2026 ED138: A Large Trans-Neptunian Object in Early Rubin Observatory Data, and How Many More May Remain

Introduction

The largest trans-Neptunian objects (TNOs), the dwarf planets and their near relatives, formed in the disk that also formed the giant planets. Their number and orbits constrain how the outer solar system formed and how Neptune migrated. They are also the easiest TNOs to find. The wide-field searches of recent decades [40, 41, 76, 13, 65, 69 D. Rabinowitz et al.] 2012; M. E. Brown et al. [15] found most of them in searches over much of the sky. Most of those searches avoided the crowded star fields near the Galactic plane, however, and reached at most $R \approx21.6$ (§ 9.1), about two magnitudes brighter than the Rubin search that found the object reported here (M. J. Holman et al. [30]). The Legacy Survey of Space and Time (LSST; Ž. Ivezić et al. [32]) at the NSF-DOE Vera C. Rubin Observatory (Rubin) will increase the number of known TNOs several-fold (M. E. Schwamb et al. [66]; J. A. Kurlander et al. [39]), and it observes those fields routinely.

This paper reports a large TNO, 2026 ED138 (PN01109 in the search), found in early Rubin data by the search pipeline described by M. J. Holman et al. [30], which found 2,440 previously unknown TNOs; 2026 ED138 has the lowest absolute magnitude of them. The search uses pointdexter software (M. J. Holman & K. J. Napier, in preparation) running as a part of Rubin Solar System Pipelines (M. Jurić et al., in preparation) that reads Rubin’s single-visit source catalogs and groups detections from different nights that belong to one slow-moving distant object. The search reported 2026 ED138 as a candidate.

This paper has three aims:

1. to describe the independent checks that show 2026 ED138 is real and new, and to measure its orbit, size and colors from every Rubin and archival detection;

2. to understand why it had not been found before, including why Rubin’s own catalogs missed it in most of the visits whose images show it;

3. to estimate how many more TNOs with $H_V < 4$ may remain undiscovered, and where.

§ 2 describes the Rubin data, the initial identification, and our orbit fitting. § 3–§ 5 describe the checks, and § 6–§ 8 give the orbit, colors and size, and whether 2026 ED138 can have kept volatile ices. § 9 explains why earlier surveys and Rubin’s single-visit catalogs missed it. Section 10 uses the search’s recoveries of the known TNOs with $H_V < 4$ to estimate how many such objects remain, and § 11 summarizes.

Data

Rubin data

All data are from Rubin’s 2026 ED138 data release.¹ They consist of images taken with LSSTCam (SLAC National Accelerator Laboratory & NSF-DOE Vera C. Rubin Observatory [70]) at Rubin Observatory’s Simonyi Survey Telescope (MPC code X05) between 2025-05-20 and 2026-07-12. We use Rubin’s calibrated single-visit images (NSF-DOE Vera C. Rubin Observatory [45]) and two source catalogs measured on them: AP-S, from Daytime Alert Production (Daytime AP; NSF-DOE Vera C. Rubin Observatory [46]), which detects sources at $5\sigma$ and gives a position error for each, and NV-S, from the nightly validation pipeline, which gives no position errors. Both are catalogs of detections in direct images. None of the visits of the observations that the search reported, which we call the delivered observations (§ 2.2), has difference-image sources.

Both catalogs are made automatically from the calibrated images, by two different runs of the LSST Science Pipelines (Vera C. Rubin Observatory Science Pipelines Developers [78]): AP-S by Daytime Alert Production, which reprocesses each night’s exposures the following day, and NV-S by the nightly validation run, which has been utilized to provide quick reductions for situational awareness and has since been retired. Neither was tuned for crowded fields, so both miss many detections of 2026 ED138 (§ 9.2). Where a catalog has no entry for an object, we measure it ourselves on the calibrated image (§ 4).

Positions are calibrated to Gaia DR3 (Gaia Collaboration et al. [26]). For each AP-S position the delivered observations give the catalog error plus 0.015″ in quadrature; we report the catalog error plus 0.010″, the convention of Rubin’s submissions to the MPC. NV-S gives no position errors, and we leave those uncertainties empty. The orbit fits use the observations as submitted to the MPC, weighted by these uncertainties and their RA–Dec correlations (Appendix § 15.§ 2). Appendix § 15.§ 1 gives further details of the products.

The initial identification

2026 ED138 was first detected by pointdexter as a cluster of 10 observations on 8 nights, spanning 35 days from 2026-05-01 to 2026-06-05 in the ugrizy bands. These delivered observations were the starting point for orbital arc extension and all further studies. The median $i$ band magnitude of 19.4 pointed to the lowest absolute magnitude of the new objects found by the search of early operations Rubin data. The object lies far south, at declination about $-43^\circ$ and ecliptic latitude about $-20^\circ$, $12^\circ$ south of the Galactic mid-plane (§ 9.1). Table 1 summarizes the observations of 2026 ED138 by data set and band, and Table 11 lists every position used in this paper.

Data set	$u$	$g$	$r$	$i$	$z$	$y$	$VR$	Total	Nights	First	Last	Arc (d)
Delivered	2	2	$\cdots$	1	4	1	$\cdots$	10	8	2026-05-01	2026-06-05	35
Rubin, all	3	3	4	6	9	5	$\cdots$	30	20	2026-03-04	2026-07-11	129
DECam	$\cdots$	7	6	6	7	6	12	44	14	2016-02-25	2023-04-07	2598
Adopted fit	3	10	10	12	16	11	12	74	34	2016-02-25	2026-07-11	3789

Time (UTC)	R.A.	Decl.	$\sigma_{\alpha}$	$\sigma_{\delta}$	mag	$\sigma_m$	Band	Source	Origin
(deg)	(deg)	(″)	(″)
2016-02-25 07:49:48.7	260.3187519	-44.7826230	0.0519	0.0518	19.751	0.0166	VR	DECam	D
2016-02-25 07:51:31.7	260.3187745	-44.7826351	0.0537	0.0530	19.769	0.0161	VR	DECam	D
2016-02-25 07:53:14.8	260.3187903	-44.7826461	0.0530	0.0525	19.740	0.0184	VR	DECam	D
2016-02-25 07:54:59.0	260.3187980	-44.7826574	0.0526	0.0522	19.730	0.0144	VR	DECam	D
2016-08-11 02:37:59.6	257.8336367	-44.9161799	0.0515	0.0516	19.386	0.0189	i	DECam	D
2016-08-11 02:38:58.5	257.8336257	-44.9161732	0.0521	0.0521	19.261	0.0268	z	DECam	D
2016-08-11 02:39:57.7	257.8336280	-44.9161708	0.0571	0.0546	19.217	0.1875	Y	DECam	D
2016-08-24 03:50:21.4	257.7360115	-44.8086511	0.0516	0.0515	20.272	0.0151	g	DECam	D
2016-08-24 03:51:52.3	257.7359814	-44.8086388	0.0516	0.0515	19.649	0.0153	r	DECam	D
2016-08-24 03:58:56.8	257.7359532	-44.8085991	0.0520	0.0519	19.674	0.0165	r	DECam	D
2016-08-24 04:00:29.4	257.7359842	-44.8085970	0.0523	0.0519	20.253	0.0163	g	DECam	D
2016-08-25 04:28:49.8	257.7317654	-44.8000678	0.0533	0.0524	20.236	0.0217	g	DECam	D
2016-08-26 04:37:40.3	257.7280846	-44.7916420	0.0514	0.0520	19.636	0.0147	r	DECam	D
2016-08-26 04:39:11.4	257.7281019	-44.7916334	0.0507	0.0575	20.281	0.0166	g	DECam	D
2017-05-04 05:57:10.3	261.5597919	-45.0599636	0.1078	0.1780	19.129	0.0546	Y	DECam	D
2017-05-04 05:58:08.8	261.5597833	-45.0599611	0.0703	0.1948	19.207	0.0219	z	DECam	D
2017-05-04 05:59:07.4	261.5597653	-45.0599684	0.0523	0.0678	19.293	0.0152	i	DECam	D
2017-05-04 06:06:01.6	261.5596721	-45.0599819	0.0522	0.0656	19.296	0.0149	i	DECam	D
2017-05-04 06:06:59.6	261.5596548	-45.0599803	0.0698	0.1984	19.189	0.0246	z	DECam	D
2017-05-04 06:07:58.4	261.5596465	-45.0599880	0.1032	0.1641	19.037	0.0511	Y	DECam	D
2017-05-04 07:44:47.9	261.5583237	-45.0602510	0.0516	0.0516	19.596	0.0201	r	DECam	D
2017-05-04 07:46:18.8	261.5582925	-45.0602640	0.0513	0.0513	20.213	0.0208	g	DECam	D
2017-05-04 07:54:31.1	261.5581905	-45.0602605	0.0512	0.0513	20.211	0.0173	g	DECam	D
2017-05-04 07:56:02.4	261.5581644	-45.0602680	0.0518	0.0514	19.572	0.0206	r	DECam	D
2017-07-02 01:47:44.6	260.1349406	-45.0432077	0.0512	0.0513	19.559	0.0107	VR	DECam	D
2017-07-02 01:49:27.8	260.1349088	-45.0432022	0.0512	0.0512	19.562	0.0105	VR	DECam	D
2017-07-02 01:51:11.4	260.1348795	-45.0431965	0.0512	0.0512	19.555	0.0103	VR	DECam	D
2017-07-02 01:52:56.0	260.1348521	-45.0431889	0.0513	0.0513	19.575	0.0124	VR	DECam	D
2017-07-03 02:14:18.0	260.1104213	-45.0387656	0.0511	0.0511	19.611	0.0099	VR	DECam	D
2017-07-03 02:16:00.5	260.1103923	-45.0387557	0.0511	0.0512	19.611	0.0088	VR	DECam	D
2017-07-03 02:17:44.0	260.1103632	-45.0387480	0.0511	0.0511	19.633	0.0092	VR	DECam	D
2017-07-03 02:19:27.9	260.1103360	-45.0387467	0.0512	0.0512	19.614	0.0110	VR	DECam	D
2018-05-12 09:30:18.5	262.9473439	-44.9080449	0.0520	0.0519	19.197	0.0280	z	DECam	D
2018-05-12 09:31:18.1	262.9473351	-44.9080449	0.0517	0.0516	19.322	0.0180	i	DECam	D
2018-05-19 08:16:41.7	262.7943405	-44.9255037	0.0515	0.0515	20.221	0.0157	g	DECam	D
2018-05-19 08:18:13.3	262.7943040	-44.9255042	0.0526	0.0517	19.551	0.0193	r	DECam	D
2019-04-28 10:04:53.7	264.7398643	-44.6549333	0.0710	0.0883	19.122	0.0766	Y	DECam	D
2019-04-30 08:19:09.2	264.7089166	-44.6639005	0.0564	0.0528	19.150	0.0576	Y	DECam	D
2019-04-30 08:20:07.2	264.7088969	-44.6639087	0.0778	0.0516	19.255	0.0215	z	DECam	D
2019-04-30 08:21:06.6	264.7088858	-44.6639064	0.0569	0.0528	19.342	0.0165	i	DECam	D
2019-05-01 09:13:48.8	264.6916101	-44.6685837	0.0627	0.0615	19.149	0.0590	Y	DECam	D
2019-05-01 09:14:47.9	264.6916117	-44.6685946	0.2069	0.1194	19.248	0.0348	z	DECam	D
2019-05-01 09:15:46.6	264.6916290	-44.6685956	0.1980	0.0851	19.361	0.0266	i	DECam	D
Time (UTC)	R.A.	Decl.	$\sigma_{\alpha}$	$\sigma_{\delta}$	mag	$\sigma_m$	Band	Source	Origin
	(deg)	(deg)	(″)	(″)
2023-04-07 06:11:02.9	270.8309540	-43.5573462	0.0592	0.0551	19.414	0.0416	z	DECam	D
2026-03-04 09:33:14.7	274.7906291	-42.4543647	0.0131	0.0130	19.319	0.0218	y	local fit	R
2026-03-04 09:35:07.4	274.7906470	-42.4543812	0.0127	0.0134	19.286	0.0221	y	local fit	R
2026-03-31 08:28:06.4	275.0026189	-42.6096994	0.0115	0.0115	19.414	0.0142	z	local fit	R
2026-03-31 09:00:12.6	275.0026441	-42.6098348	0.0108	0.0106	19.411	0.0134	z	local fit	R
2026-04-16 07:56:43.5	274.9631508	-42.7055538	0.0368	0.0336	21.941	0.0565	u	local fit	R
2026-05-01 06:13:30.8	274.8202383	-42.7885683	0.0105	0.0104	19.328	0.0127	z	local fit	R
2026-05-02 06:03:12.7	274.8074060	-42.7936882	0.0111	0.0112	19.350	0.0146	z	local fit	R
2026-05-12 10:00:13.8	274.6542322	-42.8423077	0.0104	0.0104	19.450	0.0119	i	local fit	R
2026-05-13 08:44:09.8	274.6380746	-42.8464499	0.0104	0.0105	19.285	0.0121	z	AP-S	P
2026-05-13 08:44:49.6	274.6380717	-42.8464509	0.0106	0.0104	19.281	0.0120	z	AP-S	P
2026-05-17 09:15:23.4	274.5661101	-42.8631783	0.0453	0.0446	21.920	0.0571	u	AP-S	P
2026-05-18 09:19:25.0	274.5473351	-42.8671390	0.0264	0.0221	21.774	0.0360	u	AP-S	P
2026-05-22 08:59:35.7	274.4698391	-42.8819353	0.0109	0.0107	20.265	0.0120	g	AP-S	P
2026-05-22 09:00:22.4	274.4698273	-42.8819401	0.0110	0.0109	20.230	0.0123	g	AP-S	P
2026-05-23 07:29:07.2	274.4509893	-42.8852067	0.0112	0.0113	19.580	0.0123	r	local fit	R
2026-05-23 08:10:34.5	274.4504135	-42.8852918	0.0139	0.0140	19.368	0.0145	i	AP-S	P
2026-05-25 08:11:42.9	274.4093250	-42.8919531	0.0108	0.0108	19.432	0.0123	i	local fit	R
2026-05-25 08:16:21.4	274.4092572	-42.8919625	0.0110	0.0110	19.677	0.0125	r	local fit	R
2026-05-27 05:48:28.0	274.3694115	-42.8979098	0.0112	0.0110	19.248	0.0135	z	local fit	R
2026-05-27 05:49:20.7	274.3694008	-42.8979164	0.0118	0.0119	19.259	0.0139	z	local fit	R
2026-05-30 04:31:31.1	274.3058211	-42.9064128	0.0177	0.0192	19.059	0.0330	y	AP-S	C
2026-05-31 05:09:15.4	274.2831986	-42.9091496	0.0128	0.0131	19.266	0.0186	z	local fit	R
2026-06-05 04:09:17.7	274.1710311	-42.9207871	·	·	19.236	0.0254	y	NV-S	C
2026-06-05 04:09:59.8	274.1710207	-42.9207888	0.0157	0.0154	19.234	0.0200	y	AP-S	P
2026-06-12 07:03:52.1	274.0027490	-42.9326467	0.0104	0.0105	19.373	0.0115	i	local fit	R
2026-06-27 05:23:48.3	273.6353771	-42.9376950	0.0103	0.0103	19.310	0.0114	i	AP-S	C
2026-07-09 04:04:54.3	273.3429974	-42.9219206	0.0103	0.0102	19.396	0.0113	i	AP-S	C
2026-07-09 04:49:07.0	273.3422495	-42.9218574	0.0104	0.0104	19.635	0.0124	r	AP-S	C
2026-07-11 06:43:24.1	273.2926341	-42.9173097	0.0110	0.0112	19.680	0.0120	r	local fit	R
2026-07-11 07:21:14.5	273.2920220	-42.9172443	0.0154	0.0165	20.358	0.0178	g	local fit	R

