An Archival Optical Counterpart Search for Extragalactic Fast X-Ray Transients Discovered by Einstein Probe

To ensure the completeness of our X-ray sample, we used a list of all high-confidence eFXT candidates (Q. Wu et al. 2026, in preparation) and unverified sources detected by EP between 2024 January 9 and 2025 June 20.

Generally, eFXTs are defined as non-Galactic sources that manifest as nonrepeating, soft X-ray flashes (D. Alp & J. Larsson 2020; J. Quirola-Vásquez et al. 2023). To be included in our high-confidence eFXT candidate sample, a source must satisfy the following criteria based on the method described by Q. Wu et al. (2026, in preparation):

1. 1.  
   High significance. The detection signal-to-noise ratio (S/N) must generally exceed 6.0 for ground-based triggers and 7.0 for onboard triggers.
2. 2.  
   No X-ray history. There must be no known X-ray source previously detected at the position.
3. 3.  
   Not a Galactic origin. The source must not be associated with any known Galactic object, e.g., X-ray binaries, stars.
4. 4.  
   Short-duration X-ray flash. The source must exhibit a distinct enhancement in X-ray fluence confined within a single orbit observation (approximately 3.6 ks).

In addition to high-confidence eFXT candidates, we also include unverified sources (subthreshold sources) in our analysis.³⁰ These events are labeled as Unverified during postprocessing, often due to their low S/N, the absence of clear flaring behavior in the light curve, or the lack of prior X-ray detections. Such characteristics suggest that they are likely genuine detections, although their physical nature remains uncertain. In general, unverified sources display realistic point-spread functions (PSFs) and reasonable spectral properties, but their classification is hindered by the relatively large WXT positional uncertainties, which stem from the limited data quality. We include them to avoid missing potential low-S/N eFXT candidates.

After applying these filtering criteria, our selected X-ray sample comprises 95 eFXT candidates and 5140 unverified sources. The sky distribution can be seen in Figure 1.

Zoom In Zoom Out Reset image size

We constructed an initial sample of optical transients by applying a filtering process to the Zwicky Transient Facility (ZTF) alert stream (M. T. Patterson et al. 2018), supplemented with objects reported in the Transient Name Server³¹ (TNS). To identify extragalactic optical transients, we utilized the Lasair broker system,³² which provides real-time access to ZTF alert data via structured SQL-based queries (R. D. Williams et al. 2024). As part of this process, we leveraged the Sherlock sky context package, an integrated database framework that enables rapid and reliable classification by spatial cross-matching with a wide range of astrophysical catalogs, facilitating efficient selection of transients of interest (D. Young 2023). The optical candidates should satisfy the following criteria:

1. 1.  
   The object is either cross-matched with a source in the TNS or has a Sherlock classification label as one of (NT, SN, ORPHAN). These labels correspond to the following categories: (1) transients that fall within the 1$\mathop{.}\limits^{\unicode{x02033}}$5 of the core of a resolved galaxy (NT), (2) transients that are not classified as NT but are located within a magnitude-, morphology-dependent radius or are physically separated from a galaxy (SN), and (3) transients that fail to match any known cataloged source (ORPHAN).
2. 2.  
   The object is >2″ away from star-like objects (${\mathtt{sgscore}}\gt 0.5$) in the Pan-STARRS (PS) catalog (K. C. Chambers et al. 2016).
3. 3.  
   The discovery date reported by TNS is later than 2024 January 9 (UTC), corresponding to the launch date of EP.
4. 4.  
   The earliest detection in ZTF is later than 2024 January 9 (UTC).

It is important to note that EP commenced routine observations on 2024 March 1, following the gradual activation of all 48 CMOS modules. Therefore, the selected optical transient sample fully encompasses the temporal window during which associations with X-ray sources could plausibly occur. This filtering strategy enabled the construction of a sample of extragalactic optical transients from the ZTF alert stream. After applying all selection criteria, we identified a total of 12,346 candidates, which form the basis for subsequent multiwavelength cross-matching and analysis. In addition, we incorporated transients reported to the TNS since the launch of EP, resulting in a supplementary sample of 33,048 objects. This sample partially overlaps with the ZTF-selected candidates, thereby improving the overall completeness of our transient catalog. The combined sky distribution of these samples is shown in Figure 1. There is a concentrated region for unverified sources at the Galactic plane between 15^h and 19^h, attributed to stellar flares and X-ray binaries.

