{"id":"f77d596d-2d11-42cb-aa2c-a9f61311e492","arxiv_id":"2607.17204","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A Rabi drive during dispersive readout hides a transmon's computational states from the probe while leaving the |f⟩ leakage state visible, giving single-shot heralded leakage detection at 97.1% fidelity with 92.9% post-detection state fidelity.","lead":"Researchers show that a microwave 'locking' pulse applied during measurement makes a superconducting qubit's two computational states invisible to the probe, while its unwanted high-energy 'leakage' state stays visible — so leakage can be flagged in 80 nanoseconds at 97% accuracy without erasing the stored quantum information. The method needs no extra hardware and could plug directly into quantum error-correction and erasure-decoding pipelines.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Measured demonstration is sound, but the coherence-preservation mechanism and all projected performance claims rest on a photon-noise model that the paper's own data contradict by ~3× with unexplained origin; the mechanism is quantitatively incomplete as written.","rationale":"The reader's weakest_assumption correctly identifies the quantitative noise model as the most fragile pillar. The central measured quantities—detection fidelity, false-flag rate, and conditioned state fidelity—do not depend on the theory being perfect, so the demonstration itself remains credible. However, the title claims 'preserved computational-state coherence' via a specific suppression mechanism, and the summary projects device-level improvements based on that mechanism. The paper openly concedes that the measured dephasing exceeds the model by ~3× with an unknown origin and that removing it is a prerequisite for the projections. That is an internally flagged missing support, not an external disagreement with consensus. A conditional verdict is therefore appropriate; no rejection is warranted because the raw measured results are self-consistent and the error budget reproduces the p_f→0 limit. My read does not change the reader's verdict.","tokens_in":26645,"tokens_out":12643,"duration_ms":117128,"concrete_test":"Use the qubit as an in-situ spectrometer of S_nn(ω): measure measurement-induced dephasing at zero drive (Ramsey) and under drive (Rabi-envelope decay) while sweeping probe power n̄ and Rabi frequency Ω_q across a range that includes the operating point. Compare the extracted Γ_2ρ^meas(Ω) with the two-pole prediction of Eq. (S39) using independently measured κ_eff, κ_f, J, and χ. If the measured S_nn is uniformly ~3× the model across Ω and n̄, the missing noise source is confirmed and the model-based projections cannot be accepted until it is identified. If the ratio is strongly Ω-dependent, the excess is drive-induced rather than a static noise floor, which would point to a different physical origin.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The measured core—97.1(3)% detection fidelity, 2.3(3)% false-flag rate, 92.9(5)% conditioned state fidelity—is well supported by the data and I do not question it. The load-bearing gap is the quantitative claim that the Rabi drive suppresses measurement-induced dephasing according to the photon-number shot-noise spectrum S_nn(ω) = n̄κ/(ω²+(κ/2)²) (Eqs. 1–2), extended to the two-pole Purcell-filter form in Sec. IX A. The paper's own Sec. IX C states that the measured dressed-state dephasing Γ_2ρ^meas exceeds this prediction by approximately a factor of three at the operating point, and that even without the Rabi drive the weak-probe dephasing exceeds the standard Γ_m = 2χ² n̄/κ by about 1.6×, with the excess of unidentified origin. The forward-looking claims—threefold infidelity improvement at κ_eff ≈ |χ|, false-flag rate below 1%, and 10^-3-level back-action—are extrapolations from this same noise model, and the text explicitly states that removing this excess is 'a prerequisite for the estimates above.' Thus, while the central experimental demonstration stands, the paper's explanation of why coherence is preserved, and its projected performance, are quantitatively incomplete without identifying and accounting for this extra noise source.