{"id":"b9e5b678-5999-446e-bf38-12f5df0bb7b4","arxiv_id":"2607.24926","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"Syntriod recovers Keplerian orbital periods from sparse RV data at ~94% success with six points and ~83% with five, outperforming Lomb–Scargle and also recovering mass ratio and systemic velocity below the sampling limit.","lead":"Syntriod is a template-matching tool that guesses binary-star orbital parameters from sparse radial-velocity data, even with only five or six measurements. It is meant as a fast starter for full orbit fitters used on large survey catalogs.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"Headline sparse-regime success rates are measured under a two-guess scoring rule on noise-free data drawn from the same model class as the template library; the single-guess, noisy rate is never reported.","rationale":"I read the paper in good faith: the method is a sensible deterministic pre-solver, the real-system validation (Tables 1–2) and open code are genuine evidence, and the LS superiority for eccentric/sparse cases is plausible and independently supported by the morphological argument (harmonic leakage vs. phase-domain matching). The load-bearing soft spot is not the algorithm's existence but the interpretation of the headline percentages. The reader identified the two-guess rule and noise-free synthetics; I concur and sharpen it: the noise model in Appendix B (σ∝|RV|) is itself favorable, and the penalty prior is matched to the test population's eccentricity distribution, so even the Appendix B degradation likely understates the real-world drop. This is a correctness-risk concern about the quoted numbers, not about the method's validity, so it does not move the verdict below CONDITIONAL — it specifies exactly which condition should be attached: report top-1 recovery fractions under constant-σ noise before the 83–94% figures are used as survey-ready expectations. The real-binary subsampling tests (with five repeats) partially mitigate this, but those are 12 systems with heterogeneous literature errors, not a calibrated rate. Verdict: unchanged at CONDITIONAL.","tokens_in":28647,"tokens_out":2716,"duration_ms":100420,"concrete_test":"Using the released code/configs, rerun the N_obs=6 and N_obs=5 synthetic campaigns with two modifications: (1) score only the single top-ranked main-template solution (drop the PSin second guess), and (2) inject fixed-σ noise in km/s (e.g., σ=1 and σ=3 km/s) rather than σ∝|RV|. Report top-1 ≤10% period-recovery fractions. If N_obs=5 top-1 noisy recovery falls below ~60–65%, the abstract's ~83% figure should be relabeled as a two-candidate, low-noise upper bound; if it stays ≥75%, the concern largely dissolves.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim — ~94% success at N_obs=6 and ~83% at N_obs=5 — rests on an evaluation protocol that is optimistic in three stacked ways. (1) For N_obs<7, success counts a recovery if EITHER of two returned candidates (main template search or auxiliary PSin) is within 10% (§3.2 footnote); the paper itself states the second guess adds 5–10 percentage points, so the single-guess rate at N=5 is plausibly ~73–78%, not 83%. (2) The primary campaign injects no noise (§3.1); the noisy tests in Appendix B use σ scaled to |RV|, which makes noise vanish near systemic-velocity crossings — an unrealistically favorable noise model for spectroscopy, where error floors are roughly constant in km/s. The σ=2 fixed case exists but its headline fractions are not quoted in the text. (3) The synthetics are exact Keplerians, i.e., the true signal lies inside the template family being searched, and the eccentricity prior encoded in the penalty term (Eq. 12, e0=0.4, suppressing high e) matches the low-e-weighted distribution used to generate the test sample (Fig. 3); the tuned weights (W_φ=1.5, W_tr=2, Δ0=3, W_Δv=0.3) were presumably adjusted on this same population. None of this makes the method unsound — the qualitative LS comparison and real-system tables are independently supportive — but the specific numbers 94%/83% are upper bounds from a best-case protocol, and the abstract presents them (and the >99% q/γ claim, which §5 shows drops to 73–90% with modest noise) without those caveats. A reader deploying Syntriod as a survey pre-solver needs the single-guess, fixed-σ noise rate.","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript presents Syntriod, a phase-domain Keplerian template-matching algorithm intended to supply initial orbital parameters (P, e, ω, T0, K, γ, and for SB2s q) for RV orbit fitters, with an auxiliary harmonic periodogram (PSin) and a linear SB2 fallback for N_obs ≤ 4. The authors validate on 10,000 synthetic Keplerian orbits across N_obs = 10, 8, 6, 5, reporting period-recovery success rates of ~99.9% (N=10), ~94% (N=6), and ~83% (N=5) against a 10% relative tolerance, versus ~94%/62%/42% for Lomb–Scargle, and demonstrate recovery of literature solutions for 12 real systems under random subsampling. For N_obs ≤ 4 they estimate q and γ via the Wilson (1941) linear