With the compiled catalogs of X-ray and optical transients, we performed a cross-match to identify potential SN candidates exhibiting prompt X-ray emission. For each X-ray source, we identified optical transients residing within the positional uncertainty region of the WXT detection. A conservative matching radius of $r\le {3.5}^{{\prime} }$ was adopted to accommodate potential localization uncertainties, particularly given that the positional accuracy during the early commissioning phase of EP was approximately ${3.5}^{{\prime} }$ (W. Yuan et al. 2022). Subsequently, we applied a temporal constraint by selecting candidates within a time offset δt, defined as the time difference between the EP detection and the optical discovery. A positive δt indicates that the X-ray detection precedes the optical transient, aligning with the expected sequence for phenomena such as SN SBO. Considering the expected delay between the prompt X-ray emission and the emergence of optical SN signatures (S. Campana et al. 2006; A. M. Soderberg et al. 2008b; F. Förster et al. 2018; W.-X. Li et al. 2025b; H. Sun et al. 2025b), as well as the survey cadence and occasional gaps in optical coverage—which can cause the discovery time to lag behind the true explosion time t₀ by missing the early rise or peak—we adopt a conservative matching window of 0 < δt ≤ 30 days. This choice is motivated by the fact that core-collapse SNe typically show rest-frame rise times of t_r ≲ 20 days (S. J. Prentice et al. 2016; F. Taddia et al. 2019). For the maximum redshift (∼0.4) of ground-based EP SN discoveries, this corresponds to an observer-frame rise time of t_r ≲ 28 days, well within our adopted window.

It is important to note that the candidates identified through this process include not only genuine candidates but also known events (i.e., GRB afterglows), contaminations (i.e., Galactic sources), and chance coincidences. Therefore, a thorough vetting and multiwavelength analysis are essential for assessing the nature of each candidate.

After applying the cross-matching procedure described in Section 2, we identified 16 candidates that satisfy our selection criteria: nine high-confidence eFXTs candidates and seven unverified sources, as listed in Tables 1 and 2. Several candidates in our list have been extensively followed up and analyzed in the literature (H. Z. Wu et al. 2024; R. Eyles-Ferris et al. 2025; H. Hamidani et al. 2025; S. Jiang et al. 2025; W.-X. Li et al. 2025b; Y. Liu et al. 2025c; G. P. Srinivasaragavan et al. 2025a; H. Sun et al. 2025b; J. N. D van Dalen et al. 2025; Z. Zhu et al. 2025b). For the remaining events, we carried out case-by-case inspections to evaluate the likelihood of a genuine association. Candidates lacking supporting multiwavelength evidence—such as UV or radio detections—or those coincident with cataloged X-ray sources possessing sub-10″ localizations are attributed to chance coincidence.

We conclude that none of the unverified sources are of extragalactic origin: WXT J160346+193555 is likely associated with a cataclysmic variable (CV), while the remaining six candidates are consistent with temporal or spatial coincidences (see Appendix A), and therefore fall outside the scope of this study. In addition to the well-studied eFXTs, we highlight EP240506a, which appears to be an orphan event: despite extensive multiwavelength follow-up during the first few days, only upper limits were obtained. Remarkably, it is spatially and temporally associated with the optical transient AT 2024ofs, which was reported on TNS approximately 19 days after the EP trigger. We also identify prior UV emission at ∼2 days and early optical emission at ∼8 days post-trigger. Given this temporal offset, the chance-coincidence probability estimated in Section 4 is P_cc ∼ 3 × 10⁻⁴, strongly suggesting that the association is genuine rather than a random alignment. The WXT image and the position of AT 2024ofs are shown in Figure 2. Following a comprehensive analysis of archival data and the properties of its host galaxy, as detailed below, we confirm that EP240506a is associated with the SN candidate AT 2024ofs. This makes EP240506a another EP-detected eFXT linked to an SN.

Zoom In Zoom Out Reset image size

In this section, we present the observations and analysis of the newly discovered SN candidate EP240506a/AT 2024ofs.