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a heralded leakage-detection protocol for a fixed-frequency transmon: during dispersive readout, a strong Rabi drive on the |g⟩–|e⟩ transition makes the computational states indistinguishable to the probe while leaving |f⟩ and higher states distinguishable. The authors characterize the scheme with 30,000-shot single-shot histograms, obtaining an assignment matrix with false-flag rate ε_FF = 2.3(3)% and undetected-leakage rate ε_UL = 3.5(2)%, corresponding to F_d = 97.1(3)% in an 80-ns window. They further prepare incoherent mixtures p_f |f⟩⟨f| + p_ge ρ_ge and, after the detection and a calibrated recovery gate, perform QST; for p_f = 0.5 the post-selected computational-subspace state has average fidelity 92.9(5)%. The paper also develops a dressed-state noise model in which measurement-induced dephasing is suppressed by sampling photon-number noise at the Rabi frequency (Eqs. 1–2), and uses this model, together with an error-budget decomposition, to analyze the detection errors and project performance for optimized device parameters.","tokens_in":27004,"tokens_out":5733,"duration_ms":59710,"significance":"The experimental core is valuable and credible: the central numbers are direct measurements with statistical uncertainties, the error decomposition is careful, and the QST-based conditional fidelities support the claim that the scheme can herald leakage while largely preserving computational-subspace states. The protocol requires no additional hardware beyond standard dispersive readout, which is an attractive practical feature. However, the paper's quantitative explanation of the coherence-preservation mechanism is incomplete: the measured measurement-induced dephasing exceeds the shot-noise model by roughly a factor of three at the operating point, and by ~1.6× even without the drive, with the origin unidentified. Because the projected performance in Secs. VIII–IX explicitly depends on removing this excess, the forward-looking claims are extrapolations rather than supported predictions. The demonstrated measurement itself stands, but the mechanism and projections need revision.","major_comments":[{"comment":"The quantitative coherence-preservation claim rests on the photon-number shot-noise spectrum S_nn(ω)=n̄κ/(ω²+(κ/2)²) sampled at the Rabi frequency. The paper's own data, however, show that the measured dressed-state dephasing exceeds this prediction by roughly a factor of three at the operating point, and that even without the Rabi drive the weak-probe dephasing exceeds the standard Γ_m=2χ²n̄/κ by about 1.6×, with the origin unidentified. The excess is absorbed into C_back in Eq. (S32), and Sec. IX C states that removing it is \"a prerequisite for the estimates above.\" Thus the model as written does not quantitatively explain the observed coherence preservation, and the projected 10^-3-level back-action, the κ_eff≈|χ| optimum, and related estimates are not supported predictions. This is load-bearing for the central explanatory claim. I recommend either identifying and including the excess","section":"Supplemental Sec. VIII, Eqs. (S30)–(S35)"},{"comment":"The projection model is calibrated entirely at the present operating point: η=0.25 is chosen to match the measured separation error, C_back is chosen to match T_Rabi, and γ_rel, γ_leak are chosen to reproduce Table III. The linear photon-number scaling of the flip rates (p_ge→f = p_0 + γ_leak n̄_c τ_tot; p_f→ge = Γ_1,ef t_R + γ_rel n̄_f τ_tot) is introduced as a phenomenological model without derivation or validation at a second linewidth. Consequently, Fig. S8 is not a parameter-free prediction; its quantitative behavior is set by the calibration assumptions. The text should state this limitation explicitly so that readers do not mistake the projected false-flag rates and infidelities for validated device-independent scalings.","section":"Supplemental Sec. IX A and main text summary"}],"minor_comments":[{"comment":"The anomalous Rabi-oscillation points are attributed to an unidentified \"spurious mode near the qubit frequency\" and omitted, and this omission caps the drive at Ω_q/2π=186 MHz. Please provide the affected data and whatever frequency/power dependence is available so readers can assess the attribution. As written, the explanation is an ad hoc entity introduced to justify excluding the strongest-drive data.","section":"Supplemental Sec. II G and Fig. 2(c)"},{"comment":"The analytical model's agreement with Fig. 4(e) is presented as \"validating the detection protocol and the error-budget analysis.