relation, reporting >99% success in the ideal case.","tokens_in":29057,"tokens_out":4210,"duration_ms":147046,"significance":"If the quantitative claims hold, this is a useful, practical contribution: a fast, deterministic pre-solver for the sparse-RV regime that large surveys (Gaia, SDSS, LAMOST, DESI) routinely produce, where LS aliasing is a real failure mode. The paper ships several genuine strengths: an open-source implementation with configuration files and synthetic datasets on GitHub; a 10,000-orbit injection–recovery campaign with recovery scored against externally known injected parameters; a direct like-for-like LS comparison; twelve real benchmark systems spanning 0.42–3748 d; template-resolution sensitivity checks (Appendix A); and noise-injection tests (Appendix B). The template library, penalty weights, and tolerances are fixed a priori rather than fitted to the success metric, so the evaluation is not circular in construction. The concern is not the architecture but the calibration of the headline numbers, several of which are produced under best-case scoring rules and idealized data.","major_comments":[{"comment":"The headline sparse-regime rates (~94% at N=6, ~83% at N=5, quoted in the abstract and §7) count a recovery as successful if EITHER of two returned candidates (main template search or auxiliary PSin) is within 10% (footnote 4, §3.2). The paper itself states the PSin second guess adds 5–10 percentage points, implying single-solution rates closer to ~85–90% (N=6) and ~73–78% (N=5). Since the method's stated purpose is to provide initial guesses to a downstream optimizer, the single-best-candidate rate is the operationally relevant number. Please report single-candidate and two-candidate rates separately in Fig. 4 and Table 3, and state the scoring rule in the abstract.","section":"§3.2, Fig. 4, footnote 4"},{"comment":"The primary 10,000-orbit campaign is noise-free (§3.1); noise tests are deferred to Appendix B, where the dominant model scales σ to |RV| — heteroscedastic and atypical for spectroscopy, where error floors are roughly constant in km/s. The one realistic case (fixed σ = 2) exists but its global recovery fractions are never quoted in the text; Fig. 11 shows only binned curves. The abstract's 94%/83% figures therefore describe idealized data without saying so. Please add a compact table of global success rates vs. N_obs for each noise prescription (including σ = 2 and σ/|RV| = 0.05), and qualify the abstract numbers accordingly.","section":"§3.1 and Appendix B, Fig. 11"},{"comment":"The abstract claims q and γ are recovered 'with success rates exceeding 99%' for N_obs ≤ 4, but §5/Fig. 9 show that with modest noise (σ = 0.05×RV) the 10%-tolerance rates drop to 90% (N=3) and 73% (N=2) for q, and only ~72% of systems fall within 20% for γ. Moreover, in the ideal N_obs = 2 case the OLS problem has two unknowns and two equations, so the noise-free solution is exact by construction and the >99% ideal rate is nearly tautological rather than evidence of skill. The abstract must carry the noise-dependent numbers, and the ideal-case framing should be corrected.","section":"§5, Fig. 9, abstract"},{"comment":"The validation is in-family in three stacked ways: (i) synthetic signals are exact Keplerians, i.e., inside the searched template family; (ii) the eccentricity penalty (Eq. 12, e0 = 0.4, N0 = 8) suppresses high-e solutions in the same regime where the generation distribution (Fig. 3) is low-e weighted; (iii) the tuned weights (W_φ = 1.5, W_tr = 2, Δ0 = 3, W_Δv = 0.3) show no stated tuning protocol or holdout. Please document how the weights were chosen (and on what sample), and add at least one out-of-family stress test — e.g., orbits with e > 0.8 (outside the library grid) or perturbed/non-Keplerian curves — so readers can gauge performance when the truth is not representable by the template set.","section":"§2.2, §2.4.2, §3.1"}],"minor_comments":[{"comment":"Fig. 8 and §3.4 describe template-matching outputs as 'joint posterior distributions' with '1σ confidence intervals', yet §6 correctly states the method does not sample posteriors or produce formal uncertainties. Relabel as recovery/error distributions across the synthetic ensemble and remove 'confidence interval' language.","section":"§3.4, Fig. 8"},{"comment":"Notation inconsistencies: Eq. (1) uses W_Δλ L_Δλ while §2.4.1 and Eq. (9) define W_Δv L_Δv; Fig. 9 axis labels read 'e = 0' where the caption means σ_e = 0; 'σ_e = 0.05' in Fig. 9 vs 'σ/|RV|' in App. B should be unified. Also Eq. (6) uses a single σ although real errors are heteroscedastic — state what σ was used for the noise-free synthetics.","section":"§2.4, Fig. 9, Eq. (6)"},{"comment":"Eq. (14): 'Gaussian noise (scaled by 0.1)' — 0.1 of what quantity? Give σ_perturb in absolute or fractional units and state how many perturbation realizations are run.","section":"§2.5, Eq. (14)"},{"comment":"Table 1, HD 160934 at N_obs = 5: the period is recovered (3529 vs 3748 d) but e = 0.50 vs 0.65, ω = 20° vs 218°, K1 = 14.8 vs 7.9 km/s. The text's