3.2.1. X-Ray/γ-Ray Detection

EP240506a was discovered by WXT in the 0.5–4 keV band at 05:01:39 UTC on 2024 May 6, during the commissioning phase of EP (D. Y. Li et al. 2024). The WXT data were processed with wxtpipeline, an analysis chain of the WXT Data Analysis Software (WXTDAS) developed by the EP Science Center. The transient was detected with a significance of 7.8, which was below the onboard trigger³³ threshold of 8.0 at that time, and therefore no automatic follow-up observations were triggered; the latter are essential for capturing the rapid decay phase. Nevertheless, it was reported via the VHF Alert (W. Yuan et al. 2022), with a sky position of R.A. = 14^h 15^m 56$\mathop{.}\limits^{{\rm{s}}}$88, decl. = −16° 14′ 16$\mathop{.}\limits^{\unicode{x02033}}$8 (J2000), with position uncertainty of 3$\mathop{.}\limits^{{\rm{^{\prime} }}}$5.

The flare, however, was missed in the default telemetry data as it occurred shortly after the passage through the South Atlantic Anomaly (SAA) and was truncated. The data were subsequently recovered by rerunning the pipeline manually, yielding an event exhibiting a duration of ${T}_{90}=41.{1}_{-4.0}^{+2.0}\,{\rm{s}}$ and a peak flux of ∼1 × 10⁻⁸ erg cm⁻² s⁻¹ and total 36.8 ± 6.5 photons, as seen in the light curve shown in Figure 3. T₉₀ is defined as the time during which the central 90% of the fluence is observed. The derived X-ray peak luminosity is (2.3 ± 0.6) × 10⁴⁷ erg s⁻¹ in the WXT 0.5–4.0 keV band at z = 0.12 (see Section 3.2.4). The burst spectrum can be fitted by an absorbed power-law model with a photon index of Γ = 0.85 ± 0.44, assuming a fixed Galactic hydrogen column density of N_H = 9.24 × 10²⁰ cm⁻² (R. Willingale et al. 2013). The corresponding averaged unabsorbed flux is estimated to be (1.61 ± 0.41) × 10⁻⁹ erg cm⁻² s⁻¹ and corresponding averaged luminosity is (6.45 ± 1.65) × 10⁴⁶ erg s⁻¹. As shown in Figure 4, the X-ray luminosity of EP240506a significantly exceeds that of XRO 080109 and is comparable to GRB 060218, EP250108a, and EP240414a, implying the energy falls into the regime of relativistic SBO.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

A Swift follow-up observation commenced at 02:45:40 UTC on 2024 May 8, approximately 45 hr after the EP trigger, yielding a total exposure of 2.59 ks. No uncataloged X-ray source was detected within the 3$\mathop{.}\limits^{{\rm{^{\prime} }}}$5 EP/WXT error circle. Assuming the best-fit absorbed power-law model derived for the prompt emission, we estimate a limiting flux of 7.9 × 10⁻¹⁴ erg cm⁻² s⁻¹ in the 0.5–4 keV band, calculated using the HEASARCWebPIMMS tool (K. Mukai 1993).

Although the sky position of the transient was visible to Fermi-GBM, no significant signal was detected within the time interval T₀ ± 100 s. By combining the nondetection from Fermi-GBM down to ∼7 × 10⁻⁸ erg cm⁻² s⁻¹ (assuming α = −0.85 and β = −2.3) with the flux measured by WXT, and assuming the spectrum follows a Band function, we constrain the peak energy to E_peak < 140 keV and derive an isotropic-equivalent energy of E_iso < 9.3 × 10⁴⁹ erg, again adopting z = 0.12 (see Section 3.2.4). The derived upper limits on the peak energy and isotropic-equivalent energy fall within the region where type I and type II GRBs overlap on the Amati relation (L. Amati et al. 2002). However, these limits rule out a large portion of the parameter space typically occupied by type I GRBs, providing meaningful constraints despite the remaining ambiguity.

