Instrument Science Report COS 2025-18(v1) Overview of the New Hubble Spectroscopic Legacy Archive Abstract A BSTRACT The new Hubble Spectroscopic Legacy Archive (HSLA) provides coadded spectra of individual targets that have been observed with the Cosmic Origins Spectrograph (COS) and the Space Telescope Imaging Spectrograph (STIS) over their operating lifetime. HSLA uses data available in the Mikulski Archive for Space Telescopes (MAST). It automatically produces coadds whenever new data become publicly available or when there is newly recalibrated data. HSLA defines individual targets by their associated coordinates, accounting for proper motions, and uses SIMBAD, NED and the Phase II observing proposals to obtain astronomical classifications for each object. Coadded spectra are produced for each observing mode. In the case of COS far-ultraviolet observations there is one coadded spectrum for each lifetime position (LP). Additionally, a spectrum spanning the entire wavelength range covered by the observations is produced by abutting the spectra from a selection of individual modes. For each individual target, HSLA also provides a human-readable metadata file with key information that can be used in searches or for further exploration of the data. The HSLA project also makes the code used for coadding spectra publicly available along with several other tools (using Jupyter notebooks) for custom coaddition required in special cases. In this report we will describe the main components of HSLA and provide a brief description of how the data and metadata can be accessed. Contents 1. Introduction The Hubble Space Telescope’s (HST’s) two primary spectroscopic instruments, the Space Telescope Imaging Spectrograph (STIS, operational since 1997) and the Cosmic Origins Spectrograph (COS, operational since 2009), have obtained over 64,000 spectra of nearly 7000 individual astronomical objects to date. These instruments provide access to ultraviolet (UV) wavelengths, with STIS coverage extending from 1150 Å to 10,300 Å covering the far-ultraviolet (FUV) to the near-Infrared (NIR), and COS coverage from 912 Å to 3300 Å spanning the Lyman-ultraviolet (LUV), FUV and the near-ultraviolet (NUV) regimes. The wealth of UV spectra from COS and STIS are hosted by the Mikulski Archive for Space Telescopes (MAST) and are available to the public, but they are primarily organized by links to individual observing programs in which they were obtained. Therefore finding and selecting data for individual objects or by source type for several objects is not straightforward. The need to organize the spectra in a simpler way was recognized years ago, and resulted in the creation of the original Hubble Spectroscopic Legacy Archive (o-HSLA; Peeples et al. 2017). The o-HSLA represented a significant step forward in the organization of HST UV spectra and provided a template for organizing a legacy archive. However, the data products were static and did not include STIS observations. Only COS NUV and FUV data obtained before July 2018 are in the o-HSLA, and they have not been updated with any newer data products or with improved calibrations. The new Hubble Spectroscopic Legacy Archive (HSLA) was motivated by the same underlying reasons as the earlier version, but it was designed from the start to include spectra from all COS and STIS channels and to automate the process to keep the archive current. One of the drivers leading to the implementation of the new HSLA was the recently completed HST Director’s Discretionary program, Hubble UV Legacy Library of Young Stars as Essential Standards (ULLYSES). ULLYSES provided the impetus for creating coaddition software for STIS and COS spectra and providing scientifically useful data products that could be easily accessed by the astronomical community (Roman-Duval et al. 2025). The data in the ULLYSES program are relatively constrained both in terms of target type (young hot stars) and in the observation modes used, but it was realized that the software tools were sufficiently flexible that they could be adapted more generally. The HSLA was conceived as a multi-stage project. The first stage was to provide coadditions of all spectra of each target obtained as part of a single observing program at the visit and program level, sorting them by instrument and grating. Additionally, at each level abutted spectra combining instruments and modes are provided. These show the full wavelength coverage of the data, and are meant for “quick-look” purposes rather than for use in scientific analysis. The independent value of these visit and program level data products was realised early during implementation, and this first stage was completed as the Hubble Advanced Spectral Products (HASP) project (Debes et al. 2024). This document describes the completion of the second stage and the release of the full HSLA, in which coadded spectra for each individual target across multiple programs are created. These are made available along with target classifications and metadata files and enable archive-wide searches based on individual source names, or broader categories of source types. The following sections describe the key components of the HSLA: target association (§2), target classification (§3), and the data products (§4). Following that, in §5, we describe the testing procedure used to verify and validate the HSLA products. Information on catalog and data access as well as a list of cases requiring special consderation are given in §6, and §7 describes python scripts (implemented as Jupyter notebooks) that have been written to perform custom coadditions of the data. 