Orbit fitting

We iteratively refitted the orbit of 2026 ED138, beginning with the delivered observations and then each time we added detections in the arc extension process (§ 4–§ 5), first with Rubin and then with DECam observations. Two fits recur in the paper: the Rubin-only fit, to all the Rubin observations, and the adopted fit, which adds precovery detections in images from the Dark Energy Camera (DECam; [22]). We fit with our own least-squares code (Levenberg–Marquardt), which propagates orbits with ASSIST ([29]), built on REBOUND ([58]) and its IAS15 integrator ([59]), with the DE440 planetary ephemeris ([48]). The model includes the 16 most massive asteroids as perturbers and the light-travel time, and computes each observation from its own observing site. The state is parametrized in topocentric spherical coordinates at a mid-arc epoch ([9]); we quote $1\sigma$ and $3\sigma$ ranges from $\chi^2$ profiles along the poorly constrained radial velocity ($\Delta\chi^2 = 1$ and $9$) and derive posterior probabilities by statistical ranging ([79]). We weight the observations by their reported uncertainties; repeating every fit with a $0.1^{\prime\prime}$ floor added gives the same conclusions (Appendix § 15.§ 2).

Validation

We checked that 2026 ED138 is real and that it was not already known. No observation in the MPC database lies within $2'$ and $\pm1$ h of any reported detection, no known orbit passes near it, and no tracklet in the MPC’s Isolated Tracklet File (ITF), which holds observations not linked to any object, matches its track (Appendix § 15.§ 3). Every reported row matches a source in the Rubin catalogs, with signal-to-noise ratios (S/N) of 19.3–163, and Figure 1 shows 2026 ED138 in single-visit images, absent from the same sky on another night (Appendix § 15.§ 4).

Additional detections in Rubin data

We call any Rubin detection that is not among the delivered observations an additional detection, and keep precovery for the archival DECam detections (§ 5). Searching the catalogs of every LSSTCam visit near the predicted track added 5 additional detections of 2026 ED138, which extend its arc to 69 days (Appendix § 15.§ 5, Table 4). In most other visits whose images show 2026 ED138, the catalogs are incomplete around it (§ 9.2).

Class	Meaning	$N$	Img.	Add.	Rej.
Delivered	delivered observation	10	10	0+1r	0+1d
Detected	cataloged detection found by the search	5	4	0	0
On a detector, no cataloged detection:
Incomplete catalog	catalog incomplete around the position	18	14	12	2
Edge	near a detector edge	9	5	1	4
Masked	hole in the catalog at the position	3	3	3	0
Blended	cataloged source of the wrong brightness	2	2	0	2
Shallow	image too shallow for the object	0	0	0	0
Missed	not cataloged; no other explanation	0	0	0	0
No catalog	neither catalog covers the position	4	0	0	0
Off every detector:
Not covered	predicted position off every detector	12	0	0	0
Total		63	38	16+1r	8+1d
Final observations	delivered + detected + added − dropped	30

We then measured 2026 ED138 directly at its predicted positions with a local PSF fit (Appendix § 15.§ 6): the point-spread function (PSF) model that the pipeline stored with the image, placed at the object and at each detected neighbor, and a background plane, fitted to a small cutout of the calibrated image, with no catalog and no deblending step. Where a clean catalog detection exists, the fit agrees with it to 2.6 mas rms. It adds 16 measurements of 2026 ED138, the earliest from 2026-03-04, in visits where no catalog has the object (Figure 2). After inspection, we replaced one cataloged row of 2026 ED138 with its local PSF fit and dropped another (Appendix § 16). The complete Rubin data set has 30 observations of 2026 ED138 on 20 nights, from 2026-03-04 to 2026-07-11.

Vector (PDF)

Precovery in DECam images

The Rubin data span only 129 days. Earlier detections of 2026 ED138 would constrain its orbit far better, and at an $i$-band magnitude of about 19.4, 2026 ED138 is within reach of several earlier surveys of the southern sky.

We first searched with the precovery service of the Asteroid Institute’s ADAM platform ([38]), which returns the cataloged sources near an orbit’s predictions in the surveys it indexes. Given the Rubin-only orbit, it returned candidates from DECam, in the NOIRLab Source Catalog (NSC DR2; [43]), and from SkyMapper ([36]) and ATLAS (J. L. Tonry et al. 2018)). After we checked each DECam candidate in the images, 11 observations on 5 nights link to the Rubin orbit; the earliest lay $1.8^{\prime\prime}$ from the Rubin-only prediction, whose formal uncertainty there was $66^{\prime\prime}$. With the orbit refit to include these observations, we searched every DECam image in the NOIRLab archive that contains the ephemeris, measuring the image at that position rather than relying on the catalogs. Following this process, we find that 2026 ED138 is detected in 45 exposures on 14 nights, from 2016-02-25 to 2024-04-07 (Figure 3).

We tied each position locally to the Gaia DR3 stars ([26]) in a cutout around 2026 ED138, fitted it with an empirical PSF together with its known static neighbors, and corrected the bias from uncataloged faint stars, which we measured by injecting 2026 ED138 into images of the same sky taken on other nights. We kept 44 detections on 14 nights, with a median uncertainty of 52 mas per axis. With each night left out in turn, the fit predicts that night’s positions with an rms of 37 and 22 mas in RA and Dec. Timing errors are far smaller. Differential chromatic refraction is 4 mas in the median and exceeds 20 mas in only 4 exposures, reaching 40 mas in $g$ at high airmass (Appendices § 18 and § 19). The positions are given in Table 11.

Orbits

Table 2 gives the elements of the adopted fit (§ 2.3), heliocentric and barycentric. We quote and classify with barycentric elements, which change little between the different epochs of the fits; heliocentric elements oscillate with the Sun’s motion about the barycenter. Appendix § 15.§ 7 shows how the orbit evolved as detections were added (Table 5, Figure 6) and the residuals of the Rubin observations (Figure 7).

2026 ED138
Heliocentric
Epoch (MJD TDB)	59337.8171
$a$ (au)	$43.269536^{+0.000055}_{-0.000059}$
$e$	$0.1339372^{+0.0000075}_{-0.0000075}$
$q$ (au)	$37.47414^{+0.00027}_{-0.00028}$
$i$ (deg)	$24.3846266^{+0.0000018}_{-0.0000018}$
$\Omega$ (deg)	$322.797458 \pm0.000029$
$\omega$ (deg)	$169.4550 \pm0.0029$
$M$ (deg)	$120.4534 \pm0.0041$
Barycentric
$a$ (au)	$43.090975^{+0.000056}_{-0.000060}$
$e$	$0.1387280^{+0.0000074}_{-0.0000074}$
$q$ (au)	$37.11305^{+0.00027}_{-0.00027}$
$i$ (deg)	$24.4338512^{+0.0000018}_{-0.0000018}$
$\Omega$ (deg)	$322.606057 \pm0.000029$
$\omega$ (deg)	$168.6998 \pm0.0028$
$M$ (deg)	$121.0797 \pm0.0040$
Fit
$r$ at epoch (au)	46.726
$\Delta$ at epoch (au)	46.052
Observations	74
Arc (d)	3789.0
$\chi^2/\mathrm{dof}$	84.72/142
Dynamics
12:7 resonant fraction	0.46
libration amplitude (deg)	95–115
Tisserand parameter $T_N$	2.835
Proper inclination (deg)	25.7

Data	Weights	$N$	Arc (d)	$\chi^2$/dof	$a$ (au)	$a$ ($1\sigma$) (au)	$a$ ($3\sigma$) (au)	$e$	$q$ (au)	$i$ (deg)
Delivered	reported	10	35	17.8/14	43.63	43.37–43.93	43.0–44.7	0.167	36.34	24.416
Delivered	floor	10	35	0.5/14	43.51	42.55–46.54	42.5–64.3	0.159	36.61	24.419
+ cataloged	reported	15	69	21.7/24	43.18	43.15–43.22	43.1–43.3	0.142	37.06	24.427
+ cataloged	floor	15	69	0.7/24	43.16	42.91–43.48	42.6–44.3	0.141	37.10	24.427
+ local fits	reported	30	129	65.5/54	43.09	43.08–43.10	43.06–43.12	0.139	37.11	24.434
+ local fits	floor	30	129	2.2/54	43.09	43.02–43.17	42.9–43.3	0.139	37.11	24.434
+ DECam	reported	74	3789	84.7/142	43.090975	43.090915–43.091032	43.09080–43.09115	0.1387280	37.11305	24.4338512
+ DECam	floor	74	3789	6.2/142	43.09139	43.09066–43.09211	43.0892–43.0936	0.138729	37.113350	24.4338493

Vector (PDF)

Vector (PDF)

The adopted fit of 2026 ED138 spans 3789 days, from 2016-02-25 to 2026-07-11, with 74 observations; no residual exceeds 3.8 $\sigma$. It gives barycentric $a = 43.0910$ au (1$\sigma$ range 43.0909–43.0910 au). With the Rubin data alone (129 days) the 1$\sigma$ range was 43.08–43.10 au, and on the delivered observations the 3$\sigma$ range was 43.0–44.7 au (Figure 6). The DECam detections fit the Rubin-only orbit’s predictions (Appendix § 18.§ 1), and the two fits agree. With barycentric $a = 43.0910$ au, $i = 24.4339^\circ$ and $q = 37.1130$ au, 2026 ED138 lies near the 12:7 mean-motion resonance with Neptune. We integrated the nominal orbit and clones drawn from its covariance, 201 particles in all, for 10 Myr. The model has the Sun, with the mass of the inner planets added to it, and the four giant planets. We work in the invariable plane, the plane perpendicular to the total angular momentum of the solar system, and measure the elements from the system’s center of mass. In this integration the resonant angle $\phi= 12\lambda- 7\lambda_N - 3\varpi- 2\Omega$, where $\lambda$ and $\varpi$ are the mean longitude and the longitude of perihelion, $\Omega$ is the longitude of the ascending node and $\lambda_N$ is the mean longitude of Neptune, librates about $180^\circ$ with a large amplitude ($95$–$115^\circ$, the median over 200-kyr windows for the resonant particles), so the orbit lies at the edge of the resonance. By the criteria of T. Khain et al. [37], 93 of the 201 particles are resonant over 10 Myr and the rest are not. That fraction, 0.46, cannot be distinguished from the 50% at which T. Khain et al. [37] call an object a resonant candidate, and it is far below the 80% for a secure class. In earlier integrations with 31 particles, it ranged from about two fifths to about three quarters between models and between refits of the orbit that differ by a hundredth of its uncertainty. About half of the orbits that fit the data are therefore resonant. The orbit is now so well determined that the clones differ only by chaotic spreading at the resonance’s edge, so a longer arc will not settle the class. We tested all 28 arguments of this fifth-order resonance allowed by the d’Alembert rules, including those that involve Neptune’s longitude of perihelion $\varpi_N$ and node $\Omega_N$. Only two librate: $\phi$, and $12\lambda- 7\lambda_N - \varpi_N - 4\Omega$ in 10 of the same particles. No particle librates in any argument without also librating in $\phi$.