3.2.2. UV/Optical/Near-IR Observations

Analysis of the initial Swift/UVOT target-of-opportunity (ToO) data show a marginal UVM2-band excess at the position of the optical counterpart, measured at 22.57 ± 0.43 mag (S/N = 2.5) in a 903.8 s exposure (Figure 5(a)), with a 5σ limiting magnitude of >21.59 mag. This potential new detection, which has not previously been reported, lies near the sensitivity limit of the observation and may be affected by background fluctuations. To evaluate its reliability, we conducted a series of follow-up Swift ToO observations between 2025 May 20 and 2025 June 24. A stacked UVM2-band image constructed with uvotimsum (total exposure of 2181.5 s) achieved a 5σ limiting magnitude of >22.31 but revealed no significant emission at the position of AT 2024ofs (Figure 5(b)). The absence of persistent emission in the deeper template suggests that the earlier excess, if real, was transient in nature; however, given its low significance, the possibility of a spurious fluctuation cannot be excluded. Nonetheless, we adopt this measurement in our analysis owing to its temporal and spatial consistency with the optical counterpart.

Zoom In Zoom Out Reset image size

Despite extensive follow-up campaigns within the first 3 days after the EP trigger, no optical/IR counterpart was identified (A. Aryan et al. 2024, 2025; N. Chang et al. 2024; J. J. Jin et al. 2024; N. Pankov et al. 2024; I. Perez-Garcia et al. 2024; S. Tinyanont et al. 2024). However, nearly two weeks later, a new optical transient, AT 2024ofs/ZTF24aaowcoo, was independently discovered by ZTF and PS within the 3$\mathop{.}\limits^{{\rm{^{\prime} }}}$5 localization uncertainty of the WXT detection (K. C. Chambers et al. 2024). The transient is spatially coincident with the UVOT source and shows a fading trend in the ZTF forced-photometry light curve (see below; F. J. Masci et al. 2023).

From Figure 2(b), AT 2024ofs is offset by 5$\mathop{.}\limits^{\unicode{x02033}}$32 from the galaxy SDSS J141550.45−163937.0 (hereafter J141550.45), consistent with a location in the galaxy’s halo region. To further assess its origin, we revisited archival observations from multiple facilities—including ZTF, ATLAS, PS, GROND, Thai Robotic Telescope (TRT), and Small & Light Telescope (SLT), and the Xinglong 2.16 m telescope (XLT)—and compiled them to reconstruct the evolution of the optical light curve of AT 2024ofs. An even earlier signal was revealed in ATLAS forced photometry at 7.7 days after the WXT trigger.

To improve the photometric quality, we employed the Forced Photometry Service (F. J. Masci et al. 2023) for ZTF data between T₀ + 13 days (MJD 60450.27), when the transient initially emerged on the subtracted ZTF image, and T₀ + 28 days (MJD 60464.27), which significantly enhanced the accuracy of the light curves. For the ATLAS light curves, we used the publicly available ATClean³⁴ package (S. Rest et al. 2023, 2025) to retrieve the forced photometry from the ATLAS forced photometry³⁵ server (L. Shingles et al. 2021) and clean it following the recipe described by S. Rest et al. (2025). For PS data, photometric monitoring started on MJD 60460.327, 24.1 days after the discovery of EP240506a. The images were obtained in the w filter (J. L. Tonry et al. 2012). Images were processed with the Image Processing Pipeline (E. A. Magnier et al. 2020a; C. Z. Waters et al. 2020), astrometrically and photometrically calibrated (E. A. Magnier et al. 2020b), and individual frames were coadded with median clipping to produce stacks, on which PSF photometry was performed (E. A. Magnier et al. 2020c). In addition, we coadded PS images taken around T₀ + 35.8 days (MJD 60472), yielding a clear detection of the transient at w = 21.83 ± 0.15 mag using AutoPhOT. The source subsequently faded and was not detected in later epochs.

The sky position of AT 2024ofs was also observed as part of our J-band follow-up of EP240506a using the 7-channel imager GROND (J. Greiner et al. 2008) mounted on the MPG 2.2 m telescope at ESO’s La Silla Observatory. The observation was carried out on 2024 May 9 at UT 05:02, corresponding to 72 hr after the EP discovery. Image differencing was performed using the AutoPhOT pipeline (S. J. Brennan & M. Fraser 2022), with a J-band image from the VISTA Hemisphere Survey³⁶ obtained in 2012 May serving as the reference. No source was detected at the target position, with 5σ limiting Vega magnitudes of J > 19.1, H > 18.3, and K > 17.3.

