The Chooz reactor cores are pressurized water reactors (PWRs), each containing 205 fuel assemblies composed of approximately 600 kg of enriched uranium dioxide (UO2), primarily U238 with a few percent of U235. During operation, additional fissile isotopes, Pu239 and Pu241, are produced through neutron capture and subsequent decay processes involving U238. The fission of these four isotopes (U235, U238, Pu239, Pu241) accounts for more than 99.7% of the core’s thermal power. The reactors typically operate at full power for over a year in what is known as an irradiation cycle, followed by a refueling period during which the reactor is shut down for 6–8 weeks. During this period, about one third of the spent fuel assemblies are removed and transferred to storage pools located in an adjacent building, approximately 38 m from the reactor cores. Each assembly typically undergoes three irradiation cycles and reaches a burnup of approximately 45  GW·days/ton before being removed. For UO2 fuel with 4% enrichment, this burnup typically corresponds to the consumption of ∼75%–80% of the initial fissile material. The spent fuel assemblies are then cooled in storage pools for several years before being transported off site for reprocessing. Electron antineutrinos (ν¯e) are mainly produced by the β- decay of FPs and, to a lesser extent, by neutron capture reactions forming isotopes such as U239 and Np239. However, only ν¯e from FPs are detectable due to the 1.8 MeV threshold of the IBD reaction. Most fission-induced ν¯e are emitted promptly following fission events. During reactor operation, long-lived FPs gradually accumulate in the fuel. After shutdown, these isotopes continue to decay, producing a small residual ν¯e flux that decreases over time. Around a reactor, the intensity and spectral shape of this residual flux depend on both the irradiation history and the time elapsed since shutdown, with contributions from fuel assemblies still in the cores and in nearby cooling pools.

A two-detector setup was used to measure the θ13 neutrino mixing angle, with detectors placed at average distances of 400 m and 1.05 km from the reactor cores to observe neutrino oscillations [7]. The experiment operated from 2011 to 2017, with the far detector (FD) starting in 2011 and the near detector (ND) joining in 2014 to complete the two-detector setup. Both detectors, shielded underground by 115 (ND) and 300 m (FD) water equivalent, were identically designed to minimize systematic uncertainties and enable precise near and far comparisons of ν¯e rates and energy spectra [9]. The experimental layout, showing detector positions relative to the reactor cores and spent fuel pools, is illustrated in Fig. 1. Unlike multicore reactor experiments such as Daya Bay [10] and RENO [11], where simultaneous core shutdowns never occurred, the two-core configuration of the Chooz B plant enabled in situ background measurements during such events. This unique feature not only strengthened the original oscillation analysis [7,12] but also provides a rare opportunity to measure residual ν¯e emission. Each detector consisted of four concentric volumes topped by an outer muon veto system. The innermost volume (neutrino target, NT) was filled with 10.3  m3 of Gd-loaded (1  g/l) liquid scintillator, surrounded by a 22.6  m3 gamma catcher (GC) filled with unloaded scintillator to ensure full calorimetry of interactions associated with neutron capture on Gd. These two volumes, along with the 105 cm-thick buffer tank filled with non-scintillating mineral oil and instrumented with 390 low-background 10-inch photomultiplier tubes (PMTs), formed the inner detector (ID). The ID was surrounded by a 50 cm-thick liquid scintillator inner muon veto (IV) equipped with 78 8-inch PMTs. The far detector was shielded from rock radiation by 15 cm of demagnetized steel, while the near detector was shielded by 1 m of water. An outer muon veto (OV), consisting of segmented scintillator modules positioned above the detector, provided additional rejection of cosmic muons. Reactor ν¯e were detected via the IBD process, primarily occurring in the NT and GC. At full reactor power, the near and far detectors recorded about 900 and 140 IBD events/day, respectively [7].

Between 2011 and 2017, the Double Chooz experiment recorded rare reactor-off periods when both reactor cores were simultaneously offline. This scenario was uncommon due to the plant’s alternating refueling schedules. A first analysis of such conditions, based on 7.5 days of reactor-off data collected in 2011–2012 with only the far detector, validated the predicted background model for the oscillation analysis [13]. However, the limited duration of that period and the absence of near-detector data resulted in only ∼20 IBD candidates, preventing any extraction of a residual reactor ν¯e spectrum. In the present Letter, we exploit a much larger dataset collected in 2017, during four reactor-off intervals in which both cores were simultaneously shut down for refueling or maintenance. As shown in Fig. 2, these intervals totaled 24.4 days (1.6, 1.1, 1.0, and 20.8 days), providing the statistics needed for a quantitative residual-flux measurement. After accounting for muon veto–induced dead time, the detector live time amounted to 17.2 days for the near detector and 22.2 days for the far detector.