The orbit of 2026 ED138 is close to that of the collisional family of the dwarf planet Haumea ([14, 57, 53, 52]), and Haumea is in the 12:7 resonance. Family members share two properties ([57]): a surface of nearly pure water ice, with neutral visible colors ([14, 63, 71]), and an orbit that the collision could have produced. The color of 2026 ED138 rules out membership: with $g-r = 0.63 \pm0.07$ (§ 8.1), it is redder than the Sun ($g-r = 0.41$ in Rubin’s bands; [30]) by about three times its uncertainty, whereas Haumea, whose visible reflectance is neutral ([55]), and the confirmed members have nearly solar colors. The orbit is consistent with this, though on its own it would not decide membership. We integrated Haumea and its 9 members confirmed by their spectra for 10 Myr together with 2026 ED138, and computed proper inclinations (inclinations with the periodic perturbations by the planets averaged out) relative to the invariable plane. Reaching the orbit of 2026 ED138 from the center of the family (proper $a = 43.10$ au, $e = 0.118$, $i = 28.2^\circ$, as computed by D. Ragozzine & M. E. Brown [57] with their own method) needs an ejection velocity of at least 214 m s$^{-1}$, against 21–142 m s$^{-1}$ for the members and about 150 m s$^{-1}$ for the family as a whole ([57]); Haumea itself, which has drifted in the resonance, needs 390 m s$^{-1}$. Because of such drift, the family is also compared through the Tisserand parameter with respect to Neptune, a combination of $a$, $e$ and $i$ that such drift nearly conserves: it is 2.835 for 2026 ED138, above the family’s 2.798–2.820 (Haumea 2.796). The proper inclination of 2026 ED138, 25.7°, is lower than Haumea’s (26.6°) and than that of every member (26.8–28.1°), and its mean eccentricity, 0.139, is just above theirs (0.087–0.136). 2026 ED138 is therefore not a family member. It is one of the unrelated objects expected in this part of the belt [57]. A near-infrared spectrum without water-ice absorption would add direct evidence.

Photometry

Rubin

We use the Rubin magnitudes of 2026 ED138 from the catalogs and the local PSF fits (§ 4), as submitted to the MPC. Unlike the DECam magnitudes (§ 7.2), these are measured on direct images, where unresolved stars could bias them. The local PSF fits model each detected neighbor and agree with the catalog magnitudes to 0.014 mag rms where both exist (Appendix § 15.§ 6), and the one catalog row blended with a static star involves a star of $i \approx22.2$, against $i \approx19.4$ for 2026 ED138 (Appendix § 15.§ 4). We compute reduced magnitudes $m-5\log_{10}(r\Delta)$, where $r$ and $\Delta$ are the distances from the Sun and from the observer in au, from the adopted orbit geometry. At the small phase angles $\alpha$ (the Sun–object–observer angle) of a distant object, the reduced magnitude grows almost linearly with $\alpha$, so we model it as $H+\beta\alpha$, where the phase slope $\beta$ is in mag deg$^{-1}$ and $H$ is the absolute magnitude, at zero phase. The phase angles of 2026 ED138 span 0.42–1.21°, enough to measure $\beta$: a fit of one $\beta$ and one $H$ per band to the 27 grizy measurements, each weighted with 0.03 mag of rotational variation added in quadrature, gives $\beta= 0.13 \pm0.05$ mag deg$^{-1}$ ($\chi^2 = 49.4$ for 21 degrees of freedom; the error is scaled by $\sqrt{\chi^2/\nu}$; Figure 4). This is typical of TNOs [56, 3]. Phase slopes can differ between bands [4, 2], which a single $\beta$ does not capture; over the small phase angles of our observations the effect on the colors is small. We reduce 2026 ED138 to zero phase with it, which lowers its $H_V$ from 3.27 at the mean observed phase to 3.19 at zero phase.

Vector (PDF)

**Note—**Rubin and DECam observations (“+ DECam” in Table 5). Ecliptic J2000 osculating elements at each fit’s epoch; ASSIST with DE440 and sb441-n16. Heliocentric elements use the Sun’s mass; barycentric elements use the mass of the Sun, the planets, the Moon and Pluto (DE440), about the solar-system barycenter. Uncertainties on $a$, $e$, $q$, $i$ are the $\Delta\chi^2 = 1$ profile range; on $\Omega$, $\omega$, $M$ the linearized $1\sigma$ from the fit covariance. The barycentric state vector and its covariance are in the data release. Dynamics: from 10 Myr integrations of the nominal orbit and clones drawn from its covariance (§ 6); the resonant fraction is that of the particles resonant over the integration by the criteria of T. Khain et al. [37], and the libration amplitude is the median over 200-kyr windows, for the resonant particles; the proper inclination is relative to the invariable plane.

Rubin’s calibration in the visits of 2026 ED138 agrees with The Monster, its reference catalog (P. S. Ferguson et al. [19]), to within 0.007 mag in every band; against synthetic photometry from the low-resolution Gaia DR3 XP spectra (Gaia XP; Gaia Collaboration et al. [25]) its $g$ differs by $-0.020$ mag, which we treat as a systematic uncertainty (Appendix § 17).

DECam

The DECam magnitudes of 2026 ED138 are PSF fits on difference images, in all 44 exposures: each exposure minus PSF-matched templates of the same sky from other nights (Appendix § 17.§ 2). The difference removes the static stars of this crowded field, cataloged or not; at its own position 2026 ED138 is blended with stars that bias a direct PSF fit by up to 0.16 mag in single exposures. We tied each exposure to Gaia XP synthetic photometry of stars in the same cutout (Appendix § 17; Table 6). We do not use the NSC DR2 zero points: most reference stars in these fields are reddened, which biases those zero points by up to 0.17 mag. Reduced to zero phase and converted to the LSST bandpasses, the DECam magnitudes of 2026 ED138 differ from Rubin’s by $-0.00$, $+0.03$, $-0.00$, $+0.04$ and $-0.00$ mag in $g$, $r$, $i$, $z$ and $Y$ (Table 7). A joint fit of the Rubin and DECam magnitudes, with one $\beta$ and one DECam offset, gives $\beta= 0.14 \pm0.03$ mag deg$^{-1}$, consistent with the Rubin value of § 7.1, and an offset of $+0.01 \pm0.01$ mag.

Band	$N$	Nights	$m_{\mathrm{XP}}$	$m_{\mathrm{NSC}}$	XP $-$ NSC	$H_{\mathrm{XP}}$	$H_{\mathrm{XP}}^{\mathrm{LSST}}$	$H_{\mathrm{Rubin}}$	$\Delta H$
g	7	5	20.24	20.33	$-0.09$	3.48	3.50	3.50	$+0.00$
r	6	4	19.62	19.79	$-0.17$	2.85	2.90	2.87	$+0.03$
i	6	5	19.33	19.42	$-0.09$	2.58	2.61	2.61	$+0.00$
z	7	6	19.25	19.37	$-0.12$	2.48	2.53	2.49	$+0.04$
$Y^{a}$	6	5	19.14	19.20	$-0.08$	2.38	2.40	2.40	$+0.00$
VR	12	3	19.61	19.63	$-0.02$	2.92	$\cdots$	$\cdots$	$\cdots$

Band	$N$	NSC, $\alpha_{\mathrm{obs}}$	Phase	NSC, $\alpha= 0$	Calib.	XP, $\alpha= 0$
g	7	$+0.11$	$-0.07$	$+0.04$	$-0.04$	$+0.00$
r	6	$+0.21$	$-0.05$	$+0.16$	$-0.13$	$+0.03$
i	6	$+0.09$	$-0.03$	$+0.06$	$-0.06$	$+0.00$
z	7	$+0.12$	$+0.00$	$+0.12$	$-0.08$	$+0.04$
$Y^{a}$	6	$+0.04$	$-0.02$	$+0.03$	$-0.03$	$+0.00$

Physical properties

Table 3 collects the physical properties of 2026 ED138 derived in this and the previous sections. **Note—**Absolute magnitudes are at zero phase, with the measured phase slope; colors are differences of them. $H_V$ is from $g$ and $i$ with the relation of M. N. Porter et al. [49]; the MPC-style range applies the MPC’s fixed band-to-$V$ offsets, without a color term, to each band. Diameters are $D = 1329\ \mathrm{km}\,p_V^{-1/2}10^{-H_V/5}$. Temperatures and volatile retention are for an isothermal body with the albedo range shown (Figure 5).

Quantity	Value	Section
Photometry, zero phase
$H_u$ (mag)	5.05 ± 0.07	8.1
$H_g$ (mag)	3.50 ± 0.05	8.1
$H_r$ (mag)	2.87 ± 0.04	8.1
$H_i$ (mag)	2.61 ± 0.04	8.1
$H_z$ (mag)	2.49 ± 0.05	8.1
$H_y$ (mag)	2.40 ± 0.06	8.1
$H_V$ (mag)	3.19 ± 0.04	8.1
$H_V$, MPC band offsets (mag)	2.76–3.22	8.1
$u-g$ (mag)	1.55 ± 0.09	8.1
$g-r$ (mag)	0.63 ± 0.07	8.1
$r-i$ (mag)	0.26 ± 0.05	8.1
$i-z$ (mag)	0.12 ± 0.06	8.1
$z-y$ (mag)	0.08 ± 0.07	8.1
Phase slope $\beta$ (mag deg$^{-1}$)3	0.14 ± 0.03	7.2
Phase angles (deg)	0.42–1.21	7.1
Median $i$, 2026 (mag)	19.4	2.2
Rotational variation	not measured4	7.1
Size
$D$, $p_V = 0.25$ (km)	612	8.2
$D$, $p_V = 0.10$ (km)	968	8.2
$D$, $p_V = 0.04$ (km)	1530	8.2
Albedo, rotation period, mass	not measured	···
Temperature and volatiles, $p_V = 0.25$–0.03
Perihelion temperature (K)	44.0–45.6	8.3
Equivalent temperature (K)	41.8–44.1	8.3
$\mathrm{CH}_4$	lost at every $p_V$	8.3
$\mathrm{CO}$	kept if $p_V < 0.037$	8.3
$\mathrm{N}_2$	kept if $p_V < 0.033$	8.3

Colors and absolute magnitude

The absolute magnitudes below are at zero phase (§ 7.1). Per band, we average the reduced magnitudes: $H_g = 3.50 \pm0.05$, $H_r = 2.87 \pm0.04$, $H_i = 2.61 \pm0.04$, $H_z = 2.49 \pm0.05$ and $H_y = 2.40 \pm0.06$. The errors are standard errors of the mean, from the scatter within each band, and include the uncertainty of $\beta$. 2026 ED138 is redder than the Sun: $g-r = 0.63 \pm0.07$ (0.68 from a same-night pair).

We convert to $V$ with the Rubin relation of M. N. Porter et al. [49], $V = g - 0.302\,(g-i) - 0.041$, fitted to stars in the bands of ComCam, Rubin’s commissioning camera; its rms scatter is 0.022 mag. Propagating the $g$ and $i$ errors and adding that scatter in quadrature gives $H_V = 3.19 \pm0.04$. These errors do not include the difference between the ComCam and LSSTCam bandpasses or the photometric zero-point uncertainty; Rubin’s $g$ is $-0.020$ mag offset from Gaia XP in the field of 2026 ED138 (§ 7.1), which would change $g-r$ by as much. Applying a transformation fitted to stars to the spectrum of a TNO is itself uncertain: for bright TNOs with published colors, M. J. Holman et al. [30] find Rubin’s $g-r$ about a tenth of a magnitude bluer than the literature values transformed to Rubin’s bands. If that holds here, 2026 ED138 is redder still, and the color argument of § 6 is stronger. The Minor Planet Center computes $H_V$ differently: it converts each band to $V$ with a fixed offset (for example, +0.37 mag for $z$), with no color term.⁵ For an object as red as 2026 ED138, that makes the red bands too bright in $V$. Applied to the $H$ values above, the offsets give $H_V$ from 2.76 (from $y$) to 3.22 (from $g$). An $H_V$ computed mostly from the $i$, $z$ and $y$ bands, as the MPC’s is likely to be, will therefore be brighter than ours.