The 40 cm SLT at the Lulin Observatory was utilized to obtain images in the Sloan Digital Sky Survey r filter as part of the Kinder collaboration (T.-W. Chen et al. 2025a). We coadded 30 × 300 s images from SLT, reaching a 3σ limiting magnitude of ∼22 in the r band.

The collected light curve is presented in Figure 6, which we contrast against the light curves of SN Ibn and SN Ibc templates, as well as the prototypical FBOT AT 2018cow and Ic-BL SN 2024gsa associated with EP240414a. The evolution of AT 2024ofs exhibits distinct features relative to these classes: it shows a longer rise time but a comparable postpeak decay rate to AT 2018cow and SNe Ibn, along with a lower peak luminosity. Compared to typical SNe Ibc (as well as SN 2024gsa), it evolves more rapidly, indicating that AT 2024ofs belongs to a class of intermediate-timescale transients rather than representing the extremes of either rapidly evolving or normal stripped-envelope SNe.

Zoom In Zoom Out Reset image size

3.2.3. Multiwavelength Fitting

The observed data were corrected for a line-of-sight Galactic extinction of E(B − V) = 0.106 mag (D. J. Schlegel et al. 1998). The UV detection was excluded from the fit because it deviates significantly from the model light curve (see Figure 7) as well as its low S/N, and because the fitting framework may not be applicable at such early epochs.

Zoom In Zoom Out Reset image size

We modeled the rest of the multiwavelength light curve of AT 2024ofs using the semianalytical Arnett model (W. D. Arnett 1982) as implemented in the Redback package³⁷ (N. Sarin et al. 2024). The model assumes homologous ejecta expansion, with heating powered by the radioactive decay of ⁵⁶Ni and ⁵⁶Co, while photon diffusion regulates the emergent luminosity (W. D. Arnett 1982). Bayesian inference was performed via nested sampling with dynesty³⁸ (J. S. Speagle 2020), using a likelihood function that incorporates both detections and nondetections. The analysis highlights the sparse temporal coverage—particularly the lack of photometry around peak luminosity—which produced poorly constrained posteriors. Although the model broadly reproduces the observed light curves (Figure 7), the inferred parameters drift toward unphysical regions of parameter space, as illustrated in Figure 10 in Appendix C.

For context, we compare our results with ejecta properties inferred for other SNe associated with eFXTs detected by EP (H. Sun et al. 2025b; W.-X. Li et al. 2025b), which indicate ⁵⁶Ni masses of ∼0.8 M_⊙ and relatively high M_Ni/M_ej ratios. Assuming this ⁵⁶Ni mass, the expected peak bolometric absolute magnitude is ${M}_{\mathrm{peak}}=-({\mathrm{log}}_{10}{M}_{\mathrm{Ni}}+8.184)/0.415=-19.5$, which is consistent with our observations (J. D. Lyman et al. 2016). Motivated by these studies, we fixed the ⁵⁶Ni mass to a plausible value and allowed other parameters to vary within reasonable ranges, ensuring that the model predictions remained broadly consistent with the limited data. While this approach yields a qualitative description of the event, it should not be regarded as unique or statistically robust.

The optical light curves are reasonably reproduced by a radioactive-decay-powered model, yielding an ejecta mass of M_ej ≈ 2.5 M_⊙, v_ej = 1.6 × 10⁴ km s⁻¹, and kinetic energy of ${E}_{{\rm{K}}}=3{v}_{{\rm{sc}}}^{2}{M}_{{\rm{ej}}}/10\approx 1.9\times 1{0}^{51}\,{\rm{erg}}$. The photospheric velocity at peak light, v_sc, and the peak bolometric luminosity, L_peak ≈ 2.3 × 10⁴³ erg s⁻¹, are inferred from the fitted parameters. These values are broadly consistent with those of similar events, although the adopted ⁵⁶Ni mass is higher than that of typical stripped-envelope SNe. However, it remains compatible with the yields of energetic Type Ic-BL SNe given the high peak luminosity, supporting the interpretation that the emission was primarily powered by ⁵⁶Ni and ⁵⁶Co decay (S. Campana et al. 2006; A. M. Soderberg et al. 2008a; J. D. Lyman et al. 2016; S. J. Prentice et al. 2016; F. Taddia et al. 2019; R.-Z. Li et al. 2025a; H. Sun et al. 2025b).

