The COS Legacy Archive Spectroscopy SurveY (CLASSY) Treasury Atlas

Far-ultraviolet (FUV; ~1200-2000 angstroms) spectra are fundamental to our understanding of star-forming galaxies, providing a unique window on massive stellar populations, chemical evolution, feedback processes, and reionization. The launch of JWST will soon usher in a new era, pushing the UV spectroscopic frontier to higher redshifts than ever before, however, its success hinges on a comprehensive understanding of the massive star populations and gas conditions that power the observed UV spectral features. This requires a level of detail that is only possible with a combination of ample wavelength coverage, signal-to-noise, spectral-resolution, and sample diversity that has not yet been achieved by any FUV spectral database. We present the COS Legacy Spectroscopic SurveY (CLASSY) treasury and its first high level science product, the CLASSY atlas. CLASSY builds on the HST archive to construct the first high-quality (S/N_1500>~ 5/resel), high-resolution (R~15,000) FUV spectral database of 45 nearby (0.002<z<0.182) star-forming galaxies. The CLASSY atlas, available to the public via the CLASSY website, is the result of optimally extracting and coadding 170 archival+new spectra from 312 orbits of HST observations. The CLASSY sample covers a broad range of properties including stellar mass (6.2<logM_star(M_sol)<10.1), star formation rate (-2.0<log SFR (M_sol/yr)<+1.6), direct gas-phase metallicity (7.0<12+log(O/H)<8.8), ionization (0.5<O_32<38.0), reddening (0.02<E(B-V<0.67), and nebular density (10<n_e (cm^-3)<1120). CLASSY is biased to UV-bright star-forming galaxies, resulting in a sample that is consistent with z~0 mass-metallicity relationship, but is offset to higher SFRs by roughly 2 dex, similar to z>~2 galaxies. This unique set of properties makes the CLASSY atlas the benchmark training set for star-forming galaxies across cosmic time.


INTRODUCTION
The interplay between gas and stars is one of the most fundamental, yet unsettled, drivers of galactic evolution. In the basic picture, gas is accreted onto galaxies from the cosmic web, settles into their gravitational wells, and is converted into stars. The massive stars in star-forming galaxies ionize the surrounding gas, producing nebular emission and driving outflows and radiative shocks. Such feedback drives chemical evolution and can modulate or limit accretion processes, thereby regulating the subsequent growth of galaxies. The far-ultraviolet (FUV), defined here as ∼ 1200 − 2000 Å, is arguably the richest wavelength regime in diagnostic spectral features characterizing these processes and will provide an important window onto the first generation of galaxies in the forthcoming era of extremely large telescopes (ELTs) and the James Webb Space Telescope (JWST).

The Need for a Benchmark FUV Spectral Atlas
While the full FUV spectrum is rich in diagnostic power, studies have thus far largely focused on specific and individual components of the galaxy evolution puzzle. As a result of differing requirements of various studies, the archive has been a piecemeal collection, missing significant pieces that reside in different parts of the FUV spectrum. This incompleteness has prevented us from understanding the interdependent nature of the physical processes in star-forming galaxies and their UV spectroscopic properties. Recently, spectroscopic studies of 2 < z < 4 star-forming galaxies have emerged as our most comprehensive rest-FUV spectral datasets (e.g., Rigby et al. 2018;Steidel et al. 2016). Of particular importance to developing a detailed understanding of UV emission lines has been the large sample VUDS (e.g., Le Fèvre et al. 2015), VANDELS (e.g., McLure et al. 2018), and MUSE (e.g., Schmidt et al. 2021) surveys. However, the proper tool-set to interpret these rich datasets has been lacking.
A scientifically useful FUV spectral atlas requires three components. First, it must have broad wavelength coverage to simultaneously probe the continuum, absorption, and emission features that characterize the stellar populations and gas of the galaxy. Second, theory predicts that the velocities of galactic outflows scale with the circular velocities of their host galaxies (e.g., Murray et al. 2005). This means that very high spectral resolution (R > 10, 000 or velocity resolution CLASSY I 3 less than 30 km s −1 ) is required to measure the impact of stellar feedback in galaxies with low stellar-mass (see, e.g., Mc-Gaugh et al. 2001). Third, spectra with high signal-to-noise (S/N) in the continuum are needed to measure faint features.
There are several existing compilations of FUV spectra of local star-forming galaxies. These collections include the the Kinney et al. (1993) atlas, which is perhaps the most impactful FUV atlas to date, of 143 star-forming and active galaxies observed with the International Ultraviolet Explorer (IUE), the Leitherer et al. (2002) atlas of 19 galaxies observed with the 0.9-m Hopkins Ultraviolet Telescope (HUT), the Grimes et al. (2009) atlas of 16 galaxies with archival Far Ultraviolet Spectroscopic Explorer Spectrograph (FUSE) spectra, and the Leitherer et al. (2011) atlas of 28 galaxies observed with the Goddard High Resolution Spectrograph (GHRS) and the Faint Object Spectrograph (FOS). Unfortunately, each of these atlases is limited by the properties of the UV spectrograph used and so is insufficient for current science objectives.
The properties characterizing important compilations of FUV spectra are listed in Table 1 in the order of publication. While the Kinney et al. (1993) atlas is the largest and still the most impactful, the resolution (∼ 6 Å) and S/N (∼ 1) of its spectra are too low to resolve the FUV spectra features and utilize their diagnostic power. The Leitherer et al. (2002) atlas using the HUT was ground-breaking for the FUV at the time because it pushed blueward down to 912 Å, but suffered from even lower spectral resolution (∼ 1 Å) than the Kinney et al. (1993) atlas. In contrast, the Grimes et al. (2009) atlas was the first FUV atlas to achieve both high S/N and spectral resolution, but observed only sub-Lyα (< 1215 Å) wavelengths. Finally, the Leitherer et al. (2011) atlas made progress in combining both large wavelength coverage and moderately high spectral resolution (0.9 Å), but only for a small subsample of 12 galaxies with GHRS observations. Sadly, the limited wavelength coverage, spectral resolution, and S/N of these past compilations are insufficient for the simultaneous study of the stellar, nebular, and outflow properties in star-forming galaxies needed to advance our understanding of galaxy evolution.

The Cosmic Origins Spectrograph
The installation of the Cosmic Origins Spectrograph (COS) on HST in May 2009 during Servicing Mission 4 ushered in a new era of higher-sensitivity FUV spectroscopy. It is optimally designed for moderate-resolution UV spectroscopy of faint point sources with a resolving power of R ∼ 15, 000 and a FUV spectroscopy channel covering approximately 900 − 2000 Å. Not only does COS have excellent FUV wavelength coverage and resolution, as shown in Table 1, it also benefits from superior FUV sensitivity over all previous FUV spectrographs. In fact, COS boasts more than 30 times the sensitivity of the Space Telescope Imaging Spectrograph (STIS) for FUV observations of faint objects, allowing its spectroscopic observations to unify the study of FUV stellar and gas-phase properties in a way that would not be possible with STIS 1 or any other past FUV spectrograph.
While COS has been used widely to observe many samples of star-forming galaxies and has enabled key scientific discoveries, it lacks a high caliber atlas in the class of those discussed above. This is due in large part to the piecemeal nature of the current archive, which, as a result, suffers from a variety of deficiencies. For one, archival COS data usually cover only a portion of the FUV owing to the diversity of science goals, the competitive nature of HST, and the narrow wavelength coverage of the individual mediumresolution gratings (∼ 300Å). In addition, strong stellar features and interstellar absorption lines are more concentrated in the bluer end of the FUV, while strong emission lines are clustered in the redder portion of the FUV. Consequently, it is impossible to fully connect properties of the gas and stars within a single grating spectrum of a given galaxy. Second, the S/N in the continuum, and/or the resolution of archival COS spectra is often too low to accurately measure the ISM absorption or stellar features. For example, there are ample programs of low-resolution G140L grating spectra that offer more efficient exposure times over the higher-resolution gratings, but data taken with G140L do not have sufficient spectral resolution to characterize the kinematics of the gas or disentangle multi-component emission and absorption features. Third, the highest S/N FUV HST spectra come from nearby galaxies, which can suffer the consequences of foreground Milky Way absorption lines contaminating those intrinsic to a galaxy. These problems are illustrated by the state of the Hubble Spectral Legacy Archive (HSLA; Peeples et al. 2017), which "provides the community with all raw and associated combined spectra for COS far-ultraviolet data publicly available in MAST". In the July, 2018 HSLA release, the number of nearby (z < 0.3) galaxies with high-S/N ( 5 per 100 km s −1 resolution element) FUV COS spectroscopy totaled only 101. Of these, only 37 galaxies had observations with more than one grating setting, and poignantly, only two had full FUV 1200 − 2000 Å coverage.

The CLASSY Treasury
Here we present the COS Legacy Archive Spectroscopic SurveY (CLASSY) as the first high-S/N (> 5), highresolution (R ∼ 15, 000) FUV COS spectral atlas of star-4 BERG ET AL. The best FUV collections of spectra are listed here. Column 1 provides the reference for the identified survey or analysis associated with the data. For each of these samples, Column 2 lists the number of galaxies, Column 3 lists the redshift range, Column 4 lists the B-band magnitude range, and Column 5 lists the gas-phase oxygen abundance range. The remaining columns provide details of the spectral observations: Column 6 names the instrument/telescope (note that STIS, GHRS, FOS, and COS were all on HST) and spectral aperture used, Column 8 lists the corresponding spatial scales probed, Column 9 gives the mirror diameter of those telescopes, Column 10 lists the wavelength coverage, Column 11 provides the resolution (at 1500 Å), and Column 12 estimates the signal-to-noise ratio measured from the FUV continua of the survey spectra. Note that FUSE had 4 mirrors, so the effective diameter is listed here. If multiple resolutions were available, only the most comparable resolution mode is listed. The GHRS also had a very high resolution mode of R = 100, 000, but was limited to observing objects brighter than 14th magnitude. Considering these characterizations, only CLASSY has the combined wavelength coverage, resolution, and signal-to-noise to study both the massive stars and the gas properties in the same galaxies. a S/N given per pixel for Leitherer et al. (2011), per 0.078 Å bin for Grimes et al. (2009), and per resel for CLASSY. S/N was estimated from the average rms of the Kinney et al. (1993) templates. No S/N was provided by Leitherer et al. (2002) or Chandar et al. (2004). b Note that while a range of aperture sizes were used for the HUT observations of Leitherer et al. (2002;12 , 20 , 32 , and 10 ×56 ), 20 was adopted here as an average. c The ranges in gas-phase oxygen abundance sampled by the Grimes et al. (2009) and Leitherer et al. (2011) surveys may be misleading owing to their inclusion of only two very metal-poor galaxies (I Zw 18 and SBS 0335-052).
forming galaxies. CLASSY fills in the fundamental gaps in the 1200-2000 Å FUV range that will achieve a holistic view of star-forming galaxies and enhance the legacy and utility of the COS archive.
CLASSY combines 135 orbits of new HST data with 177 orbits of archival HST data, for a total of 312 orbits, to complete the first atlas of high-quality rest-FUV COS spectra of 45 local star-forming galaxies. In order to achieve nearlypanchromatic FUV spectral coverage with the highest spectral resolution possible, CLASSY efficiently augments existing archival data with new HST/COS observations, uniting the high-resolution G130M, G160M, and G185M gratings for a sample of galaxies spanning broad parameter space. As a result, CLASSY provides a well-controlled local FUV sample with the requisite sensitivity and spectral resolution to enable synergistic studies of stars and gas within the same galaxies. The rest-FUV coverage of the CLASSY data provides a diversity of spectral features with unparalleled diagnostic power. With a wavelength range of roughly 1200-2000 Å, the CLASSY spectra cover many emission and absorption features important for characterizing the ionizing stellar populations (e.g., metallicity, age), the properties of CLASSY I 5 gas flows (velocities, column densities, ionization phases), and the physical conditions of the multiphase interstellar gas (e.g., metallicity, temperature, ionization) within the same galaxies.
The CLASSY atlas is the benchmark training set for starforming galaxies across cosmic time, including the earliest galaxies that will be observed with the eminent James Webb Space Telescope (JWST) and the next generation of extremely large telescopes (ELTs). Therefore, CLASSY will provide a number of state-of-the-art data products to the astronomical community via MAST and HSLA to ensure their enduring value and utility. The primary data products are the high-resolution, high-S/N coadded multi-grating FUV spectral templates of the CLASSY sample of 45 star-forming galaxies presented here. Additional planned data products include stellar continuum fits, nebular abundance measurements, improved UV emission line diagnostics, feedback properties, radiative transfer models, and more. With these high-level science products (HLSPs), the CLASSY Treasury will complete the vital picture that is needed to diagnose the suite of star-forming galaxy properties that will be offered to us in the next decade. Given the uncertain future of observedframe UV capabilities, the CLASSY templates and models embody an indispensable toolset.
Here we present Paper I of the CLASSY Treasury detailing the release of the first HLSP: the CLASSY coadded spectral atlas. In Section 2 we introduce the CLASSY atlas of FUV and optical spectra. We describe the sample selection in § 2.1, the HST/COS spectroscopic observations and coaddition process in § 2.2, and the ancillary optical spectra in § 2.3. The details of the coadded spectra data release are described in Section 3, including the coadded spectra file format ( § 3.1), digital access to the resulting data release ( § 3.2) for ease of utility, and a comparison to previous FUV surveys in ( § 3.3).
In order to give a general overview of the CLASSY galaxies and their spectra, we discuss optical measurements of their host galaxy properties in Section 4. Specifically, we revisit the ancillary optical spectra used in § 4.1, the emission line measurements ( § 4.2) and their reddening corrections ( § 4.3), and calculations of the electron densities ( § 4.4), temperatures ( § 4.4), oxygen abundances ( § 4.5), ionization parameters ( § 4.6), and stellar mass and star formation rates ( § 4.7).
The resulting global properties of the CLASSY sample are discussed in Section 5. Finally, a demonstration of the subsequent scientific studies that can be performed with CLASSY is presented in Section 6. A summary of Paper I can be found in Section 7. We refer the reader to the Appendix for additional CLASSY resources. Specifically, Appendix A contains tables of the CLASSY FUV+optical spectral observations (Tables A and 4). Appendix B contains tables of the CLASSY galaxy properties derived from optical spectra (Table 5) and photometry (Table 6). Appendix C discusses the details of our spectral energy distribution fitting. Appendix D compiles useful notes on individual CLASSY galaxies. Finally, appendix E displays the CLASSY coadded spectrum for each galaxy in the sample. Note that, for consistency, the CLASSY objects are designated by an abbreviation of their sexigesimal coordinates as JHHMM±DDMM. A flat ΛCDM cosmology (H 0 = 70 km s −1 Mpc −1 , Ω m = 0.3) was assumed throughout this work.