Size and albedo

With $D = 1329\ \mathrm{km}\,p_V^{-1/2}10^{-H_V/5}$ (P. Pravec & A. W. Harris [50]), geometric albedos of 0.25, 0.10 and 0.04 give diameters of 612, 968, and 1530 km. Measured TNO albedos span this range (E. Lellouch et al. [42]), but they depend on size: the largest TNOs tend to have higher albedos than smaller ones (M. E. Brown [13]; E. Lellouch et al. [42]). For an object with an absolute magnitude as low as that of 2026 ED138, an albedo at the low end of the range is therefore less likely, though low albedos do occur among large TNOs: Salacia, about one magnitude fainter in absolute magnitude than 2026 ED138, has a geometric albedo of only about 0.04 (S. Fornasier et al. [23]). Several large TNOs whose sizes have been measured by stellar occultation or thermal emission have albedos near 0.1: Quaoar 0.109 $\pm$ 0.007 (F. Braga-Ribas et al. [12]), 2002 MS4 0.10 $\pm$ 0.025 (F. L. Rommel et al. [60]) and Gonggong 0.089 (A. Pál et al. 2016); at that albedo 2026 ED138 would be about 968 km across. Albedos can also be much higher: Pluto, Eris and Makemake, whose surfaces are rich in CH$_4$ ice, have geometric albedos of about 0.6 to 0.9 (M. E. Brown [13]). With such an albedo, 2026 ED138 would be only about 320–400 km across. This is less likely, but its surface composition is still unknown. At that size it could not have kept CH$_4$ (§ 8.3), so so bright a surface would need CH$_4$ supplied from its interior. A diameter between 612 and 968 km (albedos between 0.25 and 0.10) is more probable. Even the smaller of these sizes is above the size at which G. Tancredi & S. Favre [72] expect icy bodies to be close to hydrostatic equilibrium, about 450 km, so 2026 ED138 is a dwarf-planet candidate. Size is only the first of the tests of G. Tancredi & S. Favre [72]; a lightcurve or a shape measurement is needed to apply the others. Whether 2026 ED138 is compact and round is less certain: W. M. Grundy et al. [27] suggest that icy bodies of 400–1000 km are the transition between small, porous objects and larger ones that have collapsed, so the answer depends on the albedo. 2026 ED138 is one of the ten intrinsically brightest known TNOs (Section 10). Other large TNOs found in recent years, such as (532037) Chiminigagua (2013 FY27; S. S. Sheppard et al. [68]), 2017 OF201 (S. Cheng et al. [17]), 2010 JO179 (M. J. Holman et al. [31]) and 2015 RR245 (M. T. Bannister et al. [6]), are comparable or fainter. Thermal photometry or a stellar occultation is needed to decide between these sizes.

Volatile retention

Whether 2026 ED138 can have kept volatile ices such as CH$_4$, N$_2$ and CO on its surface depends on its size and temperature. E. L. Schaller & M. E. Brown [62] estimated this with a simple model: each ice is in equilibrium with its vapor, and that thin atmosphere escapes by Jeans escape at a rate set by the vapor pressure and the body’s gravity. They divided each orbit into short segments, added up the loss over the age of the solar system, and compared it with a primordial, cometary abundance of the volatile. A body retains a volatile only if it is large and cold enough. To place a TNO on the diagram, they converted its total loss into an equivalent temperature: the temperature of an identical body on a circular orbit that would lose the same amount. Most of the loss happens near perihelion, so the equivalent temperature is close to the perihelion temperature, and lower for more eccentric orbits. S. Protopapa et al. [51] updated this simple model of Jeans escape for surface-bounded atmospheres over 4.6 Gyr, with new vapor pressures (N. Fray & B. Schmitt 2009; W. M. Grundy et al. [28]) and revised sizes and albedos of the known TNOs. Figure 5 shows their result, with 2026 ED138 added.

We place 2026 ED138 on the diagram in the same way. We take it to be an isothermal body, $T = 278.6\,\mathrm{K}\,(1-A)^{1/4}(r/1\,\mathrm{au})^{-1/2}$, with Bond albedo $A = p_Vq$ and phase integral $q = 0.336p_V + 0.479$ (M. J. Brucker et al. [16]), and a bulk density of $1.8\,\mathrm{g\,cm^{-3}}$. For each of $\mathrm{CH_4}$, $\mathrm{N_2}$ and $\mathrm{CO}$ we average its Jeans escape rate over the orbit, with the vapor pressures of W. M. Grundy et al. [28], and find the equivalent temperature. For geometric albedos from 0.25 to 0.03 ($D = 612$–1767 km), the perihelion temperature of 2026 ED138 is 44.0–45.6 K and its equivalent temperature 41.8–44.1 K. The three species give equivalent temperatures within 0.2 K of each other.

At a typical albedo, 2026 ED138 should have lost these volatiles. If its geometric albedo is 0.04 or higher ($D \le1530$ km), it lies above the curves at which Jeans escape removes all of its $\mathrm{CH_4}$, $\mathrm{N_2}$ and $\mathrm{CO}$ (Figure 5), as do the other TNOs of its size and temperature (see also J. P. Emery et al. [18]). Only a very dark, and so larger, 2026 ED138 could have kept some. Below $p_V = 0.037$ ($D > 1590$ km) it would keep some CO, and below 0.031 ($D > 1750$ km) it would lose less than 1% of it. Below $p_V = 0.033$ ($D > 1680$ km) it would also keep some $\mathrm{N_2}$. Albedos this low would be unusual for so large a TNO; even Salacia is brighter. $\mathrm{CH_4}$ is lighter and escapes more easily, and 2026 ED138 loses all of it down to $p_V = 0.03$, the darkest albedo we consider. The less volatile $\mathrm{C_2H_6}$ should remain.

This model assumes simple Jeans escape of a primordial, cometary abundance of volatiles. On large TNOs the loss and processing of volatiles are likely more complex. Hydrodynamic escape or a giant impact would have removed more than Jeans escape alone (S. Protopapa et al. [51]). On the other hand, volatiles can be resupplied from the interior: the light hydrocarbons on Sedna, Gonggong and Quaoar suggest that their $\mathrm{CH_4}$ is resupplied (J. P. Emery et al. [18]). A near-infrared spectrum of 2026 ED138 would test this. If 2026 ED138 is dark, it lies in the transition region of Figure 5, where it should keep some volatiles but not others. Finding which ices are present would show whether processes other than escape shape its surface.

Why 2026 ED138 was missed

At $V \approx20$ ($i \approx19.4$), 2026 ED138 is about two magnitudes brighter than the depth of the wide-field searches of the past two decades, $R \approx21.4$–21.6 (S. S. Sheppard et al. 2011; {citet}D. Rabinowitz et al. 2012}), which together are thought to have found nearly all TNOs this bright over most of the sky ({citet}S. S. Sheppard et al. 2011; {citet}M. E. Schwamb et al. 2014`). Why, then, was it still unknown?

Earlier surveys

It is at declination about $-43^\circ$, ecliptic latitude about $-20^\circ$ and Galactic latitude about $-12^\circ$. It was even closer to the plane during the earlier searches: it crossed the plane in 2011 and lay within $10^\circ$ of it from 2000 to 2023. The searches from the northern hemisphere, Spacewatch (J. A. Larsen et al. [40, 41]) and the Palomar surveys (C. A. Trujillo & M. E. Brown [76]; M. E. Brown [13]; M. E. Schwamb et al. [65]), did not reach this far south, and the survey of M. E. Schwamb et al. [65] also avoided fields within $15^\circ$ of the Galactic plane. The southern and Galactic-plane survey of S. S. Sheppard et al. [69] reached $R = 21.6$, deep enough, but in 2010 2026 ED138 lay at the southern edge of the area it covered, within about a degree of its fields (their Figure 1). The La Silla–QUEST survey (D. Rabinowitz et al. [54]) searched about 7,500 square degrees of the southern sky to $R = 21.4$, but during that survey 2026 ED138 lay within a few degrees of the Galactic plane. The all-sky search of M. E. Brown et al. [15] covered only Galactic latitudes above $20^\circ$, and in the south it was nearly complete only for $V \lesssim18.6$. The Dark Energy Survey, the largest earlier search of the southern sky for TNOs (P. H. Bernardinelli et al. [8]), stayed about $19^\circ$ from 2026 ED138 throughout (M. J. Holman et al. [30]); Pan-STARRS cannot observe this far south, and the solar-system objects in Gaia’s data releases are objects already known (P. Tanga et al. [73]). Even the Rubin search that found 2026 ED138 excluded the region within $10^\circ$ of the Galactic plane, and 2026 ED138 lies only about $2^\circ$ outside it (M. J. Holman et al. [30]).

Rubin's single-visit catalogs

In early operations, Rubin’s pipelines were configured and validated for typical star densities; crowded fields near the Galactic plane were processed on a best-effort basis. Because of this, as 2026 ED138 crosses a dense star field at low Galactic latitude Rubin’s single-visit catalogs have no entry for it in 24 of the 38 visits whose images show it (Table 4). In 6 of them, Daytime AP produced no catalog for the detector at all (the nightly-validation catalog NV-S did, and Table 4 classifies those visits by it). In the other 18, the single-frame pipeline had merged the light of neighboring stars into detection regions larger than its deblender accepts, and dropped every source inside them: in 15 of the 18, 2026 ED138 lies inside such a region, flagged deblend_parentTooBig. In the other three, it is blended with a star or lies at a detector edge.

These measurements describe the catalogs of direct images that our search read, as Daytime AP made them when these data were taken, along the track of 2026 ED138. In LSST operations, once templates exist, alerts come from difference images (M. Jurić et al. [35]), where subtracting a template removes the static stars that formed the blended regions in the direct images; the completeness of difference-image catalogs for slow-moving objects near the plane has not yet been measured. For searches that rely on direct-imaging catalogs, a bright object can be imaged on many nights and still be missing from them near the Galactic plane, where many of the remaining large objects probably are (Section 10). Specialized pipelines for processing crowded-field direct images are in development.

The DECam and ATLAS catalogs

DECam did image 2026 ED138 in other programs, on 14 nights from 2016-02-25 to 2023-04-07 (§ 5), but the NSC DR2 catalog lists it in only 11 of our 45 detections, on 5 nights. Those nights fall in different seasons, never more than two in one, and on each of them the detections span only minutes, too short to show the object’s motion. A search that links detections across nights needs several nights within one season, and a search for motion within a night needs a longer sequence, so NSC DR2 provided neither. Of our other detections, 33 were made before NSC DR2 ended in 2019. In 21 of these, the Community Pipeline, NOIRLab’s standard DECam processing, flagged the pixels of 2026 ED138 in its data-quality mask with the value it gives to transients, which may be why NSC DR2 has no measurement there; for the rest we do not know the reason.

ATLAS also recorded 2026 ED138: ADAM returned 26 ATLAS detections on 24 nights, all consistent with the adopted orbit (Appendix § 19.§ 2). Because these are mostly single detections on separate nights, and near ATLAS’s detection limit, they were unlikely to improve orbit or physical property measurements and were not used in this work.

How many more remain?

As of October 2026, only 9 known TNOs have lower absolute magnitudes than 2026 ED138 in the MPC catalog (whose $H$ values come from many sources and are only approximately in the $V$ band): Eris, Pluto, Makemake, Haumea, Sedna, Gonggong, Orcus, Quaoar and (532037) Chiminigagua. Chiminigagua’s catalog $H$ (3.09) is only about 0.1 mag below our $H_{V}$, and the MPC’s own $H$ for 2026 ED138 (2.93; MPEC 2026-U85) is below Chiminigagua’s, so whether 2026 ED138 ranks ninth or tenth depends on which $H$ values are compared. The Rubin search of these early data gives a first estimate of how many objects with $H_{V} < 4$ remain. Of the 33 known TNOs with $H_{V} < 4$, 10 were discoverable: the visits searched held at least ten of their detections on at least six nights within one search region and time window, enough for an entry in the catalog of the search (M. J. Holman et al. [30]). The search recovered 8 of these 10. The two it missed were close to its limits in different ways: (55565) Aya reached exactly ten detections, and vetting removed one of them; (90482) Orcus had more but was not linked, for a reason we have not found ((90568) Goibniu, with as many detections, was recovered). It also found one new object with $H_{V} < 4$, 2026 ED138. If the objects not yet discovered were as easy to find as the known ones, about 4 were undiscovered before the search (90% range 0.2–20), 2026 ED138 among them, so about 3 remain. The number before the search is 33 ×1/8, and its range is the 90% Poisson interval on the single new object. The method does not assume that the known objects are a complete list: the ratio of new to recovered objects measures how incomplete it is.

This estimate counts only what the Rubin search could have found, and it is probably an underestimate. First, still undiscovered objects are likely to be harder to find than known ones: they are the objects earlier surveys missed, because they are fainter in the sky, more distant, or in crowded fields, as 2026 ED138 was, and the same traits make them harder for Rubin. Second, the estimate applies only away from the Galactic plane. The known objects were found mostly away from the plane, because the earlier searches avoided it, and the Rubin search measured its efficiency only there. The band within $10^\circ$ of the plane covers 2,830 square degrees, 14% of the sky within $30^\circ$ of the ecliptic, and neither the Rubin search nor the all-sky search of M. E. Brown et al. [15] covered it. Third, it misses the more distant objects described below.