2. Target Association The HST archive contains over 60,000 COS and STIS exposures distributed over a few thousand observing programs. The main tasks of target association are to identify all the unique sources observed, to determine all the exposures (and thereby calibrated data files) associated with each source, and to label each source with a unique name and provide their best possible coordinates. In principle, the main task of cross-matching is straightforward, as one can use target coordinates and names to make the identifications. In practice, there are a number of complicating factors. The primary sources of information are the names in the submitted Phase II proposals and the actual coordinates observed. It is often the case that the same source targeted in different programs will be identified using different names, and may have slightly different coordinates. There are several reasons for scatter in the coordinates for the same target. In the case of objects where proper motion data are provided, the header keywords in the calibrated data files are estimated at the most likely execution time. Other sources of potential uncertainty include the fact that different instrument/grating combinations are located at slightly different sky positions, and the possibility of incorrect specification of the reference frame by the user. The outputs of the target association step are used for target classification (§3) and for creating the coadded data products (§4), which include a human-readable metadata file for each target. 2.1 Optimal angular radius for cross-matching The first task for target association is finding the optimal angular tolerance for identifying unique targets based on sky coordinates. This is non-trivial because several observations have been made in crowded fields (such as the Magellanic Clouds) where the separation between individual sources can be smaller than the coordinate scatter for individual targets. The cross-matching radius should be large enough such that it encompasses the scatter, but not so large that it is unable to distinguish between multiple nearby sources. Using a very small cross-match radius will improve accuracy at the cost of having multiple products for a unique target, while a large cross-match radius will improve completeness at the cost of coadding unrelated exposures. For a given angular tolerance for matching, the accuracy can be improved by using the names for targets provided by the user in the Phase II proposal or from SIMBAD. To determine the optimal cross-matching radius for target association, we first performed a name-based cross-match of every individual exposure with every COS+STIS observation in the HSLA archive (excluding moving targets, which are so specified in the Phase II proposals, and those labeled as “calibration” observations). For every exposure we included the full list of SIMBAD identifiers, thereby generating a list of exposures that is potentially associated with every observation. For every exposure, we can examine the distance distribution of every potentially associated observation based on target name. This is useful to understand the typical scatter in the spatial distribution of exposures that are highly likely to be associated based on their names, and it allows us to quantify the success (or lack thereof) of name-based target association as a function of on-sky distance. To test this in practice, we need to have a ground truth value of the object coordinates available. We treated the SIMBAD coordinates (which are typically those provided in Gaia DR3) as the actual source position, and performed a cross-match (where the cross-match radius is the parameter to be optimized) with the positions listed in the headers of the individual exposure calibrated data files. We define the accuracy as the number of (SIMBAD) sources for which every associated file (based on name matching only) is located within the cross-match radius under consideration. The completeness as a function of cross-match radius is derived from the number of SIMBAD sources with associated exposures; for example, if we consider a 0.5 ′′ cross-match radius and every associated file is located at a distance > 0.5 ′′ from the SIMBAD position, there will be no associations. Finally, we quantify the contamination as the number of sources where exposures are present within the cross-match radius but are not associated through their name. This happens when multiple sources with different names are located close together (such as individual stars in a dense cluster) and would result in exposures erroneously coadded if associated. Figure 1: Completeness, accuracy and contamination of name-based target association as a function of cross-match radius. The cross-match completeness is defined as the number of apertures that are identified as cross-match divided by the total number of apertures that match one of the target SIMBAD aliases as a function of radius. Coordinate accuracy is defined as the number of apertures that, based on name and coordinates, are correctly cross-matched to a given target as a function of radius. A “mismatch” corresponds to a SIMBAD cross-match based on name, but the difference between SIMBAD coordinates and the HST observation is >2^{\prime\prime} . We tested a range of cross-match radii from 0.3–2.5 ′′ , and considered the trade-off between maximizing the completeness while simultaneously maintaining high accuracy. Figure 1 shows these quantities as a function of cross-match radius. The results are very encouraging, as a completeness and accuracy of > 95% can be achieved at a cross-match radius of 2 ′′ , while the contamination remains of order a few percent (the black curve shows the uncontaminated fraction). We therefore choose 2 ′′ as an optimal cross-match radius for target association. 