THE CLASSY ATLAS OF UV+OPTICAL SPECTRA
The primary community data product of the CLASSY treasury is high-quality, coadded FUV plus optical spectra for the 45 star-forming galaxies in the CLASSY sample. These CLASSY spectra will enable a number of important synergistic studies between the stars and gas of galaxies that will result in additional CLASSY HLSPs (see § 6) and an untapped wealth of potential community science investigations. Below we describe the CLASSY sample selection in § 2.1, the HST/COS FUV observations in § 2.2, and briefly describe the spectra coaddition process in § 2.3. We refer the reader to James et al. (2021; hereafter, Paper II) for a detailed discussion of the data reduction and technical details of the coaddition process, along with best practices for COS data of extended sources.
In order to assemble an atlas with unprecedented S/N and spectral resolution that efficiently utilizes the archive, we chose a sample of nearby, UV-bright galaxies, with existing high S/N spectra in at least one of the high-spectralresolution gratings on HST/COS. Note that we selected spectra from a single instrument to ensure a consistent aperture. We also highlight that COS was chosen over STIS due to its higher sensitivity in the FUV. Existing archival spectra were first examined using the HSLA, from which 101 nearby (z < 0.2) galaxies were selected based on their high-S/N ( 7 per 100 km s −1 resolution element) COS spectroscopy in at least one medium-resolution grating (G130M, G160M, or G185M). The following selection criteria were then implemented to ensure the efficient completion of high-quality, 6 BERG ET AL. comprehensive rest-frame FUV spectra for a large, diverse sample of star-forming galaxies: 1. Star-forming galaxies: Any targets with secondary classifications of quasi-stellar object (QSO) or Seyfert in the HSLA were removed. Remaining targets were visually confirmed to not have obvious broad QSO emission features. 2. Low redshift: Redshifts from existing optical spectra of z < 0.2 were required to capitalize on the FUV sensitivity of COS, which is significantly higher than the NUV detectors (COS and STIS), but falls off rapidly redward of 1950 Å. 3. Compact: Through a visual inspection of the existing HST/COS NUV acquisition images, we selected galaxies with single dominant emission regions and relatively compact morphologies in the sense that the Gaussian distribution of their NUV light had a full-width halfmaximum (GFWHM) <2. 5. This ensured that the COS 2. 5 aperture (optimized for point sources) will capture most of the FUV light from the galaxy. Note, however, that this is a flawed criterion, as COS UV acquisition images are vignetted and thus the optical sizes of the CLASSY sample are significantly larger for a few galaxies. 4. UV bright: We required Galaxy Evolution Explorer (GALEX; GR6; Bianchi et al. 2014) FUV magnitudes of < 20 mag to allow for an efficient observing strategy, resulting in < 3 orbits per grating setting, on average. 5. Grating efficiency: From the HSLA, we found three galaxies that had observations of sufficient quality in all of the proposed gratings (J0337-0502, J0942+3547, and J0934+5514). Additionally, 32 galaxies had observations of sufficient quality in two of the proposed gratings, allowing their rest-frame FUV coverage to be completed with the addition of a single grating. Of these, 16 galaxies had existing G130M+G160M observations and so CLASSY completed their FUV spectra with the addition of G185M and/or G225M observations, while the remaining 16 had G160M+G185M observations and were completed with the addition of G130M observations. Among the galaxies with only single-grating observations (G130M or G160M), we preferentially chose the five with existing high-quality G160M observations because new G130M observations are less time intensive due to the gratings' higher sensitivity. 6. Broad parameter space coverage: Measurements from existing optical spectra were used to select broad and efficient coverage of galaxy properties to ensure the sample is suitable for comparison to star forming galaxies near and far (z < 7): star formation rates (SFRs), specific SFRs (sSFRs), stellar masses, nebular metallicities, and ionization parameter. The ranges of stellar mass, SFR, and metallicity are shown in Figure 2. To better sample the higher nebular densities observed for z > 1 galaxies (e.g., Shirazi et al. 2014;Sanders et al. 2016;Kaasinen et al. 2017;Harshan et al. 2020) and directly study the effect of electron density on rest-FUV galaxy properties, we mined the Sloan Digital Sky Survey Data Release 12 2 (SDSS-III DR12; Eisenstein et al. 2011;Alam et al. 2015) and existing literature in order to unearth additional targets with [S II] λλ6717,6731 densities of n e > 400 cm −3 . Although such targets are extremely rare at z ∼ 0, we were able to compile a sample of 10 galaxies in our so-called "high-density sample" that met criteria 1-4. Of these, five galaxies were already in our sample, while the other five had no previous HST/COS observations, and so required new G130M+G160M+G185M observation.
As a result of the search just described, the final CLASSY sample contains 45 nearby (z < 0.2), UV-bright (m FUV < 21 AB arcsec −2 ), relatively compact (GFWHM NUV <2. 5), galaxies. Note, however, that COS Instrument Handbook defines compact objects as those having a GFWHM<0. 6; in this sense, the CLASSY sample includes a distribution of compact to extended morphologies. For comparison, we determined the optical size of each galaxy by measuring the average half light radius (r 50 ) from u and g−band SDSS, DES, and PanSTARRS imaging (listed in Table 6); the optical sizes are 0. 22 larger on average than the UV.
The HST/COS NUV acquisition images are shown in comparison to the optical imaging in Figure 3, demonstrating that the peak of the optical and UV surface brightness profiles are aligned and that most of the stellar light is captured within the COS 2. 5 aperture. While the optical imaging shown is extended due to the mixture of nebular emission (some with EW(Hα) > 10 3 Å) and light from older (relative to the UV) stellar populations, the similar 2. 5 COS aperture and 3 SDSS aperture will capture comparable nebular emission in their spectra. This allows the FUV and optical spectra to be used in concordance to characterize the stellar and nebular properties of the CLASSY galaxies. Note, however, that four of the CLASSY galaxies (J0036-3333, J0337-0502, J0405-3648, and J934+5514) have additional knots of starformation clearly outside of the aperture.
Additionally, the CLASSY sample covers a range of galaxy properties, listed in Tables 5 and 6 in the Appendix (see Section 4 for details of property calculations), including stellar masses of 6.22 < log M (M ) < 10.06, star formation rates of −2.01 < log SFR (M yr −1 ) < 1.60, direct nebular metallicities of 6.98 <12+log(O/H) < 8.77, electron densities of 10 < n e (cm −3 ) < 1120, ionizations, characterized by O 32 = [O III] λ5007/[O II] λ3727, of 0.54 < O 32 < 38.0, and 7 reddening values of 0.001 < E(B−V) < 0.673. This broad sampling of parameter space allows the CLASSY sample to uniquely serve as templates for interpreting galaxies across all redshifts. However, it is important to note that CLASSY sample is biased, on average, to more extreme O 32 values (25th and 75th percentiles of 7.70 and 8.25), low stellar masses (25th and 75th percentiles of 7.06 and 9.04), higher SFRs (25th and 75th percentiles of -0.78 and 1.03), and compact UV morphologies.
The range of the total stellar masses, SFRs, and oxygen abundances are shown in Figure 2. In comparison to other z ∼ 0 (e.g., the Low-Mass Local Volume Legacy: LMLVL, Berg et al. 2012) and z ∼ 2 (e.g., Whitaker et al. 2014) surveys, the CLASSY sample follows the expected massmetallicity trend of local galaxies (left panel of Figure 2), but is offset to higher SFRs that are more typical of galaxies at cosmic noon (e.g., Madau & Dickinson 2014; right panel of Figure 2).
New Data: In order to complete the FUV spectral coverage of the CLASSY sample, the archival observations were supplemented with 76 new HST/COS spectroscopic datasets of 42 galaxies, compiled over 135 orbits in Cycle 27 as part of PID 15840. These data consist of 27 G130M datasets, 5 G160M datasets, 29 G185M datasets, and 8 G225M datasets. Note that the G225M datasets were necessary to ensure coverage of the C III] λλ1907,1909 emission lines for the 8 galaxies with z > 0.11. Details of all individual datasets are provided in Table A. For each CLASSY galaxy, the archival and CLASSY obser-vations are denoted by their program ID (PID), followed by their respective acquisition image and FUV spectral observation details.

General Observing Procedure
Target Acquisition: The entire CLASSY sample has precise and accurate target coordinates that enabled consistent pointings between archival and new data with a precision of < 0.1 . For the 37 targets with previous HST/COS observations, the archival data coordinates were used to acquire new data. For the five CLASSY galaxies selected to target high [S II] electron densities (n e > 300 cm −3 ), no previous HST/COS observations exist. Instead, coordinates from the SDSS were used for four of these high-density galaxies (J0944+3442, J1200+1343, J1253-0312, J1323-0132) and coordinates for J0127-0619 were adopted from the HST Faint Object Spectrograph observations of Thuan et al. (1996). The overall excellent positional alignment of the SDSS and COS apertures is shown in Figure 3.
Target acquisitions were performed using the ACQ/IMAGE mode with the PSA aperture and MIRRORA when possible (87 datasets) for the COS/NUV configuration. The resulting NUV images are useful for understanding the spatial structure of light entering COS aperture. However, MIRRORB, was employed out of caution for five galaxies (J1016+3754, J1105+4444, J1150+1501, J1157+3220, J1359+5726) that are especially bright in the UV. While the use of MIRRORB serves to protect the detector, it has the drawback of producing a double peaked image that impedes our interpretation of the spatial distribution of light 4 . The CLASSY NUV acquisition images are shown in Figure 3, where targets observed with MIRRORB are easily recognizable by their more pixelated distribution of blue NUV light and their additional 'ghost' peak along the spectral axis (e.g., J1016+3754; see, also, Table A and Paper II). Considering the sample as a whole, the peaks of the UV light distributions of the CLASSY galaxies are clearly well aligned with the positioning of the COS 2. 5 aperture (blue circle). As a result, the COS aperture captures the majority of the UV emission of the CLASSY galaxies. In comparison to the GALEX fluxes of the CLASSY sample, we estimate that on average more than 60% of the total UV flux is captured by the COS aperture. Spectral Observations: The HST/COS FUV observations used in the CLASSY survey were all taken in the TIME-TAG mode using the 2. 5 PSA aperture. Note that no restrictions were placed on the position angle of the new CLASSY datasets; the implications of this decision will be discussed in Paper II. The FP-POS=ALL setting was used for most of the 8 BERG ET AL.  Clearly the CLASSY MZR is typical of local low-mass star-forming galaxies, but the sample is offset to higher SFRs that are more typical of galaxies at cosmic noon.
datasets to take four spectral images offset from one another in the dispersion direction, increasing the cumulative S/N and mitigating the effects of fixed pattern noise. TIME-TAG mode was employed, allowing abnormal detection events to be filtered out, eliminating cosmic rays in the process. For the G130M grating with the CENWAVE set to 1291, only FP-POS = 3 and 4 were available for the new observations (see the COS2025 guidelines 5 for further information). Each target was observed for the time predicted to reach a S/N 5 per resolution element in the continuum for the G130M and G160M gratings. The input continuum flux was measured at 1500 Å from, in priority order, existing COS spectra, GALEX fluxes, or an extrapolated model fit to existing optical data. For the G185M and G225M gratings, observation times were determined to detect the C III] λλ1907,1909 emission lines with a strength of S/N 3, based on observed C/O values for a given metallicity (see, e.g., Berg et al. 2019a), the observed [O III] λ5007 flux, and the theoretical emissivities. All raw data for the new and archival observations were retrieved from the HST archive and were reduced with the CALCOS pipeline (v3.3.10) using the standard twozone extraction technique. In Paper II we provide a detailed discussion on the optimized extraction techniques employed for the extended sources within the CLASSY sample.