The result agrees with earlier estimates once the quantities compared are the same. M. E. Schwamb et al. [64] estimated that about 12 excited Kuiper belt objects are brighter than $R \approx19.5$, of which 9 were known, so that the inventory of objects that are bright in the sky is nearly complete, with perhaps one or two left in the Galactic plane or the southern sky; 2026 ED138, near that limit, is such an object. S. S. Sheppard et al. [69] similarly found the Kuiper belt nearly complete to $R \approx21$. Our estimate counts objects with $H_{V} < 4$ to much fainter apparent magnitudes. Most of those that remain are more distant and fainter in the sky than 2026 ED138, so they are harder both to find and to study.

The known objects with $H_{V} < 4$ were found by surveys that could reach them only when they were near the Sun. At the depth of the earlier wide searches, about $V \approx21.7$, an object with $H_{V} = 4$ is visible only within about 60 au. Several of the largest known TNOs are on eccentric orbits that take them much farther: Eris, Sedna and Gonggong are now at about 95, 83 and 90 au, and they were found because their absolute magnitudes are lower still, which makes them bright enough even at those distances. An object with $H_{V} < 4$ on a similar orbit, now beyond about 60 au, was out of reach of every earlier wide search. S. S. Sheppard et al. [69] estimated in this way that about 35 ± 15 scattered-disk objects with $H \lesssim4.1$ exist, most of them too far from the Sun to have been seen (their Table 4). At the 50% depth of the Rubin search (M. J. Holman et al. [30]), an object with $H_{V} = 4$ can be found out to about 80 au. Our estimate does not count these distant objects; they add to it, and LSST, which reaches this depth over the whole southern sky, should find many of them.

Objects move into and out of the Galactic plane over about two decades. At 40–50 au a TNO moves about 1.0–1.4° a year against its orbit, and the ecliptic crosses the band within 10° of the plane over about 23° of longitude, so an object takes about two decades to cross it. 2026 ED138 lay within 10° of the plane from 2000 to 2023 and was found just outside it, and the L5 cloud of Neptune Trojans, hidden by the plane for more than 15 years, has likewise moved into Rubin’s view (M. E. Schwamb et al. [67]). Near the plane, even where Rubin observes, the current catalogs drop every source inside large blended regions (§ 9.2). Local PSF fits at the predicted positions recover those detections, but they need an orbit to predict from, so they can extend the arc of a known object but cannot discover new ones. A search near the plane therefore needs one of three things: catalogs made with crowded-field settings, catalogs of difference images, or a search of the images along trial orbits. With that, a search of the plane through the first years of LSST should find most of the remaining objects.

Conclusions

2026 ED138 is a new TNO with $H_V = 3.19$ and a well-determined orbit at the edge of the 12:7 resonance with Neptune, like Haumea; its color shows that it is not a member of the Haumea family. It is a dwarf-planet candidate, pending an albedo measurement.

2026 ED138 was missed by earlier searches because it lies far south and near the Galactic plane, a region those searches mostly avoided (§ 9). Rubin found it, but its single-visit catalogs have no entry for it in 24 of the 38 visits whose images show it, mostly because the deblender isn’t yet tuned for crowded fields and for performance reasons skips processing of oversized blends. Local PSF fits at the predicted positions do recover these detections once the orbit is known. If the undiscovered objects were as easy to find as the known ones, about 3 with $H_V < 4$ would remain away from the Galactic plane (Section 10). More probably they lie within $10^\circ$ of the plane, which a search through the first years of LSST can cover with image differencing or better tuned crowded-field region processing.

At $i \approx19.4$, 2026 ED138 is bright enough to be studied in detail. A near-infrared spectrum would identify the ices on its surface, and lightcurves would give its rotation period. Thermal measurements would give its size and albedo, and a satellite, if one is found, its mass. It lies toward the Galactic bulge, so it will often pass in front of stars, and occultations can measure its size and shape directly.

Later data can also test our results. If its albedo is like those of other large TNOs, an occultation or thermal measurement should give 2026 ED138 a diameter of 612–968 km, and a near-infrared spectrum should not show the deep water-ice bands of the Haumea family. At that size and temperature it should also have lost its CH$_4$, N$_2$ and CO (Figure 5), so finding any of them on its surface would test the models of volatile retention. Away from the Galactic plane, about 3 more objects with $H_V < 4$ should be found; finding many more would mean that the search found new objects less readily than it recovered known ones. More should turn up within $10^\circ$ of the plane.

Acknowledgments

M.J.H. and K.J.N. acknowledge support from the NSF (grant No. AST-2206194) and the NASA YORPD Program (grant No. 80NSSC22K0239). M.J. acknowledges support from the NSF (grant No. AST-2606453). M.J., J.A.K., A.H., J.M., P.H.B., I.S., E.C.B., C.O.C., I.C., S.G., C.L. and D.S. acknowledge the support of the University of Washington College of Arts and Sciences, Department of Astronomy, and the DiRAC (Data-intensive Research in Astrophysics and Cosmology) Institute. I.C. acknowledges the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) and is partially funded by an NSERC Postgraduate Scholarship-Doctoral. The work of S.G. is supported by NOIRLab, which is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the U.S. National Science Foundation. M.E.S. acknowledges support from UK Science and Technology Facilities Council (STFC) grant ST/X001253/1. This material is based upon work supported in part by the National Science Foundation through Cooperative Agreements AST-1258333 and AST-2241526 and Cooperative Support Agreements AST-1202910 and AST-2211468 managed by the Association of Universities for Research in Astronomy (AURA), and the Department of Energy under Contract No. DE-AC02-76SF00515 with the SLAC National Accelerator Laboratory managed by Stanford University. Additional Rubin Observatory funding comes from private donations, grants to universities, and in-kind support from LSST-DA Institutional Members. This work uses data from the NSF–DOE Vera C. Rubin Observatory, obtained during its early operations, and data and services provided by the International Astronomical Union’s Minor Planet Center. We thank the Rubin Observatory builders. This publication is based in part on proprietary Rubin Observatory Legacy Survey of Space and Time (LSST) data, and was prepared in accordance with the Rubin Observatory data rights and access policies. All authors of this publication meet the requirements for co-authorship of proprietary LSST data. This research uses services or data provided by the Rubin Science Platform at NSF–DOE Vera C. Rubin Observatory, which is jointly funded by the U.S. National Science Foundation and the U.S. Department of Energy, Office of Science. This paper makes use of LSST Science Pipelines software developed by the Vera C. Rubin Observatory. We thank the Rubin Observatory for making their code available as free software at https://pipelines.lsst.io/. We thank the Asteroid Institute, a program of B612 Foundation, for the precovery search on its ADAM platform (https://b612.ai; run c742167c-021e-48a8-a8c5-66d86865f227). This research uses services or data provided by the Astro Data Lab, which is part of the Community Science and Data Center (CSDC) of the National Science Foundation.

::: ter (CSDC) Program of NSF NOIRLab, and the NOIRLab Source Catalog. This research draws upon DECam data as distributed by the Astro Data Archive at NSF NOIRLab. NOIRLab is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the U.S. National Science Foundation. This project used data obtained with the Dark Energy Camera (DECam), which was constructed by the Dark Energy Survey (DES) collaboration. Funding for the DES Projects has been provided by the US Department of Energy, the US National Science Foundation, the Ministry of Science and Education of Spain, the Science and Technology Facilities Council of the United Kingdom, the Higher Education Funding Council for England, the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, the Kavli Institute for Cosmological Physics at the University of Chicago, Center for Cosmology and Astro-Particle Physics at the Ohio State University, the Mitchell Institute for Fundamental Physics and Astronomy at Texas A&M University, Financiadora de Estudos e Projetos, Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro, Conselho Nacional de Desenvolvimento Científico e Tecnológico and the Ministério da Ciência, Tecnologia e Inovação, the Deutsche Forschungsgemeinschaft and the Collaborating Institutions in the Dark Energy Survey. The Collaborating Institutions are Argonne National Laboratory, the University of California at Santa Cruz, the University of Cambridge, Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas–Madrid, the University of Chicago, University College London, the DES-Brazil Consortium, the University of Edinburgh, the Eidgenössische Technische Hochschule (ETH) Zürich, Fermi National Accelerator Laboratory, the University of Illinois at Urbana-Champaign, the Institut de Ciències de l’Espai (IEEC/CSIC), the Institut de Física d’Altes Energies, Lawrence Berkeley National Laboratory, the Ludwig-Maximilians Universität München and the associated Excellence Cluster Universe, the University of Michigan, NSF NOIRLab, the University of Nottingham, the Ohio State University, the OzDES Membership Consortium, the University of Pennsylvania, the University of Portsmouth, SLAC National Accelerator Laboratory, Stanford University, the University of Sussex, and Texas A&M University. This work is based in part on observations at NSF Cerro Tololo Inter-American Observatory, a program of NSF NOIRLab (NOIRLab Prop. IDs 2014A-0480, PI: R. M. Rich; 2016A-0189, PI: A. Rest; 2016B-0279, PI: D. Finkbeiner; 2017A-0389, PI: D. Tucker; 2018A-0251, PI: D. Finkbeiner; 2018B-0271, PI: D. Finkbeiner; 2019A-0265, PI: D. Finkbeiner; 2019A-0305, PI: A. Drlica-Wagner), which is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the U.S. National Science Foundation.

This work has made use of data from the European Space Agency (ESA) mission Gaia (https://www.cosmos.esa.int/gaia), processed by the Gaia Data Processing and Analysis Consortium (DPAC); funding for the DPAC has been provided by national institutions, in particular the institutions participating in the Gaia Multilateral Agreement. We used the JPL Horizons ephemeris service. This research has made use of SciX, a scientific literature search and discovery platform.

We used large language models (Claude, Anthropic; and Grok, SpaceXAI) throughout this work: to write and run the analysis pipeline, including the archive queries, the validation, the local PSF fits and the orbit fits; to make the figures and tables; to check the numbers in the text against the analysis; and to draft and edit the text of the manuscript from the authors’ analysis and notes. The authors designed the analysis and made the scientific decisions, reviewed and verified the code, results and text, and take full responsibility for them. M.E.S. and I.C. did not make use of generative AI.

Author Contributions

M.J.H. wrote the pointdexter linking code, building on an initial version by K.J.N., built the stationary-source and transient catalogs it searches, and ran the search that found 2026 ED138; and carried out the resonance, Haumea-family and dynamical-classification analyses. M.J.H. is responsible for the outer solar system search algorithms on the Rubin Solar System Operations Team. M.J. led the Rubin and DECam precovery and arc extension work; carried out crowded-field astrometry and photometry, and the orbit fits and size estimates; and built the reproducible pipeline. M.J. leads the Rubin Solar System Operations Team and also led development of the Solar System infrastructure this work relies on (the River catalog database, the MPC submission system, and the ephemeris and known-object matching services). K.J.N. wrote the initial version of pointdexter, developed the density-based thinning of the transient catalogs, and checked the discovery. J.A.K. experimented with the transient-catalog production for low-coverage regions, including the lower minimum numbers of detections and nights per source that the production now uses generally. A.H. vetted the TNO submissions, performed independent orbit fits, defined the MPC deduplication rules for overlapping linkages, and leads HelioLinC development. J.M. made the ADAM precovery search possible and built the MPC submission tracking software. M.E.S. contributed to the interpretation of the results, including the dwarf-planet size threshold and the expected population of large TNOs at low Galactic latitude. S.P. contributed the volatile-retention analysis in § 8.3, placing 2026 ED138 on the retention diagram of S. Protopapa et al. [51], and its interpretation. P.H.B. contributed TNO-survey expertise, including the comparison with the Dark Energy Survey search, and contributed to the MPC submissions. I.S. and E.C.B. lead the Rubin Alert Production pipeline, which produces some of the catalogs searched in this work and the images used for the local PSF fits. E.N. contributed to the Solar System pipeline tasks in the LSST Science Pipelines, including the Solar System tables and the HelioLinC inputs, and contributed to the MPC submissions. C.O.C., I.C., E.C., S.C., S.E., S.G., C.L., J.E.R., D.S. and D.Z. are members of the Rubin Solar System Operations Team, contributed to its commissioning-era work, and commented on the manuscript.

