The Early Data Release of the Dark Energy Spectroscopic Instrument

The Dark Energy Spectroscopic Instrument (DESI) completed its 5 month Survey Validation in 2021 May. Spectra of stellar and extragalactic targets from Survey Validation constitute the first major data sample from the DESI survey. This paper describes the public release of those spectra, the catalogs of derived properties, and the intermediate data products. In total, the public release includes good-quality spectral information from 466,447 objects targeted as part of the Milky Way Survey, 428,758 as part of the Bright Galaxy Survey, 227,318 as part of the Luminous Red Galaxy sample, 437,664 as part of the Emission Line Galaxy sample, and 76,079 as part of the Quasar sample. In addition, the release includes spectral information from 137,148 objects that expand the scope beyond the primary samples as part of a series of secondary programs. Here, we describe the spectral data, data quality, data products, Large-Scale Structure science catalogs, access to the data, and references that provide relevant background to using these spectra.


INTRODUCTION
Wide-field imaging and spectroscopy enable a host of astrophysical studies that range from the largest cosmological scales to the local environment of the Milky Way galaxy.Starting in 2000, the Sloan Digital Sky Survey (SDSS; York et al. 2000) represents the largest such program.The latest public release of SDSS data, DR17, includes 5,580,057 optical and near-infrared spectra passing quality cuts (Abdurro'uf et al. 2022).Data from SDSS has been used in more than 11,200 peer-reviewed publications.1Following the precedent of SDSS, recent releases from the Dark Energy Survey (Sevilla-Noarbe et al. 2021;Abbott et al. 2021), Gaia collaboration (Gaia Collaboration et al. 2022), Hyper Suprime-Cam Subaru Strategic Program (Aihara et al. 2022), and the Galaxy And Mass Assembly team (Driver et al. 2022) represent broader efforts of wide-field survey teams to provide well-calibrated data with comprehensive documentation to the public.
The Dark Energy Spectroscopic Instrument (DESI; DESI Collaboration et al. 2016a,b) began science observations in December 2020, making it the first Stage-IV (Albrecht et al. 2006) dark energy program to begin operations.DESI will obtain spectra of stars, galaxies, and quasars over approximately 14,000 deg 2 .Data from the Milky Way Survey program (MWS; Cooper et al. 2023), comprised of more than 7 million spectroscopicallyconfirmed stars, will be used to characterize the assembly history and mass profile of our Galaxy.The extragalactic spectroscopic sample, consisting of nearly 14 million bright galaxies (BGS; Hahn et al. 2022), 7.5 million luminous red galaxies (LRG; Zhou et al. 2023), 15.5 million emission line galaxies (ELG; Raichoor et al. 2023a), and 3 million quasars (QSO; Chaussidon et al. 2023), will be used to explore the fundamental physics that governs the evolution of the Universe.The DESI sample size will be ten times larger than the totality of the SDSS spectroscopic programs for extragalactic targets.
In this paper, we describe the public release of the first sample of DESI spectroscopic data, the "Early Data Release" (EDR).The data in this release originate from the "Survey Validation" (SV) of DESI (DESI Collaboration et al. 2023) that took place between December 2020 and May 2021, prior to the start of the DESI Main Survey.The first phase of SV, "Target Selection Validation" (abbreviated SV1), was comprised of observations made to refine and validate the selection of targets for the MWS, BGS, LRG, ELG, and QSO samples (Myers et al. 2023).Compared to the DESI Main Survey, this phase used looser target selection cuts to span a larger range of observed properties and observed these targets to higher signal-to-noise (S/N).This enabled building truth samples, optimizing target selection cuts, and tuning the necessary signal-to-noise to meet the survey requirements (DESI Collaboration et al. 2023).After a brief "Operations Development" phase (SV2), DESI finished SV with the "One-Percent Survey" (SV3), which further optimized the efficiency of observing procedures and produced samples with very high fiber assignment completeness for clustering studies over an area that is approximately 1% of the final DESI Main Survey.
This paper is organized as follows.In §2, we present the DESI instrument, the design of SV observations, and a brief summary of the target classes contained in this data release.A full description of target classes can be found in Appendix A and Appendix B. In §3, we discuss the spectral processing, data quality, and data included in this release.Next, in §4, we detail the creation and uses of the large-scale structure (LSS) value-added catalogs of the One-Percent Survey that accompany this release.In §5, we describe online access to the data and tutorials with examples of working with the data.Finally, in §6, we provide a brief summary of results produced with these SV data and the plans for future releases.

DATA ACQUISITION
The DESI spectrographs were built to obtain spectra of roughly 40 million galaxies and quasars over a fiveyear period to study dark energy through measurements of large-scale structure.The maps produced with these spectroscopic samples are expected to allow volumeaveraged measurements of the baryon acoustic oscillation (BAO) feature at a precision better than 0.5% over each of the intervals 0.0 < z < 1.1, 1.1 < z < 1.9, and 1.9 < z < 3.7.These maps will also allow percentlevel precision measurements of redshift space distortions (RSD) over each interval 0.0 < z < 1.1 and 1.1 < z < 1.9.Here, we present an overview of the instrument design, observing strategy, phases of SV observing, and the SV samples that were used to inform the strategy to make these cosmological measurements.

Instrument Design
DESI requires a wide field of view that was made possible with a new prime focus corrector at the NOIR-Lab's 2 4-m Mayall telescope at Kitt Peak National Observatory in Arizona.These optics allow spectroscopy over a 0.8-meter diameter focal plane (Miller et al. 2023) located at the prime focus, corresponding to roughly 3.2 degrees on the sky and a field of view just over 8 deg 2 .Installed in this focal plane are 5,020 roboticallycontrolled fiber positioners (Silber et al. 2023), each holding a unique fiber with a core diameter of 107 µm ∼ 1.5 arcsec (Poppett et al. 2023).Twenty fibers direct light to a camera to monitor the sky brightness, while the remaining 5,000 fibers direct the light of a targeted object from the primary focus to one of ten spectrographs.The focal plane is constructed of 10 "petals", with each petal of 500 science fibers mapping to a single spectrograph that measures all 500 targets simultaneously.These spectrographs have three cameras, denoted as B (3600-5800 Å), R (5760-7620 Å), and Z (7520-9824 Å), 3 that provide a resolving power of roughly 2000 at 3600 Å, increasing to roughly 5500 at 9800 Å (Jelinsky et al. 2023).
The full system has the sensitivity to measure and resolve the [Oii] doublet down to fluxes of 8 × 10 −17 erg/s/cm 2 in effective exposure times of 1000 seconds for galaxies 0.6 < z < 1.6.Here effective ex-2 Formerly named the National Optical Astronomy Observatory (NOAO). 3Future DESI data releases may adjust the exact wavelength grid extracted from the data.
posure time corresponds to an exposure time in reference conditions -zenith, dark sky, FWHM seeing of 1.1 arcsecond, and no Galactic extinction (see §3.1.2for a summary and §4.14 of Guy et al. 2023 for details).At this effective exposure time, and accounting for increased overhead due to Galactic extinction, airmass, weather, operational overheads, and engineering downtime, the DESI instrument can be used to complete a 14,000 deg 2 survey in five years.A comprehensive description of the completed instrument can be found in DESI Collaboration et al. (2022).

Observing Strategy
DESI has five "primary" target classes (MWS, BGS, LRG, ELG, and QSO in increasing order of mean redshift), as well as many "secondary" target classes which are generally used as filler samples for fibers that cannot reach a primary target.Observations are based upon tiles, which are a given pointing of the telescope combined with assignments of each fiber to a specific target for that telescope pointing.Tiles are associated with a single survey, or phase of the DESI operations.Tiles are further grouped within a survey by their program indicating the observing conditions under which they should be observed.For example, BGS and MWS targets are assigned to "bright" program tiles, while fainter ELG, LRG, and QSO targets are assigned to "dark" program tiles.The bright vs. dark distinction is based upon survey speed, or how quickly the instrument can accumulate signal-to-noise given the current observing conditions, as estimated by the exposure time calculator (Kirkby et al. 2023).Bright tiles are observed when the survey speed is 2.5× worse than the reference conditions described in §2.1."Backup" program tiles are used when conditions are too poor for bright tiles, 12.5× worse than reference conditions, where bright stars are targeted.Multiple programs were interleaved during each survey, selected dynamically based upon current observing conditions.During SV, some bright tiles were purposefully observed under both bright and dark conditions to provide comparison datasets for quality assurance tracking.
Each program is subdivided into one or more subprograms, called fiberassign programs (data column FAPRGRM).For the majority of programs, there is only one fiberassign program of the same name, however, some programs like the SV1 survey's dark program contained multiple fiberassign programs to differentiate the various purposes of each set of tiles.More information on fiberassign programs will be described in Raichoor et al. (2023b).Table 1 lists the surveys, programs, and fiberassign programs available in the EDR, with further details in Table 5 and the Appendices.With the exception of commissioning and specifically designed tiles in a "special" survey, each survey included dark, bright, and backup programs, while Target Selection Validation (SURVEY=sv1) also includes an "other" program for tiles dedicated to secondary targets.
DESI files typically follow the convention of uppercase column names and lowercase string values, e.g.PROGRAM=dark. 4In the data files, tiles are tracked with a unique integer TILEID, and each tile is associated with specific strings for SURVEY, PROGRAM, and FAPRGRM.
Tiles overlap on the sky, enabling both greater fiber assignment completeness for dense targets such as ELGs, as well as the opportunity to observe fainter targets to higher signal-to-noise by combining observations across multiple tiles, e.g. for high redshift quasars.Each target has a unique integer TARGETID to track their observations across multiple tiles, and targeting bitmasks5 to track the reason(s) that they were selected for observation, as detailed in Myers et al. (2023), and Appendix A and Appendix B.
A given TARGETID could be assigned to a single tile; multiple tiles of the same SURVEY and PROGRAM (such as QSO targets); or multiple tiles of different SURVEYs or PROGRAMs (such as brighter LRGs on PROGRAM=dark tiles also being selected as BGS targets on PROGRAM=bright tiles).To preserve the data uniformity within a (survey, program) combination, target selection and fiber assignments within a (survey, program) are independent of whether the target is also selected and assigned in a different (survey, program), even when this results in additional observations of the same target.Similarly, these data are also processed independently such that spectra are not co-added or fit across different surveys and programs.

SV Observations
DESI Survey Validation observations began on December 14, 2020 and nominally concluded on May 13, 2021 (see DESI Collaboration et al. 2023 and§2.3 of Myers et al. 2023 for further details).An additional 22 SV-designed tiles were observed on 5 nights after the start of the Main Survey on May 14, 2021; with the final observations taking place on June 10, 2021.The Table 1.Surveys, programs, and fiberassign programs available in the EDR, in increasing order of specificity.Each fiberassign program represents a choice for how targets were selected for a given observation and under what conditions a tile should be nominally observed.See Table 5 for more details on the non-standard fiberassign programs.additional tiles were observed to improve the completeness of the One-Percent Survey areas, and are included in the EDR.Table 2 lists the number of nights, tiles, exposures, effective exposure times, and area covered by tiles for each survey included in the EDR. Figure 1 shows the number of unique tiles per night for each of the three phases of Survey Validation.Although there are distinct boundaries for when each survey began, there is an overlap in dates as incomplete tiles from a previous survey were sometimes completed after the start of the next survey.A given tile can be observed on multiple nights, and thus contributes to multiple bins, but if it was observed multiple times on a single night it is only counted once for that night in Figure 1.

SURVEY PROGRAM FAPRGRM
The covered area listed in Table 2 is simply the unique area that is overlapped by any tile in that survey, while not double-counting area covered by more than one tile.This gives a sense of the scope of the Survey Validation observations, but note that these areas are larger than the true effective area due to gaps in the focal plane coverage, disabled or broken hardware, and target assignment priorities, as will be discussed in §4.1.

Target Selection Validation
The first phase of SV observations, Target Selection Validation, optimized the DESI survey strategy and target selection algorithms.Target Selection Validation used SURVEY=sv1. 6 key product of the observations for Target Selection Validation was the calibration of effective exposure times.As stated previously, effective exposure times account for varying throughput and background and provide a standard metric of exposure depth that corresponds to an exposure time at airmass 1, zero Galactic extinction, 1.1 arcsecond FWHM seeing, and zenith dark sky.DESI uses two effective times, EFFTIME_ETC and EFFTIME_SPEC, which are used for determining when to stop an exposure of a tile at the telescope and determining when a tile has been observed enough to meet survey specifications, respectively.EFFTIME_SPEC is based on the offline spectroscopic data and will be de-scribed in §3.1.2.EFFTIME_ETC is based on active monitoring of the sky conditions and location on the sky (Kirkby et al. 2023).The sky brightness is monitored with sky monitor fibers along the outer rim of the focal plane, which send light to a dedicated imaging system that is read out regularly during the spectroscopic exposure.The image quality and sky transparency are derived from the guide focus assembly (GFA) system.For more information about these components, see DESI Collaboration et al. (2022).For the Main Survey, a calibrated algorithm called the Exposure Time Calculator (ETC) determines when the exposure is estimated to be complete (Kirkby et al. 2023).For SV1, there was no calibration of sky conditions to effective time, so a power law: t exp = t 0 X 1.25 was used.Here X is the airmass, t 0 is the nominal time, and the relation is empirically derived from BOSS/eBOSS data.These data were used to calibrate the EFFTIME_SPEC of the offline pipeline, which in turn was used to calibrate EFFTIME_ETC.
During Target Selection Validation, requested effective exposure times were increased by approximately a factor of four relative to the survey design to provide high signal-to-noise spectra.These data were typically collected over four different nights to allow tests of redshift classification using subsets of data acquired under different observing conditions.
In addition to the four-epoch strategy, Target Selection Validation observations included even deeper exposures of some tiles.The deepest tiles for each of the 5 primary DESI target classes (MWS, BGS, LRG, ELG, and QSO) are listed in Table 3.These tiles contain spectra that are much higher signal-to-noise than most data expected from the program and can therefore be used for unique studies of stellar, galaxy, or quasar astrophysics.A subset of these deep tiles (including tiles not listed in Table 3) also have visual inspections; see §3.3.6.
In total, 137.5 effective hours during SV were dedicated to 175 Target Selection Validation (SV1) tiles.