Coaddition Process
Overview: Below we describe the main steps in the coaddition process, however, the discussion of the other interme-5 https://www.stsci.edu/hst/instrumentation/cos/proposing/cos2025-policies diate analyses and their impact on the reduced spectra are discussed in Paper II.
The primary HLSPs of the CLASSY treasury survey are high S/N multi-grating coadded spectra. However, each HST/COS grating has a different spectral resolution that must be accounted for when combining data from multiple gratings. Additionally, while the CLASSY galaxies have relatively compact stellar clusters, their nebular emission generally fills the COS aperture (at minimum, with diffuse light; see Figure 3), and, consequently, experience varying degrees of vignetting and degradation in spectral resolution depending on the shape of their overall UV light profile entering the COS aperture.
To address the challenges associated with coadding multiple datasets that consist of different grating configurations, exposure times, and resolutions, we developed a custom python reduction code (available as a jupyter notebook on the CLASSY HLSP website). The main steps followed in the code are depicted in Figure 3 and include: (1) joining the segments/stripes of individual grating datasets, (2) coadding any multiples of individual grating datasets, including all cenwave configurations, (3) coadding datasets across gratings, (4) binning the spectra, and (5) correcting for Galactic contamination. Below we describe each step in more detail.
1. Joining grating segments and stripes: The HST/COS CLASSY spectra were acquired across multiple FUV detector segments and NUV detector stripes, so the first step of the coaddition process unifies these components for each dataset. For the G130M, G160M, and G140L gratings, the COS FUV Figure 3. The HST/COS NUV acquisition images of the 45 galaxies in the CLASSY sample are shown in blue, overlaid on top of optical SDSS r-band imaging (unless otherwise noted; red-yellow color bar). The 2.5 COS aperture used for the UV CLASSY spectra is shown as a blue circle and is very similar to the SDSS 3 aperture in orange. Galaxies missing SDSS optical spectra do not have a 3 orange circle. Relative to the HST/COS aperture, most of the NUV light is captured, but a significant fraction of the extended optical light, and subsequently nebular emission from strong emitters (e.g., with EW(Hα 10 3 Å), may still be missed.  detector has two segments (FUVA, FUVB) 6 , while the NUV detector used for the G185M and G225M gratings has three stripes (FUVA, FUVB, FUVC). For each dataset, we extracted the WAVELENGTH, FLUX, ERROR, and DQ_WGT (or data quality flag weights) columns for each segment or stripe from the x1dsum files produced by the CALCOS pipeline and appended them. Because the dispersion varies with wavelength, we then used the pysynphot package in python (STScI Development Team: Lim, P. L., Diaz, R. I., & Laidler, V. 2013) to define a source spectrum and interpolate onto a new wavelength grid defined by the largest step in the original wavelength array. Note that there are wavelength gaps between each set of segments/stripes, as well as gainsag holes resulting from detector damage due to extensive airglow emission exposure, so the full spectrum produced for each galaxy is contiguous rather than continuous.
2. Coadding single grating datasets: If multiple datasets of a given grating exist for an individual target, such as observations acquired over multiple visits or targets observed at multiple central wavelength configurations (CENWAVEs), they must be combined in step two of Figure 3 before proceeding. Such collections of grating datasets exist for the G130M observations of one galaxy (J0934+5514 / I Zw 18), the G160M observations of five galaxies, and the G185M observations of eight galaxies, or ten CLASSY galaxies in total. These grating datasets were then coadded using a combined normalized data quality weight (using the DQ_WGT array; to filter out or de-weight photons correlated with anomalies/bad data) and exposure-time-weighted calibration curve (to preserve the Poisson count statistics). This weighting method was used for all instances where coadding was performed.
3. Coadding multiple grating datasets: Accounting for the change in spectral dispersion between individual gratings, and datasets, is a particularly difficult task. Not only does dispersion change as a function of wavelength across the detector, it also changes as a function of lifetime position on the COS FUV detector (the CLASSY data span all 4 lifetime positions thus far), and are degraded by extended CLASSY sources depending on their specific UV light profiles and aperture position angles. To take this into account, at each stage of the co-adding process, we performed a common resampling of the wavelengths to the highest dispersion of a given combination of gratings. The different spectral dispersion of the coadds allow CLASSY to be used for many scientific studies that require a range of spectral resolutions. Specifically, we produce four spectral coadds for each galaxy with different resolutions: Note that the flux calibration was performed for each dataset during the initial reduction by calcos, however, the relative fluxing between gratings was also carefully considered during the coadding process. In short, the G160M spectrum was treated as the flux anchor of each spectrum and the continuum of all other datasets were fit and scaled to G160M prior to coadding. See Paper II for a detailed discussion of the flux calibration and its effects on the final coadded CLASSY spectra. 4. Binning the spectra: In order to gain signal-to-noise, the coadded spectra were also binned in the dispersion direction. The COS medium-resolution gratings have a resolving power of R ∼ 15, 000 for a perfect point source, which corresponds to six FUV detector pixels in the dispersion direction. All coadds were binned by a factor of six to reflect this using the SpectRes (Carnall 2017) python function, which efficiently resamples spectra and their associated uncertainties onto an arbitrary wavelength grid while preserving flux.
5. Correcting for Galactic Contamination: Finally, proper interpretation of UV spectral properties requires the removal of potential contamination from Geocoronal emission and Milky Way interstellar absorption. It is important to note that this contamination affects the absorption spectra of the entire CLASSY sample, whether directly by deforming intrinsic features or indirectly by degrading adjacent regions of good continuum. However, the high-ionization Milky Way absorption features (i.e., Si IV and C IV) are typically weak (Savage et al. 2000) relative to the intrinsic profiles, and so can often be ignored. To correct the CLASSY spectral templates, we create mask arrays assuming central velocities of v cen = 0 km s −1 and conservative widths of ∆v = 500 km s −1 for the very-high, high, and moderate resolution coadds and ∆v = 1000 km s −1 for the low resolution coadds.
The final high-quality CLASSY coadded spectra are shown in the Appendix in Figure 12. These coadds provide both high resolution spectra over the ∼ 1200 − 1700 Å range of each object that is important to studies of stellar and ISM absorption features and contiguous FUV ∼ 1200 − 2000 Å wavelength coverage at low resolution that is necessary to simultaneously interpret the stars and gas. However, due to the varying extended nature of the CLASSY sample, the observed spectral resolutions differ considerably amongst the four coadditions for different galaxies. Specifically, the extended source distribution causes the nominal point source line spread function to broaden in complex, unpredictable ways (see Paper II for further discussion). Therefore, to estimate observed spectral resolution of each galaxy, we measure the Gaussian FWHM of the best (i.e., cleanest and highest S/N) Milky Way ISM absorption feature present. Note that this measurement assumes unsaturated, unresolved lines. Table 3 lists the nominal and measured spectral resolutions of the CLASSY coadded spectra; 41 galaxies had Milky Way ISM features that were well fit by a Gaussian, while the remaining 4 lacked good Milky Way features to fit. While the choice of coadd dataset / spectral resolution is left to the user, Table 3 suggests that the moderate resolution coadds are the most appropriate spectra for most of the CLASSY sample.
To ensure robust direct T e measurements for J0808+3948, J0944+3442, and J1545+0858, very high-S/N optical spectral were obtained using the Multi-Object Double Spectrographs (Pogge et al. 2010; MODS,) on the Large Binocular Telescope (LBT, Hill et al. 2010). Simultaneous blue and red MODS spectra were obtained using the G400L (400 lines mm −1 , R ≈ 1850) and G670L (250 lines mm −1 , R ≈ 2300) gratings, respectively. Observations used the 1 ×60 long slit for 3×900s exposures, or 45 min of total exposure per object. The slits were centered on the same coordinates as the COS apertures. Spectra were reduced, extracted, and analyzed using the beta version of the MODS reduction pipeline 8 which runs within the XIDL 9 reduction package. One-dimensional spectra were corrected for atmospheric extinction and flux-calibrated based on observations of flux standard stars Oke (1990). Further details of the MODS reduction pipeline can be found in Berg et al. (2015).
Additional optical spectra we gathered from archival VLT/MUSE IFU spectra and Keck/ESI spectra from Sanders et al. (2021), and were prioritized over the SDSS spectra due to their superior S/N and spectral resolution. For galaxies with IFU spectra, we extract a spectrum using a 2. 5 aperture to better match the COS aperture. The details of the ancillary optical spectra adopted for the CLASSY sample are listed in Table 4.

COADDED SPECTRA DATA RELEASE
Overview: The result of the coaddition process described in Section 2.3 is the first CLASSY HLSP: a set of high-quality, moderately high-resolution, uniformly calibrated spectral templates with contiguous FUV coverage. The CLASSY atlas spectra are stored as multi-extension fits (MEF) files containing up to 20 total extensions of spectra, with four possible spectral resolutions and five extensions per resolution set corresponding to combinations of Galactic reddening and redshift corrections. Any use of the CLASSY spectral atlas should cite this work (Paper I) and Paper II. Below we further describe the coadded spectra HLSPs and how to access them.

Coadded Spectra File format
The CLASSY coadded spectra HLSP consists of 45 MEF files, or one for each CLASSY galaxy. For a given galaxy, its MEF file can have up to 21 extensions (0-20), where the first extension (0) is the primary header containing general details relevant to the spectrum and the remaining 20 extensions contain spectral data tables. Information important to the spectra of a given galaxy are stored in the primary fits header keywords. As shown in Figure 4, there are two groups of keywords that describe either the target or the observations used to produce the coadded spectra.
The CLASSY coadded spectra are stored in many different forms to maximize convenience for the user depending on their scientific needs. Specifically, for each of the four potential coadd resolutions of a given galaxy, there are a set of five extensions corresponding to: Ext. 1: The observed-frame coadded spectrum.

G160M coadd
Step   . The main steps in the CLASSY coaddition process. The CLASSY spectra are coadded from a combination of G130M, G160M, and one or more of the G185M, G225M, and G140L HST/COS gratings. The first step highlights how the segments of the FUV datasets and the stripes of the NUV datasets are resampled to a common wavelength grid. Then, in step two, any sets of multiple grating datasets for a given galaxy are coadded using the weighting method described in Section 2.3. Finally, to provide the optimal combination of wavelength coverage and spectral resolution for different science cases, the gratings for a given galaxy are coadded into four spectra with different resolutions.
Extensions 2 and 5 are corrected to the observed frame using the redshifts reported in Table A, which were adopted from the SDSS, when available. For galaxies not observed by the SDSS, redshifts were sourced from another optical spectrum (see adopted references in Table 4). Extensions 4 and 5 contain coadded and binned coadded spectra that were corrected for foreground Galactic extinction. We used the Python dustmaps (Green 2018) interface to query the Bayestar 3D dust maps of Green et al. (2015) to determine the total foreground Galactic reddening along the line of sight of the coordinates of each CLASSY galaxy. We note that the Green et al. (2015) map was adopted over more recent versions due to its more optimal coverage of the CLASSY sample. Specifically, the Green et al. (2015) map provides E(B-V) values for all but two galaxies (J0405-3648 and J0036-3333) in the CLASSY sample, whereas the most recent Green et al. (2019) map only covers 16 of the 45 galaxies. Fortunately, for these 16 galaxies covered by both maps, the average difference in E(B-V) is only 0.030 magnitudes, with 25th and 75th percentiles of 0.012 and 0.040 magnitudes, respectively. For J0405-3648 and J0036-3333, the galactic E(B-V) values were taken from Schlafly & Finkbeiner (2011). The foreground Galactic extinction correction was then applied using the Cardelli et al. (1989) reddening law.
As shown in Figure 4, these five extensions form a set of coadded spectra that are repeated for each resolution that exists for a given galaxy. All 45 CLASSY galaxies have VHR and HR coadds, while 42 have MR coadds. One of the three galaxies missing an MR coadd, J1112+5503, is missing G185M observations owing to an observational failure that was not made up as it was the final CLASSY observation. The other two galaxies missing MR coadds, J1044+0353 and J1418+2102, already had significant C III] λλ1907,1909 detections in their archival G140L spectra that deemed follow-up G185M spectra unnecessary. Including these three galaxies, six of the CLASSY galaxies have G140L spectra and subsequent LR coadds. In summary, 4 CLASSY galaxies have 20 spectral extensions (VHR, HR, MR, and LR coadds) and the remaining 41 have 15 extensions (VHR, HR, and MR coadds).
Each extension in the CLASSY coadded spectra HLSPs is a simple data table with 5 columns. The columns correspond to wavelength (Å), flux density, error, S/N ratio, and a mask, where the fluxes and errors are given in units of 10 −15 erg s −1 cm −2 Å −1 . The spectral mask in the 5th column of each extension can be used to remove regions contaminated by Geocoronal emission from Lyα and O I λλ1301,1305,1306 and Milky Way interstellar absorption features. For most of the CLASSY sample, the redshift is large enough to cleanly separate the Geocoronal and Milky Way features from extragalactic features of interest. However, for galaxies with z 0.003, and depending on the extent of the stellar wind and outflow features in a given galaxy, users should examine features for contamination. Finally, each fits file has a table detailing the individual observations used in the coadds as the last extension. A detailed demonstration of the final CLASSY atlas spectra is shown for an absorption-line galaxy, J0036-3333, and an extreme emission-line galaxy, J1323-0132, in Figures 5 and 6, respectively. These moderate-resolution coadded spectra highlight the diversity of spectral features observed in the CLASSY sample.

HLSP Digital Access
All CLASSY Treasury HLSPs will be released to the astronomical community in digital form as they are completed. The publication of this paper marks the release of the first HLSP, the CLASSY FUV spectral atlas of star-forming galaxies, on two main platforms.
First, all CLASSY HLSPs will be available via the CLASSY HLSP home page: https://archive.stsci.edu/hlsp/ classy, which also serves as the main website for the CLASSY Treasury and relevant information. Importantly, this website provides simple tarball downloads of individual CLASSY HLSPs, as well as the entire CLASSY HLSP data collection. In addition to full data access, the CLASSY HLSP home page provides machine-readable table downloads for the tables of properties in this paper (Tables A,  5, and 6), interactive spectra that allow quick-look examination of the coadded spectral features, user-friendly jupyter notebook guides to accessing and using the CLASSY HLSPs, and important links such as to the NASA Astrophysics Database System (ADS) library of CLASSY publications.
The second platform for accessing the CLASSY HLPSs will be the Mikulski Archive for Space Telescopes (MAST) via the CLASSY MAST website: https://mast.stsci.edu/ search/ui/#/classy (expected early 2022). The CLASSY MAST portal will provide a unique search-and-selection method to access the CLASSY HLSPs, where users can select a subsample of the data products based on HLSP type, filename, or target properties such as coordinates or stellar mass. This search tool will continue to be developed based on the specifics of each HLSP release in order to allow more efficient and straightforward access to data products that are of interest to the user. In this manner, appropriate subsets of the CLASSY HLSPs can easily be selected for a given scientific analysis, to match specific survey sample properties, and more.