Facilities: Rubin:Simonyi (LSSTCam; SLAC National Accelerator Laboratory & NSF-DOE Vera C. Rubin Observatory 2025), Rubin:USDAC, Blanco (DECam), Astro Data Lab, Astro Data Archive, Gaia

Software: pointdexter (Holman & Napier, in preparation), ASSIST (M. J. Holman et al. [29]), REBOUND (H. Rein & S.-F. Liu [58]), LSST Science Pipelines (Vera C. Rubin Observatory Science Pipelines Developers [78]), Data Butler (T. Jenness et al. [33]), Astropy, NumPy, SciPy, Polars, Matplotlib, SkyBoT (J. Berthier et al. [10]), SEP (K. Barbary [7]), GaiaXPy (D. Ruz-Mieres et al. [61]), ADAM (K. Kiker et al. [38])

Data Availability

The astrometry of 2026 ED138, Rubin and DECam, is listed in Table 11, which is also available in machine-readable form, and provided in ADES format (pipe-separated values). The Rubin data we use are in the 2026 ED138 data release on the Rubin Science Platform (https://data.lsst.cloud), in /rubin/shared/datasets/2026-ED138: the calibrated single-visit images, the AP-S and NV-S catalogs, and the visit metadata, as a Data Butler export (T. Jenness et al. [33]) and tables. These are proprietary data, available to Rubin data rights holders. Every Rubin number, table and figure in this paper is produced from this release by one scripted pipeline; the other inputs come from read-only queries of the MPC, NOIRLab and Gaia archives and of the ADAM service.

Rubin data and checks

Products

All Rubin data products utilized in this work come from the Rubin 2026 ED138 data release. By searching this repository we find that the only exposures from 2025 that cover 2026 ED138 are 2 acquisition frames on the path of 2026 ED138, for which no catalogs exist (Appendix § 15.§ 5). During the period these images were taken, Rubin processed each exposure in several ways ([34, 35]), and we use three of the products:

• Prompt Processing (NSF-DOE Vera C. Rubin Observatory 2026b,a) runs within minutes of each exposure. We use its calibrated single-visit images (preliminary_visit_image), made with the LSST Science Pipelines ([78]). Prompt Processing itself detects sources only above $20\sigma$; we use its images, not its sources.

• Daytime Alert Production (Daytime AP; NSF-DOE Vera C. Rubin Observatory 2026b) reprocesses each night’s exposures during the following day. Its single-frame task, singleFrameDetectAndMeasure, measures sources to $5\sigma$ on each direct image (not on a difference image). We call these catalogs AP-S. They give a position error for each source.

• Nightly validation, a processing run that checked each night’s data in early operations, produces a second source catalog for each visit (single_visit_star). We call it NV-S. It gives positions but no position errors.

We read the AP-S and NV-S catalogs, and the images, from the 2026 ED138 data release (Data availability); the positions we use are in Table 11. Astrometry is calibrated to Gaia DR3 ([26]). The catalog error of each AP-S measurement plus $0.015^{\prime\prime}$ in quadrature reproduces every uncertainty in the delivered observations (§ 2.2; all 10 rows that carry one). The files we submit to the MPC, in the format of the Astrometry Data Exchange Standard (ADES), add $0.010^{\prime\prime}$ instead, as Rubin’s own submissions do, and give the correlation between the RA and Dec errors; these are the uncertainties of Table 11 and of our orbit fits. The files of the delivered and cataloged data sets (Table 5) carry the $0.015^{\prime\prime}$ term; their fits replace it with $0.010^{\prime\prime}$. Where a catalog provides no error, the uncertainty columns are left empty.

Weighting of the orbit fits

We weight the observations by their reported uncertainties as they are. With them, the Rubin-only fit has $\chi^2 = 65.5$ for 54 degrees of freedom. For the 29 observations that have a reported uncertainty, the residual rms per axis is 21 mas, compared with a median reported uncertainty of 11 mas; the rms is dominated by the few faint observations with larger uncertainties, and the $\chi^2$, close to the number of degrees of freedom, shows that the reported uncertainties describe the residuals. As a check, we repeat every fit with a $0.1^{\prime\prime}$ floor added in quadrature to the reported uncertainties. The floor inflates them by a median factor of 9.0 and gives $\chi^2 = 2.2$; Table 5 lists both weightings. The adopted fit, to the Rubin and DECam observations together, has $\chi^2 = 84.7$ for 142 degrees of freedom (6.2 with the floor). A $\chi^2$ this far below the number of degrees of freedom means that the DECam uncertainties, which include a $0.05^{\prime\prime}$ systematic term (Appendix § 18.§ 3), are conservative; we report them unchanged to the MPC.

One of the Rubin observations, from NV-S, has no reported uncertainty. In both weightings, the fit weights it as if its uncertainty were $0.2^{\prime\prime}$ per axis. This weight is not an uncertainty estimate: its ADES uncertainty columns stay empty, and it is left out of the residual statistics above.

Was it previously known?

We queried a replica of the MPC observation database (obs_sbn) for any observation within $2^{\prime}$ and $\pm1$ h of each reported detection, and found none.

We then selected the 9,003 MPC orbits that satisfy $q > 25\,\mathrm{au}$, $a > 25\,\mathrm{au}$ or $H < 6$, and propagated them with ASSIST ([29]) to 3 epochs on the arc. The nearest TNO was 2000 JG81 ($H = 8.0$), at $3.6$–$3.7^{\circ}$. A cone search with SkyBoT ([10]), a service that lists the known small bodies in a field, of radius $10^{\prime}$ at the same epochs returned 5 objects, none farther than 3.2 au from the Sun and none closer than $207^{\prime\prime}$ to 2026 ED138.

Finally, we looked for tracklets in the ITF within $20^{\prime}$ of the predicted track outside the observed arc, from 2025 to 2026-10-15. We found none.

Are the detections real?

Every reported row matches an AP-S source in the same visit (10 rows). The worst matches are within 0.2 mas and 5 ms. None of the matched sources has any of the pixel, point-spread-function (PSF) flux or centroid flags set. Signal-to-noise ratios (S/N) are 19.3–163.

Two rows needed attention:

1. The 2026 ED138 row of 2026-05-01T06:13:30.869Z is a deblend ‘parent’ (the combined footprint of overlapping sources, before the deblender splits it) flagged as extended, because it includes an equally bright neighbor 2.0″ away. We do not use extended parents, so we first replaced it with its point-like ‘child’, 21.9 mas away. The local PSF fit (§ 4), which models the neighbor, lies 4 mas from the parent and 18 mas from the child, and we use it instead of both.

2. The 2026 ED138 i-band row of 2026-05-23T08:10:34.592Z is blended with a static $i \approx22.2$ star 1.1″ away.

We kept the blended row, which fits the orbit (§ 6).

Figure 1 shows 2026 ED138 in single-visit images, together with the same sky on another night. At fixed sky positions we also measured aperture photometry in every covering visit: 560 visits at 15 positions. Flux appears only when the object is present, except at 2 of the positions, where a known static neighbor lies within the aperture.

Additional cataloged detections

We predicted the position of 2026 ED138, with its covariance, from a fit to the delivered observations with a 0.1″ floor added to their uncertainties (§ 2.3); the floor gives wider search windows. We did this for every LSSTCam exposure within 2.1° of the track. There were 63 such exposures (Table 4). Using detector outlines from the export of the Rubin consolidated database provided in the release, we found that the predicted position fell on a detector in 51 of 63 visits; in the other 12, it falls off every detector. In each of these visits, we searched the AP-S and NV-S catalogs within $\max(2'', 3\sigma)$ of the prediction, and accepted sources within 0.75 mag of the expected magnitude in the visit’s band: the median of the delivered observations in that band, or, for a band with none, the value of the neighboring band. Where AP-S had a deblended extended parent, the point-like child was preferred.

This search added 5 cataloged detections on 4 nights that were not among the delivered observations. Of these, 1 are NV-S detections without position errors. With the new detections, the arc grows to 69 days.

We classified each other covering visit by why it has no detection (Table 4). Most of the remaining visits fall in one of two classes:

• 18 have an incomplete catalog: fewer than half of the stars brighter than the object were recovered nearby;

• 9 fall at detector edges.

The incomplete catalogs are the problem described in § 9.2.

A second pass with the refitted orbit changed no classification and found no new detections.

There is no 2025 science coverage of the track. 2 LSSTCam acquisition frames from 2025 cover 2026 ED138’s predicted path, but no catalogs exist for them.

Local PSF fits

Rubin’s catalogs miss 2026 ED138 in most of the visits whose images show it (Table 4; § 9.2 explains why). Instead of reprocessing whole detectors, we measured 2026 ED138 directly at its predicted positions with a local PSF fit. The predictions came from a fit to the cataloged detections, with the same floor as in Appendix § 15.§ 5. We used the same Prompt Processing images, with the PSF models, world coordinate systems (WCS) and photometric calibration that the pipeline stored with each image.

In each visit, we cut a 41-pixel (8.2″) stamp and fitted a model with these components:

• a PSF at the object, with free position and flux;

• a PSF at every neighbor detected above $5\sigma$, with free flux, and also free position for neighbors brighter than $50\sigma$;

• a background that is a plane: a constant plus a gradient in each direction across the stamp.

We weight the fit by each image’s variance plane and mask saturated, bad, cosmic-ray, suspect and edge pixels. Positions and their covariances are converted to RA and Dec through the image’s WCS. Magnitudes come from the PSF flux, the image’s aperture correction and its photometric calibration. The method uses no catalog subtraction and no deblending step.

We validate this approach by first applying the fit to the 12 visits where AP-S already has a clean detection. The local and catalog positions differ by a median of −0.1 and +0.4 mas in RA and Dec, with an rms of Note—$N$: exposures within the search radius of the predicted track (§ 4), by the class the catalog search assigns. Img.: a single-visit image exists at the predicted position. Add.: local PSF fits (§ 4) that pass the cuts and add an observation; “r” marks fits that instead replace a cataloged row: an NV-S row without position errors, or a row overridden after inspection (§ 4). Rej.: local fits that fail the $S/N$ or $\chi^2$ cut; “d” marks cataloged rows dropped after inspection. The validation set of the local fit (§ 4) is the imaged reported and detected visits with an AP-S detection whose fit also passes the $\chi^2$ cut. Final observations count rows of the final data set (Table 11).

2.6 mas. This is consistent with the fit’s own errors of $\approx3$ mas. The magnitudes agree to 0.014 mag rms. There is no systematic offset between the image WCS and the catalogs, so we do not inflate the fit errors; as for the AP-S rows, we add 0.010″ in quadrature when we report them (Appendix § 15.§ 1).

We keep a measurement if it has $S/N \ge5$ and a reduced $\chi^2 < 1.5$. The $\chi^2$ cut rejects fits whose residuals show an unmodeled neighbor or blend (Figure 2). The fit yields 16 new measurements, with $S/N = 20$–$193$. 7 visits fail the $\chi^2$ cut and 1 fails the $S/N$ cut. The earliest new points, from 2026-03-04, were never in any catalog. Daytime AP produced no catalog for their detectors, and NV-S has no source at the object’s position. They lie within a fraction of an arcsecond of the position predicted without them.

The complete Rubin data set has 30 observations of 2026 ED138 on 20 nights, from 2026-03-04 to 2026-07-11. These are the 10 delivered, 5 additional cataloged and 16 remeasured observations, less the 1 dropped after inspection (Table 4).

Orbit fits

Table 5 and Figure 6 show how the orbit evolved from the delivered observations to the complete Rubin set and with the DECam recoveries; Figure 7 shows the residuals of the Rubin observations.

Astrometry

After inspection, two cataloged Rubin rows of 2026 ED138 were overridden (Appendix § 15.§ 6). The decisions and their reasons are listed in remeasure_overrides.csv.

• 2026 May 1 (z): the catalog row is the de-blend child that replaced the delivered parent (Appendix § 15.§ 4). The local PSF fit models the equally bright neighbor 2.0″ away (flux ratio 0.93). The fitted position lies 4 mas from the deblend parent and 18 mas from the child. Its magnitude, 19.33, matches the parent’s 19.32, not the child’s 19.39. The fit replaces the child.

• 2026 May 2 (z): the catalog position lies 47 mas from the local PSF fit. A static $\approx20.2$ mag star 1.4″ away shifts the catalog position. The fit itself fails the $\chi^2$ cut (2.7), so there is no clean measurement and we drop the row. An exposure 8 min apart on the same night is clean.

DECam photometry and checks

Rubin calibration

Rubin’s calibration is tied to The Monster, Rubin’s reference catalog for photometric calibration ([19]). Stars in the Rubin visits of 2026 ED138 agree with The Monster to within 0.007 mag in every band (Table 8). Against Gaia XP synthetic photometry in the LSST bands, the same visits’ stars differ by $-0.020$ mag (Rubin − XP) in $g$ and agree to 0.016 mag in $riz$. We treat the $g$ offset as a systematic uncertainty of Rubin’s $g$ calibration in this field and do not correct for it.