\" However, the theory curve uses the measured T_Rabi, ε_sep, and ε_flip from the same device and operating point (Sec. VI). This is a consistency check of the error sum, not an independent prediction; the wording should be softened accordingly.","section":"Main text Fig. 4(e); Supplemental Sec. VI"},{"comment":"The sentence \"Since the photon-number noise scales as S_nn(Ω_q)∝n̄κ, reducing the effective resonator linewidth κ_eff ... suppresses the back-action\" is correct only in the single-pole regime. The two-pole filter model of Sec. IX A gives a different, steeper scaling (∝Ω_q^-4 in the stop band). Please qualify the statement so that it does not overstate the transferability of the single-pole intuition.","section":"Main text summary; Supplemental Sec. IX A"},{"comment":"The word \"hardware-agnostic\" is stronger than what is demonstrated: the scheme still requires a Rabi drive on the computational transition and a dispersively coupled readout resonator. The final sentence of the abstract already says the scheme requires no additional circuit elements; it would be more accurate to phrase the contribution in those terms than as hardware-agnostic.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"I see no reason to question the measured experimental numbers; the single-shot histograms, assignment matrix, QST fidelities, and error decomposition are credible and well presented. The unresolved excess dephasing is acknowledged honestly in the manuscript, but it is not a minor caveat: it is a factor-of-three discrepancy in the central mechanism, and the projected performance estimates explicitly depend on removing it. The paper would be suitable after either a corrected noise model, or a clear reframing of the main claim as an empirical demonstration with the projections demoted to conditional estimates. I do not see grounds for rejection, but the current framing overstates the quantitative support for the coherence-preservation mechanism."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper really does what it says: single-shot, heralded leakage detection on a fixed-frequency transmon, with the |g⟩/|e⟩ subspace made transparent to the readout by a Rabi drive while |f⟩ stays visible. The headline numbers are direct measurements, not fits: 97.1(3)% detection fidelity, 2.3(3)% false-flag rate, 3.5(2)% undetected-leakage rate, and 92.9(5)% conditional state fidelity at p_f = 0.5. The error budget is careful and internally consistent — the sixteen-pathway model in the Supplement reproduces the infidelity curve against p_f, and the T_Rabi limit correctly captures the p_f→0 floor. The POVM check on a coherent qutrit superposition is a nice extra. This is exactly the kind of reproducible experimental evidence that should be taken seriously.\n\nThe genuinely new piece is applying quantum rifling to leakage detection. The two-level mechanism (sampling photon-number noise at the Rabi frequency) is prior art, and the authors credit Fedorov's group. The anharmonicity clock condition, the recovery gate to undo the dressed-frame rotation, and the error decomposition are practical contributions. The writing is honest: they explicitly say the origin of the excess dephasing is unidentified.\n\nWhere the paper is soft: the quantitative mechanism for back-action suppression does not fully hold. Their own Sec. IX C reports that measured measurement-induced dephasing exceeds the single-Lorentzian shot-noise prediction by about a factor of three at the operating point, and even without the drive the weak-probe dephasing is 1.6× above the standard Γ_m. They cannot explain it. That means the statement \"the Rabi drive suppresses measurement-induced dephasing according to Eqs. (1)–(2)\" is only qualitatively supported. The central 97%/93% numbers survive because they are measured. But the forward-looking claims — threefold infidelity improvement at κ_eff ≈ |χ|, false-flag below 1%, back-action at 10⁻³ — are extrapolations from a model with four coefficients calibrated at the operating point, and the paper itself says removing the excess is a prerequisite. The projections should be clearly labeled as conditional, not as performance guarantees.