Comparison to Previous FUV Spectroscopic Surveys
The CLASSY atlas provides spectral templates with the powerful combination of contiguous 1200-2000 Å FUV wavelength coverage, high resolution (λ/∆λ= 15, 000), and high S/N (average S/N = 6.4 per resolution element). These templates simultaneously probe the interactions between stellar winds, galaxy-scale outflows, and nebular emission and provide a unique dataset that probe spatial scales of 0.1-10.7 kpc across the COS aperture at the range of redshifts of the CLASSY sample. As we discussed in Section 1.3, this combination of qualities of the CLASSY survey is unique with respect to previous FUV spectra surveys.
Interestingly, few CLASSY galaxies are in previous FUV spectral atlases, making it difficult to directly compare them. Fortunately, IUE and CLASSY observations both exist for two galaxies: J0337-0502 and J0934+5514. For a visual comparison, the HST and IUE spectra for J0337-0502 are plotted in Figure 7, showing substantial, but expected differences. The smaller aperture of COS relative to the IUE (10 ×20 aperture) records lower UV continuum fluxes (however, this would not be expected for most of the NUVcompact CLASSY sample) and some features are present in only the IUE spectrum, which also contains the SE cluster region. For example, there appear to be weak emission features present just blueward of 1400 Å and at 1750 Å, likely N III], that are emitted by the SE region but not the NW region. Perhaps the most important difference, though, is the significant number of features and details revealed by the higher resolution of the CLASSY spectra (∼ 0.1 Å resolution compared to the ∼ 6 Å of the IUE spectrum). Specifically, a number of significant ISM absorption lines and nebular emission lines are recovered, such as the O I and Si II lines shown in the left-most inset window of Figure 7 and the C IV, He II, O III], and C III] lines shown in the other three inset windows. Figure 7 clearly demonstrates that the CLASSY survey provides a dramatic leap forward in FUV spectral atlases and is an invaluable tool for the interpretation and modeling of star-forming galaxies across redshifts. HST proposal ID for the G130M spectra G130M_CW

CLASSY GALAXY PROPERTIES
Central wavelength of the G130M spectra G160M_ID HST proposal ID for the G160M spectra G160M_CW Central wavelength of the G160M spectra

Ext Header
Ext 5

Ext Header
Ext 4

Ext Header
Ext 3

Ext Header
Ext 2

Ext Header
Ext 1

REST VHR COADD + GAL CORR + BINNED
HST proposal ID for the G185M spectra G185M_CW Central wavelength of the G185M spectra G225M_ID HST proposal ID for the G225M spectra G225M_CW Central wavelength of the G225M spectra G140L_ID HST proposal ID for the G140L spectra G140L_CW Central wavelength of the G140L spectra G130_BIN Size of bin (pixels) used for G130M coadd HR_BIN Size of bin (pixels) used for HR coadd MR_BIN Size of bin (pixels) used for MR coadd LR_BIN Size of bin (pixels) used for LR coadd

RADESYS
Coord. system used for RA and Decl.

ISM Lines Nebular Lines
Fine Structure Lines Stellar Wind Lines Photospheric Lines Fig. 5. This moderate-resolution coadded CLASSY spectrum for J0036-3333 (black line) is rich with stellar wind features, such as the Si IV and C IV features, and interstellar medium absorption lines, including the strong Si II and C II features. J0036-3333, however, contains no significant UV emission features besides Lyα. Additionally, the flux±error spectrum, shown in gray, demonstrates the high S/N of CLASSY and the ability to resolve absorption in J0036-3333 from contamination by Geocoronal and Milky Way features (regions where the spectrum/black line isn't shown).   . This moderate-resolution coadded CLASSY spectrum for J1323-0132 (black line) is strikingly different than that of J0036-3333 in Figure 5, highlighting some of the emission lines that are observed at varying strengths for a subset of the CLASSY sample. The spectrum has particularly strong emission features from Lyα, C IV, He II, O III], and C III], additional weak detections from the high-ionization O IV, S IV, and N IV] lines, but minimal absorption features. Note the high S/N of this spectrum is shown by the gray flux±error spectrum, while the spectrum/black line isn't shown for regions contaminated by Geocoronal and Milky Way features.

J1323-0132
Overview: The CLASSY sample of star forming galaxies was carefully chosen to overcome the common limitations of other surveys introduced by the biases of their samples. Specifically, while CLASSY only targets relatively FUVbright galaxies (GALEX m FUV < 20 and F 1500 1 × 10 −15 erg s −1 cm −2 ), which also ensures the presence of young stellar populations, it probes a broad parameter space in stellar mass, star formation rate, gas-phase abundance, ionization parameter, and gas-phase electron density in a nearly uniform manner.
Below we describe the measurements of CLASSY nebular quantities and their resulting trends. All physical property calculations used the emission line ratios measured from the optical spectra listed in Table 4 and the PYNEB package in PYTHON (Luridiana et al. 2012; with the atomic data adopted in Berg et al. (2019a).

Contiguous FUV+ Optical Spectra
A benefit of the CLASSY sample is the availability of ample ancillary data, including excellent optical spectra for the entire sample. Optical spectra have long been the gold standard source for calculating important nebular properties, including the oxygen abundance, density, and reddening tabulated in Table 1. As discussed in Section 2.3, the majority of the CLASSY sample was also observed as part of the SDSS, which provides full optical spectra using a similar aperture (3 ) to that of the COS UV spectra. Additionally, 8 of the CLASSY galaxies have IFU spectroscopy from either VLT/MUSE, VLT/VIMOS, or Keck/KCWI, while 9 of the CLASSY galaxies have LBT/MODS or Keck/ESI longslit spectra (see Table 4). We take advantage of these high-S/N spectra in order to calculate direct T e abundances when the SDSS spectra had insufficient signal-to-noise.
There are obvious differences in the aperture sizes between the longslit optical spectra and the circular COS aperture, yet, these differences do not result in significant changes in the nebular properties calculated. A full analysis of the aperture effects on the nebular properties of CLASSY galaxies with multiple sources of optical spectra will be discussed in a forthcoming paper by Arellano-Córdova et al. (2022). However, even when the UV and optical spectra have similar apertures, one significant difference is the vignetting of the COS aperture for the UV spectra. Fortunately, for the current purpose of deriving galaxy properties, only relative flux ratios are needed, and so the absolute flux calibration is irrelevant.

Emission Line Measurements
Several of the galaxy properties of interest are best measured from the photoionized nebular emission lines present in the adopted optical spectra of CLASSY galaxies. Here in Paper I we focus on measuring the redshift (z), intrinsic dust reddening (E(B −V )), nebular electron density (n e ) and temperature (T e ), and gas-phase metallicity (12+log(O/H)) that characterize the CLASSY sample. More detailed discussions of the optical and UV line measurements, their analysis, and their resulting HLSPs (see HLSPs 4, 6, 7, 8, and 9 in Section 5) will be presented in forthcoming papers by Arellano-Córdova et al. (2022) and Mingozzi et al. (2022). Below we briefly describe the CLASSY optical emission line measurements.
For the purposes of calculating the nebular properties of the CLASSY sample, we measured the strength of as many of the optical emission lines necessary to determine: For all of the optical spectra, regardless of telescope and instrument used, we first modeled the stellar continuum using the Starlight 10 spectral synthesis code (Fernandes et al. 2005) using stellar models from Bruzual & Charlot (2003). This step is important for removing any absorption component present in the Balmer emission lines. Emission lines were then fit in the continuum-subtracted spectrum with Gaussian profiles using MPFIT (Markwardt 2009). The fit parameters (i.e., velocity width, line center) of neighboring lines were tied together, allowing weak or blended features to be measured simultaneously. The uncertainty on the flux measurements is assumed to be the uncertainty on the fit returned from the Monte Carlo minimization code. The complete database of measured emission line fluxes will be released in a forthcoming HLSP via Arellano-Córdova et al. (2022) and Mingozzi et al. (2022).
We note that the dominant emission component is a narrow, nebular component for the optical emission lines measured for the CLASSY sample. However, broad emission can occur in the presence of stellar winds, outflows, and shocks, and is often clearly visible at the base of the strongest emission features. Because the broad components only contribute a small fraction to the total line flux, we restrict the present analysis to the narrow components only and reserve the analysis of multi-component fits for the forthcoming work of Mingozzi et al. (2022).

Reddening Corrections
Following the same method discussed in Section 3.1 for the UV spectra, flux measurements for the optical emis- Comparison of rest-frame spectra for J0337-0502 to demonstrate the improvement of CLASSY over previous FUV spectral atlases. The IUE spectrum is relatively low resolution (λ/∆λ∼ 300) and consists of the integrated light within its large 10 ×20 aperture. In contrast, the CLASSY coadded spectral templates are characterized by an unrivaled combination of contiguous 1200-2000 Å FUV wavelength coverage, high spectral resolution (λ/∆λ= 15, 000), and high S/N (average S/N = 6.4 per resel) for much more compact regions that can be targeted with the 2. 5 COS aperture. Both spectra are normalized at 1450 Å. Note that the breaks in the CLASSY spectrum are due to chip gaps and that the Geocoronal Lyα and O I emission have been whited-out. While few features are apparent in the IUE J0337-0502 spectra, the CLASSY spectrum reveals a significant number of absorption and emission features that enable simultaneous studies of the massive stellar population, ISM outflows, and nebular emission.
sion lines were first corrected for Galactic extinction using the Cardelli et al. (1989) reddening law with values queried from the Green et al. (2015) extinction map using the PYTHON DUSTMAPS interface (Green 2018). Subsequently, the emission lines were corrected for their corresponding intrinsic galaxy dust reddening. The dust reddening values, E(B − V ), were determined using the relative intensities of the strongest Balmer lines available in a given spectrum (i.e., Hα/Hβ, Hγ/Hβ, Hδ/Hβ) with an iterative application of the Cardelli et al. (1989) reddening law, parameterized by an A v = 3.1 × E(B − V ). Note that Wild et al. (2011) found that the nebular attenuation curve has a slope similar to the MW extinction curve, rather than that of the SMC or the Calzetti et al. (2000) curve, and so the Cardelli et al. (1989) extinction law is appropriate for correcting the CLASSY emission line fluxes. The iterative approach allows for a careful consideration of deviations due to electron temperature and density. The initial reddening seed was determined from the Balmer line ratios using theoretical values from PYNEB, assuming a starting temperature of T e = 10 4 K and n e = 10 2 cm −3 for Case B recombination. In turn, this reddening value is then used to determine an initial estimate of the electron density and temperature (using the best measured auroral line available in a given spectrum). This process is then repeated iteratively to modify the E(B − V ), T e , and n e values until the electron temperature changes by less than 20 K between iterations. The final reddening estimate is an error-weighted average of the Hα/Hβ, Hγ/Hβ, and Hδ/Hβ reddening values and was used to reddening-correct the optical emission line intensities.

Electron Densities and Temperatures
Detailed understanding of the nebular properties of star forming galaxies requires knowledge of the electron temperature, T e , and density n e structure in a galaxy compiled from the values in each ionization zone. Such a detailed examination of the physical properties across a nebula and comparison of different T e and n e measurements will be the focus of a later work. Here in Paper I, we seek only to measure a characteristic T e and n e value for each CLASSY galaxy. To do so, a standard 3-zone ionization model is adopted (for a thorough discussion, see Berg et al. 2021a) and the highionization zone (characterized by O ++ ) is assumed to dominate in the young H II regions of the CLASSY galaxies.
The optical spectra compiled for the CLASSY sample are not uniform -the individual spectra vary significantly in their S/N, resolution, and wavelength coverage. While it is desirable to calculate T e values in the best and most consistent manner for the entire sample, no single T e -sensitive auroral line is uniformly detected for the CLASSY sample. As such, direct T e values were determined using the highest S/N auroral line detection for each galaxy. In general, the [O III] λ4363/λ5007 ratio is prioritized for metal-poor galaxies (12+log(O/H) < 8.0) with blue wavelength coverage, while the [S III] λ6312/λλ9069,9532 ratio is prioritized in more metal-rich galaxies and/or spectra with red wavelength coverage, and the [N II] λ5755/λ6584 ratio is useful for lower-ionization and more metal-rich galaxies. The [O II] λλ3727,3729/λλ7320,7330 ratio was only used if no other T e measurements were possible, as the red [O II] lines are sensitive to contamination from dielectric recombination at relatively low densities (n e > 300 cm −3 ). Additionally, the red [S III] lines are susceptible to contamination from atmospheric lines, and so were visually checked and corrected to the theoretical emissivity ratio when needed.
For simplicity, a single, uniform density is assumed and derived from the n e -sensitive [S II] λ6717/λ6731 ratio. The [S II] ratios measured for H II regions are typically consistent with the low-density limit (n e < 10 2 cm −3 ). In this regime, fluctuations on the order of the ratio uncertainty have negligible impact on abundance calculations, and thus supports the assumption of a homogeneous density distribution expected to be n e ∼ 100 cm −3 throughout a nebula.
The uncertainties on the CLASSY T e and [S II] n e measurements were determined using a Monte Carlo simulation: An array of temperatures/densities were calculated for a 2dimensional normal distribution of values generated with 500 values in each dimension, with centers and widths corresponding to the values and one sigma uncertainties, respectively, of the measured emission line flux ratio and electron temperature/density. The uncertainty was then taken as the standard deviation of the resulting distribution calculated. The ionic temperatures used and calculated densities are reported in Table 5 in the Appendix.

