{"id":"6db3ab48-fa72-402f-acbc-3aace35d08a0","arxiv_id":"2512.21428","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"New optical clock frequency ratios between Al+, Yb, and Sr reach fractional uncertainties <=3.2x10^-18 and disagree with 2021 values by up to 14 sigma.","lead":"Three high-precision atomic clock pairs in Boulder were compared over a shared 3.6 km fiber link, giving frequency ratios with fractional uncertainties as low as about 2 parts in 10^18. The results meet a benchmark for redefining the SI second and, by disagreeing with the same team's 2021 values, expose unknown systematic errors in optical clock comparisons.","discovery_kind":"replication","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed sub-3.2e-18 uncertainties are conditional on an unproven no-constant-bias premise; the 14-sigma Sr/BACON21 offset remains unresolved and needs an independent Sr ratio.","rationale":"I agree with the reader's weakest-assumption diagnosis. The paper is unusually careful and honest: the hierarchical model, the network loopback, and the systematic modulation tests are strong internal evidence. However, the central quantitative claim depends on the assumption that no constant unknown bias affects the current campaign. The Bayesian model's zero-mean random-effect structure cannot estimate or bound such a bias, and the 14-sigma Sr discrepancy with BACON21 is exactly the kind of unexplained offset that should prevent unconditional acceptance. The paper itself concedes that the discrepancy is of unknown origin and that more interlaboratory comparisons are needed. Since the reader already reached CONDITIONAL on essentially this basis, my stress test does not change the verdict. I would not move to REJECT because the paper does not hide the limitation and the internal validations are genuine; I would not move to ACCEPT because the no-unknown-bias premise is not yet independently established.","tokens_in":25924,"tokens_out":11603,"duration_ms":125345,"concrete_test":"Use the published 2025 direct Yb/Sr values from Ref. [35] (INRIM/PTB, INRIM/SYRTE, NMIJ/NPL, etc.), which the paper cites in Fig. 5, and compare each to Eq. (1) using its reported uncertainty. If any of these independent direct measurements agrees with BACON21 and disagrees with Eq. (1) by more than the combined uncertainty, the ~1e-16 Sr revision is already contradicted by independent data; if they agree with Eq. (1), the no-constant-bias assumption is supported. If those data are insufficiently precise, perform a new direct 87Sr-vs-171Yb ratio measurement at an independent laboratory with total uncertainty below 1e-17 to settle the same question.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive premise is stated in Supplement C: 'all models here ... assume that the measurements themselves are free from any unknown systematic bias.' It is encoded in the Bayesian model: systematic corrections enter as a_i*alpha with alpha ~ N(0,1), so they are centered on the evaluated corrections, and the day-to-day 'dark uncertainty' terms lambda are zero-mean N(0, xi^2). Such terms can absorb scatter (posterior xi_87Sr ~ 2.2(1.2)e-18) but cannot absorb a constant, campaign-wide offset. The data themselves exhibit the signature of exactly such an offset: both Sr-containing ratios differ from BACON21 by ~1e-16 (14 sigma), while Al+/Yb differs by only 1.6e-17 (2.4 sigma). The three measured ratios are internally consistent, so this internal consistency cannot tell whether the offset is in BACON21 or in the present campaign. The loopback measurement bounds the optical network at 0.9(1.6)e-19 and the modulation runs validate known systematics, but neither constrains a constant atom-level bias. The paper honestly flags 'discrepancies of unknown origin'; therefore Eqs. (1) are campaign-conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports direct optical frequency ratio measurements between the 27Al+, 171Yb, and 87Sr clocks in the Boulder network, carried out on 13 days between January and March 2025. Using a cryogenic silicon cavity as a common reference distributed over a 3.6 km fiber link, the authors achieve fractional instabilities of a few times 10^-16/sqrt(τ) and quote total fractional uncertainties of 2.2×10^-18 (Al+/Sr), 3.2×10^-18 (Al+/Yb), and 3.1×10^-18 (Yb/Sr) after Bayesian aggregation. The paper emphasizes that this meets a 5×10^-18 roadmap milestone for the SI second, while also reporting large discrepancies with the BACON21 results, especially in the Sr-containing ratios (~1×10^-16, about 14σ). Extensive modulation tests, a new end-to-end network loopback, and multi-ratio analytics are presented to support the accuracy of the campaign.","tokens_in":26215,"tokens_out":7244,"duration_ms":85144,"significance":"If the quoted uncertainties are taken at face value, these are the most precise direct optical clock ratios yet reported and are relevant to the SI second redefinition process. The paper is notable for its transparency: it states the discrepancies with prior work, gives the full Bayesian model, reports alternative statistics, and includes modulation tests of several large systematic effects. The