The analysis was performed for both near and far detectors; however, we focus primarily on the near detector in the following due to its closer proximity to the reactor cores and spent fuel pools, offering higher sensitivity to this low-rate signal.

The IBD detection relies on the characteristic time and spatial correlation between a prompt signal, produced by the positron’s ionization (e+) and annihilation, and a delayed signal from neutron (n) capture. The positron’s energy deposition directly correlates with the reactor ν¯e energy, with Eν¯e≃Ee++0.78  MeV. IBD event selection follows the same criteria as the oscillation analysis in [7], employing the total neutron capture (TnC) method, which, in addition to gadolinium, includes neutron captures on carbon and hydrogen across all ID volumes. After applying vetoes based on the ID, IV and OV to reduce muon-induced backgrounds, IBD candidates are selected within a 0.5–800  μs time window and a spatial separation of 0–1.2 m between the prompt and delayed signals. The energy windows for the prompt and delayed signals are set to [1, 20] MeV and [1.3, 10] MeV, respectively. A high-efficiency artificial neural network (ANN) was used to further suppress accidental background contamination by analyzing the time, spatial, and energy correlations of the delayed signals. As detailed in Refs. [7,14], the classifier is a multilayer-perceptron with one hidden layer of nine nodes, trained on off-time accidental pairs from data and IBD Monte Carlo events. It is applied after the loose IBD preselection and used as a multivariate classifier on this sample, with the final selection defined by a cut on its discriminant output. Two types of backgrounds mimic the ν¯e signal: correlated and accidental. Due to the relatively shallow detector overburden, correlated backgrounds are dominated by muon-induced processes. These arise primarily from fast neutrons generated by muon interactions in the surrounding rock and from β-n decays of isotopes such as Li9, produced by muon spallation on C12. The rate of β-n emitters is reduced by tagging their temporal and spatial correlations with the parent muon. FN rates and energy spectra are estimated using energy depositions tagged by the IV and OV systems up to 20 MeV. Contributions from other sources, such as stopping muons, β-n decays of He8, and β- decay of B12, are found to be negligible. Accidental backgrounds, caused by random coincidences of natural radioactivity and neutron captures, are strongly suppressed by the ANN, with a rejection power exceeding 400. The residual accidental rate is measured in situ using off-time windows within the ID, following Refs. [7,14]. These selections, applied in the 1–9 MeV energy range, yield 517 IBD candidates in the near detector, corresponding to a rate of 30.0±1.4 events/day, prior to background substraction. In the far detector, 189 IBD candidates are observed, yielding 8.5±0.7 events/day. The estimated background rates in the same energy window are 26.2±1.4 and 7.7±0.2 events/day, respectively. A summary of the residual ν¯e rates, after background subtraction, is provided in Table I. Further details on the detector properties, calibration, data selection, and background estimation are provided in [7].

A detailed summation method model of the residual ν¯e rate and energy spectrum emitted from both the reactor cores and the spent fuel pools has been developed for the four reactor-off periods following an approach similar to that in [15,16]. This model relies on coupling FP activity predictions with a database of ν¯e spectra from β decay.

The FP activities are computed using the APOLLO-2.8.4 and DARWIN-3 simulation codes [17,18], based on the JEFF-3.1.1 nuclear data library [19]. These widely used and validated tools for neutron transport and residual power calculations simulate the time evolution of the fuel’s isotopic composition and activity during both irradiation phases and cooling periods.