Operations Development
After Target Selection Validation, SV continued with an Operations Development phase with SURVEY=sv2, in preparation for the One-Percent Survey.The purpose of these observations was to validate the end-to-end operational procedures needed to schedule observations of targets in a tile, process those observations, identify successfully acquired redshifts, and determine which targets were completed or should be scheduled for new observations in additional overlapping tiles.A major focus of this phase was to establish the "Merged Target List" ledgers (MTLs; see §5 of Schlafly et al. 2023), which track the observational state and redshift of each target and determine whether a target requires further observations.In total, 6.4 effective hours during SV were dedicated to 39 Operations Development tiles.

One-Percent Survey
The final phase of the SV program, the One-Percent Survey, (SURVEY=sv3) was used to validate final operational procedures and compile extensive samples of sources that could be used for clustering studies.The One-Percent Survey was conducted over the period April 5 to June 10, 2021, with the vast majority of observations occurring on or before May 13, 2021.In total, the One-Percent Survey covered 20 fields, each with a rosette pattern of ∼10-11 overlapping tiles in bright time and ∼12-13 tiles in dark time, providing high fiber assignment completeness over an area of 6.48 deg 2 .Within this high-completeness area, more than 95% of MWS and ELG targets, and more than 99% of BGS, LRG, and QSO targets, received fibers.Targets covering an additional 2 deg 2 were observed with fewer visits and lower completeness in fiber assignment because of the dithering between successive observations of each field.Large-Scale Structure catalogs were created covering the entire area and are detailed in §4.
One-Percent Survey rosettes were selected to cover major datasets from other surveys, including the Cosmic Evolution Survey (COSMOS Scoville et al. 2007) 4. In total, 102.2 effective hours during SV were dedicated to 488 tiles in the One-Percent Survey (SV3).

Other SV Observations
In addition to Target Selection Validation, Operations Development, and One-Percent Survey tiles, Survey Validation observed additional tiles dedicated to secondary targets proposed by members of the DESI collaboration to extend the scientific reach of this Early Data Release.These tiles are listed in Table 5, with the target selection bits described in Appendix B. Secondary targets were also used as low-priority filler targets on other tiles, to be used in cases when an available fiber could not reach a primary target and was not needed for calibration targets (standard stars or sky locations).DESI observed 37.9 effective hours of these 13 dedicated secondary target tiles.
EDR includes one commissioning tile (TILEID=80615, SURVEY=cmx) covering M33 that was observed during the SV time period.It also includes sixteen SURVEY=special tiles used for fiber assignment testing (TILEIDs 81100-81115).Although these special tiles used dark time targets (LRG, ELG, QSO), most were observed for less than a minute of effective exposure time and only tiles 81100 and 81112 will be retained in future data releases.In total, 0.9 effective hours of exposure time were dedicated to commissioning observations and 0.3 effective hours to SURVEY=special observations.

SV Target Samples
DESI primary targets (MWS, BGS, LRG, ELG, and QSO) are selected from Data Release 9 of the Legacy Imaging Surveys (LS/DR9; Zou et al. 2017;Dey et al. 2019;Schlegel et al. 2023), while secondary targets could come from LS/DR9 or other sources.Targets identified for spectroscopic observations are recorded from the imaging data and documented for downstream redshift and clustering catalogs (see Myers et al. 2023 and Appendix A).The target selections used for the SV data in EDR are described in Allende Prieto et al. (2020) for MWS, Ruiz-Macias et al. (2020) for BGS, Zhou et al. (2020) for LRG, Raichoor et al. (2020) for ELG, and Yèche et al. (2020) for QSO.Secondary targeting programs are briefly summarized in Appendix B and references therein.These algorithms were updated based on analysis of the EDR data resulting in the DESI Main Survey final target selections documented for the MWS program in Cooper et al. (2023), BGS in Hahn et al. (2022), LRG in Zhou et al. (2023), ELG in Raichoor et al. (2023a), and QSO in Chaussidon et al. (2023).
The Early Data Release consists of 2,847,435 spectra unique to a given survey and program, when including science targets, standard stars, and sky fiber spectra.Of those, 2,757,937 are unique locations on the sky.Selecting only spectra that do not have hardware or observing flags yields 2,183,282 unique spectra.Further sub-selecting to science targets, the EDR contains 1,852,883 unique science spectra free of hardware Table 6 summarizes the number of good objects per target class in each survey as well as in the total EDR sample, with the additional restriction that the target was classified as the intended target selection type -GALAXY for BGS, ELG, and LRG samples; QSO for QSO samples; and STAR for MWS samples.Note that the sum of a column or row may not be equal to the total due to individual objects being observed in multiple surveys or individual objects being selected for multiple target classes.The numbers as a function of redshift for the full EDR sample are shown in Figure 2.This includes both primary targets that were classified as their targeted type (colored histograms), as well as confident classifications of any targets (primary or secondary) regardless of their expected target type (gray histograms).For example, the bump at z < 0.5 in the gray histogram for QSO classifications comes primarily from BGS targets that were either quasars or Active Galactic Nucleus (AGN)-like galaxies that Redrock classified as QSO instead of galaxies.Although they may be valid redshifts, these are not included in Table 6.
Figure 3 shows the density of good target redshifts on the sky for each of the three primary phases of Survey Validation -Target Selection Validation (sv1, blue), Operations Development (sv2, green), and the One-Percent Survey (sv3, orange).Target Selection Validation has many tiles distributed over the sky, while the One-Percent Survey has a larger number of tiles over a smaller area, leading to much higher good target densities (and thus much higher target completeness for those patches of sky) for the One-Percent Survey.
Metadata about the targets included in each file are recorded in the FIBERMAP Header/Data Unit extension of the FITS7 files in the EDR.These include the input photometry used for target selection,8 target selection bitmasks recording which target classes each target was selected for, and information about which fiber each tar-0.00150.0010 0.0005 0.0000 0.0005 0.0010 0.0015 0 50000 NSTAR DESI Early Data Release N(z) The reddish-brown distribution is for objects targeted to be a star and classified by Redrock to be SPECTYPE==STAR.The purple, green, and red histograms show objects targeted as BGS, ELG, and LRG respectively, and classified as a GALAXY.
The blue distribution shows objects targeted as a QSO and classified as a QSO.The gray distributions depict all objects that were classified by Redrock as a STAR, GALAXY, or QSO for the top, middle, and bottom panels respectively.*Note that the gray differs from the colored histograms because of secondary targets and other target types that were classified to a different category (e.g. a QSO target that was classified as a STAR).Also note that an object can be targeted by two galaxy target classes, and such objects will appear in both distributions.
get was assigned to and how accurately that fiber was positioned.This information is also propagated into the final redshift catalogs.The target selection bitmasks for each of the surveys included in the EDR are described in Appendix A, with further details in §2.4 of Myers et al. (2023).The spectroscopic data reduction for DESI is performed with a newly developed, Python-based pipeline.A detailed description of the DESI pipeline can be found in Guy et al. (2023); here we provide a short summary.
Spectroscopic data is transferred from the telescope at Kitt Peak in nearly real time to the National Energy Research Scientific Computing Center (NERSC) for processing and archiving.Each night the data is processed from raw images to redshifts such that tiles identified as being observed to full depth can be approved as complete and the targets entered into the MTL ledgers as being observed (see §5 of Schlafly et al. 2023).This al-Table 6.The number of "good" spectra obtained in each phase of SV, along with details for dedicated pointings (special) and commissioning (cmx).Here each target class is selected with the bitmasks for that tracer in that survey, and "good" refers to science targets that have no Redrock ZWARN bits set and whose best fitting templates are consistent with the tracer (GALAXY for BGS, ELG, and LRG targets; QSO for QSO targets, and STAR for MWS).Each row counts unique targets, but since some targets were observed under multiple surveys, the total number of unique targets is less than the sum of the rows.lows for subsequent overlapping tiles to be designed and observed in the same region of sky with new targets (and repeated high-redshift QSO targets for those identified as Lyman-α forest sources).The EDR is a re-processing of the raw data performed using a tagged version of the pipeline code.9The directory structure and filenames retain the internal name of the release, fuji, named after the mountain and signifying the sixth internal release of data.

SURVEY
The DESI instrument consists of ten petals of 500 fibers each that send light to 10 corresponding spectrographs.Each spectrograph has three arms: blue, red, and near-infrared (denoted B, R, and Z in the data files).In its nominal configuration, the instrument generates 10 4096x4096 pixel blue images and 20 4114x4128 pixel red and NIR images, with each containing the data for 500 fibers in one of the three wavelength ranges.In the afternoon before each night of observation, calibration data are acquired and processed so that they can be applied to the scientific data taken throughout the night.Twenty-five 0-second zero exposures are taken with the shutter closed to generate a master bias for the night.This bias is removed from all other calibrations and science exposures for the night.Next, a 300-second dark exposure, again with the shutters closed, is taken to identify new bad columns or pixels in the CCD and to test the nightly master bias against a template bias to determine which gives the smallest residuals in the processed image.If the template produces smaller residuals, then it is used instead of the master bias derived on that night.
Five arc exposures are then taken with Ar, Cd, Hg, Kr, Ne, and Xe arc lamps.These are used for an initial wavelength calibration for each fiber throughout the night, in addition to generating a full two-dimensional point spread function (2D PSF) model for each image to be used for extracting the spectral signal from the 2D data using an implementation of the spectroperfectionism algorithm described in Bolton & Schlegel (2010).
These exposures are also used to determine the fiber traces representing the two-dimensional extent of each fiber on the image.Finally, four sets of three flat exposures are taken with LED lamps on a white dome screen and combined to generate a per fiber flat field correction that is applied to the subsequent science exposures on the night.
Science data taken throughout the night are first preprocessed to convert analog digital units into electron counts, identify and mask cosmic rays and bad pixels, remove detector bias and correct for dark current, apply a CCD flat field correction, and estimate the perpixel variance in the image.Next, the wavelength solution derived from the arc exposures is refined using sky lines, small trace shifts are identified, and a new 2D PSF model is generated for each exposure.Counts for each fiber are then extracted, along with estimates of the variance and the resolution matrix that encapsulates the mapping between the 2D model and the 1D, uncorrelated, linear wavelength binned output spectra.The output spectra are then flat fielded using the per fiber flat field vectors derived from the flat exposures.The sky is then removed using fibers explicitly positioned to empty sky locations.These sky fibers are combined to create one high-resolution sky model for each spectrograph, before using the resolution matrix of each fiber to subtract the sky.
After sky subtraction, main sequence F stars are used for flux calibration.The selected standard star fiber spectra are fit to all 3 cameras simultaneously using theoretical models for a wide range of stellar effective temperature (T eff ), surface gravity (log 10 g), and iron abundance ([Fe/H]).The resulting fits can be used to derive the throughput of the instrument by relating the measured counts to the expected flux from photometry, which enables the generation of calibration vectors for all fibers on the petal to convert counts into fluxes in units of 10 −17 erg/s/cm 2 / Å.To improve signal-tonoise, if multiple exposures of a single tile are observed on a night, then all of the exposures are used to jointly model standard stars, before each exposure is used independently to generate a calibration vector for that exposure.The per-star calibration is used to derive the observed flux compared to the expected flux from imaging photometry, and 3σ outliers are rejected compared to the scatter observed for all standard stars across all spectrographs.The final calibration vector per camera is an average over the remaining standard stars on that camera.Finally, a cross-talk correction is applied to account for the fact that the 2D PSF of each fiber extends into the region of its neighbors before writing out the per-exposure per-fiber calibrated fluxes and variances.
During SV1, tiles were allocated with 80 sky fibers and a goal of 20 standard stars per petal.Using these data, tests were performed to identify the minimum number of sky fibers and standard stars that could be used per petal before degradation in sky subtraction or flux calibration would be observed.It was shown that as few as 20 sky fibers per petal and 10 standard stars per petal were sufficient to maintain ELG redshift efficiency, which was used as a proxy for measuring the impact of sky subtraction residuals on the resulting spectra.For SV3 and the Main Survey, DESI requires a minimum of 40 sky fibers and 10 standard star fibers per petal to be conservative.DESI tiles are observed to have roughly equal effective exposure time rather than raw exposure time.This results in varying raw exposure times and a varying number of exposures acquired for a given tile.For the EDR the median number of exposures for a tile is 1, with a mean of ∼ 3.4, and a maximum of 30 for deep SV1 BGS tile 80613.

Spectroscopic Effective Exposure Time
The Exposure Time Calculator (ETC) provides a realtime estimated effective exposure time to determine when to end observations of a given tile, using information available during an exposure.However, a more ideal quantity would be an effective time that incorporates the instrumental effects on the spectra themselves and how a particular target class might be impacted by such effects.EFFTIME_SPEC, which is derived from the spectroscopic data themselves each night as the data are acquired, was designed to incorporate these features.This quantity is what is used by survey operations to determine if observations of a tile have achieved enough effective time to be marked as complete for the designated PROGRAM.
The spectroscopic effective time estimate is detailed in §4.14 of Guy et al. (2023).First, a template signal-tonoise ratio squared (TSNR 2 ) is defined as the mean of the squared signal-to-noise for an ensemble of templates over a representative redshift range, which incorporates instrument and observational quantities such as fiber aperture losses, detector read noise, and sky residuals.Finally, the mean TSNR 2 value for all fibers is multiplied by a constant to get the EFFTIME_SPEC for the exposure.Each tracer class has a different morphology and redshift range, so this is done for each separately, with each target class having a different constant of proportionality.The constants are empirically fit to equal the actual exposure time when observing in nominal conditions at zenith with dark sky, ideal transparency, no Galactic extinction, and median seeing of 1.1 .This is done for all target classes, but the reported EFFTIME_SPEC for dark time is based on the LRG value, while the bright time EFFTIME_SPEC is based on the BGS value.