2.2 Procedure details With the optimal cross-matching angular radius determined, the target association is done using coordinate matching. Additionally, it is assumed (as was done for HASP) that within a given program there is a one-to-one matching between unique targets and Phase II names. Initially, a list was made of all the COS and STIS datasets used in HASP visit and program level products along with the relevant metadata: right ascension (RA), declination (DEC), observation time, proper motion and observation epoch, which are stored in the archive catalog. Then, the proper motions were used to determine the coordinates at a common epoch (January 1, 2016) for all the datasets. The observations were then grouped by visit and target name supplied in the Phase II proposals, with the unique visit names determined by the first six characters in the dataset root names. This resulted in a list of visit/target combinations, each of which is called a “pointing”. The next step was to loop through the list of pointings and determine the angular separation of each pointing from all other pointings (using the astropy SkyCoords routine). All observations within 2 ′′ of the pointing were considered to be part of the same association. If no previous targets were found within 2 ′′ of a given pointing, it was designated as a new target; otherwise it was added to the existing target. The RA and DEC of the target were taken to be the mean of the RAs and DECs of all the associated observations, and when new observations were added to an existing target, the coordinates were updated. As the RA and DEC of each target could change during the process of assembling the associations, a check was made after all the datasets had been associated with a target to ensure that all the coordinates were within 2 ′′ of the target center. When this was not the case, the observations furthest away from each other (extreme pointings) were identified, and each of the remaining pointings were associated with the extreme pointing closest to them. This is an issue only in crowded fields where there are observations of distinct objects within a few arcseconds of each other. A special, possibly unique, case to be noted is the pair of targets, \alpha Cen A and B, which move relative to one another non-linearly in a way that causes coordinate confusion. After associating targets, a list was compiled of all the target names, target categories and target descriptions supplied in the Phase II proposals for each of the constituent datasets, which are stored in the headers (keywords: TARGNAME, TARGCAT, TARGDESCR, respectively) of the input FITS files. This list is not a primary HSLA data product but is used to help with checking and testing the classification and categorization of targets. The targets thus defined will remain static in HSLA operations. New targets will be added to HSLA as they are observed, but when new datasets are added to existing target associations, the coordinates will not be updated. 3. Target Classification There are a wide range of target types in the HSLA, and a robust classification scheme is essential for efficient search and retrieval of data on objects in a specific class of scientific interest, such as “White Dwarf Stars” or “Ultraluminous Infrared Galaxies”. One part of the problem is deciding what the classes should be, or in other words defining a vocabulary for the classification, and the other is to decide which class or classes a given object should be assigned to. In the o-HSLA, the classification was done by manually sorting indivdual targets into tables for several broad categories, which are described in Peeples et al. (2017). These categories and tables, like the rest of the o-HSLA, are static. For the new HSLA the primary requirement was to automate assigning target classifications to existing and new data, and an important consideration was to make the classification scheme maximally useful for archival research. 3.1 Methodology HSLA’s target classification is structured as a three-tiered hierarchy. This is to faciliate both broad and narrow searches. For example, a user can query all O-type stars, or all Early Type stars, or even all stars. This approach offers users a flexible method for interacting with the database, making it suitable for a wide variety of research use cases. We used three sources of information, SIMBAD, NED and the Phase II proposals to develop a database of all possible object types and create the classification vocabulary for HSLA. SIMBAD is a rich resource for a wide variety of astronomical objects from stars to nebulae to external galaxies. NED is an extragalactic database, and therefore more specialized and restricted in scope than SIMBAD. In the case of the Phase II proposals, the principal investigators (PIs) are required to include scientific keywords describing their targets. These have the advantage of both flexibility and the specialized knowledge of the proposing team, but this flexibility also allows for greater inconsistencies, particularly when classifications are based on the scientific aim of the proposal rather than describing the observed target. The available classifications in all three sources were scrutinized carefully, and a broad amalgamation of categories were defined in three hierarchical tiers, with Tier 1 providing the most general description, and Tier 2 and 3 increasing in specificity. This three-tiered structure is the basis for