A similar procedure was used to simulate each fuel assembly, whether located in the reactor core or stored in the spent fuel pool during shutdown periods. This involved simulating the individual irradiation history of each assembly up to the beginning of each reactor-off period, including a detailed follow-up of the thermal power it experienced throughout its lifetime as well as any cooling periods between irradiation cycles and after its removal to the spent-fuel pool. An optimized, high-resolution temporal discretization was used to accurately model the buildup and decay of both short- and long-lived fission product isotopes. Residual ν¯e spectra are obtained by coupling the predicted FP activities with the BESTIOLE library [20]. BESTIOLE models ν¯e emission by summing individual β branches using an advanced formalism based on Fermi theory, including electromagnetic corrections, finite nuclear size, atomic screening, and shape factors for both allowed and forbidden transitions. Particular care is taken in modeling first forbidden nonunique transitions, several of which, such as Pr144 a major contributor to the residual ν¯e flux below 3 MeV, incorporate detailed nuclear structure calculations. The IBD spectrum in each detector is obtained by integrating the time-dependent flux over the duration of the reactor-off periods and summing over all sources: dNν¯edEvis=∑iNp4πLi2∬ϕi(E,t)σIBD(E)⁢Pee(E,Li)D(Evis|E)dE dt,(1)where i runs over the two reactor cores and the two spent fuel pools, Np is the number of target protons, Li is the baseline to the source i, ϕi(E,t) is the time-dependent energy spectrum of the emitted ν¯e, σIBD(E) is the inverse β-decay cross section, Pee(E,Li) is the survival probability for baseline Li, and D(Evis|E) is the detector response, which folds energy scale, resolution, and selection efficiency to map the true antineutrino energy E onto the reconstructed (visible) energy Evis. The IBD cross section formalism of [21], using neutron lifetime value from [22] is used. For this calculation, the oscillation parameter sin2(2θ13)=0.105±0.014 from the latest Double Chooz analysis is used [7], yielding mean survival probabilities of 0.978±0.003 for the near detector and 0.909±0.012 for the far detector.

To first order, the rate and energy spectrum of residual ν¯e emitted by a burnt fuel assembly are primarily determined by the burnup achieved before shutdown, which sets the inventory of accumulated fission products. As shown in Fig. 3 for a representative UO2 assembly irradiated to 45 GWd/t, the predicted IBD ν¯e spectrum exhibits a rapid decline in both intensity and mean energy following reactor shutdown. As short-lived isotopes decay, the IBD mean cross section per fission (MCSPF) decreases by 1, 2, and 3 orders of magnitude after ∼10  min, ∼15 h, and ∼2.5  y, respectively. Over the same period, the mean ν¯e energy drops from ∼4.2  MeV to ∼3.2  MeV, and eventually to ∼2.7  MeV. Immediately after shutdown, the flux results from the superposition of hundreds of β branches from many fission products. Within a few hours, however, it becomes dominated by a small number of longer-lived isotopes. After several months, the primary contributors are the short-lived isotopes Y90 (T1/2=3.19 h, Qβ=2.28  MeV), Rh106 (30.1 s, 3.54 MeV), and Pr144 (17.3 min, 3.00 MeV), whose long-lived precursors have accumulated in the fuel: Ce144–Pr144 (T1/2=285 d), Ru106–Rh106 (372 d), and Sr90–Y90 (28.9 y). Beyond ten years, Y90 alone accounts for more than 90% of the residual flux.

Burnt fuel assemblies remaining in the reactor core during refueling or maintenance are a significant source of residual ν¯e in the early stages of a reactor-off period, i.e., within the first tens of hours after shutdown, when the flux is strongly influenced by short-lived isotopes. In contrast, spent fuel stored in cooling pools contributes a long-lived component to the residual flux. Unlike burnt fuel, this contribution arises from the cumulative emission of assemblies removed over multiple fuel cycles. Some have been cooling for years, while others were removed more recently, resulting in a broad distribution of cooling times and activity levels that shapes both the intensity and spectrum of the emitted ν¯e.

Figure 4 shows the predicted residual ν¯e IBD spectrum for the Double Chooz near detector, including contributions from both reactor cores and spent fuel pools. The predicted IBD spectrum extends up to approximately 4.5 MeV with 98.7% of the residual flux expected below 3 MeV. An energy-integrated contribution of 56% from the reactors and of 44% from the storage pools is obtained. While residual ν¯e emission is initially dominated by burnt fuel still in the reactors due to the presence of short-lived isotopes, the long-term emission from spent fuel pools, integrating multiple assemblies with varying cooling times, becomes comparable over extended reactor-off periods. The B1 pool (yellow area) contributes less than B2 (red area) because the longest reactor-off period coincided with the refueling of reactor B2. During this time, a new batch of spent fuel was added to the B2 pool (red line), increasing its relative contribution to the total residual ν¯e flux. Further details on the time evolution of the individual residual-signal components are given in the End Matter [23]. The total normalization uncertainty on the predicted residual ν¯e signal in the near detector is 7.4%, dominated by a 6.0% uncertainty in antineutrino spectrum modeling. This uncertainty arises primarily from nuclear-structure uncertainties affecting the dominant first-forbidden transitions of Pr144. Conservatively, the 1σ uncertainty on the Pr144 spectral shape was defined as the difference between the shape factor obtained from the detailed nuclear-structure calculations and that obtained using the simplified ξ approximation, which assumes an allowed transition. Other significant contributions come from geometric baseline uncertainties (2.9%), the simulated fission product inventory (2.1%), and the estimated amount of spent fuel assemblies in the pools (2.0%). The latter includes a conservative treatment of older assemblies whose status is uncertain, by assigning a 100% uncertainty to their contribution. Detector-related systematics are subdominant, contributing less than 0.8% in total.