Redshift Fitting and Classifications
Spectral classifications and redshifts are measured using the Redrock software package 10 ( Bailey et al. 2023).Redrock performs a χ 2 vs. redshift scan, fitting a set of Principal Component Analysis (PCA) templates to every target at every redshift.The fit with the lowest χ 2 determines the spectral classification (SPECTYPE=GALAXY, QSO, or STAR for DESI) and redshift.Each set of templates is fit on all spectra regardless of target selection type; including standard stars, sky-subtracted sky fibers, and spectra from nonfunctioning positioners that were not pointing at any known target.This procedure is similar to the method used in SDSS/BOSS (Bolton et al. 2012), with improvements to the underlying PCA templates, more exact error propagation, and more detailed modeling of the per-wavelength per-fiber spectral resolution.Although Redrock was originally developed for DESI, it was previously used by eBOSS for their final cosmology analyses (Ross et al. 2020).
The primary outputs from Redrock are the redshift (Z), redshift uncertainty (ZERR), spectral classification (SPECTYPE), a warning bitmask (ZWARN), the coefficients for the linear combination of the best fitting templates (COEFF), the χ 2 of the fit (CHI2), and the value ∆χ 2 giving the difference between the best fit χ 2 and that of the second best fit (DELTACHI2).Larger values of DELTACHI2 represent greater statistical confidence that the best fit is correct.
In addition to classifications and redshifts, Redrock includes a per-object ZWARN bitmask indicating if there are any known problems with the data or the fit.ZWARN==0 means that there are no known problems, and non-zero values encode the reasons for possible problems.The meaning of the individual bits are documented in the Redrock code at https://github.com/desihub/redrock/blob/0.15.4/py/redrock/zwarning.py#L14 and further described in Bailey et al. (2023).Some bits record problems with the input spectrum, e.g., that all flux values were masked, while other bits record problems with the Redrock fit itself, e.g., a failed parabola fit to the χ 2 vs. z minimum.Most analyses should require ZWARN==0 to obtain good quality results, which implicitly include the requirement that DELTACHI2 > 9. To obtain a purer sample of more confident redshifts, some analyses may place a higher cut on DELTACHI2.

Post Redshift Value-Added Processing
After redshift fitting, three additional steps were included in the pipeline to post-process the data and derive value-added quantities such as line identifications, line flux estimates, refined redshift fits, and further quasar classifications.
In addition to the Redrock results, emlinefit provides simple fits of the major galaxy emission lines.The approach is purposefully simple; a more refined approach is performed with e.g.Fastspecfit.11The primary motivation is to fit the [O II] doublet, and use it to identify reliable redshift measurements for ELG spectra (see §4.2.1 and Equation 2).For convenience, we also provide fits for the [O III] doublet (λλ 4960,5007 Å), and for the H α , H β , H γ , and H δ lines.All lines are computed for all spectra.However, detailed studies should be performed prior to using quantities other than [O II], or [O II] on non-ELG targets, to assess their accuracy.The fits are simple Gaussian fits at the expected position based on the Redrock best-fit results; the fitted flux is not forced to be positive, so negative values can be reported.The continuum is estimated from the wavelengths 200 Å (in rest-frame) around the emission line (bluewards for the [O II] doublet).For the [O II] doublet, the line ratio is left free during the fit; for the [O III] doublet, it is fixed.For more details, see §7 of Raichoor et al. (2023a).
To improve the classification and the redshift determination for quasars, we also provide results from two additional codes: an Mg II broadband fitter and a neural network classifier, QuasarNET (Busca & Balland 2018;Farr et al. 2020).The Mg II fitter aims to classify spectra that exhibit a broad Mg II emission line as QSO.The algorithm determines the width of the Mg II emission by fitting a Gaussian in a 250 Å window (observer-frame) centered at the position of the Mg II line given by the redshift identified by Redrock.For the QSO classification, the Mg II emission line is considered broad if the improvement of χ 2 is better than 16, the width of the Gaussian is greater than 10 Å, and the significance of the amplitude of the Gaussian is greater than 3.
Additionally, we run QuasarNET on all the targets.QuasarNET is a deep convolutional neural network (CNN) classifier designed explicitly to identify quasars and their redshifts.The input power spectrum is reduced by four layers of convolutions and is then passed to a fifth, fully connected layer before feeding into six line finder units: one for Lyα, C IV, C II, Mg II, Hα, and Hβ.Each line finder unit consists of a fully con-nected layer trained to identify a particular emission line.The output of each unit is a confidence level (between 0 and 1) to have found the desired line and the redshift at which it was found.The DESI large-scale structure analyses require at least one emission line to have a confidence level above 0.95 for a spectrum to be considered a QSO.For each target identified by Quasar-NET to be a QSO, Redrock is rerun using only QSO templates and a tophat redshift prior of ±0.05 to determine the final redshift.

Suggested Quality Cuts
The choice of a 'good' redshift is subjective and depends on the individual science case in question.In this paper, unless stated differently, we have elected to restrict to spectra with no hardware, observing, or redshift fitting flags (ZWARN==0).This is our generic recommendation, where some may choose to relax restrictions for specific bits if an analysis is robust to the implications of including such data, and some may choose to use additional selection criteria such as a cut on DELTACHI2 or e.g.TSNR2_LRG.The large-scale structure analyses within DESI use cuts that are generally more stringent than this, as outlined in §4.2.2.More details for each target class are available in the references cited in that section.
A less strict criterion would be to restrict based on the co-added spectrum's fiberstatus, COADD_FIBERSTATUS, which encapsulates hardware and observing issues for all input data for that spectrum, but does not depend on redshift fitting.The bits are defined in the desispec.maskbitscode.12Note that ZWARN includes a bit which is false if COADD_FIBERSTATUS equals 0 or 8, where 0 signifies no issues and 2 3 = 8 corresponds to a positioner that had a restricted range but could still reach the target location.Therefore selecting ZWARN==0 implies selecting COADD_FIBERSTATUS ∈ [0, 8] in addition to the redshift fitting flags.

Data Products
Full details of the directory organization and file formats in the EDR are given in the DESI Data Model at https://desidatamodel.readthedocs.io.The directory structure is summarized in Table 7.The following subsections provide a conceptual overview of the structure of the available data, starting from the root directory of the EDR; see §5 for methods to access these data.

Spectroscopic Data Processing Runs
Production data processing runs are alphabetically named after mountains and contained under spectro/redux/.Each production run uses a defined set of input data processed with a fixed set of tagged software.The EDR contains a single production named "Fuji", available under spectro/redux/fuji/.Future data releases will contain one or more production runs, differing by the input raw data, the software tags used, or both.
In the top-level production directory, spectro/redux/fuji/tiles-fuji.fitscontains a catalog of all DESI tiles included in Fuji.This can be used for a quick assessment of the footprint of the available DESI data and to filter available tiles by SURVEY and PROGRAM.Since tiles may be observed on multiple exposures spanning multiple nights, more detailed per-exposure information is available in exposures-fuji.fitsif needed for time-domain studies or systematics comparisons of data on different nights.

Spectra, Coadds, Classifications, and Redshifts
Spectra, coadditions (coadds) of those spectra, and classifications and redshifts fit to those coadds are available under two broad groups: per-tile and full-depth.Tile-based spectra under spectro/redux/fuji/tiles/ combine information across multiple exposures of the same tile, but not across different tiles even if the same target was observed on multiple tiles (such as high redshift quasars).Full-depth coadds combine exposures for targets on a given HEALPix (Górski et al. 2005) pixel of sky, including combining data across tiles if the same target was observed on multiple tiles.These coadds are referred to as "healpix" coadds and redshifts since they are stored in files based on HEALPix pixel number (nested scheme, NSIDE=64) under spectro/redux/fuji/healpix/.
However, even in the healpix case, data are not combined across surveys (e.g.sv1, sv2, sv3) and programs (e.g.dark, bright, backup) in order to prioritize the uniformity of each (survey, program) combination.We anticipate that the tile-based spectra will be of primary interest to analyses of Target Selection Validation (SURVEY=sv1), while the HEALPix-based full-depth spectra will be used more for the overlapping rosettes of the One-Percent Survey (SURVEY=sv3) and future releases of DESI Main Survey data.
Tile-based spectra, coadds, classifications, and redshifts come in additional subgroupings under spectro/redux/fuji/tiles/. The cumulative/ directory tree contains all data for each tile, coadded across exposures and nights.The pernight/ directory tree The perexp/ directory contains classifications and redshift fits to individual exposures to explore performance and reproducibility at even lower signal-to-noise.Additional custom coadds, classifications, and redshift fits are available for selected tiles to match the expected depth of the Main Survey (1x depth), four times the expected Main Survey depth (4x depth), and coadds using only data from poor observing conditions (lowspeed).If a tile was only observed on a single exposure on a single night, the cumulative/, pernight/, and perexp/ outputs are identical, but they are still kept in all three directories so that each can be used independently.
Future data releases will continue to support tiles/cumulative/ and healpix/, but other groupings of tiles-based spectra will not be included for Main Survey data and may be dropped from reprocessing runs of the SV data.

Redshift Catalogs
Redshift and classification catalogs for individual tiles and healpix are available in the same directories as the spectra and coadds to which they were fit.For convenience, these catalogs are also combined across the thousands of individual files into stacked redshift catalogs in spectro/redux/fuji/zcatalog/.Like the spectra and coadds, these come in multiple groups, e.g.combining all of the cumulative tile-based redshifts for a given (survey, program) into a single file, with different files for different (survey, program) combinations.
For analyses that simply want the recommended "best" redshift for a given target regardless of the DESI-specific (survey, program), zall-pix-fuji.fitscombines all the HEALPix-based redshifts across all programs into a single file, with a ZCAT PRIMARY boolean column indicating which row is considered the best redshift for each target.This uses the code desispec.zcatalog.find_primary_spectra,13which could also be used by any analysis to subselect multiplyobserved targets from a custom selection of spectra to determine the recommended redshift.It first prioritizes results with Redrock ZWARN==0, then sorts by the LRGoptimized template signal-to-noise TSNR2 LRG, though users can specify a different secondary sort column.A description of the template signal-to-noise ratio can be found in §3.1.2with further details in §4.14 of Guy et al. (2023).
Similarly, zall-tilecumulative-fuji.fits provides all cumulative tile-based redshifts across surveys and programs, with ZCAT PRIMARY indicating the recommended best single tile-based redshift per target.

Target Catalogs
Target catalogs used as input for DESI observations were previously published under the "Early Target Selection" release available at https://data.desi.lbl.gov/public/ets/ and described in Myers et al. (2023).Although the EDR does not include any new target selection catalogs, edr/target/ links to the same directory tree as ets/target/ so that the EDR can be used as a self-contained release including target selection infor-mation, without having to combine information across releases.

Fiber Assignment Catalogs
Fiber assignment is the process of assigning individual targets to individual fibers.In DESI, the assignment of targets to fibers for a given tile is designed by the Fiberassign program, as described in Raichoor et al. (2023b).The output files, called fiberassign files, that detail the fiber assignments used in the EDR can be found in target/fiberassign/tiles/tags/0.5/.Within each fiberassign file, the FIBERASSIGN table contains the mapping of assigned TARGETID to FIBER and LOCATION; 14 the TARGET RA and TARGET DEC coordinates and proper motions PMRA, PMDEC, and REF EPOCH; which are in the International Celestial Reference System (ICRS) tied to Gaia (Gaia Collaboration et al. 2022); the input photometry, object shapes, and quality flags used for primary target selection (Myers et al. 2023); 15 and the targeting bitmasks described in sections A.1, A.2, and A.3.
In addition to the target-to-fiber assignments, each file contains all possible assignments in the POTENTIAL ASSIGNMENTS and TARGETS extensions.The TARGETS table has one row per TARGETID and includes the targeting bitmasks for each potential target.The POTENTIAL ASSIGNMENTS table only has TARGETID, FIBER, and LOCATION columns; providing the mapping of which fibers could reach that target, with potentially multiple rows per TARGETID.Although fiberassign files contain the input photometry for targets that were assigned, they do not include full photometric information for unassigned targets covered by each tile, in order to keep the files to a manageable size.If photometric information is needed for unassigned targets, the TARGETIDs must be cross-matched back to the input targeting catalogs.This is most easily done by using the "LS/DR9 Photometry" Value Added Catalog included in the EDR and discussed in section 3.3.6,which combines and standardizes the information from the multiple input targeting catalogs.
Each raw data exposure directory also contains a copy of the fiberassign file that was used at the tele-14 FIBER [0-4999] tracks the position of the spectra on the spectrographs, with the spectrograph number = petal number = int(FIBER/500).LOCATION tracks the position of the fiber on the focal plane which is purposefully randomized with respect to FIBER to break degeneracies between systematics related to position on the focal plane vs. position on the spectrographs.Although there is a fixed 1:1 mapping between FIBER and LOCATION, both are recorded for convenience. 15Secondary targets are allowed to come from any data source, and thus their exact input selection parameters are not tracked here.
scope at the time of observation.For some tiles, the photometry for secondary targets was incorrect, and this was corrected post-facto in the files under target/fiberassign/tiles/tags/0.5/,which should be considered the most correct reference version.This was only done for catalog columns that were not actually used by observations.Spectroscopic data processing used these files to supersede the raw data versions when propagating the information downstream via the FIBERMAP extensions.