HSLA target classification and enables searches on broad and narrow categories of objects. For every target in the HSLA database, SIMBAD, NED and the Phase II proposals are queried in that order, and the object types available are assigned to the target. If an object type exists in SIMBAD for the target in question, those classifications are used for HSLA. If there is no object type, then the description returned by NED is used if available. If the target is present neither in SIMBAD nor in NED, then the Phase II keywords are used. A special case is if the Phase II classification indicates an exoplanet, which triggers a query to the NASA Exoplanet Archive. The HSLA decision tree in how an object type is assigned is graphically represented as a flow chart in Figure 2 . The three main Tier 1 classes are “Galaxy”, “ISM” and “Star”. There are 12 Tier 2 and 59 Tier 3 classes under “Galaxy”, 14 and 46 under “ISM”, and 21 and 140 under “Star”. In each of the secondary and tertiary classification lists, one of the options is “Pass”, which is assigned if there is insufficient information about the target to classify it at these narrower levels. For some targets neither SIMBAD nor NED types are available, and the Phase II keywords make the classification ambiguous. These are assigned a Tier 1 classification of “Multiple”. A handful of observations do not fit into any of the categories and are classified as “Unknown”. A potential future plan is to manually examine the targets falling in these latter two categories and reassigning them to the most appropriate classes. A hierarchical list of all the classes is available on the HSLA website . Figure 2: Flow chart detailing the decision tree of the final object classification. The databases SIMBAD, NED, and the NASA Exoplanet Archive are queried and assigned as the classification in that order of object type availability, but if the target is not present in any, then the Phase II keywords are assigned as the classification. The red asterisks denote the location in the decision tree when an additional query is performed in the case the object is an exoplanet host star. The NASA Exoplanet Archive is queried for the host star spectral type, which is then assigned as its object type. In addition to the three-tiered classification vocabulary, each Tier 1, 2 and 3 combination was mapped to a Unified Astronomy Thesaurus (UAT) concept and number (Frey & Accomazzi 2018). In most cases the UAT concepts match the Tier 3 classifications – for example HSLA Tier 3 “Seyfert galaxy” maps to UAT concept (ID) “Seyfert galaxies (1447)”. In other cases, they are closely related – “Blue supergiant” in HSLA maps to “Early-type supergiant stars (431)”. The UAT project provides brief definitions of all the object types as well as links to related types. Mapping classes to UAT allows HSLA users to explore other archives and documents based on a wider vocabulary and to access supplementary data or information on specific types of astronomical objects of interest. 3.2 Results Table 1 : Number of HSLA Targets in various Classes \begin{array}[c]{\string lr}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Classification\hfil\hfil TargetCount\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}{\vskip 5.81247pt}\hskip 0.0ptTier1\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}Galaxy2459{}\cr{\vskip 0.0pt}{}ISM179{}\cr{\vskip 0.0pt}{}Star3894{}\cr{\vskip 0.0pt}{}Multiple170{}\cr{\vskip 0.0pt}{}Unknown10{}\cr{\vskip 0.0pt}{}\par{\hline}\par{\vskip 5.81247pt}\hskip 0.0ptTier2\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}Activegalaxy2082{}\cr{\vskip 0.0pt}{}Dwarfgalaxy158{}\cr{\vskip 0.0pt}{}Nebulae145{}\cr{\vskip 0.0pt}{}StellarRemnant1004{}\cr{\vskip 0.0pt}{}Early-typestar889{}\cr{\vskip 0.0pt}{}Late-typestar766{}\cr{\vskip 0.0pt}{}Binarystar487{}\cr{\vskip 0.0pt}{}Youngstellarobject250{}\cr{\vskip 0.0pt}{}\par{\hline}\par{\vskip 5.81247pt}\hskip 0.0ptTier3\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}Quasar963{}\cr{\vskip 0.0pt}{}Seyfertgalaxy707{}\cr{\vskip 0.0pt}{}Starburst193{}\cr{\vskip 0.0pt}{}PlanetaryNebula64{}\cr{\vskip 0.0pt}{}Whitedwarf870{}\cr{\vskip 0.0pt}{}Bstar255{}\cr{\vskip 0.0pt}{}Cataclysmicvariablestar186{}\cr{\vskip 0.0pt}{}Bluesupergiant166{}\cr{\vskip 0.0pt}{}\par\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\end@float\par TheresultsinthissectionarebasedontheHSLArunimplementedonSeptember8,2025,whichuseddatacollectedthroughJune25,2025.Atotalof6712uniqueHSLAtargetswereidentified.Ofthese,anoverwhelmingmajorityhaveaTier1classificationof``Star^{\prime\prime}(3894)or``Galaxy^{\prime\prime}(2459).Onlyasmallfraction(179)fallunder``ISM^{\prime\prime}.Oftheremaining,170areclassifiedas``Multiple^{\prime\prime},and10as``Unknown^{\prime\prime}.\par ThenumberoftargetsclassifiedineachofthemainTier1classes,alongwiththenumbersforasampleofTier2andTier3classesareshowninTable\ref{tab:hsla_class_totals}.ItisworthnotingthatalargefractionoftheGalaxieshave``Activegalaxy^{\prime\prime}and``Quasar^{\prime\prime}astheirsecondaryandtertiaryclassifications,andthatamongtheStars,thelargesttypeare``StellarRemnants^{\prime\prime}(secondary),whichinclude``Whitedwarf^{\prime\prime}(tertiary).Ofthe6712totalHSLAtargets,SIMBADwasusedtoclassify5548,ofwhich58wereexoplanethoststarsrequiringNExScI{}.NEDwasusedfor38objects,andtheremaining1126usedthePhase~IIproposalsforclassification.