Metallicities
Here we determine total oxygen abundances, 12+log(O/H), for the purpose of characterizing the gas-phase metal content of the CLASSY galaxies. The determination of a complete database of UV and optical ionic abundances and relative abundances will be reserved for forthcoming papers by Arellano-Córdova et al. (2022) and Mingozzi et al. (2022).
The total oxygen abundance relative to hydrogen is calculated by summing the abundances of the individual ionic species (X i ) together relative to hydrogen as: where the emissivity coefficients, j λ(i) , are determined assuming the characteristic T e and n e of the corresponding ionization zone. To determine the temperature of an ionization zone not directly measured, we adopt appropriate temperature relationships based on the O 32 ratio (Yates et al. 2020).
For high-ionization (O 32 > 1.25), high-temperature (T e > 2 × 10 4 K; low-metallicity) galaxies, we adopt the theoretical T e − T e relationships of Garnett (1992), while for lowerionization (O 32 < 1.25), lower-temperature (T e < 2 × 10 4 K; higher-metallicity) galaxies, we adopt the empirical T e − T e relationships of Berg et al. (2020). To be precise, the total oxygen abundances (O/H) should be calculated from the sum of four ionization species that can be present and observed in H II regions:  Table 5 in the Appendix. Note that the uncertainties were simply propagated from the emission line measurements in § 4.2, accounting for errors in the continuum subtraction, flux calibration, and dereddening. It may be appropriate to consider an additional systemic uncertainty of +0.1 − 0.2 dex to account for potential depletion onto dust grains (larger at higher metallicities) and biases due to temperature inhomogeneities (larger at lower metallicities).

Ionization Parameters
An important parameter for characterizing the physical nature of the ionized gas in a galaxy is the ionization parameter, q, or the flux of ionizing photons (cm −2 s −1 ) per volume density of H, n H (cm −3 ). Typically the dimensionless, volumeaveraged ionization parameter, logU, is used and defined as U = q/c. Using the photoionization model prescriptions from Table 3 of Berg et al. (2019b), logU values for the CLASSY sample were inferred from the relationship with the observed light-weighted optical [O III] λ5007/[O II] λ3727 ratio. Typical star-forming regions have been found to have average logU values between −3.5 and −2.9 (e.g., Dopita et al. 2000;Moustakas et al. 2010). In comparison, for the CLASSY sample we measure ionization parameters of −3.1 < logU < −1.6. The range of CLASSY ionization parameters extend to much high values than typical star-forming regions, likely due to their enhanced star formation rates (see Figure 2) and ensuing extreme nebular conditions.

Stellar Masses and Star Formation Rates
Stellar mass (M ) and star formation rates (SFRs) were determined for the CLASSY sample by constraining broad-band spectral energy distribution (SED) fitting via the BayEsian Analysis of GaLaxy sEds (BEAGLE, v0.24.0, Chevallard & Charlot 2016) code with UV+optical photometry. Importantly, BEAGLE models both the stellar populations and the nebular emission. Two sets of stellar mass and SFR parameters were computed assuming a constant star formation history (SFH): (1) an aperture set focused on the star-forming regions within the COS 2. 5 aperture and (2) a galaxy set which characterizes entire host galaxies. Details of the SED fitting process, model assumptions (e.g., a Chabrier (2003) Galactic initial mass function was assumed), and tests are discussed in Appendix C.
The resulting stellar masses and SFRs are reported in Table 6 in the Appendix for both the aperture and total galaxy sets. We also measure the time since the onset of constant SFR, which we refer to here as the galaxy age. For host galaxies, we find ages of the constant SFH of our sample spanning < 5 Myr to 5 Gyr, with a 16%-50%-84% range of log Age/yr = 8.18 +0.33 −0.75 . According to these results, the youngest galaxy is J1323-0132, (an extreme emission-line galaxy with EW Hβ = 251 Å) with an age of only 3. −0.04 and σ log SFR/(M yr −1 ) = 0.17 +0.08 −0.06 , similar to the values found for the host galaxy measurements. The fact that aperture measurements lack UV photometry is partly mitigated by the large excesses in broadband flux caused by strong nebular emission from the starburst, and BEAGLES' ability to account for this emission in its models. This excess is most evident in aperture flux, while in measuring total flux it is degraded by the relatively older stellar population outside of the aperture. Note that we have adopted a simple constant SFH for simplicity and consistent comparison with high-redshift galaxies, however, this method can result in large systematic uncertainties (see, e.g., Lower et al. 2020). We will further assess the SED parameters relative to the optical spectra derived properties, as well as explore the effects of different attenuation models, in a future paper.

GLOBAL PROPERTIES OF THE CLASSY GALAXIES
By design, the CLASSY sample broadly probes many galaxy parameters. Specifically, the CLASSY sample spans approximately four orders of magnitude in both SFR and M , roughly two orders of magnitude in both sSFR and gas-phase oxygen abundance, one order of magnitude in the O 32 ionization ratio, and two orders of magnitude in gas-phase electron density. This broad sampling of parameter space is demonstrated in Figure 8 by plotting the properties calculated from the optical spectra and UV+optical photometry described in Section 4.

The SFR-M Relationship
The upper row of Figure 8 shows that the SFRs and stellar masses of the CLASSY sample form a tight, increasing trend when both the total galaxy properties (left) and the total galaxy surface density properties (right) are considered. To parameterize the CLASSY stellar mass versus SFR relationship, we used the python LINMIX Bayesian linear regression code. LINMIX implements a linear mixture model algorithm (Kelly 2007) to fit data with uncertainties on two variables, and includes an explicit treatment of the intrinsic scatter. Using this analysis, we fit the linear relationship and determined the 1σ dispersion. As a result, we find a relationship for the total galaxy properties of log SFR tot. = (0.91 ± 0.07) × log M ,tot. − (7.25 ± 0.61), with a total scatter of σ tot. = 0.49 dex and an intrinsic scatter of σ int. = 0.42 ± 0.07. For the surface density properties, we find logΣ SFRtot. = (0.44 ± 0.13) × logΣ M ,tot. − (0.92 ± 0.25), with a total scatter of σ tot. = 0.37 dex and an intrinsic scatter of σ int. = 0.31 ± 0.05. Due to the relatively small errors determined for the stellar mass and SFR measurements, the scatter seems to be largely intrinsic, accounting for roughly 80% of the total scatter. However, these uncertainties do not incorporate contributions from the uncertainties in the stellar models, initial mass function, or attenuation curve assumptions, and so may be misleading.
The small dispersion in the CLASSY SFR-M relationship is unsurprising: star-forming galaxies are known to follow the Star-forming Main Sequence (MS; e.g., Brinchmann et al. 2004;Noeske et al. 2007;Salim et al. 2007). The MS has further been identified as a plane in the fundamental metallicity relationship (FMR; Mannucci et al. 2010). The tightness of the MS is thought to result from self-regulation that significantly slows the conversion of gas to stars (e.g., Bouché et al. 2010;Davé et al. 2012). Galaxies with elevated SFRs appear as shifted-up outliers from the star-forming MS. It is, therefore, informative to compare CLASSY to the MS.
To this end we consider both nearby and intermediate redshift samples for comparison MSs. At z ∼ 0 we consider two samples: first, the Chang et al. (2015) sample, .0°8 .5°8 .0°7 .5°7 .0°6 . .0°8 .5°8 .0°7 .5°7 .0°6 .5 log sSFR aper. (yr°1) .0°8 .5°8 .0°7 .5°7 .0°6 .5 .0°8 .5°8 .0°7 .5°7 .0°6 .5 log sSFR aper. (yr°1) .0°8 .5°8 .0°7 .5°7 .0°6 .5 log sSFR aper. (yr°1) The CLASSY sample properties, as measured from the optical spectra and UV+optical photometry described in Section 4, span a broad range of parameter space. Top row: Star formation rate is plotted versus stellar mass for both the entire galaxy (left) and in terms of surface density (right). The left plot shows a trend that extends to lower masses and star formation rates, but lies above the previous galaxy sample studies at z ∼ 0 (SDSS: Chang et al. (2015); blue dashed line and LMLVL: Berg et al. (2012) and Lee et al. (2009); cyan dot-dashed lines) and z ∼ 1 − 2.5 (Whitaker et al. 2014; yellow lines). The total and intrinsic scatter of the observed trend are given as σtot. and σint., respectively. Middle row: Specific star formation rate is plotted against gas-phase oxygen abundance, using the total stellar mass and SFR on the left and the aperture values on the right. Here, the majority of the CLASSY sample aligns with or extends from the Curti et al. (2020) relationships for 10 8 < M /M < 10 11 (orange to yellow lines, respectively) star-forming SDSS galaxies, but lie well above the dwarf galaxies of the LMLVL (Berg et al. 2012;Lee et al. 2009). Bottom row: Oxygen ionization ratio is plotted versus density in the left-hand plot, showing the wide range in nebular conditions of the CLASSY sample. In the bottom right-hand plot, gas-phase oxygen abundance is plotted relative to total stellar mass, showing the well-known mass-metallicity relationship. The CLASSY sample seems to follow the same trend as other normal star-forming galaxies (e.g., Berg et al. 2012;Curti et al. 2020), but with larger scatter.
which derived stellar masses and SFRs for approximately 10 6 SDSS+WISE galaxies. Second, and perhaps the best comparison sample, is the 19-galaxy low-mass subsample of the Local Volume Legacy (LVL; Kennicutt et al. 2008) analyzed to have direct abundances and improved mass measurements by Berg et al. (2012). While most other surveys comprehensively cover massive, high-surface-brightness systems in flux-limited samples, the LVL survey provides a multiwavelength inventory of a statistically robust, approximately volume-limited sample of star-forming galaxies, which provides an essential baseline of "typical" galaxies with a similar stellar mass range as the CLASSY sample. Plotted in the upper left panel of Figure (2003) IMF and SED fitting to determine masses and SFRs. On the other hand, the low-mass LVL (LMLVL) sample adopted the Salpeter (1955) IMF and used Hα+[N II] fluxes to determine the SFRs (Lee et al. 2009). We, therefore, converted the LMLVL SFRs to a Chabrier (2003) IMF using the scaling from Kennicutt & Evans (2012). Additionally, we also consider the Hα SFRs determined for the CLASSY sample (see Section 4.7) for proper comparison to the LM-LVL SFRs. Interestingly, the CLASSY Hα SFRs determined in Section C are 0.22 dex larger on average than the aperture SED-derived values, but with a large standard deviation of 1.16 dex, and so would only serve to slightly increase the average offset of the CLASSY sample from the z ∼ 0 MS in Figure 8.
The CLASSY SFR-M relationship in the upper left panel of Figure 8 lies above the z ∼ 0 and 0.5 < z < 2.5 MSs trends. The CLASSY trend has a similar slope to the z ∼ 0 MS, but lies roughly 2 dex higher in SFR, while the intermediate-z MS relationships begin to approach the CLASSY relationship with increasing redshift. This shows that CLASSY galaxies, selected to be compact and UV bright, have significantly enhanced sSFRs relative to their star-forming MS counterparts at z ∼ 0; rather the CLASSY sSFRs are more comparable to a z ∼ 2 galaxy population.
The middle row of Figure 8 shows the total specific SFRs (left) and the aperture sSFRs (right) versus the direct T emethod gas-phase oxygen abundances of the CLASSY sample. Here, the majority of the CLASSY sample aligns with or extends from the Curti et al. (2020) relationships for 10 8 < M /M < 10 11 (orange to yellow lines, respectively) star-forming SDSS galaxies, but lies 1 − 3 dex above the LM-LVL trend. Given that both Curti et al. (2020) 11 and Berg et al. (2012) used direct T e methods for their oxygen abundance determinations, this offset in sSFR is likely due to the boosted SFR of the CLASSY sample (see the SFR-M trend).

Ionization Parameter and Density Properties
The bottom left hand panel of Figure 8 shows the CLASSY galaxies also probe a broad range of ionization parameter, as indicated by the oxygen ionization ratio, O 32 =[O III]λ5007/[O II]λ3727, and gas-phase electron density. Most H II regions have measured densities near the lowdensity limit (∼ 100 cm −3 ); this can be seen by the LM-LVL sample (blue circles), where all but one galaxy have n e < 100 cm −3 . On the other hand, seven CLASSY galaxies have n e > 300 cm −3 (purple plus symbols; the high-density CLASSY sample), where additional effects such as dielectric recombination can become important. Given the evidence that higher nebular densities may be more common at higher redshifts (e.g., Sanders et al. 2016; median n e ∼ 250 cm −3 at z ∼ 2.3), this high-density sample was included in CLASSY to investigate the role of density.

The M -Z Relationship
In the bottom right hand panel of Figure 8, gas-phase oxygen abundance is plotted relative to the total stellar mass, showing the well-known mass-metallicity relationship (MZR; e.g., Tremonti et al. 2004). Interestingly, the CLASSY sample closely follows the Berg et al. (2012) direct-metallicity MZR measured for the LMLVL, but with larger scatter. To quantify the comparison, we employ Kelly (2007)  where M has units of M , and the total scatter is σ tot. = 0.28 dex and consistent intrinsic scatter. This suggests that the CLASSY sample, while biased to target UV-bright galaxies, is consistent with typical star-forming galaxies evolving along the MZR. The trends plotted in Figure 8 demonstrate the expansive galaxy properties and nebular conditions of the CLASSY sample that allow it to uniquely serve as templates for interpreting galaxies with significant star formation episodes across all redshifts.