Value Added Catalogs
DESI data releases will include "Value Added Catalogs" (VACs), which are additional data products contributed by the DESI science collaboration.VACs are built upon the core data products (spectra, classifications, redshifts) from this spectroscopic data release.
EDR will include a set of VACs with the initial release, but will also add additional VACs based upon EDR when they become available from the DESI science collaboration in the future.The website https: //data.desi.lbl.gov/doc/vac/ will be kept up to date with details of the available VACs, including documentation and references to relevant journal articles.The files can be accessed in the same manner as the EDR data under the vac/ subdirectory.In the rest of this section, we present two VACs in the EDR that are broadly applicable to many analyses using the EDR data -the LS/DR9 Photometry VAC and the Survey Validation Visual Inspection VAC.The Large-scale Structure (LSS) Catalogs VAC used for DESI LSS analyses will also be described in §4.
LS/DR9 Photometry VAC: This VAC delivers merged targeting catalogs (targetphot) from DESI target selection (Myers et al. 2023) and Tractor16 catalog photometry (tractorphot) from the DESI Legacy Imaging Surveys Data Release 9 (LS/DR9; Dey et al. 2019) 17 for all observed and potential targets (excluding sky fibers) in the EDR.The observed targets in this VAC correspond to objects with at least one spectrum in the EDR, while the potential targets are the targets that DESI could have observed in a given fiber assignment configuration (including the objects which were actually observed). 18he construction and organization of the LS/DR9 VAC is fully documented at https://github.com/moustakas/desi-photometry; here, we briefly describe its contents.
The LS/DR9 VAC includes tractorphot catalogs, which contain Tractor catalog photometry for every unique target in the targetphot catalogs.These catalogs are "value-added" compared to the information in catalogs described in §3.3.4 in two key ways.First, the tractorphot catalogs contain all the photometric quantities measured by Tractor in the LS/DR9, not just the measurements included in the light-weight sweep catalogs used to select DESI targets (see also Myers et al. 2023).Second, the tractorphot catalogs include photometry for targets which were not necessarily targeted from the LS/DR9, such as secondary targets and targets of opportunity, by finding the LS/DR9 object within 1 of the observed (or potential) DESI target.
Survey Validation Visual Inspection VAC: During the Survey Validation period, in order to validate the performance of the DESI pipeline and assist the target selections of the Main Survey, DESI members visually inspected (VI'ed) the deep coadded spectra of 16,594 galaxy targets; including 2,718 BGS; 3,561 LRG; and 10,315 ELG targets; in addition to 5,496 QSO targets.Each spectrum had at least two inspectors.Each inspector reported the VI redshift, the redshift quality, the source type, issues observed in the spectrum, and any extra comments for each spectrum.The results from the inspectors were combined and reconciled by the VI chairperson if needed.This VI information was used to quantify the performance of the DESI operation and validate the design of the survey.The details of the compilation of the catalogs and the corresponding analyses are summarized in Lan et al. (2023) for galaxies and Alexander et al. (2023) for quasars.
These VI catalogs are provided as a VAC in the EDR.The catalogs are organized based on the target types.For galaxies, we provide the catalogs for BGS, LRG, and ELG separately.For QSO, we provide two catalogs: the quasar-survey deep-field VI catalog analogous to that provided for the galaxies, and the missed QSO catalog from a sparse VI campaign (see Tables 1 and 4 and §2.3 of Alexander et al. 2023 for details).Each catalog contains target information, including TARGETID, TILEID, FIBER, TARGET RA, and TARGET DEC; and VI information, including the VI redshift (VI Z), the redshift quality (VI QUALITY), and the type of the source (VI SPECTYPE).Note that we do not include issues or comments reported by the inspectors since most of them reflect the status of the spectra processed by an early version of the pipeline, rather than the improved pipeline used for the EDR.With the target information, one can crossmatch the VI catalogs with other catalogs in the EDR and obtain information on the sources such as the photometric properties as well as the redshift information from the DESI pipeline.We recommend using VI QUALITY ≥ 2.5 as a selection criterion for sources with confident VI redshifts; see section 2 of Lan et al. (2023) and section 2.2 of Alexander et al. (2023) for details of the VI QUALITY flags within the context of the galaxy and quasar catalogs, respectively.There are 2,640 BGS; 3,513 LRG; 7,856 ELG; and 4,890 QSO targets with any VI classification and VI QUALITY ≥ 2.5.In total there are 14,939 objects identified as a VI SPECTYPE==GALAXY; 3,182 objects as a QSO; and 778 as a STAR with VI QUALITY ≥ 2.5.

Other Files
In addition to the high-level user-facing data products, DESI data releases also contain raw data and intermediate data products.
The original raw data are available in spectro/data/NIGHT/EXPID/ subdirectories where NIGHT is the YEARMMDD date of sunset,19 and EXPID is the zero-padded 8-digit monotonically increasing exposure identification number.These directories contain the original fiber assignment observing request, raw data from the spectrographs, guiders, fiber view camera, and sky monitors, and fits to those data performed as part of fiber positioning and field acquisition.For discussion of the instrument components, see DESI Collaboration et al. (2022), and for discussion of the survey operations, see Schlafly et al. (2023).
For completeness, the EDR contains all inputs used by the spectroscopic processing pipeline, including CCD calibration files in spectro/desi spectro calib/ and spectro/desi spectro dark/; stellar templates used to fit standard stars in spectro/templates/basis templates/; and survey progress bookkeeping files in survey/ops/ used to track if a tile is "done" and should be included in a release.

Known Issues
While working with the internal pre-release of EDR, the DESI collaboration has identified several issues with the data produced by the pipeline, which we report here for completeness.
• Redrock templates do not include Active Galactic Nucleus (AGN)-like galaxies with a mixture of broad and narrow lines.As a result, these types of galaxies are often fit equally well (or equally poorly) with either GALAXY or QSO templates at the same redshift, which can also trigger ZWARN bit 2 (value 2 2 = 4) for LOW_DELTACHI2 since the χ 2 difference between the two fits is small, indicating an ambiguous answer.
• There are cases where Redrock is overconfident and reports ZWARN==0, i.e. no known problems, even though the fit is incorrect.This can include unphysical fits due to the over-flexibility of PCA template linear combinations.This is particularly true for sky fibers which have a higher fraction of ZWARN==0 than would be expected from purely random fluctuations.Users should be especially cautious in any search for serendipitous targets in nominally blank sky fibers.
• The Redrock galaxy fits extend to redshift z = 1.7, though the range 1.6 < z < 1.63 is only constrained by the [O II] doublet in the midst of significant sky background while 1.63 < z < 1.7 has no major emission line coverage.Thus 1.6 < z < 1.7 is particularly susceptible to unphysical fits.This was the motivation for the LSS catalogs to only consider galaxies with z < 1.6 (see §4).
• Negative TARGETIDs indicate positioners that were non-functional and thus were not pointing at a known science target.Although these are unique within a given TILEID, they are not unique values across different TILEIDs.Since these are not science targets, most users can discard any negative TARGETID targets.This has been fixed for Main Survey non-functional positioners in future data releases, where negative TARGETIDs will be unique.
• For most of the tiles in Target Selection Validation, proper-motion corrections were applied in Fiberassign when the tile was designed.values.The information is correct and consistent with the photometry, however.
• In the coadded FIBERMAP tables, 0.03% of targets incorrectly have COADD FIBERSTATUS==0 even though all of their data are masked.These result in ZWARN = 0 in the redshift fits, but quality cuts based solely upon COADD FIBERSTATUS have a tiny amount of contamination.Additional known issues and clarifications will be documented at https://data.desi.lbl.gov/doc/releases/edr when needed.

ONE-PERCENT SURVEY LSS CATALOGS
DESI creates large-scale structure (LSS) catalogs from its data in order to facilitate clustering measurements.The overall process is similar to that applied to SDSS (most recently eBOSS; Ross et al. 2020).We determine the area on the sky where good observations were possible for each tracer, applying criteria on the DESI data to select reliable redshifts, and provide weights that correct for variations in observing completeness.The end results are data and matched random catalogs, split into the various relevant DESI tracer classes, that can be passed directly to any common software for calculating redshift-space clustering measurements.
These catalogs for the One-Percent Survey are ideal for studying small-scale clustering, as the tiling strategy makes them highly complete for all tracer types (DESI Collaboration et al. 2023).Some studies using the results derived from these catalogs include Gao et al. ( 2023 (2023).The catalog files are available in the EDR at vac/edr/lss/v2.0;see §5 for data access details.

Gathering Assignment and Observation Information
The DESI LSS catalogs begin by gathering the information that describes where and what on the sky DESI could observe.Two fundamental pieces of this information were generated by the DESI targeting team (Myers et al. 2023).These are: 1) the data chosen ('targeted') for spectroscopic follow-up, including photometric properties of the targets and meta-data related to their observation (hereafter simply 'data') and 2) a uniform random distribution of points on the sky occupying the area covered by Legacy Survey imaging that also includes the meta-data related to the photometric observations at the given location (hereafter 'randoms'; see §4.5 of Myers et al. 2023).Both the data and randoms were given a unique identifier, TARGETID, by the DESI targeting team and this identifier is used to match between relevant data files in all relevant cases.We use 18 random files (more are available), each with a density 2500/deg 2 .The 2500/deg 2 is convenient, as it allows for quick determination of footprint area after various cuts.
With these data and randoms, we first track the locations on the sky where DESI observations could have happened.We do this using the outputs of the DESI Fiberassign software (Raichoor et al. 2023b).As described in §3.3.5, prior to the observation of each DESI tile, targets are processed through Fiberassign in order to assign targets to particular fibers (with each unique fiber corresponding to a unique robotic positioner and a unique location in the spectrograph CCDs).In addition to the particular assignment, the information on all potential assignments is also stored for each tile.Further, all settings used when running the Fiberassign software are stored so that the assignments can be reproduced.This means that we can also run the randoms through Fiberassign, with the matched settings.The positions of the randoms in the potential assignments thus provide a superset of the geometric area observable by DESI on every tile.We thus concatenate the potential assignment information across all tiles for both data and randoms. 21 The total set of concatenated potential assignment information includes many duplicated targets.In the One-Percent Survey, each location on the sky was potentially covered 13 times.Further, many sky locations are accessible to more than one robotic positioner.Initially, we keep all of the information on repeated potential assignments.Downstream, many will be vetoed, e.g., due to hardware performance or low data quality.At this 21 They are available in vac/edr/lss/v2.0/potential_assignments.  4).
stage, we also match to the information on redshifts in the cumulative tile-based redshift catalog provided in the EDR (see §3.3.2).For the data, we match not just to the target, but to the particular tile and fiber on which it was observed. 22Many rows for the data will have no match and the corresponding entries will simply be null.For the randoms, we match in order to obtain the metadata related to the particular spectrum.We, therefore, match only to the tile and fiber and will have a match for all random targets. 23 The information on the number of overlapping tiles allows us to divide the observed area by coverage.The covered area listed in Table 2 is simply the unique area that is overlapped by any tile in that survey, but it does not include the detailed accounting for focal plane geometry, disabled/broken hardware, or higher priority targets blocking lower priority targets from being observed.Using data and randoms and repeated realizations of fiber assignment provides a much more accurate and geometrically detailed measurement of the true coverage.
Fig. 4 displays this information for rosette number 1 (SV3 R1 in Table 4) of the bright time program.Each point is a BGS target at a location covered at least once by the One-Percent Survey.One can see that some areas were covered up to 11 times.For dark time observations, 22 The particular columns used for the match are TARGETID,TILEID,LOCATION; LOCATION refers to the robotic positioner, which corresponds to a unique fiber. 23The files with matches to spectroscopic information are available at vac/edr/lss/v2.0/inputs_wspec.
areas were covered up to 13 times.This dense coverage provides highly complete samples.
In order to jointly model the completeness of all targets with respect to each other, we simulate the observation of targets in the One-Percent Survey using multiple realizations of their assignment histories.The process and its results are described in more detail in Lasker et al. (2023).Briefly, all DESI targets were initially assigned a random SUBPRIORITY, which is used to determine which target gets assigned a fiber when available targets have the same overall priority (see Raichoor et al. 2023b;Myers et al. 2023).We create 128 alternative assignment histories by randomly shuffling the subpriorities 128 times. 24For each of the 128 realizations, the One-Percent Survey observations are simulated by following the same order of tile observations and targeting feedback loop. 25From these simulations, we package the results as a bit value for each target that encodes the realization numbers it was observed in.From these bit values, one can determine the individual and pairwise probabilities required to obtain unbiased clustering statistics (Bianchi & Percival 2017).

LSS Catalogs
The LSS catalogs are cut (subselected down) to unique TARGETID per supported tracer type.They come in two flavors.The 'full' catalogs contain an entry for all reachable targets, whether or not they were observed26 .They also include all columns believed to possibly be relevant.The 'clustering' catalogs cut to good spectroscopic observations and the redshift range intended for clustering analysis27 .They include weights to account for variations in the selection function and only include the columns required to calculate clustering statistics.
Catalogs were created for the four extragalactic DESI target types: BGS, LRG, ELG, and QSO.For all except QSO, catalogs are produced for additional sub-type definitions.In the cases where the sub-type corresponds to a bitname from SV3 targeting (see Table 15 and surrounding text), we use that name.The additional LRG selection, named LRG main, keeps only targets that satisfy the Main Survey selection (see Zhou et al. 2023 for details).The ELG sample is cut in three additional ways.First, ELG HIP contains only the ∼75% of the ELG sample assigned higher priority (see Raichoor et al. 2023a for more details).Then, for each of ELG and ELG HIP, we also remove QSO targets.QSO targets have the highest priority and it may, therefore, be useful to treat any targets that satisfy both the QSO and ELG selections within the QSO analysis.Finally, there are two BGS samples: BGS ANY and BGS BRIGHT.BGS ANY is the combination of both the BGS BRIGHT and BGS FAINT BGS selections.See Hahn et al. (2022) for more details.A summary of the statistics for all tracer types is given in Table 8.

Full Catalogs
Starting from the compilation of all potential observations, the first step in creating the full catalogs for the data is to cut to targets of a given target type.We then cut to unique targets.We cannot do so randomly, as, e.g., we must keep the cases that have been observed.We define the following boolean quantities: • H good : The particular fiber on the given tile was observed with good hardware.
• L fa : The particular target, fiber, and tile was assigned and observed.
• T fa : The particular fiber and tile (but not necessarily target) was assigned and observed.
• S good : The particular fiber and tile (but not necessarily target) was determined by the spectroscopic pipeline to have a template squared signal-to-noise ratio (TSNR2, referred to here as S ratio ) above the vetoing threshold (defined below).
We clip S ratio to be within the range (0,200) and then these quantities are combined to create a value to sort by v sort = L fa S good H good (1 + S ratio ) + T fa H good + H good .
(1) We sorted by this v sort in ascending order and then cut to unique targets by selecting the last entry for each unique target.After cutting to unique targets, we join to the redshift determined in the HEALPix-based redshift catalog and use Z HP as the column name.
For ELG catalogs, we join the information on the [O II] emission line fits, produced with the spectroscopic release using the HEALPix-based coadd.We combine the information on [O II] flux and its inverse variance to obtain the signal-to-noise ratio of the [O II] flux for each observed spectrum, which we denote S [O II] .Following Raichoor et al. (2023a), this information is combined with the ∆χ 2 obtained from the redshift fitting pipeline between the best and next-best-fit redshifts (the Table 8.Statistics for each of the DESI tracer types for which One-Percent Survey LSS catalogs were created.We list the number of good redshifts included, the redshift range we included them from, the sky area occupied, and the observational completeness within that area.The area is slightly different for different tracer types due to priority vetoes (e.g., a QSO target can remove sky area from lower priority samples).The completeness listed is the number of targets observed divided by the number of targets within the entire observable area.The completeness can be increased by cutting on the rosette radius (see text and Fig. 5).