loopback measurement bounding the network at 0.9(1.6)×10^-19, the in-situ correction validations, and the three-cornered-hat analysis are commendable and strengthen the case that no single known systematic at the 10^-17 level is unaccounted for. However, the central claim of sub-3.2×10^-18 uncertainty is conditional on an unproven premise—the absence of any unknown constant systematic bias—and the unresolved 14σ Sr discrepancy makes this premise load-bearing rather than cosmetic.","major_comments":[{"comment":"The model encodes the assumption, explicitly stated in Supplement C, that 'the measurements themselves are free from any unknown systematic bias.' The systematic corrections are central (α,β,γ ~ N(0,1)) and the day-to-day dark-uncertainty terms are zero-mean (λ ~ N(0,ξ^2)). Such terms can inflate the reported uncertainty but cannot absorb a constant, campaign-wide offset. Given the ~1×10^-16 (14σ) discrepancy in both Sr-containing ratios, a constant bias in the present campaign is a viable explanation of the data. The abstract's statement that these measurements 'meet an important milestone criterion for redefinition of the second' therefore overstates the robustness of Eqs. (1). Please either add a global-offset parameter with a sensitivity analysis, or explicitly qualify the abstract and conclusion so that the quoted uncertainties are described as conditional on no uncharacterized comm","section":"Supplement C, Comprehensive Bayesian model"},{"comment":"The statement that 'all known systematic effects that shift any of the three clocks by 10^-17 or more have been tested' supports the individual corrections, but the discrepancy is an order of magnitude larger than the quoted uncertainties. The modulation tests cover known effects; they do not, and cannot, rule out a combination of smaller unknown effects or a single effect not included in the model. The conclusion already notes that repeatability below 10^-17 remains outstanding, but this caveat appears only after the central claims. Please add an explicit paragraph stating that the unresolved discrepancy could reside in the present campaign and discussing the consequences for the SI-redefinition roadmap. The accompanying Bayesian model should also be applied with a constant-offset scenario so readers can see how the quoted uncertainties change under that assumption.","section":"Main text, Discussion and Conclusion"},{"comment":"The Yb/Sr ratio has χ2_red = 6.4, with between-day variability of 3.3(9)×10^-18. The Bayesian model accounts for this through zero-mean random effects (ξ). For 13 daily points, a time-correlated drift in the Yb/Sr residuals—for example from a slow environmental change or an operational parameter that evolves over the campaign—could bias the mean without violating the zero-mean assumption. Please report the daily residuals as a function of date, check for autocorrelation, and include a sensitivity test with a linear trend or an AR(1) term in the dark-uncertainty model. This directly affects the central value of Yb/Sr and the claimed 3.1×10^-18 uncertainty.","section":"Measurement results; Supplement C, dark uncertainty"}],"minor_comments":[{"comment":"The sentence in the Discussion says the loopback test evaluates 'every element from Si cavity to 87Sr atoms,' but the description shows the 698 nm comparison is made against the 87Sr clock laser 'picked off just before the atoms.' This is an overstatement; the loopback validates the network and laser delivery up to the atoms, not the atom-light interaction. Please rephrase to reflect the actual measurement point.","section":"Main text, loopback description"},{"comment":"The caption refers to 'color filled points' and colored shaded regions but does not identify which color corresponds to which ratio. Please add a legend or explicit color definitions so the lower panel is interpretable.","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"This is a strong experimental paper with unusually transparent treatment of systematics and a hard, unresolved tension with prior work. My main concern is that the headline uncertainty claim is presented without the conditional clause that the paper's own Supplement C identifies. A major revision that adds a global-offset sensitivity analysis and qualifies the abstract/conclusion would resolve this. I do not think rejection is warranted: the measurements are valuable regardless of the BACON21 discrepancy, and the authors have made a serious, detailed case that their known systematics are under control."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a high-quality, careful measurement campaign, and the three ratios are likely close to correct, but the quoted sub-3.2e-18 uncertainties are conditional on a no-constant-bias assumption that the 14-sigma Sr discrepancy undermines. Don't treat Eqs. (1) as final consensus values yet.