\n\nThere is also a spurious mode near the qubit frequency that forced them off the strongest drive and caused some omitted data points. That is a minor instrumentation nuisance, not a flaw in the scheme.\n\nThis paper deserves a serious referee, not a desk reject. The measured demonstration is solid and the limitations are honestly stated. The referee should ask the authors to characterize the excess dephasing source (or at least report its power dependence) and to separate measured results from model projections in the abstract. I would bring it to a reading group and I would cite it. My recommendation: send to peer review, expect heavy revision on the theory section, but keep the experimental core intact.","headline":"Solid measured demonstration of heralded leakage detection; the coherence-preservation mechanism is quantitatively incomplete, but the core numbers stand.","tokens_in":27610,"tokens_out":2017,"would_cite":true,"duration_ms":20694,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A resonant Rabi drive during dispersive readout makes a fixed-frequency transmon's computational states transparent to the probe while leaving leakage states visible, yielding 97.1% heralded leakage-detection fidelity with preserved computa","keywords":["leakage detection","superconducting qubits","transmon","dispersive readout","Rabi drive","measurement-induced dephasing","heralded error","quantum error correction"],"falsifier":"Measure Γ_1ρ^meas as a function of Rabi frequency Ω_q at fixed photon number n̄ and compare with the predicted Lorentzian χ² n̄ κ / [2(Ω_q²+(κ/2)²)]. If the extracted 'κ' from the decay curve disagrees with the independently measured resonator linewidth, or if the dephasing does not drop as 1/Ω_q² at large drive, the central suppression mechanism is wrong. A simpler check: with the Rabi drive off, the model predicts measurement-induced dephasing Γ_m = 2χ² n̄/κ; the paper finds ~1.6× this value, so eliminating that unexplained excess in a clean device would validate—or falsify—the quantitative","tokens_in":26376,"feed_emoji":"⚛️","tokens_out":9042,"duration_ms":73811,"temperature":0.7,"pith_summary":"Leakage—population leaking from a qubit's 0 and 1 states into higher levels—is a quiet killer in superconducting quantum processors: it escapes the error model of quantum error correction and can persist for many cycles. This paper demonstrates a way to catch it on the fly without destroying the quantum information being processed. The trick is to shine a strong Rabi drive on the 0–1 transition while the usual dispersive measurement is running. The drive makes the two computational states look identical to the readout resonator, so the probe no longer collapses or dephases them; higher excited states keep their distinct pull and are flagged as leakage. In 80 ns the detector correctly identifies leakage in 97.1% of shots, and when it reports no leakage the qubit's state survives with 92.9% average fidelity. Because it requires no extra circuit elements, this could become a plug-in feature of standard transmon readout, giving error-correction decoders a direct, hardware-efficient leakage flag.","feed_headline":"Rabi drive hides qubit states, flags leakage at 97%","feed_subtitle":"Flags leakage without destroying the qubit, using hardware that is already inside every transmon readout.","key_machinery":"The workhorse is the dressed-state picture of a Rabi-driven qubit plus the resonator's photon-number noise spectrum. The drive splits the computational subspace into dressed states separated by Ω_q; the dispersive coupling χ n̂|e⟩⟨e| becomes transverse in this basis, so measurement-induced dephasing samples the noise at Ω_q rather than at zero. With the Lorentzian spectrum S_nn(ω)=n̄κ/(ω²+(κ/2)²), the induced relaxation rate is Γ_1ρ^meas=(χ²/2)S_nn(Ω_q), suppressed when Ω_q≫κ. Finite anharmonicity adds longitudinal noise, which the paper cancels by a drive detuning δ≈−Ω_q²/(2α), the spin-locking 'clock condition'. A recovery gate R_z(φ)R_x(θ)R_z(λ) undoes the drive's rotation of the computat","core_discovery":"The central discovery: driving the |g⟩–|e⟩ transition during dispersive readout makes the computational subspace transparent to the probe—the |g⟩ and |e⟩ resonator responses merge—while off-resonant |f⟩ and higher states remain visible. This yields 97.1(3)% leakage-detection fidelity in an 80-ns window (false-flag 2.3(3)%, undetected