claim of 'consistently reproduces literature solutions' should be scoped to the period (and quantified with tolerances per parameter), or such rows flagged as partial recoveries.","section":"Table 1, §4.1"},{"comment":"§7 calls the LS ~46% rate at N=5 the 'random-selection limit'. A random period draw over [0.1, 100] d would fall within 10% of truth far less often than 46%; either justify the phrase with an explicit random-guess baseline or delete it.","section":"§7"},{"comment":"The 'theoretical sampling limit' N=6 is derived for SB1 (6 parameters); SB2 has 7, yet SB2 systems are included in the N=6 tests. Clarify the framing. Also the real-system subsampling is repeated only five times (§4.1); a larger number of draws (e.g., 50) would make the claimed stability more convincing.","section":"§1, §4.1"},{"comment":"Typos: 'acurracy' (§5); Eq. (10) defines I_ϕ using X(ϕ) = |dX/dϕ| but the same symbol X is used for the template itself — use a distinct symbol for the gradient map.","section":"§5, Eq. (10)"}],"recommendation":"major_revision","confidential_remarks":"The method is sound in conception and the code/data availability is exemplary; my concerns are about calibration of the advertised numbers, not the architecture. Two presentation flags for the editor: the acknowledgments already thank 'the anonymous referee' in this submitted version, suggesting text recycled from another venue's revision; and the companion paper cited as 'Barbaros et al. 2025' is in a low-visibility journal, so this manuscript should stand alone. With the scoring-rule disclosure, noisy headline rates, and abstract corrections, I would expect to support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"Punchline: this is a practical methods paper that packages phase-domain Keplerian templates, an adaptive period grid, a harmonic GLS-like PSin branch, and a clean handoff to linear q–γ when N_obs≤4. It is aimed at initialization, not full inference, and that framing is honest.\n\nWhat is actually new is the combination and the sparse-regime engineering, not template matching or periodograms alone. They back it with 10k synthetics binned in P, e, T_span/P and phase gap, direct LS comparisons, twelve real binaries with random subsampling (including hard cases like HD 160934 and short-period contact systems), multi-grid template checks, a noise appendix, and a public GitHub repo. Runtime claims are modest and believable. For N_obs≥8 the period recovery looks tight (~10^{-3}); the qualitative win over LS on eccentric and long-baseline cases is real and visible in the figures and tables.\n\nSoft spots, in proportion: the stress-test note is mostly right. For N_obs<7 success counts either of two returned candidates, and the paper itself says the second guess adds ~5–10 points, so the single-guess N=5 rate is lower than the abstract’s ~83%. Primary synthetics are noise-free; Appendix B has noise but the fixed-σ case is not the number sold up front, and σ∝|RV| is a friendly model near γ crossings. Penalty weights (e0, W_ϕ, W_tr, etc.) are hand-set and the e-penalty aligns with the synthetic e distribution. None of that collapses the central claim—LS still aliases harder, and the real-system tables survive subsampling—but the abstract oversells survey-ready single-guess rates. A referee should ask for single-candidate noisy fractions and a short weight-sensitivity check.\n\nMath and citations look ordinary and fine for this genre; circularity is low (recovery vs injected/literature truth). Who it is for: people running SB/exoplanet RV pipelines or survey triage who need a cheap basin finder before MCMC/ASA. I would send it to peer review, expect revision on the success definition and noise presentation, and cite it if I were initializing sparse SB fits. Engage.","headline":"Solid pre-solver package with real sparse-regime gains over LS; the 94%/83% headline rates are optimistic (two-guess, mostly noise-free) but the method and open code still hold up.","tokens_in":29801,"tokens_out":580,"would_cite":true,"duration_ms":19307,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Syntriod recovers Keplerian orbital periods from as few as five radial-velocity points by matching data to a library of phase-domain templates, staying accurate where classical periodograms fail.","keywords":["spectroscopic binary stars","radial velocity","orbit determination","time series analysis","template matching","period search","sparse sampling"],"falsifier":"Re-run the 10,000-orbit recovery campaign with realistic measurement noise and count only the single top-ranked period: if success at N_obs = 6 falls near or below Lomb–Scargle levels, or if randomly thinned real binaries no longer match literature periods, the central claim fails.","tokens_in":29458,"feed_emoji":"🔭","tokens_out":982,"duration_ms":16620,"temperature":0.7,"pith_summary":"Spectroscopic binaries need good starting orbital parameters before expensive Bayesian orbit fits will converge, but many modern surveys deliver only a handful of irregular radial-velocity measurements. Classical