Nonetheless, these estimates remain subject to large uncertainties because of the sparse temporal sampling and the absence of spectroscopy for AT 2024ofs. Future events with denser photometric coverage and prompt spectroscopy will be crucial for placing robust constraints on explosion mechanisms and energy sources.

3.2.4. Host Galaxy

To determine the redshift of EP240506a/AT 2024ofs, we obtained a long-slit spectrum of its host galaxy with FORS2 on the ESO Very Large Telescope (VLT) (UT1) at Paranal on 2025 April 26 (Program ID: 115.287Q.002, PI: Ling-Zhi Wang). The data were processed with the ESO Reflex workflow (W. Freudling et al. 2013) using standard procedures (bias subtraction, flat-fielding, wavelength calibration, and sky subtraction). The flux calibration was performed by applying a scaling factor to match the photometric measurements, as the spectroscopic extraction encompassed only a fraction of the flux from this extended source. The host-galaxy spectrum exhibits a continuum with prominent emission features, including Hα and [N II], as shown in Figure 8. By fitting the emission lines, we derived a redshift of z = 0.120 ± 0.002. We then performed full spectral fitting using the method of N. Li et al. (2020). Prior to fitting, all detected emission lines were carefully masked following the procedure outlined by C. Li et al. (2005). Our fitting employs the simple stellar population (SSP) models of G. Bruzual & S. Charlot (2003), which comprise 1326 SSPs spanning 221 ages from 0 to 20 Gyr and six metallicities from Z = 0.005 Z_⊙ to Z = 2.5 Z_⊙. These models are based on the Padova evolutionary tracks (G. Bertelli et al. 1994) and a G. Chabrier (2003) stellar initial mass function. From this fitting, we derive a dust attenuation of the host galaxy of E(B − V) = 0.19.

Zoom In Zoom Out Reset image size

Assuming the cosmological parameters from the Planck Collaboration (2020), the luminosity distance and angular-diameter distance to the host galaxy are 578.5 and 461.2 Mpc, respectively. Given an angular offset of 5$\mathop{.}\limits^{\unicode{x02033}}$32, the projected physical separation between EP240506a/AT 2024ofs and the center of its host galaxy is estimated to be 11.9 kpc.

We also estimated the probability of a chance coincidence between AT 2024ofs and J141550.45. Using the apparent r-band magnitude of 18.2 mag from the Legacy Survey and the angular separation of 5$\mathop{.}\limits^{\unicode{x02033}}$32, we derived a chance-coincidence probability of P_ch = 4.8 × 10⁻³ following the method of J. S. Bloom et al. (2002). Such a low probability suggests that the association between the SN and the galaxy is unlikely to be a random alignment.

Our search for optical counterparts to eFXT candidates yields more than ten matches. However, the relatively large localization uncertainties of WXT imply that a substantial fraction of these associations are consistent with chance alignments, which dominate the contamination. Quantifying the rate of chance coincidences (i.e., false associations) is therefore essential for estimating the expected number of genuine counterparts as the sample size increases.

To evaluate the likelihood of an X-ray–optical match arising from chance coincidence (P_cc), we construct a mock sample of optical transients by resampling the discovery dates from the real optical sample in Section 2 and applying a random positional offset within an annulus with inner and outer radii of 3$\mathop{.}\limits^{{\rm{^{\prime} }}}$5 and 1^∘, to prevent true association in our simulation. This approach introduces spatial and temporal randomness while preserving the underlying sky and time distributions. We then performed one-dimensional Kolmogorov–Smirnov (KS) tests on the distributions of discovery date and sky position (t_dis, R.A., decl.) for the real and mock samples. The null hypothesis that the two samples are drawn from the same distribution could not be rejected, with p-values well above 0.05 (corresponding to a 90% confidence level).

We estimate P_cc via 200 Monte Carlo simulations (P. Munar-Adrover et al. 2011; G. S. Vance et al. 2021; Y.-F. Wang et al. 2022). In each simulation, we randomly select 500 X-ray sources (≈10% of the full EP eFXT candidate catalog) and cross-match them with the mock optical sample following the procedure described in Section 2. We define P_cc as the fraction of X-ray sources that yield at least one counterpart in the mock optical sample. By varying the temporal offset threshold δt, we derive P_cc(δt) for both the full eFXT candidate sample and the subset of high-confidence events, as shown in Figure 9.