\n\nWhat's genuinely new: the common cryogenic Si-cavity reference distributed to all three clocks over the 3.6-km fiber (2-3x stability improvement), the end-to-end loopback test at 698 nm bounding network offset at 0.9(1.6)e-19, and the three-cornered-hat analysis that gives individual clock stabilities. The systematic evaluations are unusually thorough: each clock modulated its leading shifts (Sr BBR, lattice, density, Zeeman; Al+ Doppler motion; Yb bias field, cycle time, BBR), and the results are consistent. The statistics are also careful: the Bayesian model is explained, three aggregation methods agree within 1.2e-18, and the paper honestly discusses the limitations of Birge and Mandel-Paule approaches.\n\nThe soft spot is not the engineering; it's the inference. The model explicitly assumes 'the measurements themselves are free from any unknown systematic bias' (Supplement C). The dark-uncertainty terms are zero-mean, so they absorb day-to-day scatter but not a constant, campaign-wide offset. The 14-sigma disagreement with BACON21 in both Sr-involving ratios, while Al+/Yb differs by only 2.4 sigma, is exactly the signature of a constant offset somewhere. Internal consistency of the three new ratios can't localize it. The loopback bounds the network and the modulation tests constrain known atom-level shifts, but they don't rule out a new constant atomic bias. The paper says this plainly and calls for interlaboratory comparisons — good, but that means the sub-3.2e-18 milestone claim is not yet repeatability-demonstrated. Minor: raw data and code aren't public, and one Yb density-shift correction leans on an unpublished manuscript [42].\n\nWho should read it: anyone tracking optical clock ratios or the SI second redefinition. It deserves a serious referee — the measurement effort is substantial and the discrepancy itself is scientifically important. Referees should push the authors to present a conservative analysis incorporating a possible constant offset and to release the data. The next independent Sr comparison should tell us whether BACON21 or this campaign carries the offset. Until then, cite it as the new state of the art, but with that caveat.","headline":"Careful, state-of-the-art measurement whose quoted sub-3.2e-18 uncertainties rest on a no-constant-bias assumption that the 14-sigma Sr discrepancy makes doubtful.","tokens_in":26865,"tokens_out":2980,"would_cite":true,"duration_ms":30711,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["06.30.Ft"],"model":"deepseek-v4-flash","headline":"Three optical clock frequency ratios measured to below 3.2 × 10^-18, meeting a milestone precision for redefining the SI second.","keywords":["optical atomic clocks","frequency ratios","SI second redefinition","strontium clock frequency","ultrastable laser reference","phase-stabilized fiber link","quantum projection noise","dark uncertainty"],"falsifier":"An independent measurement of these three ratios by a different laboratory, ideally using a different ultrastable reference and a different network path (or a satellite link), would settle the central claim: if it reproduces the new values, the 2021 campaign was biased; if it reproduces the 2021 values, the new campaign carries the hidden offset. A simpler first test is a direct optical comparison of a distant, independently evaluated strontium clock against the same aluminum-ion clock, without the shared fiber network.","tokens_in":25781,"feed_emoji":"⏱️","tokens_out":7453,"duration_ms":70566,"temperature":0.7,"pith_summary":"This paper reports new, direct frequency-comparison measurements among three of the most precise optical atomic clocks: a single aluminum-ion clock and ytterbium and strontium optical lattice clocks. All three frequency ratios—Al+/Sr, Al+/Yb, and Yb/Sr—have total fractional uncertainties at or below 3.2 × 10^-18, with the smallest at 2.2 × 10^-18, making them the most precise direct optical clock ratios reported to date and meeting a 5 × 10^-18 threshold identified as a milestone for redefining the SI second. The key innovation is a common ultrastable optical reference, derived from a cryogenic silicon cavity and delivered to both laboratories over a phase-stabilized 3.6 km fiber link, which improves comparison stability by factors of two to three. The results, however, disagree with the same collaboration's 2021 measurements in the strontium-containing ratios by about 14 standard deviations, implying a roughly 1 × 10^-16 revision of the strontium clock frequency. The paper argues from a battery of systematics tests and an end-to-end network loopback that its new values are accurate, while conceding that the discrepancy is unresolved and needs independent cross-checks.","feed_headline":"Three clock ratios measured below 3.2e-18, a milestone for the second","feed_subtitle":"A shared ultrastable laser made the fast comparisons possible; a 14-sigma strontium shift now demands independent checks.","key_machinery":"The central mechanism is a phase-stabilized optical distribution network: a 1542 nm laser locked to a cryogenic single-crystal silicon cavity at one laboratory is sent over a 3.6 km fiber to the other laboratory, where it phase-locks each clock laser via frequency combs; a hydrogen-maser RF signal is multiplexed on the same fiber for a common microwave reference. This common reference suppresses the Dick effect and quantum projection noise, lowering the instability floor. The validity