Comparison	$g$	$r$	$i$	$z$	$Y/y$
Monster − NSC, Rubin fields	−0.030 (1072)	−0.030 (893)	−0.032 (796)	−0.005 (778)	· · ·
Rubin − Monster, stars	−0.004 (2423)	−0.002 (5817)	+0.007 (6618)	+0.007 (12156)	· · ·
Rubin − XP, stars	−0.020 (113)	+0.001 (237)	+0.009 (177)	+0.016 (289)	· · ·
XP − NSC, DECam zero points	−0.087 (7)	−0.174 (6)	−0.092 (6)	−0.116 (7)	−0.082 (6)
XP − Monster, DECam zero points	+0.035 (3)	+0.009 (3)	+0.010 (3)	−0.003 (4)	· · ·
Same method − Rubin, 2026 ED138	+0.005 (3)	−0.001 (4)	−0.004 (6)	+0.005 (9)	· · ·

DECam calibration

We measured 2026 ED138 in all 44 DECam exposures by PSF photometry on difference images. For each exposure we took 1–3 templates: the same sky in the same band on other nights, when 2026 ED138 was elsewhere. Each template is resampled onto the exposure’s pixels, and the sharper of the two images is convolved with a kernel ([1]), Gaussians of three widths times polynomials, fitted together with a linear background to all the pixels of a 79″ box except those near 2026 ED138 and saturated stars. 2026 ED138 is then fitted on the difference with the exposure’s PSF (convolved with the kernel when the exposure was the sharper image), so its flux is on the scale of the stars measured in the exposure itself. The uncertainty is empirical: the scatter of the same fit at 300 blank positions of the difference, which includes the template’s noise and the residuals of subtracted stars. The templates are combined by inverse variance, and the combined uncertainty is the scatter of the same combination at the same blank positions, because all the differences share the exposure’s noise. Copies of 2026 ED138 injected into each exposure 7–12″ from it are recovered with a median bias of −0.001 mag (at most 0.040 mag in one exposure). We would use the direct fit instead where no template is usable, or where the difference fails a check (an uncertainty more than 3 times the direct fit’s, or injected copies recovered off by more than 0.05 mag plus three times the uncertainty of their median); for 2026 ED138 none does.

The direct PSF fit of Appendix § 18.§ 3, with every cataloged neighbor held at its cataloged position, differs from the difference-image photometry by 0.02 mag (robust scatter) and by up to 0.16 mag in single exposures. Injections elsewhere in the image do not reveal this: it is specific to the position of 2026 ED138 in each exposure, where it is blended with uncataloged stars or with cataloged ones close enough to share its light. The Gaia DR3 stars in each cutout are measured with the direct fit. We tie each exposure to the Gaia DR3 XP synthetic DECam magnitudes (Gaia Collaboration et al. [25]) of the 9 (median) stars in the cutout that have XP spectra (183 stars in all). GaiaXPy (D. Ruz-Mieres et al. [61]) computes the $grizY$ magnitudes; the broad DECam VR band, which it lacks, we integrated from the calibrated XP spectra over the CTIO DECam VR throughput.

The same integration reproduces GaiaXPy’s $griz$ to 0.001 mag. Per exposure, a robust linear fit of $m_{\mathrm{XP}}-m_{\mathrm{inst}}$ against the stars’ $g-i$, evaluated at the color of 2026 ED138, gives the zero point. The stars scatter about the fit by 0.01 mag (median), and repeated exposures in the same band and night agree to 0.013 mag. The XP spectra end at 1020 nm, inside the $Y$ band, so we flag $Y$ and add 0.02 mag, the robust exposure-to-exposure scatter of its difference from the NSC-tied zero points, to its uncertainties. The robust scatter keeps one exposure from dominating: in 1 exposure the reference stars are heavily reddened (median $g-i > 2.5$), and there we use 0.18 mag, which includes that exposure’s own difference. Table 6 gives the results. Our ADES file uses these magnitudes.

We do not use the NSC DR2 mean magnitudes (D. L. Nidever et al. [43]), which are tied to ATLAS-Refcat2 (J. L. Tonry et al. (2018)) without color terms. In the DECam fields, at Galactic latitudes −10 to −3°, most reference stars are reddened. The NSC zero points differ from the XP ones by up to 0.17 mag (in $r$), and their stars scatter about them by 0.09 mag. Along the Rubin **Note—**Medians over the exposures in each band. XP (adopted): zero points from Gaia DR3 XP synthetic DECam magnitudes of stars in the same image, at the object’s color; NSC: from their NSC DR2 magnitudes (§ 7.2).

$H = m - 5\log_{10}(r\Delta) - \beta\alpha$, with the slope $\beta$ measured from the Rubin data.

$H_{\mathrm{XP}}^{\mathrm{LSST}}$: converted to the LSST bandpass at the object’s color; $H_{\mathrm{Rubin}}$: the Rubin value in that band (none for VR); $\Delta H = H_{\mathrm{XP}}^{\mathrm{LSST}} - H_{\mathrm{Rubin}}$.

track of 2026 ED138 ($b = -12^\circ$), NSC agrees with The Monster to 0.03 mag (Table 8).

DECam versus Rubin

Table 7 follows the difference between the DECam and Rubin reduced magnitudes of 2026 ED138. With NSC-tied DECam magnitudes at the observed phases, DECam is 0.04–0.21 mag fainter. Reducing both to zero phase removes part of that: the DECam phase angles, 0.7–1.2°, are on average larger than the Rubin ones. Tying DECam to XP and converting it to the LSST bandpasses at the object’s color removes most of the rest, and measuring 2026 ED138 on difference images removes the light of its neighbors. What remains is −0.00, +0.03, −0.00, +0.04 and −0.00 mag in $g$, $r$, $i$, $z$ and $Y$; with the direct PSF fit instead it would be −0.00, +0.03, +0.06, +0.05 and +0.05 mag. A joint fit of the Rubin and XP-tied DECam magnitudes with one phase slope and one DECam offset gives $\beta= 0.14 \pm0.03$, consistent with the Rubin value, and an offset of $+0.01 \pm0.01$ mag.

**Note—**Medians over the DECam exposures in each band, minus the Rubin value, in mag. NSC, $\alpha_{\mathrm{obs}}$: NSC-tied, both at their observed phase angles; Phase: the change on reducing both to zero phase with the slope measured from the Rubin data; Calib.: the change on tying DECam to Gaia XP and converting it to the LSST bandpass at the object’s color. The last column is the remaining difference (Section C.3).

Our direct PSF fit, applied to the Rubin images with zero points from the Rubin catalog fluxes of the same stars, reproduces Rubin’s magnitudes of 2026 ED138 to 0.005 mag in every band (Table 8), so the two cameras’ photometry agree where crowding does not intervene. On two consecutive nights 2026 ED138 changed by 0.02 mag in $i$ and 0.00 mag in $z$, comparable to the 0.013 mag repeatability within a night.

Checks

Table 8 collects the checks behind § 7. The Monster $-$ NSC: stars along the Rubin track of 2026 ED138 with $0.6 < g - i < 1.2$, with the PS1-to-ComCam color terms of M. N. Porter et al. [49] applied to NSC (whose southern calibration is ATLAS-Refcat2, itself tied to PS1). Rubin $-$ Monster: Rubin’s calibrated single-visit fluxes of the stars in the visits and detectors of 2026 ED138; the per-visit medians lie within 0.03 mag of zero in every band. Rubin $-$ XP: the same stars with XP spectra (320 stars; up to 20 bright, well-measured stars per visit), against their GaiaXP LSST magnitudes. XP $-$ NSC and XP $-$ Monster: the zero points of the DECam exposures from XP, from NSC, and from The Monster (the last converted to the DECam system with the XP LSST $-$ DECam relation), all evaluated at the color of 2026 ED138; The Monster is limited to the brighter stars, and gives a usable fit in only a few exposures. Same method: 2026 ED138 measured on 22 Rubin preliminary_visit_image cutouts with the DECam method (empirical PSF, static neighbors, Gaia reference stars, zero point from the Rubin catalog fluxes of the same stars), minus Rubin’s magnitude.

Note—Median differences in mag; in parentheses the number of stars, or of exposures for the zero points and the same-method test. Monster − NSC includes the RTN-099 PS1-to-ComCam color terms. The DECam zero points are evaluated at the color of 2026 ED138 (Section C).

DECam precovery

Candidates from ADAM

We first searched for archival detections with the precovery service of the Asteroid Institute’s ADAM platform ([38]). Given an orbit, it predicts the object’s position, with its uncertainty, in every exposure of the surveys it indexes, and returns the cataloged sources that lie within the uncertainty region and have a plausible brightness. We gave it the Rubin-only orbit (§ 2.3), and it searched from 2012-09-01 to 2026-10-18 (run of 2026-10-18). It examined 3292 exposures and returned 57 candidate detections:

• 28 in 16 DECam exposures on 6 nights (2016-02-25 to 2019-05-01), from the second data release of the NOIRLab Source Catalog (NSC DR2; [43]);

• 3 from SkyMapper on 2 nights (2014-08-10 and 2016-08-29);

• 26 from ATLAS on 24 nights (2019-07-06 to 2026-07-07).

We followed up the DECam candidates. DECam ([22]), on the 4-m Blanco telescope at Cerro Tololo, reaches several magnitudes deeper than SkyMapper and ATLAS, so it gives the highest signal-to-noise ratio and the most precise positions. We compare the ATLAS and SkyMapper candidates with our adopted orbit in Appendix § 19.§ 2, but do not use them in it.

At these epochs the Rubin-only prediction is uncertain by 33–74″, and in this crowded field such a region holds many cataloged sources. We therefore checked each DECam candidate in the images. For each one we cut out its DECam CCD from the NOIRLab Astro Data Lab ([21, 44]), measured aperture photometry and the brightest peak at the candidate’s position in NSC DR2, and measured

the same aperture in an image of the same sky taken on another night, when 2026 ED138 was far away (at least 253 days earlier or later for every candidate we accepted). We accepted a candidate as a moving point source if $S/N \ge8$, its peak lies within 1″ of that position, and the other night’s image holds less than 0.5 of its flux. 11 of the 28 candidates pass. Of the rest, 6 are static sources and 11 show no significant source at their NSC DR2 positions; all of these are among the 17 candidates of 2017-05-04.

We then added the nights one at a time to the Rubin orbit, starting from the night closest in time to the Rubin data. On each night we kept the candidates that agree with the current prediction (at $3\sigma$, with $0.2''$ position errors), refitted, and kept the night only if every new point fitted the new orbit at $3\sigma$. All 11 vetted candidates, on 5 nights from 2016-02-25 on, link.

Search of the DECam archive

The orbit from the ADAM detections predicts the position of 2026 ED138 in the DECam era to within 0.11″, so we measured every image at the predicted position instead of relying on catalogs. We listed every DECam exposure in NSC DR2 (1056 in a box around the track) and every DECam image in the NOIRLab Astro Data Archive that NSC DR2 does not contain (341, mostly from after 2019, when NSC DR2 ends). The prediction falls within 1.15° of the pointing in 89 exposures on 19 nights, from 2016 to 2026 (Table 9). Using the CCD outlines, it falls on a CCD in 52; the other 37 fall in gaps between CCDs or off the array.

Night (UTC)	Field	CCD	Acc.	Used	Bands	$S/N$	NSC	rms	Not used
2016-02-25	4	4	4	4	VR	54–59	4	24	–
2016-08-11	3	3	3	3	izY	17–48	3	18	–
2016-08-16	9	3	0	0	· · ·	· · ·	0	· · ·	offset peak
2016-08-24	4	4	4	4	gr	55–87	0	35	–
2016-08-25	4	2	2	1	gr	70–85	0	52	seeing
2016-08-26	4	2	2	2	gr	71–86	0	42	–
2017-05-04	20	10	10	10	grizY	13–46	0	24	–
2017-05-05	10	4	0	0	· · ·	· · ·	0	· · ·	edge
2017-07-02	4	4	4	4	VR	60–65	0	18	–
2017-07-03	4	4	4	4	VR	51–59	0	15	–
2018-05-10	2	0	0	0	· · ·	· · ·	0	· · ·	–
2018-05-12	4	2	2	2	iz	23–28	2	16	–
2018-05-19	2	2	2	2	gr	29–36	0	13	–
2018-05-21	4	0	0	0	· · ·	· · ·	0	· · ·	–
2019-04-28	2	1	1	1	Y	12–12	1	20	–
2019-04-30	3	3	3	3	izY	19–51	0	18	–
2019-05-01	3	3	3	3	izY	15–32	1	55	–
2023-04-07	2	1	1	1	z	34–34	0	34	–
2026-06-25	1	0	0	0	· · ·	· · ·	0	· · ·	–

In each of the 52, we measured the image at the predicted position as in Appendix § 18.§ 1, with the data-quality mask of NOIRLab’s DECam Community Pipeline ([77]), and a template from another night, at least 1 day earlier or later (median 271 days), when 2026 ED138 had moved well away. We accepted the detection if:

• $S/N \ge8$ in an aperture at the prediction;

• the brightest peak is within 1′′ of the prediction;

• the position is at least 5′′ from the CCD edge;

• the template holds less than 0.5 of the flux.

2026 ED138 is detected in 45 exposures on 14 nights, from 2016-02-25 to 2023-04-07, with $S/N = 12$–87 (Figure 3). Of the rejected exposures, 4 lie at a CCD edge and 3 have their peak on a brighter star next to 2026 ED138.

We then repeated the image check at the positions predicted by the orbit fitted to these detections, as a check only (Appendix § 19.§ 1). For 2026 ED138, it found no new detections and lost none. Only 11 of the 45 detections are in the NSC DR2 catalog (§ 9.3).