leakage 3.5(2)%). Conditioned on no leakage, tomography on an equal computational/leakage mixture gives 92.9(5)% average fidelity over six cardinal states. The detection is projective with POVM {|f⟩⟨f|, I−|f⟩⟨f|}; the drive's rotation is compensated by a recovery gate; no additional hardware is required.","pith_inferences":["A direct next step suggested by the mechanism is to sweep the Rabi drive strength against a two-pole Purcell filter design: the theory predicts S_nn(Ω_q) falling as 1/Ω_q⁴ in that regime, a sharp, testable signature distinct from the single-pole 1/Ω_q².","One could apply the same 'make the computational subspace transparent' trick to other dispersively coupled multilevel qubits or to higher leakage manifolds, flagging each excited level at a different readout frequency.","Interleaving this detector with a leakage-reduction unit would let the processor reset only the flagged qubits, potentially lowering average reset overhead; the flag also gives decoders a direct erasure signal for dual-rail microwave-photon communication.","Because the drive makes the qubit invisible to probe photons, the same technique could protect an idling qubit while neighbouring qubits on the same chip are being measured."],"forward_implications":["Fixed-frequency transmon processors can add single-shot, heralded leakage detection without adding any circuit elements; only the pulse schedule changes.","Post-selecting on the no-leakage outcome removes leakage population: for p_f=0.5 the conditional infidelity drops to 7.1%, and the measured p_f-dependence matches an analytical sixteen-pathway error model.","With an optimized resonator linewidth (κ_eff≈|χ|) the projected false-flag rate falls below 1% and conditional infidelity reaches ~1%; with tenfold longer T1 the back-action reaches ~2×10⁻⁴.","The protocol acts as a projective leakage measurement, directly useful for erasure-qubit architectures and dual-rail microwave-photonic communication where photon loss appears as |f⟩ population.","At small leakage fractions the residual error is set by T_Rabi (measurement-induced dephasing under the drive), giving a clear figure of merit for further hardware improvements."],"fun_headline_variants":["Non-destructive leakage detection via Rabi drive","Drive-assisted readout flags leakage, preserves qubit","97% fidelity leakage check without qubit destruction","Heralded leakage signal while keeping quantum state intact","Rabi drive makes errors visible without breaking qubit"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The quantitative coherence-preservation story rests on the assumption that measurement-induced back-action is set by the single-Lorentzian photon-number noise spectrum S_nn(ω)=n̄κ/(ω²+(κ/2)²) sampled at the Rabi frequency; the paper's own data show ≈3× (and ≈1.6× without drive) more dephasing than this model predicts, with the origin unidentified, so the projected 10⁻³-level performance requires removing that excess.","fun_headline_variants_meta":{"raw":{"variants":["Non-destructive leakage detection via Rabi drive","Drive-assisted readout flags leakage, preserves qubit","97% fidelity leakage check without qubit destruction","Heralded leakage signal while keeping quantum state intact","Rabi drive makes errors visible without breaking qubit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000251,"raw_usage":{"total_tokens":1404,"prompt_tokens":766,"completion_tokens":638,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":574}},"tokens_in":510,"tokens_out":638,"duration_ms":6503,"temperature":1.0,"reasoning_tokens":574,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T18:44:15.310765+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure Γ_1ρ^meas as a function of Rabi frequency Ω_q at fixed photon number n̄ and compare with the predicted Lorentzian χ² n̄ κ / [2(Ω_q²+(κ/2)²)]. If the extracted 'κ' from the decay curve disagrees with the independently measured resonator linewidth, or if the dephasing does not drop as 1/Ω_q² at large drive, the central suppression mechanism is wrong. A simpler check: with the Rabi drive off, the model predicts measurement-induced dephasing Γ_m = 2χ² n̄/κ; the paper finds ~1.6× this value, so eliminating that unexplained excess in a clean device would validate—or falsify—the quantitative","supporting_citations":[],"review_version":1}