\par\par 3.3 Summary HSLA is providing automated target classifications for all the targets with COS or STIS spectral observations. The classification scheme uses three hierarchical tiers, and enables searches based on object type at all three levels, e.g. from “Stars”, to “Early-type star” to “O star”, or “Galaxy” to “Active galaxy” to “Starburst”. The classifications at each tier were developed based on a careful scrutiny of the object descriptions in SIMBAD, NED and the Phase II proposals. The process of assigning categories is automated and will be applied to all new targets observed as part of the HSLA process. The scheme and implementation are sufficiently flexible and the vocabulary is sufficiently comprehensive that they may be used for high-level data product archives from other general purpose instruments. 4. Data Products Once a selection of one-dimensional spectra are successfully associated with a target, coaddition occurs following the same procedure as described for HASP (Debes et al., 2024). Input spectra are drawn from the pre-filtered datasets identified for HASP, followed by a coaddition of all input spectra from multiple CENWAVES or apertures within a single grating across one or more programs that have targeted the same object. The largest source of potential errors comes from observations in which the source is not well centered or has drifted out of the slit/aperture, and the quality issues have gone unreported. To avoid these errors, spectra with anomalously low flux are excluded from the coadds. As with HASP data, this procedure makes default HSLA data products inaccurate for variable sources (Debes et al., 2024), but custom coaddition notebooks in the HASP Jupyter notebook repository provide guidance about how to create a useful coadd from flux variable spectra. No flux checking is performed across gratings. Observations of extended sources that have spectra with different aperture sizes may have flux offsets between gratings. As a reminder, our flux and wavelength requirements do not apply to extended sources. Additionally, variable sources that vary significantly between epochs may show flux offsets between gratings. The coadd process produces data products across multiple HST programs for each grating, each lifetime position (LP) for COS FUV modes, and a target quicklook spectrum that covers the full wavelength range of the available observations. For well-studied objects this can span from \sim 900-10,000 Å, covering the far-UV to the near-IR. The data products are in FITS table format with two extensions. The first extension includes a wavelength column in units of Angstroms, flux and error columns in units of erg s -1 cm -2 Å -1 , a Signal-to-Noise Ratio (SNR) column, and an effective exposure time column in units of seconds. The wavelength bin size for single grating spectra is set by looking at the wavelength differences between adjacent bins of all the input spectra, and setting the product bin size to the largest of these differences. The wavelength bins are of equal width in these spectra. The quicklook spectra inherit the wavelength bins from the constituent single-grating spectra, and therefore the bin widths will generally change with the transition from one single-grating spectrum to the next that make up the abutted spectrum. The second extension is the provenance table, which is described in more detail in Section 4.2 . 4.1 Coadd Naming Conventions The naming convention for data products is different for single grating products and for the quicklook spectra: Single Grating: hst \_ \_ \_ - \_ cspec.fits Target Quicklook: hst \_ \_ aspec.fits In the above naming scheme, instrument is either COS or STIS, target refers to the target association name, and opt \_ elem is the grating name for the relevant mode(s). If a grating name is shared by COS and STIS a ‘‘c’’ or ‘‘s’’ will denote which instrument it is associated with. Observations with COS gratings can be obtained at a variety of lifetime positions 1 1 1 https://hst-docs.stsci.edu/cosdhb/appendix-a-cos-lifetime-positions/a-1-cos-lifetime-positions . The lifetime position is denoted in the filename with lp# . 4.2 Provenance Extension Both HASP and HSLA products contain a provenance table that contains useful information about the input filenames, proposal IDs, and instrument details. Table 4.2 gives a detailed breakdown of what information is included. The coadd code reports all input data files in the provenance table regardless of whether some input files are removed due to certain conditions as described in Debes et al., (2024). A careful parsing of both the provenance file and the output trailer file described in Section 4.5 can help determine what input files were combined in the the different data products. Some examples of how to parse HASP trailer files, which will have similar formatting to HSLA trailer files, are available as a Jupyter notebook . Table 2 : Provenance Table Columns \begin{array}[c]{\string clll}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Field\hfil\hfil Name\hfil\hfil Units\hfil\hfil Description\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}1FILENAME{}\cr{\vskip 0.0pt}{}2EXPNAMEDatasetName{}\cr{\vskip 0.0pt}{}3PROPOSIDProgramID{}\cr{\vskip 0.0pt}{}4TELESCOPE{}\cr{\vskip 0.0pt}{}5INSTRUMENT{}\cr{\vskip 0.0pt}{}6DETECTOR{}\cr{\vskip 0.0pt}{}7DISPERSERGratingUsed{}\cr{\vskip 0.0pt}{}8CENWAVE{}\cr{\vskip 0.0pt}{}9APERTURESlitUsed{}\cr{\vskip 0.0pt}{}10LIFE\_ADJDetectorLifetimePosition{}\cr{\vskip 0.0pt}{}11SPECRESSpectralResolution{}\cr{\vskip 0.0pt}{}12CAL\_VERPipelineVersion{}\cr{\vskip 0.0pt}{}13MJD\_BEGdExposureStart{}\cr{\vskip 0.0pt}{}14MJD\_MIDdExposureMidpoint{}\cr{\vskip 0.0pt}{}15MJD\_ENDdExposureEnd{}\cr{\vskip 0.0pt}{}16XPOSUREsExposureTime{}\cr{\vskip 0.0pt}{}17MINWAVE\AA\ MinimumWavelength{}\cr{\vskip 0.0pt}{}18MAXWAVE\AA\ MaximumWavelength{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\end@float\par 