of the network is established by a loopback test that sends the silicon-cavity light to the second laboratory, uses it to lock a second comb, returns a 1397 nm laser, and compares the frequency-doubled light (6","core_discovery":"On the paper's own terms: with all three clocks running simultaneously from January to March 2025, the collaboration measured the three ratios given in Eq. (1), with total fractional uncertainties of 2.2 × 10^-18 (Al+/Sr), 3.2 × 10^-18 (Al+/Yb), and 3.1 × 10^-18 (Yb/Sr). The Yb/Sr ratio instability reached 1.3 × 10^-16 at one second, and the Al+-involving ratios are limited by single-ion quantum projection noise at 3.9 × 10^-16; the common silicon-cavity reference reduces averaging time by roughly an order of magnitude compared with the previous campaign. The paper reports a 14-standard-deviation shift in the two ratios involving strontium relative to the 2021 values, corresponding to about","pith_inferences":["If the ~1 × 10^-16 strontium revision is confirmed by an independent group, the prior 2021 strontium evaluation most likely contained an unaccounted shift (e.g., in the blackbody radiation correction), rather than the new network; the revised value would become the de facto standard for strontium-based chronometric leveling and searches for time-variation of fundamental constants.","The 14σ disagreement between two campaigns that both claim sub-1 × 10^-17 accuracy is itself evidence that at least one of the two error budgets is incomplete; the paper's dark-uncertainty model absorbs day-to-day scatter but cannot absorb a constant offset, so the discrepancy cannot be resolved by re-weighting the data.","A testable extension: run one of the lattice clocks against two independent silicon cavities (or against a distant clock via satellite link) to separate network-origin from atom-origin offsets; the loopback used here covers only the two-laboratory fiber path.","The three-clocks-simultaneously design is the seed of a network-level metrology: with three-cornered hat analysis the paper already isolates per-clock stabilities, and with a multi-ion Al+ clock the same network could directly target the Yb/Sr between-day scatter."],"forward_implications":["The three ratio values in Eq. (1) are, to the authors' knowledge, the most precise direct optical clock ratios published, with total fractional uncertainties of 2.2, 3.2, and 3.1 × 10^-18.","The measured precision meets the ≤5 × 10^-18 criterion identified by the community as a milestone for deciding whether to redefine the SI second.","The common silicon-cavity reference lowers the instability of the Yb/Sr ratio to 1.3 × 10^-16/√τ and of the Al+ ratios to 3.9 × 10^-16/√τ, a factor-of-3 improvement that cuts the time needed to reach a given statistical uncertainty by roughly a factor of ten.","The strontium-containing ratios imply the strontium clock frequency should be revised by about 1 × 10^-16 relative to the 2021 values; this revision would bring the measured ratios into agreement with the new data.","The Yb/Sr excess scatter (χ²_red = 6.4) shows that uncharacterized shifts remain in at least one lattice clock, and the paper states that repeatability below 1 × 10^-17 remains an outstanding issue for the redefinition effort."],"fun_headline_variants":["Clock ratios hit 3.2e-18, milestone for redefining the second","Shared laser boosts clock comparisons to 3.2e-18 precision","14-sigma strontium shift in clock ratios demands recheck","Fiber-linked clocks reach 1e-16 per second stability"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The analysis stands on the assumption that the new campaign's measurements are free of any unknown systematic bias — if a hidden offset affected the new data, the quoted sub-3.2 × 10^-18 uncertainties and the implied 1 × 10^-16 strontium revision are not reliable.","fun_headline_variants_meta":{"raw":{"variants":["Clock ratios hit 3.2e-18, milestone for redefining the second","Shared laser boosts clock comparisons to 3.2e-18 precision","14-sigma strontium shift in clock ratios demands recheck","Fiber-linked clocks reach 1e-16 per second stability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000643,"raw_usage":{"total_tokens":2828,"prompt_tokens":814,"completion_tokens":2014,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":558,"completion_tokens_details":{"reasoning_tokens":1934}},"tokens_in":558,"tokens_out":2014,"duration_ms":15307,"temperature":1.0,"reasoning_tokens":1934,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T14:05:36.652650+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An independent measurement of these three ratios by a different laboratory, ideally using a different ultrastable reference and a different network path (or a satellite link), would settle the central claim: if it reproduces the new values, the 2021 campaign was biased; if it reproduces the 2021 values, the new campaign carries the hidden offset. A simpler first test is a direct optical comparison of a distant, independently evaluated strontium clock against the same aluminum-ion clock, without the shared fiber network.","supporting_citations":[],"review_version":1}