say so and justify why MAPE should carry more weight than BIC. As written, the selection is internally contradictory. On top of that, the printed formula for Model E is wrong: both branches are identical and the constraint theta4 > theta1 is undefined. This makes the comparison between B and E opaque, and it is hard to know whether the data actually favor a discontinuity around 1832 over a single quadratic.\n\nThe other soft spot is the circularity inherent in the method. The FoR exponents and the temporal law are fitted jointly on the same 1200–2015 data, so the resulting regularity is a selected best fit rather than an independent benchmark. That does not destroy the empirical claim—the out-of-sample MAPE checks help—but it does mean the paper should present the result as a useful fitted summary, not as a discovered law of technology. The data also come from a separate technical report with no error bars, and for medieval weapons the effective range is a rough historical judgment. The 2050 forecast is a direct extrapolation of the fitted curve, and the feasibility check uses the same fitted exponents, so it is not independent validation.\n\nIn short: the paper is a good case study in how to fit a composite index to noisy historical data, but the model-selection contradiction and the data-availability issue need to be fixed before the 800-year law is credible. I would send this to a serious referee, not desk reject it. With revisions—correct the model-selection argument, provide the data, and temper the \"law\" language—it could become a useful contribution to technology forecasting.","headline":"A cleanly presented 800-year regularity that is likely a fitted artifact rather than a discovered law; the model-selection contradiction in Table 2 is the soft spot that needs fixing before the headline claim can be taken seriously.","tokens_in":13067,"tokens_out":2312,"would_cite":false,"duration_ms":25298,"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":"A weighted product of four weapon attributes follows one accelerating curve from 1200 CE to today.","keywords":["technology trends","Figure of Regularity","small arms","military technology","superexponential growth","Moore's law","technological forecasting","performance measures"],"falsifier":"Re-fit the FoR and temporal models while perturbing the pre-1600 values of maximum effective range within plausible historical uncertainty, say ±25 to 50 meters; if the exponent on effective range moves by more than a few tenths, or if the quadratic-exponential model no longer wins on the Bayesian Information Criterion, the 800-year law is an artifact of those estimates.","tokens_in":11952,"feed_emoji":"📈","tokens_out":12261,"duration_ms":103823,"temperature":0.7,"pith_summary":"This paper proposes a way to find a composite measure, called a Figure of Regularity (FoR), for technologies that lack a single accepted performance measure, and demonstrates it on infantry small arms from medieval bows to modern assault rifles. It reports that a weighted product of muzzle velocity, effective range, projectile mass, and rate of fire follows a smooth accelerating curve for roughly 800 years. The fitted law is $\\log(FoR) = 1.0 + 8.27\\times 10^{-6}(t-1200)^2$, with $FoR = (1.1\\times 10^{-6}) V^{2.0}D^{2.35}M^{0.61}R^{0.39}$, selected over four alternative temporal-growth models. If the result stands, it gives technology analysts a quantitative basis for long-range forecasting and for checking whether a research portfolio can plausibly meet the forecast.","feed_headline":"One curve tracks 800 years of small-arms progress","feed_subtitle":"A weighted blend of velocity, range, mass, and fire rate has been accelerating since 1200.","key_machinery":"The central object is the Figure of Regularity (FoR), a multiplicative composite measure $FoR = k\\prod_i x_i^{\\alpha_i}$ built from a technology's attributes. The paper's mechanism is to fit the attribute exponents $\\alpha_i$ simultaneously with the parameters of a set of candidate temporal-growth models—exponential, quadratic-exponential, cubic-exponential, double-exponential, and piecewise-exponential—choosing among them by the Bayesian Information Criterion and related fit statistics. This joint fitting, which generalizes the earlier weighted-attribute regression approach, is what lets the composite measure and its growth law be discovered together rather than assumed in advance.","core_discovery":"The central claim is that infantry small-arms technology, although it lacks an agreed measure of performance, has a regular 800-year trend when described by the composite Figure of Regularity $FoR = (1.1\\times 10^{-6}) V^{2.0}D^{2.35}M^{0.61}R^{0.39}$, whose logarithm grows as $1.0 + 8.27\\times 10^{-6}(t-1200)^2$. The authors select this quadratic-exponential (superexponential) temporal model over exponential, cubic-exponential, double-exponential, and piecewise-exponential alternatives, preferring it by parsimony, the Bayesian Information Criterion, $R^2$, and mean absolute percentage error, and they reject the plain exponential law because it assigns an implausibly small weight to rate of fire. They present the formula as a description of the data rather than a