4.3 Coadd abutment prioritization Different gratings within the full wavelength quicklook spectrum are abutted. Abutment is when two or more gratings are joined at a single transition wavelength, rather than merged (for more details on how abutment works for HASP and HSLA, see Debes et al., 2024). Abutment is used in both HASP and HSLA products due to the varying resolutions of the various modes and the difficulty involved in coadding overlapping regions. The transition wavelengths and the priority of a given grating is provided by a separate JSON prioritization table. In some cases, a target has been observed at the same wavelength by multiple gratings. In these situations, only a subset of gratings populate the final quicklook spectrum. The motivation for which gratings are prioritized is based on optimizing line detection for a median observation within a given grating. The line detection sensitivity of a grating, ‘gk’, can be estimated based on the limiting equivalent width of an observation ( \mathrm{EQW_{limit}} ) for an unresolved spectral line: \mathrm{EQW_{limit}}(\lambda)\propto\sigma_{\mathrm{cont}}\frac{\lambda}{R_{\mathrm{gk}}} (1) where \sigma_{\mathrm{cont}} is the uncertainty per resolution element, \lambda is the wavelength, and R_{gk} is the resolution of the grating. For spectra with heterogeneous SNR, it is preferable to recast Equation 1 in terms of instrument sensitivity( S_{\mathrm{{gk}}} ), the exposure time ( t_{\mathrm{exp,gk}} ), and the source flux ( F(\lambda) ): \displaystyle\mathrm{EQW_{limit,g1}}(\lambda)= \displaystyle\left[S_{\mathrm{g1}}(\lambda)t_{\mathrm{exp,g1}}F(\lambda)\right]^{-1/2}{\frac{\lambda}{R_{\mathrm{g1}}}} (2) \displaystyle\mathrm{EQW_{limit,g2}}(\lambda)= \displaystyle\left[S_{\mathrm{g2}}(\lambda)t_{\mathrm{exp,g2}}F(\lambda)\right]^{-1/2}\frac{\lambda}{R_{\mathrm{g2}}} (3) where S_{\mathrm{g1}}(\lambda) and S_{\mathrm{g2}}(\lambda) are the wavelength dependent sensitivities of gratings g1 and g2, respectively, and t_{\mathrm{exp,g1}} and t_{\mathrm{exp,g2}} the median exposure time across all observations obtained with these gratings. Grating g2 will have higher priority if the following inequality is satisfied: \left[S_{\mathrm{g2}}(\lambda)t_{\mathrm{exp,g2}}R_{\mathrm{g2}}^{2}\right]^{-1/2}<\left[S_{\mathrm{g1}}(\lambda)t_{\mathrm{exp,g1}}R_{\mathrm{g1}}^{2}\right]^{-1/2} (4) A table of median exposure times for each grating was generated from archived COS and STIS data, and used in equation (4) to determine the prioritization. These are listed in Table 4.3 for HSLA products. The publicly available coadd code allows for the customization of prioritization for specific science cases. In particular, the overlapping regions of COS/G130M and COS/G160M may have line sensitivities that depart from the simple formulation above and require special handling (Section ). Table 3 : Grating Abutment Order Priorities (Highest to Lowest Priority) \begin{array}[c]{\string cccc}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Instrument\hfil\hfil Grating\hfil\hfil WavelengthRange(\AA)\hfil\hfil\#CENWAVEs/Tilts\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}STISEchelleE140H1140--17003{}\cr{\vskip 0.0pt}{}STISEchelleE230H1620--31506{}\cr{\vskip 0.0pt}{}STISEchelleE140M1144--17101{}\cr{\vskip 0.0pt}{}COSFUVG130M900--14708{}\cr{\vskip 0.0pt}{}COSFUVG160M1342--18006{}\cr{\vskip 0.0pt}{}STISEchelleE230M1605--31102{}\cr{\vskip 0.0pt}{}COSNUVG185M1664--213415{}\cr{\vskip 0.0pt}{}COSNUVG285M2474--322117{}\cr{\vskip 0.0pt}{}COSNUVG225M2069--252613{}\cr{\vskip 0.0pt}{}STISMAMAG140M1140--174012{}\cr{\vskip 0.0pt}{}STISMAMAG230M1640--310018{}\cr{\vskip 0.0pt}{}STISCCDG430M3020--561010{}\cr{\vskip 0.0pt}{}STISCCDG750M5450--101409{}\cr{\vskip 0.0pt}{}COSFUVG140L900--21503{}\cr{\vskip 0.0pt}{}STISCCDG230MB1640--319011{}\cr{\vskip 0.0pt}{}STISMAMAG140L1150--17301{}\cr{\vskip 0.0pt}{}STISMAMAG230L1570--31801{}\cr{\vskip 0.0pt}{}STISCCDG430L2900--57001{}\cr{\vskip 0.0pt}{}STISCCDG230LB1680--30601{}\cr{\vskip 0.0pt}{}STISCCDG750L5240--102701{}\cr{\vskip 0.0pt}{}COSNUVG230L1650--32004{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\end@float\par 4.4 Metadata Files In addition to the spectra, each HSLA target association has a companion metadata text file that compiles information about the target and the input data in a simple human-readable format. We present two examples and describe the contents of these files. Figure 3: Example metadata file for the LMC O star SK -69 212. Figure 4: Example metadata file for the exoplanet hosting A star \beta -Pictoris. The first is the file for SK -69 212, an O star in the Large Magellanic Cloud (LMC; Fig. 3 ). The first part of the file contains naming and coordinate information. A primary target name and set of coordinates is derived from the target association process described in Section 2. Target Association . J2000 coordinates at epoch 2016 are derived from either SIMBAD, NED, or the Phase II astrometry. Next, there is a unique HSLA catalog name with the primary target name plus a unique numerical identifier, along with names, if available, from SIMBAD, NED, and the Phase II naming. If SIMBAD reports other sources within 2 ′′ of the target, this is noted along with the name. This is useful for checking for possible contaminants. In this case there is an X-ray source identified by Chandra about 0.3 ′′ away, which does not impact the HSLA catalog nor the data. The next few entries deal with the target class, including each of the three tiers of classification, the primary classification source and the UAT equivalent classification. Following that is