Zoom In Zoom Out Reset image size

The lower P_cc inferred for the full eFXT candidate sample is primarily driven by the inclusion of X-ray sources located near the Galactic plane: because the mock optical transients are distributed predominantly at high Galactic latitude, the effective sky overlap is reduced, leading to a lower chance-coincidence rate (see Figure 1). Adopting δt = 30 days, the expected number of random matches is ∼5, fully consistent with the seven unrelated associations identified in this work (see Section 3.1 and Appendix A).

For the subset of high-confidence eFXTs, we estimate an expected number of random associations of only ∼0.1 at δt = 30 days. These events generally exhibit much shorter X-ray–optical time delays—typically within ten days—owing to rapid multiwavelength follow-up efforts. Such prompt observations substantially suppress the probability of spurious positional and temporal coincidences.

EP240315a was discovered by EP/WXT at 20:10:44 UTC on 2024 March 15, with a duration of T_90,X = 1034 s. The gamma-ray counterpart, GRB 240315C, was subsequently detected by Swift’s Burst Alert Telescope (BAT) and Konus-Wind 372 s after the initial WXT trigger. The optical counterpart, AT2024eju, was identified by ATLAS approximately 1.1 hr later. Spectroscopy established a redshift of z = 4.859, indicating that EP/WXT captured the soft X-ray prompt emission of a high-redshift GRB (J. H. Gillanders et al. 2024; A. J. Levan et al. 2024; Y. Liu et al. 2025c).

EP240414a was discovered by EP/WXT at 09:49:10 UTC on 2024 April 14, and no contemporaneous gamma-ray signal was detected. The optical counterpart, SN 2024gsa, was subsequently identified in the luminous host galaxy SDSS J124601.99−094309.3 and classified as (Ic-BL) SN. The light curve of SN 2024gsa exhibits a distinctive multibump evolution. The observed soft X-ray spectrum and luminosity at z = 0.401 suggest that EP240414a originated from the interaction of a weak relativistic jet with an extended envelope surrounding the progenitor star (H. Hamidani et al. 2025; H. Sun et al. 2025b; J. N. D. van Dalen et al. 2025).

EP240425a was discovered by EP/WXT at 20:50:33 UTC on 2024 April 25, with no prompt gamma-ray signal detected. A rapidly fading optical counterpart was subsequently identified, while the radio emission continued to brighten for ≳170 days. In the absence of a reliable redshift measurement, the physical origin of the transient cannot be established definitively. Nevertheless, the observed multiwavelength behavior is broadly consistent with expectations for a jetted tidal disruption event (Q. Zhao et al. 2026, in preparation).

EP240506a was triggered via the VHF EP/WXT alert channel at 05:01:39 UTC on 2024 May 6 during the commissioning phase of EP. However, as it did not activate the BeiDou alert system, no automated follow-up observations were initiated, resulting in a missed decay phase of the transient. The transient lasted for approximately 50 s, reaching a peak flux of 1 × 10⁻⁸ erg s⁻¹ cm⁻² (D. Y. Li et al. 2024). No counterpart was identified in the immediate follow-up observations (A. Aryan et al. 2024; N. Chang et al. 2024; J. J. Jin et al. 2024; N. Pankov et al. 2024; I. Perez-Garcia et al. 2024; S. Tinyanont et al. 2024). In this work, we find that the event is probably associated with an SN (AT 2024ofs). Spectroscopic observation of its host galaxy yields a redshift of z = 0.12. Further details of EP240506a are discussed in Section 3.2.

EP241030a was detected by EP/WXT at 06:33:18 UTC on 2024 October 30, with a duration of approximately 50 s. The event was found to be spatially and temporally coincident with GRB 241030A, and was subsequently identified as its X-ray afterglow counterpart (H. Z. Wu et al. 2024). The prompt gamma-ray emission was observed earlier by Fermi-GBM and Swift-BAT at 05:48:03 UTC on the same day, about 45 minutes before the EP detection (N. Klingler et al. 2024; Fermi GBM Team 2024). A bright optical counterpart was also detected by Swift-UVOT, and follow-up spectroscopic observations with Keck/LRIS measured a redshift of z = 1.411 (W. Zheng et al. 2024).