design guide, and use it to forecast an increase of about 0.80 in $\\log(FoR)$ by 2050, consistent with their estimate of the combined effect of currently reported rifle-development directions.","pith_inferences":["A testable extension of the paper's logic would be to fit the same FoR form to weapons data outside the paper's Western European and U.S. scope, or before 1200 CE; large changes in the fitted exponents would indicate the trend is tied to one regional trajectory rather than to small arms generally.","The method can be read as constructing a hidden combined measure from attributes, so similarly smooth combined measures may exist in other domains with no accepted performance metric, such as naval gunnery or hand tools.","The 2050 forecast assumes the same FoR formula extrapolates beyond the fitted range; a natural check is to compare the actual attributes of any new infantry rifle introduced in the 2020s and 2030s against the trend band implied by the model.","The large fitted exponent on effective range (2.35) suggests range has been the dominant selection pressure over the eight centuries studied, an inference about historical drivers that the paper does not itself draw."],"forward_implications":["Absent a discontinuity, the model implies small-arms $\\log(FoR)$ will grow by about 0.80 above today's best value by 2050, reaching roughly 6.97.","Assuming a Moore-law exponential trend a priori is misleading for this domain, because that model assigns almost no weight to rate of fire, contradicting the historical importance of that attribute.","Currently discussed rifle improvements—increased velocity, effective range, and projectile mass—would together raise $\\log(FoR)$ by about 0.78, close to the 0.80 forecast from the long-term trend.","The same joint-fitting procedure can be applied to other technology families lacking a defined performance measure, with the FoR serving as a Moore-law-like quantity for each.","The FoR should not be read as a design guide; the paper states explicitly that improving the FoR is not the same as designing a better weapon."],"supporting_citations":[{"why":"Supplies the full dataset of over 120 weapon-attribute data points and the source-interpretation notes on which the analysis rests.","marker":"Kott [2019]"},{"why":"Establishes the original weighted-log-attribute regression that this paper generalizes to non-exponential temporal models.","marker":"Alexander and Nelson [1973]"},{"why":"Provides superexponential long-term trend models and evidence that motivate the superexponential temporal candidates.","marker":"Nagy et al. [2011]"},{"why":"Documents exponential laws across many technologies, serving as the default trend hypothesis the paper tests against.","marker":"Nagy et al. [2013]"},{"why":"Defines the Bayesian Information Criterion and the fitting statistics used to select the quadratic-exponential model over the alternatives.","marker":"Friedman et al. [2001]"},{"why":"Discusses composite measures of technology and how weighted attribute sums act as substitutes for a missing performance measure.","marker":"Martino [1993b]"},{"why":"Supplies the double-exponential growth hypothesis that forms one of the superexponential candidate models.","marker":"Kurzweil [2001]"},{"why":"Motivates the 1832 breakpoint used in the piecewise-exponential competitor, an important alternative in the model comparison.","marker":"Lienhard [1979]"}],"fun_headline_variants":["One equation captures 800 years of small-arms progress","Small-arms evolution: a single curve since 1200","Superexponential growth: the hidden 800-year weapon law","From bows to assault rifles: one regular trend","A composite figure reveals 800-year arms regularity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The historical attribute values for medieval and early modern weapons—especially maximum effective range, which for longbows and early guns is a rough historical judgment rather than a measured quantity—are accurate and comparable enough that the fitted exponents and the shape of the 800-year trend are trustworthy.","fun_headline_variants_meta":{"raw":{"variants":["One equation captures 800 years of small-arms progress","Small-arms evolution: a single curve since 1200","Superexponential growth: the hidden 800-year weapon law","From bows to assault rifles: one regular trend","A composite figure reveals 800-year arms regularity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000299,"raw_usage":{"total_tokens":1719,"prompt_tokens":925,"completion_tokens":794,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":715}},"tokens_in":541,"tokens_out":794,"duration_ms":8964,"temperature":1.0,"reasoning_tokens":715,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:13:10.691305+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-fit the FoR and temporal models while perturbing the pre-1600 values of maximum effective range within plausible historical uncertainty, say ±25 to 50 meters; if the exponent on effective range moves by more than a few tenths, or if the quadratic-exponential model no longer wins on the Bayesian Information Criterion, the 800-year law is an artifact of those estimates.","supporting_citations":[],"review_version":1}