Astrometry

We measured each detection on a 3′ cutout of its CCD. We do not rely on the image’s WCS beyond the cutout: we tie it locally to the Gaia DR3 stars (Gaia Collaboration et al. [26]) in the cutout, propagated to the exposure’s epoch with their proper motions and measured in the same way as 2026 ED138; 79–278 stars per exposure are used. The tie is a constant offset; an affine term never predicted the stars significantly better in a leave-one-out test. 2026 ED138 is measured by an empirical PSF, built from bright isolated stars in the cutout and fitted together with every known static neighbor (Gaia DR3 sources, those without a proper motion taken as stationary, and NSC DR2 objects seen on several nights). The neighbors stay at their cataloged positions and only their fluxes are fitted; with free positions, overlapping sources trade flux and the fit becomes degenerate.

In this field even a PSF fit is biased by stars too faint to be cataloged. We measured this bias directly: we injected 2026 ED138, with its measured flux and the template’s PSF and noise, into two templates at its position, on a $5 \times5$ grid of sub-pixel offsets, and measured it as in the data. Where the two templates agree to 40 mas (24 exposures, and 6 with one usable template), we subtract the mean bias, at most 87 mas; where they disagree (14), we leave the position uncorrected and add half their difference to its uncertainty. The uncertainty of each position combines the PSF fit, the tie, the bias correction and a 0.05′′ systematic term; its median is 52 mas per axis. The times are the exposure midpoints: the start time in the image header (DATE-OBS) plus half of the exposure time and of the 1.05 s shutter travel (B. Flaugher et al. [22]; M. V. Banda-Huarca et al. [5]).

We kept 44 of the 45 detections, on 14 nights. We dropped the one with seeing worse than 2′′. No detection failed the other test, a fit with the Rubin data at $3\sigma$.

The positions are in Table 11 and in ADES format in the accompanying repository.

Checks of the DECam astrometry

This appendix gives the details behind § 5.

The second pass

The second pass of the DECam search (Appendix § 18.§ 2) repeats the image check at the positions predicted by the orbit fitted to the first pass’s detections. We use it only as a check, not as a cut: near the threshold, measuring again at a better position lowers the $S/N$ of real faint detections. A detection that the second pass loses would be reviewed, and a new one measured independently (Appendix § 18.§ 3).

Validation

Against Gaia DR3, the image WCS is offset by up to 107 mas, with 6–54 mas scatter per star; our tie removes the offset. NSC DR2’s own positions of the same stars agree with Gaia to 11 mas (median offset) and scatter about them by 23–61 mas per CCD (Table 12), so NSC DR2’s frame is sound; our tie uses Gaia directly.

Time (UTC)	Band	$N_\star$	Header − Gaia	rms (mas)	NSC − Gaia	rms (mas)	Ours (mas)	Shutter (s)	Bias (mas)	T	NSC − ours (mas)	O−C (mas)
2016-02-25 07:49	VR	137	−23, 75	14	12	32	18	0.04	−66, 56	a	−123, 102	−39, −4
2016-02-25 07:51	VR	126	14, 57	16	4	38	20	0.04	−34, 34	a	−144, 144	−18, −14
2016-02-25 07:53	VR	107	8, 57	13	10	33	18	0.04	−35, 35	a	−158, 113	−15, −20
2016-02-25 07:54	VR	138	−4, 79	12	2	32	15	0.05	−36, 35	a	−137, 115	−33, −27
2016-08-11 02:37	i	93	49, 24	17	4	38	21	0.06	6, −1	a	145, 91	−25, −1
2016-08-11 02:38	z	94	76, 43	12	1	41	15	0.07	−2, 5	a	238, 135	−34, 3
2016-08-11 02:39	Y	96	52, 30	8	11	40	13	0.26	27, −6	a	649, 162	−9, −9
2016-08-24 03:50	g	120	24, 35	29	32	47	43	−0.54	−7, 0	s	···	57, 14
2016-08-24 03:51	r	120	75, 52	15	23	39	21	−0.53	18, −18	s	···	−8, 26
2016-08-24 03:58	r	116	32, 44	23	21	41	29	−0.54	16, −21	s	···	−23, 18
2016-08-24 04:00	g	116	29, 57	47	48	61	58	−0.53	−14, 2	s	···	69, −8
2016-08-25 04:28	g	99	23, 43	49	61	58	69	−0.63	−7, 3	a	···	69, −26
2016-08-26 04:37	r	103	21, 69	15	6	31	20	−0.66	−4, −15	a	···	20, −24
2016-08-26 04:39	g	104	27, 53	26	33	41	34	−0.66	···	d	···	74, −25
2017-05-04 05:57	Y	84	−1, 0	9	0	39	12	1.02	···	d	···	1, 12
2017-05-04 05:58	z	81	6, −2	10	11	33	15	1.01	···	d	···	13, 30
2017-05-04 05:59	i	80	5, −22	9	11	38	14	1.01	···	d	···	1, 13
2017-05-04 06:06	i	90	4, −1	13	···	···	17	1.01	···	d	···	2, 29
2017-05-04 06:06	z	90	19, 20	13	···	···	21	1.01	···	d	···	−8, 44
2017-05-04 06:07	Y	90	15, −17	13	···	···	19	1.01	···	d	···	5, 25
2017-05-04 07:44	r	80	17, −18	13	6	26	19	1.01	8, 6	s	···	−7, −24
2017-05-04 07:46	g	79	0, 6	11	5	23	15	1.01	19, 15	s	···	−34, −57
2017-05-04 07:54	g	93	8, −22	17	···	···	22	1.00	7, 4	a	···	−9, 31
2017-05-04 07:56	r	91	−3, 7	12	···	···	18	1.01	11, 3	a	···	−22, 18
2017-07-02 01:47	VR	133	36, −35	12	16	53	22	0.03	−3, 5	a	···	−18, 7
2017-07-02 01:49	VR	133	30, −2	14	12	47	22	0.03	−1, 5	a	···	−25, 9
2017-07-02 01:51	VR	133	11, −15	13	16	52	18	0.03	−1, 6	a	···	−24, 11
2017-07-02 01:52	VR	134	25, −23	13	12	42	22	0.03	−1, 7	a	···	−18, 20
2017-07-03 02:14	VR	134	35, 11	8	8	28	11	0.03	−3, −2	a	···	−17, −3
2017-07-03 02:16	VR	136	19, 5	9	11	30	12	0.03	−4, −3	a	···	−17, 14
2017-07-03 02:17	VR	134	24, 2	8	9	33	12	0.02	−4, −2	a	···	−17, 23
2017-07-03 02:19	VR	134	33, 1	13	8	34	18	0.03	−5, −2	a	···	−11, 9
2018-05-12 09:30	z	135	−107, 4	13	6	41	18	0.04	−2, 10	a	−6	14, −4
2018-05-12 09:31	i	134	−75, 20	18	4	37	24	0.06	0, 6	a	8, 26	29, 3
2018-05-19 08:16	g	108	18, 9	15	6	36	22	0.06	0, 4	a	···	23, −6
2018-05-19 08:18	r	110	12, 27	15	15	35	21	0.06	−11, −5	a	···	−7, 0
2019-04-28 10:04	Y	142	14, 17	19	4	42	29	−0.18	···	d	30, 19	−2, −28
2019-04-30 08:19	Y	96	−22, 13	11	18	30	17	−0.18	···	d	···	34, 16
2019-04-30 08:20	z	95	−4, 21	9	4	27	12	−0.17	···	d	···	12, −2
2019-04-30 08:21	i	91	−12, 35	10	15	31	13	−0.09	···	d	···	13, 17
2019-05-01 09:13	Y	98	1, 33	15	14	42	23	−0.18	12, −19	a	83, −35	11, −3
2019-05-01 09:14	z	87	43, −3	21	18	49	29	−0.16	···	d	···	45, 31
2019-05-01 09:15	i	89	−23, 20	25	33	53	36	−0.17	···	d	···	119, −24
2023-04-07 06:11	z	278	−28, −52	15	···	···	21	−24.72	···	d	···	44, 21

The NSC DR2 catalog has 2026 ED138 in 11 exposures on 5 nights. On 2016-02-25, NSC DR2 and our positions differ by (−141, 119) mas in (RA, Dec). A SExtractor-equivalent windowed centroid (E. Bertin & S. Arnouts [11]; K. Barbary [7]), the method behind NSC DR2, reproduces NSC DR2’s positions there to 37 mas. The injections show that every method is pulled by the crowding at that position: a Gaussian centroid by (−99, 85) mas, the PSF fit by (−43, 40) mas and the windowed centroid by (−168, 143) mas, which accounts for the difference. On 2016-08-11, NSC DR2’s positions are 171–669 mas from ours, more than any bias the injections show there; on the other nights they agree to 90 mas.

Our main test refits without each night and predicts it (Table 10). With our adopted positions the left-out nights are predicted with an rms of 37 and 22 mas in RA and Dec; with NSC DR2’s positions, 125 and 48 mas. For the earliest night, which has the least support from the others, our positions are predicted to (−32, −19) mas and NSC DR2’s to (−185, 93) mas, against a prediction uncertainty of 12 mas. All the variants give the same orbit ($a_\mathrm{bary} = 43.0910$–43.0911 au).

Positions	$\chi^{2}$/dof	rms$_{\alpha}$	rms$_{\delta}$	LOO$_{\alpha}$	LOO$_{\delta}$	$a_{\mathrm{bary}}$
		(mas)	(mas)	(mas)	(mas)	(au)
PSF + neighbors, bias-corrected (adopted)	84.7/142	34	21	37	22	43.0910
PSF + neighbors	95.8/142	37	21	43	22	43.0910
Gaussian, bias-corrected	93.4/142	42	21	46	22	43.0910
Gaussian	127.7/142	57	26	64	27	43.0911
NSC DR2 catalog where available	138.9/142	118	46	125	48	43.0911

2026 ED138 moves 1.1–2.7′′ per hour, so a timing error of one second moves it by at most 0.74 mas. The shutter-open times in the headers agree with DATE-OBS to between $-0.66$ and $1.02$ s, except for the 2023-04-07 exposure, whose header is inconsistent (OPENSHUT $-24.7$ s from DATE-OBS, and a dark time of $66.4$ s for a $90$ s exposure); its timing cannot be checked. To test the timing independently, we compared the NSC DR2 positions of 184 numbered asteroids (SkyBoT, $V < 22.5$, on 13 nights, in our exposures and in DECam exposures of the same nights nearer the ecliptic) with JPL Horizons ephemerides at our midpoints. A timing error would shift every asteroid along its track in proportion to its rate of motion (here $5$–$64^{\prime\prime}\ \mathrm{h}^{-1}$). The median along-track residual is $1 \pm3$ mas and the cross-track one $2 \pm3$ mas; the implied clock offset is $0.4 \pm0.5$ s, and a robust (Theil–Sen) slope of the along-track residual against the rate, a less precise estimate, gives $-2.9$ s. We find no timing error, and we take the larger of the two as the bound: any error is below about 3 s, a few milliarcseconds for 2026 ED138.

The object is bluer than the typical reddened star in this field, so differential chromatic refraction displaces it relative to the tie stars [20]. We estimate at most 40 mas (40 mas in $g$ at high airmass; median 4 mas), and do not correct for it.

26 of the 26 ATLAS candidates lie within $3^{\prime\prime}$ of the adopted orbit (median $0.8^{\prime\prime}$, at most $2.6^{\prime\prime}$), consistent with their astrometric precision, and so does 1 of the 3 SkyMapper candidates ($0.1^{\prime\prime}$). The other 2, on 2014-08-10 at magnitude $\approx17.1$, lie $51^{\prime\prime}$ from the orbit, whose uncertainty there is far below an arcsecond: they are a different object. The consistent candidates are probably detections of 2026 ED138; we leave them for future work.

With the adopted orbit, the position of 2026 ED138 since 1995 is predicted to better than $1.5^{\prime\prime}$. None of the 9,084,064 observations in the ITF (§ 3) lies within $10^{\prime\prime}$ of it; the closest is $4.2^{\circ}$ away.

Table 12 lists, for each DECam exposure used, the checks above, and Table 10 the orbit fits with each centroiding method. The PSF positions were tied to Gaia DR3 with the same stars as the Gaussian centroids; the windowed (SExtractor-like) centroids, computed with SEP and with the pixels of other detected sources masked, are used only for the comparison with NSC DR2.

**NOTE—**Each variant is fitted with the Rubin observations (reported uncertainties). rms: residuals of the DECam rows in RAcos $\delta$ and Dec. LOO: rms of the DECam rows of each night predicted by a fit without that night.

Note—$N_\star$: Gaia DR3 stars measured in the cutout. Header − Gaia: the image WCS’s offset from Gaia at the object (R.Acos $\delta$, Dec) and the scatter per star. NSC − Gaia: NSC DR2’s positions of the same stars (offset, scatter). Ours: leave-one-out scatter of our PSF fit. Shutter: OPENSHUT − DATE-OBS. Bias: the injection bias removed from the PSF position; T: the two templates agree (a), only one was usable (s), or they disagree (d; no correction, larger uncertainty). NSC − ours: NSC DR2’s catalog position of 2026 ED138 minus ours. O−C: residual from the adopted fit.