Measured and predicted residual ν¯e rates are summarized in Table I for different energy ranges. In the near detector, a significant excess of IBD events is observed in the 1–3 MeV window, where the residual signal is expected to peak, as clearly shown in Fig. 5. In this range, 106±18 residual candidates are observed, in good agreement with the prediction of 88±7 (Δ≡data-prediction=18±19), both in rate and in spectral shape. The measurement uncertainty is dominated by statistical fluctuations in the observed IBD candidates, with a total of 244 events recorded in the 1–3 MeV range, before background subtraction. For the combined 17.2 days of reactor-off livetime in the near detector, this corresponds to an averaged predicted residual ν¯e flux at the detector of Φres≃3.6×108  cm-2 s-1 (spent-fuel pools: 1.6×108  cm-2 s-1, reactor cores: 2.0×108  cm-2 s-1) for antineutrino energies above the 1.8 MeV IBD threshold. In this energy range, the residual background is dominated by 86±8.3 correlated events, with 52.1±0.1 accidental events after ANN suppression. The flux in this region is expected to be dominated by Pr144 (∼54%) and Rh106 (∼38%), with minor isotopes accounting for the remainder. In the far detector, a smaller excess of 27±13 events is observed in the same energy range, compared to the predicted 14±1. Although this measurement is also in good agreement with the prediction, the larger baseline suppresses both the event rate and the statistical significance relative to the near detector. Above 3 MeV, no significant excess is observed in either detector. In this region, the background-subtracted yields are consistent with zero (see Table I), validating the background model and confirming that the small high-energy residual tail, predicted at a rate below 0.1 events/day, lies well below the current sensitivity.

Over the past two decades, the IAEA safeguards community and the neutrino-physics community have jointly explored how reactor antineutrinos could be used for nonintrusive monitoring of nuclear facilities [2]. Several reactor-on demonstrations have been performed, but the specific case of spent-fuel verification had, until now, remained untested experimentally [24].

The present Letter shows that dedicated reactor-off operation of the Double Chooz near detector delivers sufficient statistics (17.2 days) and low backgrounds to extract, for the first time, a quantitative residual reactor antineutrino spectrum. In combination with a spent-fuel simulation, this provides the first direct validation of residual-flux predictions, including contributions from both reactor cores and spent-fuel pools. In the 1–3 MeV range, where the residual neutrino signal is strongest, the observed 106±18 events represent a 5.9σ excess over background, in very good agreement with the predicted 88±7 events. The significance is obtained directly from event counting, using Z=Nres/σres. The measurement uncertainty is of the order of one half of the predicted spent-fuel-pool contribution to the residual signal. It therefore indicates that the present measurement is sensitive to the existence of this residual signal and to large changes in the overall inventory, but not to the diversion of a small number of assemblies.

This result establishes a robust experimental benchmark for residual-reactor and spent-fuel neutrino monitoring at baselines of a few hundred meters. From a safeguards perspective, it turns earlier conceptual studies into an experimental proof of principle: neutrinos can be used to monitor reactor conditions during shutdown and to obtain information on nearby spent fuel in a direct and nonintrusive way. On this basis, realistic performance targets can now be set for future, more portable and lower-cost detectors. The most natural use cases are compact fuel-storage configurations, such as spent-fuel pools or small dry-cask pads holding a limited number of assemblies, where neutrino-based nonintrusive and nondestructive assay would provide a quantitative, independent complement to today’s largely accounting-based verification, albeit at the price of sizable detector systems and infrastructure [25]. Extending neutrino monitoring to large, deep geological repositories remains a distant prospect given the impractical detector masses, background suppression, and lifetimes required [24].