Tracer
For quasars, we join to the HEALPix-based quasar catalog, which contains extra information related to classifying the spectra as QSO or not and improved redshift estimates.These catalogs will be released as a VAC and will be fully documented in Canning et al. (2023).The fiducial pipeline estimates of the redshifts are replaced by those from the quasar catalog. 28The flavor of the quasar catalog that we use is the one that only contains entries for objects believed to be quasars. 29Thus, there will be many more rows with null entries for the quasar information than for those with Redrock information.
For randoms, we must also cut to unique TARGETID and we do so separately for each tracer type, as the tracer information is included in the sort.We also apply the imaging mask bits that were applied to the target samples.These are Legacy Survey bits 30 1 and 13 applied to bright time targets and additionally bit 12 for 28 We replace the original Z HP column with that from the quasar catalog and rename it Z RR (to indicate 'Redrock').ZERR from the quasar catalog is renamed as ZERR QF and ZERR remains the ZERR from Redrock. 29We use the file: QSO cat fuji sv3 dark healpix only qso targets.fits.The quasar catalog will also provide a flavor that includes the diagnostic information for all targets. 30https://www.legacysurvey.org/dr9/bitmasks/#maskbitsdark time targets. 31In addition to the boolean quantities used to sort the data, we use: • P good : The PRIORITY of the target that was assigned at the given tile and fiber was less than or equal to the PRIORITY of the given target class.
• Z poss : The tile and fiber was either assigned to a target of the given target class or no unassigned targets of the given target class were reachable by the fiber on this tile.
The randoms are then sorted by and we cut to unique random points by selecting the highest v sort value for each.The final step for the 'full' catalogs is to apply vetoes.32All of the boolean columns defined above for data and randoms must be True to pass the veto.Note that the combination of the P good and Z poss criteria act as a priority veto mask.Thus, the area occupied by the lowest priority targets (as traced by the number of randoms) will be less than the highest priority ones.
The sky area occupied for each target class in the One-Percent Survey is given in Table 8.The difference between the highest (QSO) and lowest (ELGnotqso) priority targets is only 7% due to the survey strategy to achieve high coverage.The ELG catalogs with QSO targets removed have a smaller area than those including the QSO targets because the priority used for the P good determination is that of QSO when they are included but that of ELG HIP targets when they are not.Corre-  4) from the DESI One-Percent Survey.
spondingly, there is a small increase in the completeness of the ELG samples with the QSO targets removed, as the area removed was at sky locations where only QSO targets were observable.
Finally, additional vetoes are applied based on the imaging data.For LRGs, we apply a custom mask, described in Zhou et al. (2023).For the other tracers, Legacy Survey bit 11 is always used.For QSO, bits 8,9 are also applied (these are specific to WISE data).
Given the vetoed catalogs, we determine a completeness both per fiber, C fib , and per unique tile grouping, C tile .Per fiber is simply the inverse of the number of data points at a given tile and fiber. 33The values of C fib are similar to the P obs value obtained from 128 alternative assignment histories.The completeness per tile grouping is simply the observed number divided by the total number within each tile group, i.e., C tile = N observed /N total .As an example, Fig. 5 shows the completeness, C tile , of ELG targets in rosette number 1.The pattern observed there is typical of all rosettes.The ELG completeness in the plotted rosette is 87% and across all rosettes it is 86%.The ELG sample has the lowest completeness, as they had the lowest target priority.The completeness decreases towards the edges and the center, as the total number of overlapping tiles decreases in these regions.The completeness increases to 95% if one cuts to 0.2 < r rosette < 1.45, 33 In the catalogs, the column is FRACZ TILELOCID.
where r rosette is the angular distance from the center of the rosette, in degrees. 34

Clustering Catalogs
For both data and randoms, the clustering files take the full files and reduce them to a subset of columns necessary for calculating 2-point statistics.The full data file is cut to only those objects with good redshifts and weights to account for selection function variations.
The catalogs are provided for the 'S' (DECaLS) and 'N' (BASS/MzLS) photometric regions (see §4.1.3 of Myers et al. 2023).Given that the photometry is different because different cameras/filters were used for each, we expect at least slight differences in the selection function between the two regions and thus will always estimate it separately.The area in the S and N regions is nearly identical for the One-Percent Survey.The difference varies from the S region being less than one percent greater for the ELG samples to four percent greater for the LRG sample. 35 Only data with good redshifts are kept.We use the HEALPix-based redshift (see §3.3.2) as the estimate for the redshift. 36Each tracer has a different definition for a 'good' redshift, as detailed in the respective target selection papers (BGS, Hahn et al. 2022;LRG, Zhou et al. 2023;ELG, Raichoor et al. 2023a;and QSO, Chaussidon et al. 2023).The criteria were motivated by maximizing the completeness while minimizing the fraction of catastrophic failures expected for main survey observations and used a combination of comparisons of Redrock redshift to a visually inspected redshift or multiple Redrock redshifts from repeated observations of the target.The estimated catastrophic failure fractions are less than 0.5% for all tracer types after restricting to 'good' redshifts.Key quantities used to select good redshifts are the redshift pipeline flag ZWARN and the ∆χ 2 (DELTACHI2) obtained from the redshift fitting pipeline between the best and next-best-fit redshifts.We also restrict to a given redshift (z) range that is different for each tracer.The combined criteria are: • BGS: ZWARN==0, ∆χ 2 > 40, 0.01 < z < 0.6 ] crit > 0.9, 0.6 < z < 1.6 (with [O II] crit defined by Eq. 2).
• QSO: Not already rejected by the quasar catalog, 0.6 < z < 3.5 34 The corresponding column name is ROSETTE R in the catalogs. 35The N region in the One-Percent Survey has more area affected by stars that are bright in the infrared. 36Thus, for the 'clustering' catalogs, the column 'Z HP' is changed to 'Z'.
The quasar catalog requires that either Redrock or QuasarNET identified the object as a QSO, while the galaxy catalogs (BGS, LRG, ELG) do not explicitly require that the targeted objects are spectrally classified as galaxies.
For the DESI One-Percent Survey, we provide two weights to be used with the clustering catalogs.One, w comp , accounts for fiber assignment incompleteness.The other, w FKP , optimizes against expected signal-tonoise in 2-point clustering measurements as a function of redshift and is based on Feldman et al. (1994).We will describe both below.In previous SDSS (most recently, Ross et al. 2020) and future DESI LSS catalogs, weights that account for fluctuations in both target density due to imaging quality and redshift success due to the signalto-noise of spectroscopic observations are added.We do not include such weights for the DESI One-Percent Survey LSS catalogs.For target density fluctuations, their effects typically manifest on large angular scales and the relatively small area of the One-Percent Survey footprint is not ideal for the regression methods typically used to define them.For the redshift success, the effective exposure time reached during the One-Percent Survey was such that (after selecting the greatest signal-to-noise measurement out of any repeat observations) the variation in success rate is relatively low.Determining these weights for DESI Main Survey data is a primary focus of DESI year 1 analyses.
The completeness weights,37 w comp , are determined from the 128 realizations of alternative assignment histories.The process of generating these realizations is detailed in Lasker et al. (2023).Given there is one data realization, we have 129 total realizations.The number of realizations in which a target was assigned is thus 128P obs + 1.The probability of assignment is N assigned /N tot , and we wish to use the inverse probability as the weight.Thus, w comp = 129/(128P obs + 1). (4) These weights can be used to obtain unbiased statistics for any one-point measurement or for clustering measurements on projected scales that are large relative to the fiber patrol radius, which is at most 89 (it depends on e.g., the focal plane position due to the optics).For unbiased N -point clustering statistics in general, one should use the bit values to determine the joint probability for any configuration, and, e.g., for 2-point statistics follow the process outlined in Bianchi et al. (2018).
Comparisons of clustering results applying (or not) various weighting prescriptions to the One-Percent Survey data are presented in Lasker et al. (2023) and Rocher et al. (2023).
The clustering randoms contain the same rows as the full random files.Redshifts and weights are added to the randoms by randomly sampling the data.In this way, the weighted dN/dz of the data and random should match (and the weights on the random points are there only for this purpose).Other columns, such as photometry, that vary with redshift are similarly sampled.In all cases, any cuts that are applied to the data sample should also be applied to the random sample.Potential choices include, e.g., cuts on r rosette , N tile , or redshift.Any number of random files (recall, 18 total are available, each with a projected density of 2500/deg 2 ), or even a sub-selection of a random file, can be used without biasing any potential statistic, with the caveat that using less random points means a higher shot-noise contribution from the randoms.
The comoving number density as a function of redshift, n(z), is determined for each tracer by applying the completeness weights, and represents the estimated density for a complete sample.In order to calculate the comoving volume, we use a cosmological model based on the Planck 2018 results38 (Aghanim et al. 2020) and calculate all comoving distances in the units h −1 Mpc.The n(z) are determined separately for the 'N' and 'S' regions.Fig. 6 shows the measured n(z), taking a simple mean of the 'N' and 'S' results.The n(z) are used to determine the w FKP weights that are included in the clustering catalogs. 39We use We use a value of P 0 that is different for each tracer type and is approximately equal to the power spectrum amplitude at k = 0.15hMpc −1 ; the values are 10 4 , 7000, 6000, and 4000 Mpc 3 h −3 for LRG, BGS, QSO, and ELG.

BGS k-and e-Corrections
The BGS sample is approximately flux-limited, thus its galaxies will have an especially large range in their intrinsic luminosity.In order to compare the clustering of BGS galaxies at different redshifts and luminosities, we provide 'k' and 'e' corrections.Typically, absolute magnitudes are corrected by k-corrections to account for 0.5 1.0 1.5 2.0 2.5 3.0 redshift bandshifting effects, specifically that the observed flux distributions in a given passband will be different in the rest frames of galaxies at different redshifts.We thus provide r-band absolute magnitudes, M r , with the BGS clustering catalogs, 40 defined via Here, the subscript r represents the r-band, k(z) represents the k-correction of the galaxy, and d L (z) is the luminosity distance to the redshift z, determined using the same cosmology defined in the previous subsection (see Hogg et al. 2002 for a good overview of k-corrections).Optionally and in addition, an e-correction may be applied in order to account for the intrinsic luminosity evolution of a galaxy over time.Further, we derive the reference-frame g − r color and thus also the g-band k correction.Results are provided using both z = 0 and z = 0.1 as the reference-frame. 41 The methods we use to determine the BGS EDR k + e corrections are detailed in Moore et al. (2023). 42To begin with, we make use of the Galaxy and Mass Assembly (GAMA) DR4 dataset to create estimates of the k-and e-corrections (Driver et al. 2022).Each galaxy has an individual k-correction polynomial, calculated using KCORRECT v4.2 (see Blanton &Roweis 2007 andLoveday et al. 2012 for further details).As such, each galaxy has an individual k(z), given its observed 40 The column name is ABSMAG R and is for the z = 0.1 referenceframe. 41Denoted via 0P0 and 0P1 in the column name for z = 0 and z = 0.1, respectively. 42The code is at https://github.com/SgmAstro/DESI.g − r.These k(z) values are divided into seven equalwidth (g − r) 0 color bins.Within each of these color bins, a fourth-order polynomial is fitted to the data points corresponding to the median (g − r) 0 color of the bin (see Fig. 7).The polynomial is based on the functional form shown in Equation 7.
The coefficients of the polynomials for z ref = 0.1 are shown in Table 9.Moreover, a linear interpolation between these color polynomials is done such that the kcorrection of a galaxy is found based on its rest-frame (g − r) 0 color and its redshift.The rest-frame (g − r) 0 colors for the DESI galaxies are found using an iterative root-finding method (Brent's method as the default).Note that for all k-corrections, is true.As such, the zeroth-power coefficient (a 4 ) enforces this condition for all individual k-correction polynomials, explaining why they are all the same value for each color bin.
Finally, the e-corrections we provide are determined using the same functional form as McNaught-Roberts et al. (2014): Specifically, we assume that the density evolution (P ) is zero and the luminosity evolution (Q) is the only factor to be considered.We provide results for the r-band, for which we use Q 0 = 0.97 following the result found empirically in McNaught-Roberts et al. (2014).One can subtract the resulting E(z) from M r in order to obtain, 5. DATA ACCESS DESI data access is currently very file-oriented, reflecting the manner in which most DESI collaborators access and use the data.In addition to these files, valueadded data services and products are under development including a database and a suite of tutorials.The latter are provided on a best-effort basis to the community for convenience.We describe these data access methods as well as the data license and acknowledgments below.Documentation of DESI data access is further maintained at https://data.desi.lbl.gov/doc/access/.

File Access
For researchers who are members of other collaborations with NERSC account access, files are directly available at /global/cfs/cdirs/desi/public/edr/, without requiring DESI membership.The exact same directory structure can be inspected and individual files downloaded via https://data.desi.lbl.gov/public/edr without requiring a NERSC account.Efficient bulk download of larger amounts of data is available using the Globus 43 endpoint "DESI Public Data", also without requiring a NERSC account.All 3 of these methods access the same files on disk at NERSC.Previous descriptions of file and directory locations in this paper started at the public/edr/ level, regardless of whether that is prefixed by direct file access, https, Globus, or possibly some future data access method.
The entire EDR is over 80 TB, so users are encouraged to be selective in downloading only what they really need for an analysis.We anticipate that most users will start with one of the redshift catalogs in vac/edr/lss/ or spectro/redux/fuji/zcatalog/, sub-select to the 43 https://globus.orgobjects of interest, and then proceed to download only the files containing spectra of those targets.

Other Interfaces
Catalog-level data (target photometry, fiber assignments, exposure metadata, spectral classifications, and redshifts, but not the spectra themselves nor most of the value added catalogs) are available in a Postgres database with tables and access credentials described at https://data.desi.lbl.gov/doc/access/database.Users with NERSC accounts can directly query this database with SQL, or use SQLAlchemy wrapper objects provided with pre-installed specprodDB Python code.For users without NERSC credentials, a public copy of the database is expected to be made available from the Astro Data Lab as described at https://datalab.noirlab.edu/desi.This platform includes anonymous public access via a web query interface and authenticated access via a JupyterLab server.Additionally, the Astro Data Lab plans to serve a subset of DESI spectra via the SPectra Analysis and Retrievable Catalog Lab (SPARCL 44 ), which consists of a spectral database with a programmatic interface.The subset is limited to healpix-coadded spectra which have been combined across cameras.Other types of spectra and files are only available at the primary file-based archive at NERSC as described previously.
Other access methods may be provided in the future, e.g., webpages for interactively browsing the data, or interfaces for downloading individual spectra or custom collections of spectra.If and when these become available, they will be documented at https://data.desi.lbl.gov/doc/access/.