tools such as Lomb–Scargle assume a sine wave and collapse into aliases when orbits are eccentric or sparsely sampled. This paper introduces Syntriod, a method that folds the data into orbital phase and scores them against a fixed library of precomputed Keplerian velocity curves. On 10,000 synthetic orbits it recovers periods to roughly one part in a thousand when eight or more points are available, still succeeds about 94 percent of the time at the formal six-parameter limit, and about 83 percent of the time with only five points. On twelve real binaries spanning periods from hours to a decade it reproduces literature solutions even after random thinning. When fewer than five points remain it drops the full orbit and still extracts mass ratio and systemic velocity from the linear relation between the two stellar velocities. The practical claim is that a fast, deterministic pre-solver can hand modern pipelines a physically consistent starting guess across both rich and sparse regimes.","feed_headline":"Orbital periods from five RV points via phase-template matching","feed_subtitle":"Syntriod beats Lomb–Scargle on sparse, eccentric binaries and still yields mass ratios with two epochs","key_machinery":"Syntriod: an adaptive orbital-phase-domain template matcher that scores observations against a precomputed library of Keplerian RV morphologies (e, ω grids), refined by an effective likelihood with phase-coverage, eccentricity, and trend penalties, plus an auxiliary harmonic periodogram (PSin) and dual-candidate reporting when N_obs < 7.","core_discovery":"Matching sparse radial-velocity observations to a library of normalized Keplerian templates in orbital-phase space yields reliable initial orbital parameters, including periods accurate to order 10^{-3} for N_obs ≥ 8, success rates near 94 percent at N_obs = 6 and near 83 percent at N_obs = 5, while classical Lomb–Scargle periodograms become alias-dominated; below N_obs = 5 the same framework recovers mass ratio and systemic velocity from linear SB2 relations at high success rates.","pith_inferences":["The same phase-domain library idea could be extended to multi-planet or hierarchical triple RV signals if the template set is enlarged to superposed Keplerians.","Coupling Syntriod’s point estimates directly to a lightweight importance sampler might give approximate uncertainties without a full MCMC chain.","Survey pipelines that already store sparse Gaia or LAMOST RVs could batch-run Syntriod to flag high-priority SB2 follow-up targets by recovered q."],"forward_implications":["Orbit-fitting pipelines can seed MCMC or nested sampling from Syntriod guesses instead of fixing or narrowly restricting the period.","Large spectroscopic surveys with few epochs per star can still extract usable initial periods and, for SB2s, mass ratios and systemic velocities.","Eccentric and long-baseline systems that defeat sine-based periodograms become tractable for automated first-pass characterization.","When photometry already supplies the period, two to four RV epochs can still yield dynamical mass-ratio and γ constraints.","Runtime of seconds per system makes the method practical as a survey-scale pre-solver."],"fun_headline_variants":["Syntriod recovers periods from 5 RV points at 83% success","Phase-template matching beats Lomb-Scargle on sparse binaries","Initial orbits from 6 RV epochs with 94% success via Syntriod","Keplerian templates yield 10^{-3} period accuracy at N_obs≥8","Sparse RV to mass ratio and gamma below 5 epochs"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The headline sparse-data success rates treat a recovery as correct if either of two returned candidate periods is close enough, and the main synthetic tests use noise-free velocities, so the quoted percentages partly rest on a two-guess allowance and idealized data.","fun_headline_variants_meta":{"raw":{"variants":["Syntriod recovers periods from 5 RV points at 83% success","Phase-template matching beats Lomb-Scargle on sparse binaries","Initial orbits from 6 RV epochs with 94% success via Syntriod","Keplerian templates yield 10^{-3} period accuracy at N_obs≥8","Sparse RV to mass ratio and gamma below 5 epochs"]},"model":"grok-4.5","effort":"low","cost_usd":0.003904,"raw_usage":{"total_tokens":1321,"prompt_tokens":951,"num_sources_used":0,"completion_tokens":82,"cost_in_usd_ticks":39044000,"prompt_tokens_details":{"text_tokens":951,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":288,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":951,"tokens_out":82,"duration_ms":4940,"temperature":1.0,"reasoning_tokens":288,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T05:40:20.555440+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-run the 10,000-orbit recovery campaign with realistic measurement noise and count only the single top-ranked period: if success at N_obs = 6 falls near or below Lomb–Scargle levels, or if randomly thinned real binaries no longer match literature periods, the central claim fails.","supporting_citations":[],"review_version":1}