a line reporting either the radial velocity or the redshift (for external galaxies) if available in SIMBAD, which it is not for SK -69 212. The last part of the metadata file lists the proposal IDs from which the data were derived, followed by a summary of the constituent modes along with the wavelength coverage, and LP information for COS FUV observations. The second example is the well-studied A star, \beta -Pictoris, which hosts exoplanets (Fig. 4 ). The metadata file shows at a glance that this is a popular target, having been observed in 11 programs and with several COS and STIS modes. As it is an exoplanet host, NExScI was queried to obtain the classifcation in addition to SIMBAD (see Fig. 2 ), and this information is included in the Tertiary Classification as well as the UAT Equivalent fields. The target also has a reported radial velocity in SIMBAD, which is included in the metadata file. The use of HSLA data for studying \beta -Pictoris is discussed at more length in §7.2, below. 4.5 Trailer Files In addition to the metadata file and the coadds that are delivered with an HSLA target, users can also download a trailer file (denoted with the “.trl” file extension) that contains all the text output from the coadd code, in a format similar to HASP products. The outputs are particularly useful for determining whether particular datasets were included or rejected during the coaddition process by searching for particular key words. Details of how to inspect trailer files for dataset rejection are described in the HASP Jupyter notebook on Data Diagnostics and with some modification can be applied to HSLA trailer files when combined with information from the metadata file. 5. Testing We conducted several tests to ensure the robustness of the HSLA products. First, a random sample of 100 HSLA targets was inspected in detail to verify the accuracy of their coordinates, the accuracy and completeness of the target associations, and the accuracy of their classifications. Then, following the procedure used for HASP, we tested the agreement in flux and wavelength between input spectra and the output coadds for each HSLA target with a requirement that > 75% of spectra matched to within 5% in flux and within half a resolution element in wavelength. Finally, we used a subset of the CALSPEC standard stars to verify the absolute flux calibration accuracy of the HSLA coadded spectra. 5.1 Target Association and Classification Accuracy Investigating over 6000 individual targets for accuracy of classifications, naming, coordinates and association accuracy reliably would be prohibitively time intensive. Therefore we instead conducted random sampling of the HSLA targets. We set our accuracy requirements to be consistent with > 90% for target association completeness and accuracy, and a best effort with respect to automated classifications. A random sample of 100 targets was selected for detailed examination. This sample size ensured that under the assumption of binomial statistics, we could demonstrate that we had met our requirements provided the number of failures were small and under the assumption of 95% confidence when calculating a cumulative probability distribution. Our procedure for testing the accuracy of the coordinates, association datasets, naming, and classifications was to use the information in the metadata files. The HSLA SIMBAD, NED, or Phase II names were input into the HST Mission Search form and were compared to the returned datasets and programs with the program IDs listed in the metadata file. The HST observation coordinates were compared to the SIMBAD or NED object entry if it existed and investigated if it seemed discrepant with the details and intended target of the given HST programs. The vast majority of HSLA targets are isolated objects with a well classified nature. Most classification and naming failures occurred in the situation where a structure within a nearby, spatially resolved galaxy was targeted and where SIMBAD has multiple unique objects listed within 2 ′′ of an association’s average coordinates. Users should use extra caution when including a target association when multiple unique SIMBAD or NED objects are present within 2 ′′ , which is flagged in the metadata file for the target. Table 4 : Target association coordinate accuracy, target association input dataset accuracy, target name accuracy, and target classification accuracy. Limiting accuracies are calculated assuming binomial probability distribution functions at 95% confidence. HSLA Property Accuracy Limiting Accuracy Coordinates 99% 95% Association Inputs 98% 94% Naming 93% 87% Classifications 92% 86% Table 4 provides our estimated accuracies for association coordinates, association accuracy, naming accuracy, and classification accuracy of the HSLA. For target association and coordinate accuracies, we exceed our requirements, achieving 94% accuracy. For Classification and Naming, we find that we are better than 86% accurate. 