Tutorials
There is an internal effort within the DESI collaboration to design notebooks to help introduce specific data products or ways of accessing specific types of data.These are aggregated in a GitHub repository: https: //github.com/desihub/tutorials.The tutorials are divided into thematic and topical sub-directories.Of most relevance is the getting_started sub-directory, which includes introductions to the types of files found in the EDR, and a notebook on working with the zcatalog files.

Data License and Acknowledgments
DESI Data are released under the Creative Commons Attribution 4.0 International License. 45This allows users to share, copy, redistribute, adapt, trans-44 https://astrosparcl.datalab.noirlab.edu 45https://creativecommons.org/licenses/by/4.0/form, and build upon the DESI data for any purpose, including commercially, as long as attribution is given by citing this paper and including the acknowledgments text listed at https://data.desi.lbl.gov/doc/acknowledgments/.

CONCLUSION
This paper describes the release of the first science data taken with the DESI instrument.This new sample consists of all commissioning and Survey Validation data taken between December 14, 2020 and June 10, 2021; with the majority of the data acquisition ending on May 13, 2021.These observations were taken to validate the survey design and observing strategy for the DESI Main Survey that commenced on May 14, 2021.The release includes deep spectra with robust visual inspection classification of 769 stars, 14,735 galaxies, and 3,121 quasars from extended selection algorithms that were tested during the Survey Validation period.The release includes highly complete, good-quality, spectroscopic samples of 306,052 stars; 721,026 galaxies; and 44,151 quasars obtained over an area of roughly 170 deg 2 in the One-Percent Survey.In total, after accounting for additional spectra from secondary programs and all other DESI surveys, this release includes good calibrated spectra and catalog information for 496,128 stars; 1,125,635 galaxies; and 90,241 quasars that were spectroscopically classified and free of any known hardware, observational, or redshift fitting issues.
These data were all reprocessed with the DESI data reduction pipeline and redshift classification algorithms (Guy et al. 2023;Bailey et al. 2023).Updated versions are expected to make only minor changes to the quality of one-dimensional spectra and redshift classifications.
The early data from DESI can be accessed as described in §5.Plots and key numbers in this paper were produced from notebooks and code in https://github.com/desihub/edrpaper using data from this Early Data Release.The Digital Object Identifier (DOI) for EDR is doi:10.5281/zenodo.7964161,which includes both the EDR dataset and the data for the plots in this paper.
The redshift range, depth, and variety of spectroscopic targets make the data in this release an ideal sample for studies of stellar astrophysics, galaxy and quasar astrophysics, and early studies of the clustering of matter.Representing the range of potential studies, the DESI collaboration has already used these data to measure the 1D and 3D Lyα forest (Göksel Karaçaylı et al. 2023;Ramírez-Pérez et al. 2023;Ravoux et al. 2023), model the connection between galaxies and halos (Gao et al. 2023;Prada et al. 2023;Yu et al. 2023;Yuan et al. 2023), derive the probabilistic stellar mass functions from hun-dreds of thousands of BGS galaxies (Hahn et al. 2023), and identify very metal-poor stars in the Milky Way (Allende Prieto et al. 2023), among other topics.
The first year of the full five-year DESI survey concluded on June 13, 2022 and the DESI collaboration has finalized the internal data reductions for that sample.Those data will be publicly released after the completion and publication of the BAO and RSD measurements that motivated the construction of DESI.Meanwhile, DESI continues successful and efficient operations, planning its next major data sample when a third year of observation is complete.That threeyear sample is expected to include more than 30 million spectroscopically-confirmed galaxies and quasars, with a full 14,000 deg 2 footprint of the BGS and MWS programs.DESI will make the first-year, the three-year, and eventually its final sample, publicly available following the release of the key cosmological analyses for each corresponding dataset.

APPENDIX
A. PRIMARY TARGETS This appendix includes tables for the primary targeting bits, some of which are replicated from Myers et al. (2023), for convenience and described in §2.4.

A.1. Target Selection Validation (SV1) Targeting Bitmasks
Target Selection Validation (SURVEY=sv1) bitmasks are recorded in fibermap columns SV1 DESI TARGET, SV1 BGS TARGET, and SV1 MWS TARGET.Table 10 lists the SV1 DESI TARGET bits for dark-time targets, Table 11 lists the SV1 DESI TARGET bits for general calibration targets such as standard stars and sky locations, Table 12 lists the SV1 BGS TARGET bits for the Bright Galaxy Survey, and Table 13 lists the SV1 MWS TARGET bits for Milky Way Survey targets.These target selection bits are also defined programmatically in the open-source desitarget46 software package.A YAML-format file describing the bits is in subdirectory py/desitarget/sv1/data/sv1_targetmask.yaml, with convenience wrapper objects in the Python module desitarget.sv1.sv1_targetmask.desi_mask.Examples of using these bitmasks with this code are given in §2 of Myers et al. (2023).

A.2. Operations Development (SV2) Targeting Bitmasks
The relevant targeting bits for Operations Development (SV2) are outlined in Table 14.Note that some targeting bits were retained moving from Target Selection Validation to SV2 and, therefore, only new bits and changes to existing bits are included in Table 14.The bit-mask used to track SV2 sub-programs is called sv2 desi mask, and the bit-values for each SV2 target can be accessed via the SV2 DESI TARGET, SV2 BGS TARGET, and SV2 MWS TARGET column in data files (again, for more details, see §2.4 of Myers et al. 2023).

A.3. One-Percent Survey (SV3) Targeting Bitmasks
The relevant target selection bitmasks for the One-Percent Survey (SV3) are outlined in Table 15.Many targeting bits were retained moving from earlier phases of SV to the One-Percent Survey and, therefore, only new bits and changes to existing bits (when compared to Tables 10,11,12,13,and 14) are in-cluded in Table 15.The bit-mask used to track sub-programs for the One-Percent Survey is called sv3 desi mask, and the bit-values for each target can be accessed via the SV3 DESI TARGET, SV3 BGS TARGET, and SV3 MWS TARGET columns in data files (again, for more details, see §2.4 of Myers et al. 2023).

B. SECONDARY TARGETS
In addition to its primary science goals, the DESI survey incorporates a range of "secondary" targets to pursue bespoke research.In this appendix, we describe the secondary target campaigns included in the EDR and outline how the bit-values in their scnd mask and SCND TARGET column (see §2.4 of Myers et al. 2023) can be linked back to the relevant program.
DESI pursued secondary targets during both its "Target Selection Validation" and "One-Percent Survey" phases -which we refer to here as "SV1" and "SV3," respectively, for consistency with how the bit-masks are named.In Table 16 we list the bit-names and bitvalues for secondary targets that were scheduled during SV1.In Table 17 we indicate how these bits changed for SV3. 47The target classes listed in Table 16 were either assigned to fill "spare" fibers on regular SV1 tiles, or were assigned to their own "dedicated" campaign on custom tiles listed in Table 5. Targets that were intended for dedicated observations are marked with a * in Table 16.
In the rest of this appendix, we outline each secondary target class, moving through the bit-names in Tables 16  and 17.Further details of the selection of each type of secondary target are available at the associated docs link for SV148 or SV3.49During SV1, secondary targets were permitted multiple observations.But, during SV3, most secondary targets were limited to a single observation, unless otherwise detailed below.

B.1. VETO
This targeting bit was reserved to designate targets as unnecessary or problematic.In practice, VETO was never used, and, for later DESI programs, flagging targets as bad is done in the MTL ledgers described in Schlafly et al. (2023) instead.16) Bits are stored in the sv1 desi mask and accessed via the SV1 DESI TARGET column (see Myers et al. 2023, for more details).Additional standard star targets based purely on Gaia are included in Table 13.

B.2. UDG
This program was designed to obtain redshifts for, and hence distances to, a sample drawn from the few thousand known ultra-diffuse galaxies (UDGs) across the entire DESI footprint (∼ 0.5 deg −2 ; Zaritsky et al. 2019Zaritsky et al. , 2022)).The targeted UDGs were field galaxies selected in sparse environments using imaging from the Legacy Survey.Such galaxies are expected to be among the least efficient large galaxies hitherto known.Their distances are essential for understanding their environments, and inferring their sizes and luminosities.Because the bluest UDGs in the field are expected to have emission lines that DESI could detect (see, e.g.Kadowaki et al. 2021), the sample was limited to g − r < 0.3.

B.3. FIRST MALS
This project followed up sources from the MeerKAT Absorption Line Survey (MALS, see, e.g.Gupta et al. 2016), an L-and UHF-band survey targeting ∼ 150,000 radio-loud AGN.In the DESI footprint ∼ 60% (∼ 6 deg −2 ) of these AGN are expected to have optical counterparts to r < 23.The project sought to obtain optical spectra for sources that have 21-cm and OH absorption spectra from MALS.The main goals were to help to characterize the redshift distribution of MALS AGN and intervening and associated absorption systems, to quantify biases due to dust in optically selected AGN samples, and to facilitate photometric redshift estimates for larger radio-selected samples.

B.4. WD BINARIES
This targeting bit was only briefly used for observations associated with version 0.48.0 of the desitarget code before being replaced by WD BINARIES BRIGHT and WD BINARIES DARK (see §B.31 for a more detailed description).

B.5. LBG TOMOG
This campaign requested ∼10 hours of dedicated dark time on a single DESI tile in the COSMOS or XMM-LSS fields of the CFHT Large Area U -band Deep Survey (CLAUDS; Sawicki et al. 2019) to target ∼4,500 Lyman Break Galaxies (LBGs) and quasars in the redshift range 2 < z < 3.5.A major goal was to map the Lyman-α Forest in detail by creating a 3D tomographic map of neutral hydrogen absorption (see, e.g., Ravoux et al. 2020).Additional goals included finding high-redshift protoclusters and voids.Quasars were targeted to r ∼ < 23.5 using the standard DESI targeting approach (Chaussidon et al. 2023) and LBGs were selected to r ∼ < 24.5 using a U -dropout method applied to the CLAUDS imaging catalogs.The LBG TOMOG bit was used in files associated with versions 0.48.0, 0.49.0, 0.50.0 and 0.51.0 of the desitarget code but was gradually deprecated by the LBG TOMOG XMM, LBG TOMOG COSMOS, LBG TOMOG W3 and LBG TOMOG COSMOS FINAL bits (see §B.33).

B.6. QSO RED
The QSO RED sample targeted mildly dusty quasars that are too red to meet standard DESI criteria.The project sought to ascertain whether obscuration in quasars is explained by viewing the broad-line region through a dusty torus at a grazing angle or by an early, dusty phase in the lifetime of quasars (see, e.g., Fawcett et al. 2020).The QSO RED targets were selected from point-sources in Legacy Surveys imaging using the WISE W 1W 2W 3 color wedge of Mateos et al. (2012).Targets consistent with the "bluer" colors of existing DESI quasar targets (Chaussidon et al. 2023) were removed, producing a sample of ∼41,000 QSO RED targets over the DESI footprint (∼ 3 deg −2 ).As with all quasarlike target classes throughout the DESI survey (see §5 of Schlafly et al. 2023), QSO RED sources were scheduled for 4 total observations (starting with SV3).

B.7. M31 KNOWN, M31 QSO, M31 STAR
The DESI Andromeda Region Kinematic ("DARK") survey comprised three complementary programs aimed at studying the dynamics of our nearest large neighbor galaxy.The M31 KNOWN bit covered previously identified, bright targets -such as objects from the SPLASH sur-vey (Dorman et al. 2012), globular clusters, HII regions, planetary nebulae and variable sources.The M31 QSO bit flagged quasar targets behind M31 selected using Gaia and WISE.The M31 STAR bit indicated sources selected from a combination of PAndAS (McConnachie et al. 2018), Gaia and WISE.The dedicated observations of, and scientific results from, the DARK survey are detailed in Dey et al. (2023).

B.8. MWS DDOGIANTS
The MWS DDOGIANTS targeting bit was ultimately never used by DESI.

B.9. MWS CLUS GAL DEEP
This campaign requested dedicated dark-time tiles to obtain spectra of open clusters, globular clusters (GCs), and dwarf spheroidal galaxies in the outskirts of our Galaxy.A major goal was to combine radial velocities with Gaia astrometry for faint (19 < r < 21) stars to constrain cluster membership and measure chemical abundances.Ultimately, the program aimed to characterize the initial mass function for clusters, the kinematics of stellar streams associated with GCs and density profiles for dwarf spheroidals.Targets were selected us- ing Legacy Surveys imaging and Gaia astrometry at a density of a few thousand per DESI tile.

B.10. LOW MASS AGN
The LOW MASS AGN program targeted faint (r > 20), low-redshift AGN in dwarf galaxies, selected using optical and infrared photometry from the eighth data release of the Legacy Surveys (LS DR8).The targets were preselected to be at low redshift (0.02 ≤ z phot ≤ 0.3) based on photometric redshifts from Zhou et al. (2021), and to have faint z-band absolute magnitude (M z ∼ > 21), which was adopted as a proxy for stellar mass.Candidates were then targeted as AGN on the basis of their z − W 1, W 1 − W 2, and W 2 − W 3 colors, resulting in a sample with a density of ∼ 20 deg 2 .The main scientific goal of the LOW MASS AGN program was to identify ∼100 AGN driven by intermediate mass ( ∼ < 10 6 M ) black holes (see, e.g.Mezcua & Domínguez Sánchez 2020).The program also aimed to test the validity of a new AGN selection criterion similar to that from Hviding et al. (2022), and extend it to low-mass galaxies for which the application of infrared AGN diagnostics is debated (Hainline et al. 2016;Satyapal et al. 2018).

B.11. FAINT HPM, BRIGHT HPM
The HPM project sought to measure radial velocities and spectroscopic types for ∼1,000 faint, high-propermotion stars drawn from Gaia and the NOIRLab Source Catalog (NSC; Nidever et al. 2018Nidever et al. , 2021)).Targets were selected based on high proper motion (µ > 100 mas yr −1 ), supplemented with reliable parallax measurements from Gaia and red colors from WISE and NSC.The assembled sample -extending to G ∼ 21 (Gaia) and r ∼ 23 (NSC) -was designed to find and study ejected white dwarfs (from double-degenerate binaries), ancient white and brown dwarfs, and hypervelocity stars.Fainter candidates were observed in dark time (FAINT HPM) and brighter candidates in bright time (BRIGHT HPM).