POTENTIAL SCIENCE WITH CLASSY
Overview: The main objective of CLASSY is to use FUV spectra to unify stellar and gas-phase physics, allowing a 24 BERG ET AL.
holistic understanding of massive stars as the drivers of the gaseous evolution of star-forming galaxies. Naturally, the pursuit of this goal will produce a number of additional HLSPs, beyond the coadded spectra, that will be useful to the astronomical community. The enduring value and utility of CLASSY will reside in these state-of-the-art HLSPs products. Below we summarize the main HLSPs the CLASSY team plans to produce and the scientific objectives they will enable studies of.
Summary of Planned CLASSY HLSPs: 1. CLASSY spectral atlas 2. Compiled ancillary data 3. CLASSY stellar continuum fits 4. Database of emission and absorption feature properties 5. Database of Lyα emission profile fits Additionally, user-friendly CLASSY tutorials on interacting with and utilizing the data products will be released in tandem with corresponding HLSPs. These HLSPs will be used to diagnose a vast array of science objectives, include the (1) massive star astrophysics, (2) physical properties of outflows, (3) Lyα physics, (4) chemical evolution of galaxies, and (5) physics of reionization. Below we provide an early glimpse to the individual science cases that CLASSY will explore and motivate some of the upcoming science that can be done with the data.

Understanding The Massive Star Properties
Massive stars influence all facets of their host star-forming galaxies including the shape of their UV spectra: massive stars produce copious amounts of high energy photons that are reprocessed in the interstellar medium and power the nebular emission lines that trace the gaseous physical conditions (e.g., Strömgren 1939;Seyfert 1943;Tinsley 1980); massive stars may have been important contributors to the ionizing photon budget that reionized the early universe (e.g., Ouchi et al. 2009;Robertson et al. 2013;Finkelstein et al. 2019); massive star winds disrupt the gas cycle in galaxies and drive out multiphased material and more. Despite the obvious importance of massive stars in understanding the evolution of galaxies and the early universe, their ionizing spectra are not well understood. Thus, uncertainties in the shape of the ionizing spectrum have a significant effect on the interpretation of UV spectra, including gas properties, stellar feedback, the production of H-ionizing photons, and the effects of dust. While the implementation of new physics in stellar population synthesis (SPS) models (i.e., rotation, binaries) continues to improve predictions of the extreme UV (EUV) radiation field (e.g., Levesque et al. 2012;Eldridge et al. 2017;Götberg et al. 2018), the shape of the ionizing spectrum remains poorly constrained for the metal-poor (Z/Z < 0.2) stellar populations that come to dominate at high-z. This shortcoming thwarts our understanding of current moderate redshift (z ∼ 2 − 5) studies and future spectra of high redshift galaxies expected from ELTs and the JWST.
The CLASSY survey provides an important opportunity for progress: with high-resolution, high signal-to-noise spectra covering the entire FUV (although, not the ionizing continuum directly), CLASSY probes a large suite of emission and absorption features shaped by massive stars. These spectra include the N V λλ1238,1242, Si IV λλ1393,1402, and C IV λλ1548,1550 stellar wind features and the weak Si III λλ1290,1417, C III λλ1247,1426,1428, and S V λ1502 photospheric features, both of which are distinctly sensitive to either age or metallicity of the stellar population (e.g. Such stellar fits can determine the light-weighted metallicity, age, and reddening of the the massive star population. Recent z ∼ 0 − 3 studies have presented the first demonstrations that the combination of UV stellar and nebular spectral features can be used to constrain the shape of the otherwise unseen ionizing spectra in the EUV (e.g., Steidel et al. 2016;2018;Olivier et al. 2021). Building on this powerful tool, the CLASSY sample and its superior spectra will open an unprecedented window on the stellar populations that form in a range of environments, including the low metallicity systems that dominate at high redshift. The CLASSY spectra will enable measurements of the stellar photospheric and wind features that can be used as important direct constraints on the metallicity and age of the massive stars and the EUV radiation field they power. In turn, the CLASSY modeled intrinsic ionizing continua can provide the community with the key to understanding how massive stars influence feedback, the ionization and enrichment of nebular gas, and the conditions leading to the escape of ionizing radiation.

The Physical Properties of Outflows
The kinematics of the galaxy-scale outflows of gas driven by the massive stellar populations are encoded into the Lyα profiles and ISM resonant absorption lines observed in the FUV. These features currently provide the majority of the data on outflows at both low and high redshift. The crucial question is how these outflows affect the evolution of galaxies. Answers require determination of the basic properties of the outflows (velocities, outflow rates, ionization state, etc.)    Fig. 9. Demonstrations of science that can be derived from model fits to the CLASSY spectra. Top row: The high-resolution coadded spectrum of J0036-3333 is well-fit by a light-weighted combination of Starburst99 stellar population synthesis models (turquoise line). The regions of the continuum included in the fit are shown in purple, while masked regions are plotted in grey. The high S/N and spectral resolution of the CLASSY spectra allow both stellar wind (N V λλ1238,1242, Si IV λλ1393,1403, C IV λλ1548,1550) and stellar photospheric (C III λ1247, Si III λλ1295,1297,1299) features to be fit. Middle left: The Si II λ1260 interstellar absorption feature can be characterized in the high-resolution coadded spectrum of J1416+1223 using a double Gaussian fit. This total fit (solid green line) includes a static component (dashed blue line) that is matched to the line spread function (LSF) and an outflowing component (dot-dashed gold line) that characterizes the kinematics of the low-ionization gas. Middle right: The Lyα profile in the very-high-resolution coadded spectrum of J1025+3622 is well-fit by a TLAC (Gronke & Dijkstra 2014) radiative transfer model. This model can be used to constrain the depth and kinematics of the neutral H gas. Bottom row: The high-resolution spectrum of J1044+0353 contains many high-ionization UV emission lines that are well-fit by narrow, nebular Gaussian profiles. These lines, including the Si IV λλ1393,1403, O IV λλ1401,1405,1407, S IV λλ1405,1406,1417, C IV λλ1548,1550, He II λ1640, and O III] λλ1661,1666 lines shown, can be used as nebular diagnostics. and how these depend on the properties of the starburst and its host galaxy (star-formation rate, stellar mass, size, etc.).
The FUV spectral region covered by CLASSY contains a suite of resonance lines from atoms and ions spanning a large range in ionization potential. While several studies addressing the properties derived from the FUV exist (Heckman et al. 2015;Chisholm et al. 2015;, CLASSY represents the opportunity for a major advance. Because CLASSY spans about four orders-of-magnitude in both stellar mass and SFR (see Figure 7), it is the benchmark sample for investigations of how outflow and galaxy properties scale together in the low redshift universe. Very few measurements of outflow velocities and mass outflow rates exist for galaxies with masses below 10 9 M (e.g., Martin 1999;Martin et al. 2005;Heckman et al. 2015;Heckman & Borthakur 2016;Bordoloi et al. 2016), where gravitational potential wells are the smallest and thus the galaxies are most impacted by supernovae feedback (Dekel & Silk 1986).
Some absorption lines also have associated non-resonant fluorescence lines produced by radiative decay into a fine structure level just above the ground state. These lines are seen at both low and high redshifts (e.g., Shapley et al. 2003;Martin et al. 2012;Erb et al. 2012;Jones et al. 2012;Scarlata & Panagia 2015;Henry et al. 2015;Bordoloi et al. 2016;Finley et al. 2017;Wang et al. 2020), where the relative strengths of the fluorescent and absorption lines encode information about the size/structure, geometry, and dust content of the outflow (Prochaska et al. 2011;Scarlata & Panagia 2015;Carr et al. 2018;Wang et al. 2020;Carr et al. 2021). The CLASSY spectra will enable robust measurements of fluorescent emission that can provide a powerful diagnostic of the structure and physical properties of outflows. The ability to model complex outflow signatures in the CLASSY galaxies is demonstrated in the middle-left panel of Figure 9. The combination of CLASSY's high spectral resolution and S/N enables outflowing gas to be distinguished from static gas in the ISM and absorption profiles to be corrected for the effects of in-filling by resonance scattering (Prochaska et al. 2011;Scarlata & Panagia 2015). For the Si II λ1260.42 absorption feature in J1416+1223, a double Gaussian can be fit to the continuum-normalized spectrum to account for the static ISM component (dashed blue line) and characterize the outflow (dot-dashed gold line) kinematics of the low-ionization gas. Absorption line fits will be presented in the forthcoming CLASSY paper by Xu et al. (in preparation). Such detailed absorption line analysis can then be used to analyze covering fractions, ion column densities, and outflow rates. These measurements will constrain stellar feedback on a wide-assortment of galaxy types, opening a new window onto the impact of stellar feedback on galaxy evolution with a specific emphasis on very low stellar mass galaxies.

Lyα Physics
Outflows are predicted to promote the escape of Lyman-α (Lyα) photons before they are absorbed by interstellar dust. The shapes of the emergent Lyα line profiles therefore contain information about the physical properties of outflows. A comparison of outflow properties measured from Lyα emission lines and resonance absorption lines is a key CLASSY objective.
The CLASSY spectra cover Lyα for all targets, revealing a broad diversity of Lyα profiles. Geocoronal emission from Lyα and O I compromise a few Lyα profiles. Additionally, a significant fraction of the CLASSY spectra show broad Lyα absorption troughs underneath the Lyα emission that make continuum placement challenging, particularly in the lowest redshift targets where it blends with damped H I absorption from the Milky Way. For the majority of the CLASSY sample, however, radiative transfer models can be used to fit their complex Lyα profiles. This technique is demonstrated for J1025+3622 in the middle-right panel of Figure 9, where a Voigt profile was used to remove the underlying absorption and the Lyα emission was modeled using the Monte-Carlo radiative transfer code, TLAC (Gronke & Dijkstra 2014). The details of such Lyα models offer the opportunity to constrain how the ISM and CGM modify the intrinsic Lyα emission through resonance scattering (Gronke et al. 2015).
Photons emitted by a central source in a simple shell model are scattered by an expanding shell of neutral hydrogen with variable dust opacity (Gronke et al. 2015). The fitted shell velocity and H I column density have frequently been used to constrain galaxy properties (e.g., Kulas et al. 2012;Martin et al. 2015;Gronke 2017). Yet direct comparisons to outflow properties derived from other spectral diagnostics remain limited to small samples (Orlitová et al. 2018;Carr et al. 2021). The combined Lyα, resonance absorption, and fluorescent emission profiles of the CLASSY sample will therefore allow discrepancies in outflow parameters to be addressed via forward modeling of Lyα profiles with, e.g., the Semi-Analytical Line Transfer code (SALT; Scarlata & Panagia 2015)

and the RAdiation SCattering in Astrophysical Simulations code (RASCAS; Michel-Dansac et al. 2020).
A key goal of CLASSY is to determine the strongest scaling relations between Lyα emission and galaxy properties. Many CLASSY spectra show double-peaked Lyα emissionline profiles, an indication of low H I column density channels (Verhamme et al. 2017). These profiles can be used to measure peak-to-peak velocity separations and, more generally, Lyα velocity offsets relative to galaxy redshifts. Given that the peak-to-peak separation appears to be a promising indirect indicator of Lyman continuum leakage (Verhamme et al. 2015;Izotov et al. 2018;Jaskot et al. 2019), the CLASSY peak separations along with measurements of partial covering (via UV resonance lines), the age and metallicity of the massive star population, and the ionization structure inferred from UV and optical emission lines can be used to investigate this escape fraction indicator. Examining the same galaxies from all these perspectives will provide unique insight into escape fractions, clarifying their relationship to radiative and mechanical feedback. A limitation imposed by the targets' proximity and the solid angle subtended by the COS PSA is that the measured equivalent widths and Lyα escape fractions will be aperture limited.
Looking back to the epoch of reionization (EoR), the increasing fraction of neutral hydrogen blocks Lyα transmission. The declining volume density of Lyα emitters at high-z constrains the reionization history, and a better understanding of Lyα velocity offsets, specifically their dependence on galaxy properties, will reduce potential systematic errors on the inferred neutral fractions (Mason et al. 2018;Naidu et al. 2020). The CLASSY sample includes extreme emission-line galaxies, the closest local analogs of EoR galaxies, making them excellent environments to study the star-gas interplay that shaped cosmic reionization.

UV Diagnostics of Chemical Evolution
The CLASSY spectra contain strong UV emission-lines that characterize a plethora of gas properties, including temperature, density, and metal content (both nebular and ISM, e.g., James et al. 2014;Byler et al. 2018;2020), as well as reflecting the shape and hardness of the ionizing spectrum. The bottom panel of Figure 9 demonstrates Gaussian fits to several high-ionization nebular emission lines in the continuum-subtracted spectrum of J1044+0353. The full suite of emission-line measurements for the CLASSY sample will be released as HLSPs as part of the forthcoming CLASSY papers by Mingozzi et al. (2022) and Arellano-Córdova et al. (2022).
Such detailed emission line analyses can be used to provide diagnostics of the chemical and physical properties of the CLASSY systems. These include several temperature and density-sensitive line ratios. When combined with optical line diagnostics afforded by the ground-based data, the CLASSY spectra will provide guidance on interpreting purely UV-based diagnostics of high-redshift galaxies. For example, the agreement of electron temperatures derived from the UV auroral O III] λλ1660,1666 emission lines with those derived from [O III] in the optical can be tested.
Recently, significant efforts have been invested into photoionization modeling of nebular UV emission lines to constrain the radiation fields, ionization sources, and metal content within star forming galaxies (e.g., Byler et al. 2018;Nakajima et al. 2018;Berg et al. 2021a;Olivier et al. 2021). The CLASSY UV emission-line ratios can be calibrated against well-understood diagnostics from the existing optical spectra to provide the first stringent tests of proposed UV diagnostics, over a large range of physical environments (i.e., SFR, stellar mass, metallicity, gas density). The high-S/N detection of important combinations of UV emission lines, such as C IV λλ1548,1550/He II λ1640 versus O III] λλ1661,1666/He II λ1640, are shown in Figure 9. With these lines CLASSY can assess the utility of proposed UV diagnostics such as the UV versions of the canonical "BPT" diagram (e.g., Feltre et al. 2016;Nanayakkara et al. 2019) to distinguish ionizing sources, C III] λλ1907,1909 equivalent widths to infer stellar age and nebular metallicity (e.g., Rigby et al. 2015;Jaskot & Ravindranath 2016;Senchyna et al. 2017;Ravindranath et al. 2020), and resonant emission lines, such as Lyαλ1215 and C IV λλ1548,1550 as indicators of ionizing photon escape (e.g., Verhamme et al. 2015;Berg et al. 2019a).
Uniquely, the CLASSY sample also includes a sub-set of high-density galaxies (n e > 300 cm −3 ), with which a detailed study of the effects of density on FUV properties can be investigated. All together, CLASSY seeks to establish the essential UV diagnostic toolset needed to understand fundamental questions concerning the evolution, physical conditions, and ionization structure of star-forming galaxies across cosmic time.