EP250108a was discovered by EP/WXT at 12:30:28.34 UTC on 2025 January 8, with no significant gamma-ray counterpart detected (R.-Z. Li et al. 2025a). The optical counterpart, SN 2025kg, was identified through follow-up observations, although no associated radio emission was observed (T. An et al. 2025; R. Eyles-Ferris 2025; R. A. J. Eyles-Ferris et al. 2025; L. Izzo 2025; G. Schroeder et al. 2025; Z. Zhu et al. 2025b; X. Zou et al. 2025). Subsequent spectroscopic analysis classified SN 2025kg as a Type Ic-BL SN, originating from the core collapse of a Wolf–Rayet star at redshift z = 0.176 (W.-X. Li et al. 2025b; D. Xu et al. 2025; Z. Zhu et al. 2025b). Further investigations suggest that the observed X-ray emission likely arose from the interaction between a mildly relativistic outflow and the surrounding circumstellar material (W.-X. Li et al. 2025b; G. P. Srinivasaragavan et al. 2025a; J.-P. Zhu et al. 2025a; Z. Zhu et al. 2025b).

EP250226a was triggered by EP/WXT at 06:35:16 UTC on 2025 February 26, followed by automatic FXT observations. The event was associated with GRB 250226A, which had been detected by Fermi-GBM, Konus-Wind, Swift-BAT, and GECAM-B. The EP X-ray detection occurred about 20 s after the Fermi-GBM trigger (Fermi GBM Team 2025; S. Jiang et al. 2025). An optical counterpart was subsequently identified using the VLT, and spectroscopic observations determined a redshift of z = 3.315 (Z. Zhu et al. 2025c).

EP250304a was discovered by EP/WXT at 01:32:30 UTC on 2025 March 4, with a duration of approximately 1200 s (W. Chen et al. 2025b; Y. Zhang et al. 2025). An optical counterpart was detected about 1.12 hr after the WXT trigger (X. Liu et al. 2025a). Subsequent spectroscopic observations with the VLT measured a redshift of z = 0.2 and classified the transient as a Type Ic-BL SN (L. Izzo et al. 2025; A. Saccardi et al. 2025a). This makes EP250304a the third SN associated with an eFXT detected by EP.

EP250427a was detected by EP/WXT at 03:38:45 UTC on 2025 April 27. The true onset time (T₀) remains unconstrained, as the satellite was within the SAA region prior to the trigger. The event is associated with a subthreshold burst, GRB 250427A, detected by Fermi-GBM (M. Ravasio et al. 2025; Y. Wang et al. 2025). An optical counterpart was detected 4 hr after trigger (X. Liu et al. 2025b; I. Perez-Garcia et al. 2025). Spectroscopic observations conducted with Keck/LRIS and VLT/X-Shooter confirmed a redshift of z = 1.52 (R. Chornock et al. 2025; A. Saccardi et al. 2025b).

EPW20240326aa was initially flagged as a transient due to a prominent flare lasting approximately 10⁴ s, accompanied by apparent periodic oscillations. However, this source was later determined to be spurious, likely caused by background fluctuations, as similar patterns were observed across other CMOS detectors. Consequently, it was reclassified as an unverified source attributed to the potential of instrumental background variations.

WXT J160346+19355 was identified as an unverified source, exhibiting a relatively steady X-ray flux over several days, followed by nondetections. There is a historical X-ray source 2CXO J160346.6+193540 in the Chandra Source Catalog. The position of the corresponding optical object, as reported by Gaia and suggested to be a brightening CV, is spatially consistent with the known CV CRTS CSS160906 J160346+193540 and 2CXO J160346.6+193540. We conclude that the X-ray emission and the optical counterpart might originate from the same source. The detections on those days were made possible by ultralong effective exposures of ∼2 × 10⁴ s, accumulated during the instrument calibration phase.

Based on the large position uncertainties of WXT and low signal-to-noise ratios, the remaining unverified sources are likely attributable to chance spatial and temporal coincidences, given the marginal significance of their X-ray detections and the absence of follow-up observations or any corroborating evidence supporting a physical association.