B.12. GW190412, IC134191
These bits were intended to be used for dedicated, rapid-turnaround observations of DESI "Targets of Opportunity" (ToOs) in the vicinity of a gravitational wave signal or an IceCube high-energy neutrino event.The archival gravitational wave alert chosen to mimic a real trigger was GW190412 (Abbott et al. 2020), and a realtime follow-up was performed for the "gold" neutrino event 134191 17593623.50GW190412 and IC134191 targets were assigned in some files associated with versions 0.48.0, 0.49.0 and 0.50.0 of the desitarget code.The follow-up of GW190412 was performed two years after the gravitational wave event, since gravitational  10, 11, 12 or 13, the bit was deprecated and updated for SV2.
wave detectors were not operating during SV, both as a test and to provide spectroscopic redshifts for a standard siren measurement (Palmese et al. 2021a).On the other hand, real time follow-up of ToOs by DESI (e.g.Palmese et al. 2021b) was achieved by prioritizing tiles near IC134191 during afternoon planning (see Schlafly et al. 2023) or by assigning the BRIGHT TOO LOP, BRIGHT TOO HIP, DARK TOO LOP and DARK TOO HIP bits discussed in §B.40.

B.13. PV BRIGHT, PV DARK
The low-redshift (z < 0.15) DESI Peculiar Velocity (PV) Survey was designed to improve constraints on the growth rate of structure (see, e.g.Howlett et al. 2017;Kim & Linder 2020).The survey comprised three samples (see, e.g., Saulder et al. 2023).First, the "FP" sample, which included bright (r < 18), elliptically shaped, galaxies, to help characterize the Fundamental Plane.Second, the "TF" sample, which included locations along the major axes of SGA galaxies (Moustakas et al. 2023) with spiral-like colors, to probe the Tully-Fisher (TF) relation.Third, the "extended" sample, which covered positions across the surfaces of large (> 2 × 1.4 ) SGA galaxies, to fill in areas that have no primary DESI science targets due to fiber-patrol limitations.The PV BRIGHT targets were observed in bright time, and included TF and "extended" targets.The PV DARK sample was observed in dark time, and included FP and TF targets.

B.14. LOW Z
This campaign used imaging and photometric data from the Legacy Surveys to identify moderately faint (19 < r < 21) very-low-redshift (z < 0.03) galaxies.

B.15. BHB
The BHB sample extended the MWS blue horizontal branch program (Cooper et al. 2023) to fainter (19 < g < 21) targets that required dark-time observations.The BHB sample was designed to probe stellar populations and kinematics at distances of ∼150 kpc to constrain the dark matter mass distribution in the outermost Galaxy.Targets were color-selected at a (subsampled) density of ∼ 2 deg −2 using a combination of g − r and r − z from the Legacy Surveys (similar to, e.g., Li et al. 2019) to separate BHB stars from blue stragglers, quasars and white dwarfs.Further cuts on WISE W1 and Gaia G were applied to remove residual quasars.A few targets from RR Lyrae catalogs derived by the Gaia collaboration or Sesar et al. (2017) were included in the BHB sample.

B.16. SPCV
This project aimed to catalog -and obtain multiepoch spectra of -short-period cataclysmic variable stars (spCVs).Science goals included characterizing sources in the "CV period gap" of ∼2-3 hours (e.g.Knigge et al. 2011), and finding reference spCVs for the LISA mission to use as verification binaries (e.g.Cornish & Robson 2017).Targets were selected using Gaia colors combined with variability amplitudes of > 0.25 in Gaia G (see Abrahams et al. 2020).The resulting sample was limited to 16 < G < 21, producing ∼1300 candidate spCVs spread across the Milky Way.B.17. DC3R2 GAMA Spectra were obtained to characterize the relation between ugriZY JHK multi-color and redshift for photoz calibration across ≈ 50% of the color space visible to the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST, Ivezić et al. 2019) and Euclid (Euclid Collaboration et al. 2022).Targets were selected from public KiDS+VIKING optical-NIR imaging (Wright et al. 2019;Kuijken et al. 2019) in the GAMA 9h, 12h, and 15h equatorial fields at z fiber < 22.1 and assigned to the self-organizing map of the C3R2 survey (Masters et al. 2017(Masters et al. , 2019) ) transformed to KiDS-VIKING color space.From these, a sample of 13270 spectra from 10376 unique targets was observed on dedicated tiles 80971-80975.Targets were prioritized to maximize the ability to constrain the slope of redshift with respect to magnitude at fixed color, which is a main uncertainty of the C3R2 approach for direct calibration of redshift distributions for faint photometric galaxy samples.Some additional DC3R2 GAMA targets were observed using spare fibers during the course of SV.The first results from this campaign are presented in McCullough et al. (2023).

B.18. UNWISE BLUE, UNWISE GREEN
This dedicated dark-time program was designed to calibrate the redshift distribution of galaxies for CMB lensing tomography measurements.Targets were randomly sub-selected from the "blue" (UNWISE BLUE) and "green" (UNWISE GREEN) samples of Krolewski et al. (2020), which were derived from W 1 − W 2 color cuts in the unWISE catalog of Schlafly et al. (2019).The 6" WISE PSF is too broad to reliably center DESI fibers on the galaxy of interest, so targets were additionally matched to the nearest Subaru Hyper Suprime-Cam (HSC) imaging (with a 2.75 radius) and limited to y < 22.5 (y < 24) for the blue (green) sample, resulting in ∼9,000 (∼4,500) UNWISE BLUE (UNWISE GREEN) targets.The main goal of the program was to improve cosmological constraints from Planck-unWISE lensing measurements by reducing uncertainty in the unWISE redshift distribution, a primary contributor to S 8 uncertainty (see, e.g., Krolewski et al. 2021).This program was refined in later iterations of SV1 -files associated with versions 0.51.0 and 0.52.0 of the desitarget code were supplemented with the additional bits listed in §B.34.

B.19. HETDEX MAIN, HETDEX HP
The HETDEX dedicated dark-time campaign pursued higher-resolution spectra of a few thousand Lyman-α emitters (LAEs) from the Hobby-Eberly Telescope Dark Energy Experiment (Gebhardt et al. 2021).HETDEX targets LAEs in the redshift range 1.9 ∼ < z ∼ < 3.5 but its relatively meager spectral resolution (∼800; Hill et al. 2021) spawns contamination by low-redshift [O II] emitters.The DESI observations sought to characterize this contamination, while also preparing for future DESI-like experiments using LAEs.HETDEX MAIN targeted HET-DEX sources with spectral signal-to-noise > 5.2, supplemented by a few hundred lower-significance emitters.HETDEX HIP targets comprised a few dozen faint, highredshift LAEs to help characterize DESI detection limits for HETDEX sources.

B.20. PSF OUT BRIGHT, PSF OUT DARK
The PSF outliers campaign targeted point sources (TYPE=="PSF") in the Legacy Surveys that lie more than 10σ from the stellar locus in grz (in the spirit of, e.g., Covey et al. 2007).The PSF OUT BRIGHT (PSF OUT DARK) targets were intended for bright-time (dark-time) observations and were limited to 15 < r < 19 (16 < r < 22).The program was designed as a "filler" survey -it comprised a little more than one hundred total targets per sq.deg.and was scheduled at very low priority to mop up spare fibers.The true target density was lower as ∼ 80% of outliers from the stellar locussuch as candidate quasars, white dwarfs and compact galaxies -were already targeted by DESI.

B.21. HPM SOUM
The HPM SOUM survey pursued spectroscopy of high proper motion stars across the DESI footprint.The scientific goals of the program were similar to those for the FAINT HPM campaign (see §B.11).The target list comprised ∼2,900 faint (r > 19.5) stars with high proper motion ( ∼ > 200 mas yr −1 ).These targets were drawn from the sample of Segev & Ofek (2019), which derived proper motions by comparing the positions of sources between the Sloan Digital Sky Survey (SDSS) and Pan-STARRS1.

B.22. SN HOSTS
This program targeted ∼20,000 supernova hosts and nuclear variables from the Nearby Supernova Factory (e.g.Aldering et al. 2002), Palomar Transient Factory (e.g. Law et al. 2009), SDSS-II Supernova Survey (Frieman et al. 2008) and Zwicky Transient Factory (e.g.Bellm et al. 2019;Fremling et al. 2020).DESI spectroscopy is particularly warranted as the ZTF "SED machine" (Blagorodnova et al. 2018) can only obtain low-resolution (R ∼ 100) spectra.As about half of the SN HOSTS data set was already targeted by the DESI BGS, the true SN HOSTS sample only comprised ∼10,000 targets.Scientific goals included; probing correlations between the properties of supernovae and their host galaxies; improving cosmological constraints from supernovae; using direct, supernova-based distance measurements to characterize peculiar velocities and the Fundamental Plane of host galaxies; and enhancing populations of changing-look AGN and Tidal Disruption Events.

B.23. GAL CLUS BCG, GAL CLUS 2ND, GAL CLUS SAT
This dark-time campaign sought to build a volumecomplete sample of galaxy clusters with spectroscopically confirmed members.Targets were compiled from galaxies with a probability of cluster membership of P mem > 0.90 in version 6.3 of the SDSS redMaPPer catalog (see, e.g., Rykoff et al. 2014).The GAL CLUS BCG bit denotes Brightest Cluster Galaxies -i.e. the redMaP-PER most-probable central galaxy -to a redshift of z < 0.35.The GAL CLUS SCND bit signifies the second brightest cluster member candidate.The GAL CLUS SAT bit denotes all other (P mem > 0.90) candidate cluster members to z < 0.30.GAL CLUS BCG targets were prioritized for DESI observations over GAL CLUS 2ND, which were in turn prioritized over GAL CLUS SAT targets.After removing existing DESI targets, the sample comprised a few 100 (each) GAL CLUS BCG and GAL CLUS SCND targets and ∼10,800 GAL CLUS SAT targets.

B.24. HSC HIZ SNE
This dedicated dark-time program focused on obtaining redshifts for supernova host galaxies identified in a deep cadenced HSC survey in the COSMOS field (see Yasuda et al. 2019).The target sample consisted of 1036 supernova candidates (out of the 1824 candidates detailed in Yasuda et al. 2019) that lacked spectroscopy when the HSC HIZ SNE observations were proposed.A little more than 400 of these candidates were expected to be cosmologically important Type Ia supernovae.The main scientific goal of this program was to double the number of known Type Ia supernovae at redshifts of z > 1 and hence improve constraints on the dark energy equation of state.

B.25. ISM CGM QGP
The ISM CGM QGP campaign sought to probe the circumgalactic medium (CGM) by targeting 114 quasars (S/N > 3) with sight-lines that pass within 30 (∼250 kpc at z = 2.0) of a galaxy in the COSMOS (Laigle et al. 2016) or HSC Ultra-Deep (Aihara et al. 2018b) fields.Targets were selected from SDSS DR14 quasars (Pâris et al. 2018) with g < 22.The cool gas galaxy counterparts is selected from the COSMOS2020 catalog (Weaver et al. 2022).The main goals of the program (see also Zou et al. 2023) were to probe the metal budget in the CGM, and to characterize how the CGM is influenced by the properties of proximate galaxies.

B.26. STRONG LENS
This program sought to obtain redshifts for strong gravitational lenses identified in DESI Legacy Surveys imaging (see Huang et al. 2020Huang et al. , 2021)).The brightest image of each lensed source and ∼20% of putative lensing galaxies were scheduled for observations, resulting in a total sample of 3588 targets spread throughout the DESI footprint.The main purpose of obtaining spectroscopic redshifts for the STRONG LENS sample was to improve lensing models for these systems.Scientific goals included probing dark matter halo density profiles and sub-halo abundances, and identifying superior systems to search for multiply imaged supernovae.

B.27. WISE VAR QSO
The WISE VAR QSO bit denotes quasar targets selected via variability estimated using the WISE "light curve sweeps" supplied with DR9 of the Legacy Surveys. 51 The selection technique was based on cuts in structurefunction-space (represented by the parameters A and γ) in a similar fashion to Myers et al. (2015) (see §4.

B.28. MWS CALIB, BACKUP CALIB
These target classes indicate calibration sources that were adopted for the SV1 stellar survey described in Cooper et al. (2023).MWS CALIB and BACKUP CALIB targets were selected from publicly available survey catalogs (e.g.SDSS Segue, APOGEE, GALAH, and the Gaia ESO Survey).BACKUP CALIB targets were limited to the magnitude range 10 < G < 16 and MWS CALIB targets were limited to 16 < G < 19.

B.30. MWS RRLYR
This target class formed part of the SV1 stellar survey described in Cooper et al. (2023).The selection targets stars that are likely RR Lyrae variables based on Gaia DR2.It combines sources that were labeled as RR Lyrae by the Specific Object Study pipeline (Clementini et al. 2019) and the general variability processing pipeline (Holl et al. 2018) This campaign pursued a representative sample of all types of white dwarf binaries.The broad scientific focus was to characterize the entire dynamical range of bound white dwarfs, from intrinsically bright, high-mass transfer binaries to extremely faint, highly evolved systems.Targets were selected by cross-matching the GALEX source catalog with Gaia and retaining sources with an absolute magnitude of M F U V > 1.5(F U V − G) − 0.3, which generally lie below the main sequence.Here, M F U V is calculated using distances derived from Gaia parallaxes that are measured to ≥ 5σ, and F U V and G represent magnitudes in GALEX FUV and Gaia Gband.After removing existing DESI targets, this sample comprised ∼28,300 (∼7,400) sources with 16 ≤ G ≤ 18 (G > 18).The brighter (fainter) of these subsamples is signified using the WD BINARIES BRIGHT (WD BINARIES DARK) bit and scheduled for observations in bright (dark) time.

B.32. DESILBG
This project sought ∼10 hours of dedicated darktime observations in fields covered by the CFHT Large Area U -band Deep Survey (CLAUDS; Sawicki et al. 2019).By supplementing CLAUDS with deep grz imag-ing from HSC the project aimed to target ∼5,000 Lyman Break Galaxies (LBGs) and LAEs in the redshift range 2 < z < 4. The main goal was to prepare for future DESI-like experiments by characterizing a population of high-density, high-redshift tracers with which to improve cosmological constraints at times before dark energy began to dominate the Universe.Three different approaches were adopted to target redshifts near z ∼ 2 (the BX technique, see, e.g.Adelberger et al. 2004) and z ∼ 3-4 (u-and g-dropout techniques, see., e.g.Hildebrandt et al. 2009).The DESILBG bit wasn't introduced until version 0.51.0 of the desitarget code and it was rapidly replaced by the bits described in §B.35.