Reionization Physics
At redshifts between z = 6 − 10, ionizing photons escaped from galaxies to reionize the universe (Fan et al. 2006). Determining the sources of cosmic reionization is one of four key science goals of JWST. However, neither JWST nor ELTs will directly observe the Lyman Continuum (LyC) during the EoR owing to the increasing opacity of the intergalactic medium with redshift. The CLASSY survey will provide templates to understand the gaseous conditions in galaxies similar to high-redshift galaxies. The high S/N and spectral resolution CLASSY observations will reveal complex geometric constraints and their relations to observables like the Lyα emission or the metal covering fraction. Further, the stellar continua observations of the massive star populations can be combined with stellar population synthesis models (highlighted in Figure 9 to predict the intrinsic number of ionizing photons produced by massive stars (Q) and the production efficiency of ionizing photons (ξ ion = Q/L UV ). Extrapolating from the CLASSY stellar continuum fits, we will predict ξ ion and investigate the correlation with UV emission lines (e.g., Jaskot & Ravindranath 2016;Schaerer et al. 2018;Ravindranath et al. 2020). Finally, the comparisons between the FUV reddening and the reddening from the optical emission lines will inform on complex geometries and the origin of dust extinction. These results can then be compared to predictions from simulations to test how galaxies contribute to reionization (e.g., Fletcher et al. 2019;Finkelstein et al. 2019).
Theoretical arguments and small observational samples suggest that UV nebular emission and absorption features trace the escape fraction. CLASSY will indirectly infer escape fractions of a statistically significant sample using UV diagnostics accessible by ELTs and JWST: Lyα emission (Verhamme et al. 2015;Steidel et al. 2018;Izotov et al. 2018), the depth of low-ionization absorption lines Chisholm et al. 2018), and the strength of high-ionization emission lines . The middle left panel of Figure 9 demonstrates that the CLASSY absorption lines estimates the porosity (or covering fraction) of the foreground neutral gas. In this example, the high signal-to-noise data suggests that the Si II gas does not fully cover the background stellar continuum because the flux never reaches a value of zero. This implies that there are small regions of the stellar emission that propagate out of J1416+1223 without being absorbed by Si ions. Similarly, 28 BERG ET AL. the TLAC Lyα models in the middle right of Figure 9 emphasizes how the neutral gas geometries can be estimated from the CLASSY profiles. These constraints will illustrate the neutral gas content in galaxies and provide a template to determine the escape of Lyα and ionizing photons at high redshift.

SUMMARY
We have presented a new spectral atlas consisting of highresolution, high-S/N contiguous FUV spectra of 45 local starforming galaxies obtained with HST/COS as part of the COS Legacy Archival Spectroscopic SurveY (CLASSY). As a result, the CLASSY atlas reaches far beyond previous FUV spectral libraries. One of the considerable strengths of the CLASSY project is the careful selection of the sample so that the value of the existing HST archival spectra was maximized. As such, the CLASSY atlas was constructed from 94 archival spectra from 177 orbits and 76 new spectra from 135 orbits (170 spectra total from 312 orbits) of HST observations, or more than 600 total Cosmic Origins Spectrograph (COS) spectral images. Further, we uniformly reduced and coadded the combined archival and new observations of the CLASSY atlas, producing full far-UV spectra that allow a large number and broad range of scientific analyses to follow. The resulting CLASSY spectral atlas is the first of the CLASSY Treasury high level science products to be made publicly available to the astronomical community via the CLASSY HLSP website: https://archive.stsci.edu/hlsp/ classy.
We have also compiled an ancillary set of high-quality optical spectra and photometry and used them to derive updated galaxy properties for the CLASSY galaxies. As a result, we report properties for the CLASSY sample that span broad ranges in stellar mass (6.2 < log M /M < 10.1), star formation rate (−2.0 < log SFR (M yr −1 )< +1.6), direct gas-phase metallicity (7.0 < 12+log(O/H) < 8.8), ionization (0.5 < O 32 < 38.0), reddening (0.01 < E(B-V) < 0.12), and nebular density (10 < n e (cm −3 ) < 1120). With these data, we showed that the CLASSY sample is consistent with the typical evolution of star-forming galaxies along the mass-metallicity relationship in the Local Volume, but seem to be currently experiencing a strong burst of star formation that mimics the SFRs of moderate-to high-redshift galaxies. Thus, the CLASSY spectral templates provide a unique dataset to study both local star-forming galaxies due to the specific spatial scales that they probe and distant galaxies due to their enhanced SFRs.
Looking to the future, the CLASSY Treasury will provide a suite of high level science products to help the community prepare for future observatories and empirically connect the stellar and gas-phase properties of star-forming galaxies for the first time using a suite of UV absorption and emission line diagnostics, added to optical emission line measurements.
The CLASSY team thanks the referee for thoughtful feedback that significantly improved both the paper and the HLSPs. DAB is grateful for the support for this program, HST-GO-15840, that was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Associations of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. BLJ thanks support from the European Space Agency (ESA). The CLASSY collaboration extends special gratitude to the Lorentz Center for useful discussions during the "Characterizing Galaxies with Spectroscopy with a view for JWST" 2017 workshop that led to the formation of the CLASSY collaboration and survey. Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III web site is http://www.sdss3.org/.     Tables A and 3 and the optical  spectroscopic observations in Table 4.  NOTE-Details of the complete set of archival and new HST/COS observations used for the CLASSY Treasury are listed here. The program IDs (PIDs) for both archival and new observations are listed in Column 3. For each PID, the details of the relevant NUV acquisition images are provided, with the dataset ID in Column 4, the exposure time in Column 5, the mirror used in Column 6, and the position angle (PA) of the aperture in Column 7, and the GFWHM measured from the spatial NUV light profile in Column 8. Subsequently, the details of the corresponding FUV spectral datasets are listed; the dataset ID is given in Column 8, the grating and wavelength settings used are listed in Columns 9 and 10, the detector lifetime position (LP) is given in Column 12, the total exposure times are in Column 13, and the rest-frame wavelength coverage is noted in Column 14. The full extent of the contiguous wavelength coverage is bolded in the last row for each galaxy.

B. CLASSY SAMPLE PROPERTIES
Here we present properties of the CLASSY sample galaxies that were derived from the ancillary optical spectra in Table 5 and Beagle SED fitting in Table 6. See Section 4 for details of the calculations. Additionally, we characterize the aperture bias of the CLASSY sample in Figure 10 by plotting the fraction of the optical flux measured (from the u and g−band SDSS, DES, and PanSTARRS imaging) through a 2. 5 aperture relative to the total flux of the galaxy. The points in Figure 11 are color-coded by their optical half light radii, measured from the same imaging, showing that most of the CLASSY sample is relatively compact with 66% of the galaxies having r 50,opt. 2.0 . As demonstrated in Figure 3, the extent of the UV light is much more compact than the optical light of the CLASSY sample.

C. SED FITTING
Here we further describe the SED fitting method used to determine the stellar masses and star formation rates presented in Section 4.7. SED fitting was performed using the BayEsian Analysis of GaLaxy sEds (BEAGLE, v0.24.0, Chevallard & Charlot 2016). BEAGLE is a new-generation SED tool that incorporates several novel approaches over past codes. First, the modular design of BEAGLE allows a physically consistent combination of different prescriptions for the production of starlight in galaxies and its transfer through the ISM (absorption and emission by gas, attenuation by dust) and the IGM (absorption by gas). Second, BEAGLE uses a flexible parameterization of several prescriptions to describe the star formation and chemical enrichment histories of galaxies, ranging from simple analytic functions to the predictions of sophisticated galaxy formation models, to interpret combinations of photometric and spectroscopic observation of galaxies. As a result, BEAGLE fits can provide a physically consistent description of the contributions by stars, gas, and dust to the integrated emission from a galaxy. Note that this updated SED fitting approach is particularly important for the subset of CLASSY galaxies with very strong nebular continuum and line emission as BEAGLE self-consistently accounts for these contributions in the synthetic model spectrum. Using BEAGLE, two sets of stellar mass and SFR parameters were computed: (1) an aperture set which provides insights for the star-forming regions specifically targeted within the COS 2. 5 aperture and (2) a host galaxy set which better characterizes the properties of the entire galaxies.

C.1. Choice of Photometry
The SDSS provides optical imaging in ugriz for 42 galaxies in the CLASSY sample. For the other three, J0036-3333 and J0405-3648 have grizY images from the Dark Energy Survey (DES, DES Collaboration 2021), and J0337-0502 is observed by the Pan-STARRS1 (PS1, Chambers et al. 2016). For the UV, 43 galaxies (all but J0337-0502 and J1157+3220) have imaging in the NUV and FUV available from (GALEX, Martin et al. 2005). Additionally, WISE W1 and W2 nearinfrared (NIR) data are also available for the entire CLASSY sample. However, WISE photometry introduces large model uncertainties due to the contribution of post-main-sequence stars to NIR light (e.g., Salim et al. 2016) and does not significantly affect the CLASSY stellar mass (∆ log M < 0.31 dex) and SFR (∆ logSFR < 0.21 dex) measurements, and so was not included in the CLASSY SED fitting.

C.2. New Flux Measurements
Due to the irregular morphologies and range of redshifts of the CLASSY sample, the SDSS catalog photometry can have large uncertainties when a simple, symmetric model is assumed or only a portion (e.g., a single H II region) of a host galaxy is measured. We, therefore, remeasured the GALEX and SDSS fluxes. For the aperture stellar mass and SFR determinations, ugriz or grizY photometry was measured for NOTE-Comparison of the nominal (i.e., point source) spectral resolution of the CLASSY coadded spectra to the measured resolution from Milky Way ISM absorption lines. The best Milky Way ISM feature for each galaxy is listed in Column 2, followed by the nominal resolution at that wavelength in km s −1 per resolution element for each coadded dataset in Columns 3 − 6. Finally, Column 7 lists the Gaussian FWHM in km s −1 measured from the HR, Galactic extinction-corrected, binned spectra (extension 9).  NOTE-The CLASSY sample is composed of UV bright, nearby galaxies covering a range of metallicity, mass, SFR, and gas density. Columns 1−4 give the target name used in this work, alternative names used, coordinate R.A. and Decl., and redshift. Column 5 gives the continuum flux at 1500 Å in units of 10 −15 erg s −1 cm −2 Å −1 , from the CLASSY coadded spectra. Columns 6 − 9 list list additional nebular properties derived from the optical spectra, namely the direct Te-method oxygen abundance (12+log(O/H)) and the ion temperature used, the [S II] electron density, [O III] λ5007/[O II] λ3727 flux ratio, and the E(B−V) reddening value determined from the Balmer decrement. NOTE-CLASSY sample properties derived from UV+optical photometry, as described in Section 4.6. Columns 1−3 give the target name used in this work, coordinate R.A. and Decl., and redshift. Column 4 lists the luminosity distance, corrected for the Hubble flow using the velocity field model of Masters (2005). Column 5 gives the half light radii in arcseconds, as derived from the optical imaging (see Figure 3). Columns 6 and 7 list the stellar masses derived using Beagle SED fitting for both the light within the COS aperture and the entire galaxy, respectively, of a given target. Similarly, columns 8 and 9 provide the SED-derived SFRs.
targeted star-forming regions in a 3 aperture, well-matched to the size of COS aperture convolved with typical SDSS seeing (see, e.g., Senchyna et al. 2019). The sky background was estimated in a 1 -wide circular annulus located at 1 using the median flux values after sigma-clipping (low 10σ, high 3σ). Corrections for Galactic foreground extinction were then applied, assuming an R V = 3.1 and Milky Way extinction curve Cardelli et al. (1989). The GALEX UV imaging was not used for these aperture calculations due to its relatively poor resolution (4. 5 -6. 0 FWHM). For the total stellar mass and SFR determinations of each galaxy, total fluxes in the UV and optical were measured, but with a more sophisticated method to account for irregular morphologies and potential contamination from nearby objects. First, segmentation maps were produced for each galaxy using the Photutils package (Bradley et al. 2021). For the optical imaging, g-band segmentation was computed, which includes strong optical emission lines such as [O III] and (or) Hβ for the majority of the CLASSY sample. This maximizes the ability to probe the often faint and extended nebular emission, which are incorporated into the BEAGLE models. To produce the g-band segmentation maps, a minimum of 8 connected pixels at 2σ above the background after the image is smoothed by a 2D Gaussian kernel (FWHM = 10 pix) is required. Similarly, NUV segmentation was computed for the UV photometry, but using a Gaussian kernel with FWHM = 3 pix. Each map was visually inspected to ensure an adequate segmentation that captures the total galaxy light in both the optical and UV. If a map failed the visual inspection, the segmentation parameters were adjusted and source deblending was implemented in Photutils to exclude contamination. Second, the g-band and NUV segmentations were used to measure optical photometry in ugriz (or grizY ) and NUV and FUV photometry, respectively. In a similar manner as the aperture determinations, background subtraction and Galactic extinction corrections were applied. In general, we found that the 3 aperture flux accounts for ∼ 25% of the total g-band flux at z ∼ 0.01, and ∼ 80% at z ∼ 0.17.
Finally, galaxy distances and absolute magnitudes were determined. Since nearly half of the sample (20/45) are very nearby galaxies with optical spectroscopic redshifts of z < 0.02, they may have peculiar motions that will bias the luminosity distances derived from redshifts. To uniformly estimate their distances, the heliocentric redshifts were first converted to the cosmic microwave background (CMB) frame using the CMB dipole measured by Lineweaver et al. (1996). The local velocity field modelled by Masters (2005) was then adopted to compute the flow-corrected luminosity distances, with the amount of correction ranging from < 1% to 20% (median 6%).