B.33. LBG TOMOG XMM, LBG TOMOG COSMOS, LBG TOMOG W3, LBG TOMOG COSMOS FINAL
These targeting bits denote updates to the LBG TOMOG program described in §B.5.Overall, the science program remained the same but the selection was tweaked slightly, or applied in a new field, as outlined at each bit's docs link (see the introduction to this appendix).The LBG TOMOG XMM and LBG TOMOG W3 bits were introduced in version 0.51.0 of the desitarget code (and targeting files) and were also populated in version 0.52.0.The LBG TOMOG COSMOS bit was introduced in 0.51.0 and quickly deprecated in favor of the LBG TOMOG COSMOS FINAL bit for 0.52.0.The science goals outlined in §B.18 were unchanged, but adding bits facilitated finer-grained control of how targets were prioritized for DESI observations.The UNWISE GREEN II 4000 targets were assigned the highest priority, followed, in order, by UNWISE GREEN II 3900, UNWISE GREEN II 3800, UNWISE GREEN II 3700, UNWISE BLUE FAINT II and, finally, UNWISE BLUE BRIGHT II at the lowest priority.
The UNWISE BLUE BRIGHT II and UNWISE BLUE FAINT II targets were split at a Legacy Surveys fiber magnitude of z fiber = 21.1.The green targets were split using WISE colors according to W 1 − W 2 < 0.8, W 1 < 17.0 (UNWISE GREEN II 4000); W 1 − W 2 < 0.8, W 1 < 17.2 (UNWISE GREEN II 3900); and W 1 − W 2 < 0.8 (UNWISE GREEN II 3800); with remaining sources from the original UNWISE GREEN selection signified by UNWISE GREEN II 3700.The split in the blue sample allowed for longer exposure times on faint targets, whereas the split in the green sample de-prioritized galaxies at z > 1.6, where the redshift completeness is poor due to the [O II] line and 4000 Å break redshifting out of the DESI wavelength range.

B.35. DESILBG TMG FINAL, DESILBG G FINAL, DESILBG BXU FINAL
Starting with version 0.52.0 of the desitarget code, the DESILBG sample was split into several subsamples to make it easier to track the provenance of each targeting approach described in §B.32.The BX selection and u-dropout targets were signified by the DESILBG BXU FINAL bit; the g-dropouts were signified by the DESILBG G FINAL bit; and a new selection that resembled the LBG TOMOG target class outlined in §B.5 was built.Detailed code to derive each of these subsamples is available on GitHub.52B.36. BRIGHT TOO, DARK TOO These target classes were intended to flag general "Targets of Opportunity" (ToOs) during the SV1 phase of DESI.In actuality, ToOs were not tracked until SV3 and these bits were replaced by the bits described in §B.40.

B.37. LOW Z TIER1, LOW Z TIER2, LOW Z TIER3
When DESI moved to its SV3 phase, the LOW Z targets described in §B.14 were split into three tiers to allow different target classes to be assigned different observational priorities.The highest priority targets (LOW Z TIER1) contain the most likely low-redshift candidates based on the prediction of a machine learning method (convolutional neural network; see Wu et al. 2022).The next-highest priority targets (LOW Z TIER2) comprised objects that are in a tighter photometric space where most low-redshift candidates are situated (as introduced in Mao et al. 2021).The LOW Z TIER2 sample was designed to not overlap with BGS targets.Finally, the lowest priority targets (LOW Z TIER3) were based on the remaining targets within the overall photometric cuts (referred to as "z < 0.03-complete cuts" in Darragh-Ford et al. 2022).The LOW Z TIER3 sample was allowed to overlap with BGS targets.Again, see Darragh-Ford et al. (2022) for a full description of the LOW Z program.

B.38. Z5 QSO
The Z5 QSO program targeted quasars at redshifts of 5.0 ∼ < z ∼ < 6.5 based on color cuts applied to Legacy Surveys imaging (grzW 1W 2) supplemented by i-and These bits are taken from the desitarget code a and are described in the body of the Appendix.In addition to the changes listed in the table, many bits from SV1 were not retained or set in SV3.
y-band from Pan-STARRS1 (see, e.g., Wang et al. 2017;Yang et al. 2019).Where available, J-band from public NIR surveys was also incorporated to reject stellar contaminants.The main color criteria, which are detailed in Yang et al. (2023), produce a target sample with a density of ∼ 0.5 deg Starting with SV3, the PV BRIGHT and PV DARK samples described in §B.13 were split into three sub-classes (each) to facilitate them being scheduled for observations at different priorities (for details, see Table 1 of Saulder et al. 2023).Broadly, the highest-priority targets (PV BRIGHT HIGH, PV DARK HIGH) included "FP" (Fundamental Plane) targets and the subset of "TF" (Tully-Fisher) targets that were positioned along the axes of SGA galaxies.Then, the medium-priority targets (PV BRIGHT MEDIUM, PV DARK MEDIUM) included all other "TF" targets.Finally, any additional PV targets were signified by the lowest-priority targeting bits (PV BRIGHT LOW, PV DARK LOW).PV BRIGHT HIGH and PV DARK HIGH were scheduled for extra observations during SV3 (a total of five) to improve spectral signal-tonoise.
B.40. BRIGHT TOO LOP, BRIGHT TOO HIP, DARK TOO LOP, DARK TOO HIP These bits were used to handle "Targets of Opportunity" (ToOs) in SV3 -i.e.targets such as transients that need to be observed at short notice.Transient ToOs were selected from the DECam Survey of Intermediate Redshift Transients (DESIRT; Palmese et al. 2022), and will be described in Palmese et al. (2023).In the bitnames, DARK signifies a ToO to be observed in dark time and BRIGHT denotes a ToO that can be scheduled for either dark-or bright-time observations.LOP indicates a "low priority" ToO, which would be prioritized below all primary targets and most secondaries. 53HIP signifies a "high priority" ToO, which would be prioritized above all other DESI targets, including primaries.The general mechanisms by which ToOs are handled are discussed more in §3.2.2 of Myers et al. (2023) and §5.4 of Schlafly et al. (2023).

Figure 1 .
Figure 1.The number of unique tiles per night observed during Survey Validation.The same tile can be observed on multiple nights.

Figure 2 .
Figure2.The number of good, unique target redshifts as a function of redshift for each tracer type as defined in §3.2.The reddish-brown distribution is for objects targeted to be a star and classified by Redrock to be SPECTYPE==STAR.The purple, green, and red histograms show objects targeted as BGS, ELG, and LRG respectively, and classified as a GALAXY.The blue distribution shows objects targeted as a QSO and classified as a QSO.The gray distributions depict all objects that were classified by Redrock as a STAR, GALAXY, or QSO for the top, middle, and bottom panels respectively.*Note that the gray differs from the colored histograms because of secondary targets and other target types that were classified to a different category (e.g. a QSO target that was classified as a STAR).Also note that an object can be targeted by two galaxy target classes, and such objects will appear in both distributions.

Figure 3 .
Figure3.The density of good, unique target redshifts on the sky, split by the three primary phases of Survey Validation -Target Selection Validation (sv1, blue), Operations Development (sv2, green), and the One-Percent Survey (sv3, orange).Note that the orange One-Percent Survey colormap goes to 4× higher density than the others, reflecting the much higher density of targets in the One-Percent Survey rosettes of overlapping tiles.

Figure 4 .
Figure 4.The number of overlapping bright time tiles at each location of a BGS target in 'rosette' number 1 of the DESI One-Percent Survey (SV3 R1 in Table4).

Figure 5 .
Figure 5.The observational completeness within each tile grouping of ELG targets on 'rosette' number 1 (SV3 R1 in Table4) from the DESI One-Percent Survey.

Figure 6 .
Figure6.The comoving number density for samples used to create LSS catalogs in the DESI One-Percent Survey.For this display, we have taken the mean of the results in the 'N' and 'S' regions (see text for details).

Figure 7 .
Figure 7.The seven r-band k-correction polynomials for median rest-frame (g − r)0 in each color bin.
2.1) and Palanque-Delabrouille et al. (2011).The resulting target density was a little more than 100 deg −2 across most of the DESI footprint.As with all quasar-like classes throughout the DESI survey (see §5 of Schlafly et al. 2023), WISE VAR QSO targets were scheduled for 4 total observations (starting with SV3).The main goal of the WISE VAR QSO sample was to expand the pool of quasars recovered by DESI that could be used for studies of the Lyman-α Forest.

B. 34 .
UNWISE GREEN II 3700, UNWISE GREEN II 3800, UNWISE GREEN II 3900, UNWISE GREEN II 4000, UNWISE BLUE FAINT II, UNWISE BLUE BRIGHT II These bits were introduced in version 0.51.0 of the desitarget code to augment the UNWISE BLUE and UNWISE GREEN targets.
−2 .Science goals included constraining the quasar luminosity function at high redshift, building a sample to study the IGM near the epoch of reionization, and probing supermassive black hole growth in the early Universe.As with all DESI quasar-like targets (see §5 of Schlafly et al. 2023), the Z5 QSO sample was scheduled for 4 total observations.B.39. PV BRIGHT HIGH, PV BRIGHT MEDIUM, PV BRIGHT LOW, PV DARK HIGH, PV DARK MEDIUM, PV DARK LOW

Table 2 .
Number of nights, tiles, exposures, effective exposure time, and approximate area covered by tiles for surveys included in the Early Data Release.SURVEY=sv1 includes both Target Selection Validation tiles and tiles dedicated to secondary targets.The area covered by tiles is larger than the true effective area available to targets due to bright star exclusions, focal plane geometry, hardware configuration, and higher priority targets blocking lower priority targets.

Table 3 .
The 5 deepest tiles for the 5 primary target classes in DESI.
, Hyper Suprime-Cam (HSC Aihara et al. 2018a), Dark Energy Survey (DES Dark Energy Survey Collaboration et al. 2016) deep fields, Galaxy And Mass Assembly (GAMA Driver et al. 2011), Great Observatories Origins Deep Survey (GOODS Dickinson et al. 2003), and anticipated deep fields from future Legacy Survey of Space and Time (LSST Ivezić et al. 2019) and Euclid (Euclid Collaboration et al. 2022) observations.These are summarized in Table

Table 4 .
Selected DESI tiles with visual inspections (VI) or overlapping with datasets from other surveys."SV3 Rn" denotes rosette number n from the One-Percent Survey, with RA and DEC referring to the center of the Rosette rather than the center of an individual tile.

Table 5 .
Fiberassign programs, descriptions, TILEID ranges, effective exposure time and targeting bits for all tiles dedicated to secondary targets.See Appendix B for a description of the targeting bits for each program.

Table 7 .
Summary of the directory structure of data available in the EDR.See https:// desidatamodel.readthedocs.iofor more details including subdirectory structure underneath these directories, individual file formats, and additional directories with pipeline inputs such as calibration files.
20A consequence is that the (TARGET RA, TARGET DEC, and REF EPOCH) values are altered to have a REF EPOCH of the date that the tile was designed, which makes them differ from the input photometric column 20The design date can differ from when a tile was observed.

•
In the coadded FIBERMAP tables, MEAN FIBER RA and MEAN FIBER DEC record the average asobserved position of the fibers (in comparison to the intended positions recorded in TARGET RA and TARGET DEC plus proper motions PMRA, PMDEC, REF EPOCH).The FIBERMAP coordinate coaddition incorrectly included exposures that had been excluded from the spectral coaddition, which can result in incorrect MEAN FIBER RA/DEC values.The same issue applies to the standard deviations recorded in STD FIBER RA/DEC.As a result, TARGET RA/DEC are more reliable than MEAN FIBER RA/DEC, while noting that the actual positioning can vary by O(0.1 ), which is still small compared to the ∼ 1.5 diameter fibers.

Table 9 .
Coefficient table for the r-band k-correction polynomials.
Myers et al. 2023n the sv1 desi mask and accessed via the SV1 DESI TARGET column (seeMyers et al. 2023, for more details).a"FDR"refersto the DESI Final Design Report (see DESICollaboration et al. 2016a).

Table 11 .
SV1 bits for calibration, object-avoidance, and to indicate non-dark-time programs

Table 12 .
Myers et al. 2023vey (BGS) targeting bits for SV1 Bits are stored in the sv1 bgs mask and accessed via the SV1 BGS TARGET column (seeMyers et al. 2023, for more details).

Table 13 .
Myers et al. 2023(MWS) targeting bits for SV1 Gaia targets Bits are stored in the sv1 mws mask and accessed via the SV1 MWS TARGET column (seeMyers et al. 2023, for more details).

Table 14 .
Myers et al. 2023dated targeting bits for Operations Developement (SV2).Bits are stored in the sv2 desi mask and accessed via the SV2 DESI TARGET column (seeMyers et al. 2023, for more details).Many bits from Tables 10, 11, 12 and 13 were reused for SV2, and only new or different bits are included in this table.For example, calibration bits listed in Table11remained the same moving from SV1 to SV2.Where the name and description of a bit-value has changed in this table compared to Tables

Table 15 .
Myers et al. 2023dated targeting bits for SV3.Bits are stored in the sv3 desi mask and accessed via the SV3 DESI TARGET column (seeMyers et al. 2023, for more details).Many bits from Tables 10, 11, 12, 13 and 14 were reused for SV3, and only new or different bits are included in this table.For example, calibration, BGS and MWS bits were not altered moving from SV2 to SV3.Where the name and description of a bit-value has changed in this table compared to Tables 10, 11, 12, 13 or 14, the bit was deprecated and updated for SV3.

Table 16 .
Secondary targeting bits for SV1These bits are taken from the desitarget code a and are described in the body of the Appendix.
* A dedicated target class intended to be observed on custom tiles.a , limited to 14 < G < 19.The sample can be reproduced by running the following Gaia archival query: FROM gaia_dr2.gaia_sourceas g, x where g.source_id = x.source_id and phot_g_mean_mag between 14 and 19; B.31. WD BINARIES BRIGHT, WD BINARIES DARK *

Table 17 .
Updates to secondary targeting bits for SV3