C.3. SED Model Assumptions
With the redshifts, corrected luminosity distances, and photometry for both the targeted regions and entire galaxies in hand, the M and SFRs can be self-consistently constrained with the BEAGLE code. Briefly, the Gutkin et al. (2016) model was adopted, which self-consistently incorporates the stellar emission from the latest version of the Bruzual & Charlot (2003) stellar population synthesis results and the nebular emission output by the photoionization code Cloudy (Ferland et al. 2013). We assumed a Chabrier (2003) Galactic disk initial mass function (mass range 0.1 -100 M ), the two-component attenuation model from Charlot & Fall (2000), and a constant star formation history. Free parameters are thus the stellar mass, age, metallicity, gas ionization parameter, log U, and effective attenuationτ V . The interstellar (dust+gas-phase) metallicity was matched to the stellar metallicity of the constant star formation. Of the three choices of dust-to-metal mass ratio provided by BEAGLE (ξ d = 0.1, 0.3, 0.5), we assume ξ d = 0.3 as the value closest to the solar value (ξ d = 0.36 Gutkin et al. 2016).
The BEAGLE SED fits for two CLASSY galaxies, local dwarf galaxy J0944-0383 and z ∼ 0.1 emission-line galaxy J1416+1223, are shown in Figure 11. For comparison, the fit to the host galaxy UV+optical photometry is shown in the top panels relative to the aperture optical photometry fit in the bottom panels. Contamination to the broad-band flux from strong emission lines is clearly seen in the gr bands for local systems (e.g., J0944-0038) or r band for those at slightly high redshift (e.g., J1416+1223). These models demonstrate the importance of properly accounting for strong nebular emission in our SED fitting with a self-consistent program like BEAGLE.
From the output posterior distribution, the M and SFRs were computed for both the targeted aperture regions and the entire galaxies. The resulting stellar masses and SFRs are reported in Table 6 in the Appendix for both the aperture and total host galaxy sets.

C.4. Choice of Star Formation History
It is very difficult to recover accurate values of the SFRs through SED fitting for realistic galaxies, in large part due to their unconstrained SFHs. Fortunately, previous SED fits using a constant SFH were shown to perform well for a subset of the CLASSY sample (Senchyna et al. 2017;, and allowed for a consistent comparison with the SED fitting conducted for high-redshift galaxies. Following this method, we have assumed a constant SFH, which benefits from having the least free parameters, yet provides satisfying fits to the broad-band photometry even when strong emission-lines are present. However, an important caveat is the potentially hidden old stellar populations in the CLASSY galaxies, especially outside the burst of star formation targeted by the COS Fig. 11. Example SED fits for two of the CLASSY galaxies, one local dwarf galaxy (J0944-0038, left) and one emission-line galaxy at z ≈ 0.1 (J1416+1223, right). The fits to host galaxies are shown in the top row while the aperture fits are in the bottom. Blue triangles are the measured fluxes from broad-band photometry, and red circles give BEAGLE predictions. The median BEAGLE SED and 16-84 percentile of posterior distribution are shown in black line and gray-shaded area, respectively. The significant excess to the broad-band flux in the gr bands of J0944-0038 and r band of J1416+1223 is due to strong nebular emission lines, which are accounted for by BEAGLE.
aperture. Our assumption of a simple constant SFH thus inevitably introduces additional systematic uncertainties to our estimates, often leading to a lower stellar mass compared to other more complicated SFHs.

C.5. Exploration of SED Systematics
To better understand the accuracy and precision of our SED-derived galaxy properties and to provide consistency checks, we performed the following tests: 1. Impact of assumed SFH: We investigated the impact of assuming just a simple constant SFH by comparing to more complicated models. First, we allowed composite SFHs consisting of a constant SFH plus a single burst in the fit in order to allow the maximum flexibility for BEAGLE to properly account for any possible old stellar populations. In most cases, BEAGLE favored either a dominant constant SFH component or a dominant SSP component. The resulting fitted masses are larger relative to the values assuming a constant SFH by 0.01 +0.05 −0.04 dex. Second, since a red component from an older stellar population would be primarily arise from stars outside the aperture, we also tested fitting the flux inside and outside aperture separately using two populations of constant star formation, one inside and one outside the aperture. Using the four sources in our sample showing the most extreme extended components outside, we found differences in both stellar masses and SFRs to be 1σ and within the uncertainties. However, Lower et al. (2020) recently investigated the impact of SFH on SED fitting and found that assuming a constant SFH can underestimate the stellar mass by 0.48 ± 0.61 dex. Unfortunately, other simple parametric models such as a delayed-τ SFH 12 also performed poorly in this work; only complicated, non-parametric SFHs produce realistic values. Thus, for simplicity and for consistent comparison with highredshift galaxies, a simple constant SFH was adopted for the CLASSY SED fitting, resulting in relatively large uncertainties.
2. Comparisons to other catalogs: We compared the stellar masses and SFRs to the values from the MPA-JHU cat-alog 13 . On average, the CLASSY stellar masses are smaller by −0.28 dex, while the 16th and 84th percentile differences from the MPA-JHU values are −0.60 and 0.40 dex, respectively. For the SFRs, the CLASSY values are larger on average by 0.18 dex, while the 16th and 84th percentile differences from the MPA-JHU values are −0.09 and 0.63 dex, respectively. While we find generally good agreement with the MPA-JHU values, there are some significant outliers due to many differing assumptions in the photometry and SED modeling. However, dust extinction is unlikely to be the cause given the strong agreement between the CLASSY aperture (host) values of effective optical depth (τ V,eff ) and those derived from optical spectra (∆τ V,eff = −0.06 +0.14 −0.10 (0.09 +0.53 −0.22 )). We also compared to the the GALEX-SDSS-WISE Legacy Catalog of Salim et al. (2016;GSWLC). The GSWLC contains SDSS galaxies more distant on average than the CLASSY sample at 0.01 < z < 0.30 and r petro < 18. These distances ensure that the SDSS photometry are safe to use for SED fitting. A cross-match between the CLASSY sample and GSWLC yields 22 matches, of which 12 have GSWLC SED fits with χ reduced > 30. For the remaining 10 galaxies, 4 have GSWLC SED fits with χ reduced > 10. The large fraction of poor fits in GSWLC are likely due to its inability to account for large broad-band excess caused by strong nebular emission lines. For the 6 galaxies with GSWLC χ reduced < 10, we find differences (∆ = CLASSY values − GSWLC values) of ∆M ,host = −0.27 +0.33 −0.39 dex and ∆SFR host = 0.12 +0.13 −0.11 dex. On average, the CLASSY SFRs are larger by ∼ 0.1 dex compared to GSWLC1 SED SFRs, but smaller by ∼ 0.2 dex compared to GSWLC1 WISE SFRs. Extreme outliers and the sources of such discrepancies will be explored in a future paper.
3. Comparison to Hα SFRs: We calculated SFRs in another way, using the reddening-corrected Hα flux measured from the optical spectra, the luminosity distances listed in Table 6, and the SFR calibration from Kennicutt & Evans (2012). For galaxies with optical spectra from the SDSS, we use the Hα flux values from the MPA-JHU catalog, where each spectrum has been tied back to the SDSS r-band fibermag. Note that this Hα SFR calibration assumes a Kroupa & Weidner (2003) IMF, which is consistent with the Chabrier (2003) IMF used for the SED fitting (e.g., Chomiuk & Povich 2011). In comparison to the SED-derived SFRs, we find the Hα SFRs to be larger on average, with a typical percent difference of 8%, and a few strong outliers extending from −202% to +48%.
4. Comparison to Nebular Properties: Finally, we compared the 12+log(O/H) and logU values from the BEAGLE SED fits to those derived from the optical spectra (see Sections 4.5 and 4.6). Of our three comparisons (other catalogs, SFRs, and nebular properties), the latter shows the largest discrepancies, where the SED metallicities (ionization parameters) are systematically lower (higher) than the measured values, with median differences of 0.37 dex (−0.36 dex). However, the BEAGLE values have large uncertainties (> 0.5 dex) by design that give it the flexibility necessary to test large parameter space and thus are still generally consistent with the measured CLASSY values. Note that the choice to not fix metallicity, ionization parameter, or dust extinction to spectroscopic values was deliberate in order to not introduce other systematic biases (i.e., the burst of star formation within the spectroscopic aperture may have different ISM conditions than the outer regions of a galaxy).

D. NOTES ON INDIVIDUAL GALAXIES
For transparency and to improve usability of the CLASSY HLSPs, we mention here any unique notes for individual galaxies.
J0036-3333: Also commonly known as the galaxy Haro 11, which contains multiple stellar populations. The CLASSY coadded spectrum only focuses on knot C (Vader et al. 1993). Knot C is the most UV luminous source and main contributor to the Lyman alpha emission from Haro 11 (Hayes et al. 2007).
J0127-0619: The common name of J0127-0619 is Mrk 996. This galaxy is known to have complex emission line kinematics, a large Wolf Rayet population, and a high electron density within its inner core region (James et al. 2009;Telles et al. 2014).
J0144+0453: J0144+0453 is more commonly known as UM133 and has an extended tail to the North.
J0337-0502: J0337-0502 is part of the well-known, metalpoor galaxy SBS 0335-052 that has two main star-forming clusters, E and W. The CLASSY coadded spectrum only contains light from the E region. It also combines data from COS FUV detector lifetime positions 3 and 4. SBS 0335-052E consists of six young (10 < Myr) star clusters (Adamo et al. 2010), four of which fall within the COS aperture (Wofford et al. 2021).
J0405-3648: J0405-3648 is fairly extended, with many small ionized clumps seen in the optical imaging. The CLASSY coadded spectrum combines data taken at position angles that are offset by 50 • .

J0808+3948:
The CLASSY coadded spectrum combines data taken at position angles that are offset by 84 • . J1323-0132: J1323-0132, also known as UM 570, has extended optical emission. The CLASSY coadded spectrum combines data from COS FUV detector lifetime positions 3 and 4.
J1359+5726: J1359+5726, also commonly known as Mrk 1486 and is part of the LARS sample with ID Ly52, has extended optical emission. The CLASSY coadded spectrum combines data from COS FUV detector lifetime positions 1 and 3 and data taken at position angles that are offset by 82 • .
J1418+2102: The CLASSY coadded spectrum combines data from COS FUV detector lifetime positions 3 and 4 and data taken at position angles that are offset by 26 • . This galaxy is denoted as an extreme emission line galaxy in Berg et al. (2021a).
J1444+4237: J1444+4237 is also known as HS1442+4250. The CLASSY coadded spectrum combines data from COS FUV detector lifetime positions 3 and 4 and data taken at position angles that are offset by 47 • . Note that an additional dataset, LDXT06010, was not used in the CLASSY coadd because there was a data quality note that the Take Data Flag was not on throughout observation and so the COS light path was blocked during the exposure.
J1545+0858: The CLASSY coadded spectrum combines data from COS FUV detector lifetime positions 3 and 4 and data taken at position angles that are offset by 51 • . E. CLASSY COADDED SPECTRA In Figure 12 we display one of the moderate-or lowresolution coadded spectra for each of the 45 galaxies in the CLASSY sample. The rest-frame, Galactic extinction corrected, binned, moderate-resolution spectrum is plotted in grey when available. For galaxies with archival G140L spectra, the low-resolution coadd is shown. In order to visually detect emission and absorption features more easily, the spectrum has also been convolved with a 1D Gaussian 2element kernel and overplotted in black. The most prominent absorption and emission features are labeled in the top panel of each spectrum, with nebular (dashed pink lines) and fine structure (solid pink lines) emission features marked above the spectra, while ISM (solid blue lines), stellar wind (dotdashed green lines), and photospheric (solid green lines) are marked below the spectra. Contamination from Geocoronal and O I emission and Milky Way ISM absorption are whitedout. The bottom panel plots the S/N ratio as a function of wavelength, where each vertical shaded bar shows the average S/N within its wavelength coverage. Since the plotted spectrum is binned by the 6 pixels of a resolution element, these S/N values are per resel.  Fig. 12. Coadded spectra for all 45 galaxies in the CLASSY sample. The moderate resolution coadd is shown for most galaxies, highlighting the full FUV spectral range covered. For galaxies with G140L spectra, the low resolution coadd is shown, highlighting the full FUV spectral range covered in the CLASSY atlas. The binned, Galactic extinction-corrected, rest-frame spectrum is shown in gray and smoothed version that has been convolved with a 1D Gaussian 2-element kernel is overplotted in black to make spectral features more apparent.