5.2 Flux and Wavelength Accuracy To ensure the flux and wavelength accuracy of our data products, we perform a series of tests comparing each coadded spectrum with its component xld datasets. We use the analysis software developed for HASP and described in the HASP ISR (Debes et al. 2024). Briefly, to test the flux accuracy, each xld spectrum is filtered to remove pixels with serious data-quality flags, binned into 20 wavelength bins, and compared with the similarly prepared coadded spectrum. We consider only wavelength bins with S/N >20 and only xld spectra with at least five such wavelength bins. For each wavelength bin, we compute the residual r=(F_{\rm xld}-F_{\rm coadd})/F_{\rm coadd} . For each xld spectrum, we seek to have both the mean and standard deviation of the residuals be less than 0.05. The wavelength accuracy is assessed by cross-correlating the prepared (unbinned) coadded and xld spectra of the same high-S/N spectra; our goal is for any offset to be less than 1 pixel. Table 5 : HSLA Test Results \begin{array}[c]{\string l c c c c}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Grating\hfil\hfil Totalxld\hfil\hfil Sufficient\hfil\hfil Flux\hfil\hfil Wavelength\hfil{}\cr{\vskip 0.0pt}{}\hfil Datasets\hfil\hfil Bins\hbox to0.0pt{${}^{\mathrm{a}}$\hss}\hfil\hfil Success\hfil\hfil Success\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}\par{\vskip 5.81247pt}\hskip 0.0ptCOS\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G130M1296671.45\%94.17\%98.46\%{}\cr{\vskip 0.0pt}{}G160M865267.83\%92.88\%99.76\%{}\cr{\vskip 0.0pt}{}G140L483218.63\%96.44\%84.00\%{}\cr{\vskip 0.0pt}{}G185M146259.78\%97.48\%97.14\%{}\cr{\vskip 0.0pt}{}G225M77347.09\%98.35\%94.78\%{}\cr{\vskip 0.0pt}{}G285M31292.31\%89.93\%89.58\%{}\cr{\vskip 0.0pt}{}G230L140858.95\%91.33\%89.88\%{}\cr{\vskip 0.0pt}{}\par{\hline}\par{\vskip 5.81247pt}\hskip 0.0ptSTIS\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G140M109837.16\%93.14\%99.02\%{}\cr{\vskip 0.0pt}{}G140L239164.83\%89.48\%99.03\%{}\cr{\vskip 0.0pt}{}G230L226765.06\%91.46\%99.73\%{}\cr{\vskip 0.0pt}{}G230LB67991.75\%94.86\%97.11\%{}\cr{\vskip 0.0pt}{}G230M38273.56\%94.31\%98.22\%{}\cr{\vskip 0.0pt}{}G230MB36154.85\%99.49\%95.96\%{}\cr{\vskip 0.0pt}{}G430L188676.56\%92.31\%96.19\%{}\cr{\vskip 0.0pt}{}G430M96061.88\%88.55\%91.75\%{}\cr{\vskip 0.0pt}{}G750L153989.86\%95.30\%97.90\%{}\cr{\vskip 0.0pt}{}G750M101165.38\%82.90\%87.90\%{}\cr{\vskip 0.0pt}{}E140H108396.77\%94.08\%99.05\%{}\cr{\vskip 0.0pt}{}E140M240189.09\%89.01\%96.35\%{}\cr{\vskip 0.0pt}{}E230H164395.86\%93.33\%99.49\%{}\cr{\vskip 0.0pt}{}E230M237790.66\%96.47\%99.44\%{}\cr{\vskip 0.0pt}{}\par{\hline}\par{\bf TOTAL}\hbox to0.0pt{${}^{\mathrm{b}}$\hss}5048367.2\%93.16\%97.54\%{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\par\vskip 7.74997pt{\vskip 0.0pt{\parbox{3525.18448pt}{\hskip 9.24994pt${}^{\mathrm{a}}$Flux and wavelength tests were performed only on xld files with at least five wavelength bins with S/N $>20$.}\par}\vskip 0.0pt{\parbox{3525.18448pt}{\hskip 9.24994pt${}^{\mathrm{b}}$Reported percentage values are averages.}\par}}\end@float\par ThisanalysiswasperformedforeachofthexldspectraincludedintheHSLAproject.ResultsarepresentedinTable\ref{tab:HSLA_test_results}.Weseethatthemeanandstandarddeviationoftheresidualsarelessthan0.05formorethan93\%ofxldspectrawithS/Nsufficienttomakethecomparison.Morethan97\%ofthosespectrayieldwavelengthoffsetsoflessthan1pixel.Boththefluxandwavelengthtestssubstantiallyexceedthebenchmarksuccessrateof75\%.\par 5.3 CALSPEC Standards and Flux Calibration All COS and STIS observations are flux calibrated using observations of selected standard White Dwarf stars (WDs), for which high fidelity synthetic model spectra are available in the CALSPEC database. The detector responses on both instruments change with time, and in order to maintain calibration accuracy, standard WDs are observed regularly in order to characterize the time-dependent sensitivity (TDS) for each of the modes. For most of the COS and STIS modes the flux calibration accuracy is 5% absolute and 2% relative. In other words, the fluxes reported in any calibrated spectrum is expected to be within 5% of the true value of the source flux, and for any given observation, the relative fluxes across the bandpass within a given mode and, in the case of COS, a given detector segment are accurate to 2%. The main exceptions, for which the flux calibrations have larger uncertainties, are those COS modes with wavelength coverage extending shortward of 1150 Å and the STIS high-resolution echelle modes, especially those obtained using slits narrower than 0.2 ′′ . Additionally, for several modes the flux calibration accuracies systematically degrade toward the detector edges. (See the COS and STIS Instrument Handbooks and references therein for details.) The flux calibration accuracy of the HSLA coadded spectra is limited by the accuracy of the calibrated input data. Additional sources of uncertainty are present in the coadds, since they typically include spectra obtained with several modes, and taken over longer spans of time than individual observations. In order to characterize the flux calibration accuracy of the HSLA data products we examine the residuals between coadded spectra of standard WDs and their synthetic models. We focus on three standards, G 191-B2B, GD 71 and WD0308-565, which together span all the COS and STIS modes used, and which have been observed as part of numerous programs either for flux calibration or for monitoring the TDS. The HSLA abutted spectra (stored in the “aspec.fits” files) for each of these stars overlaid on the correspending models are shown in Figs. 5 – 7 . Also shown are the residuals between binned data and model spectra. Note that the bin sizes are not uniform across the entire wavelength range for abutted spectra (§4) and we have chosen a binning value suitable for display in each case. The overlays show that the data and models generally agree well across the spectrum. One exception that stands out is the short wavelength region ( < 1100 Å) for G 191-B2B (Fig. 5 ). This is a case where there are only a handful of observations using the COS modes that cover this region, and it is known that the model does not provide a good fit at these wavelengths. More subtle issues are seen in the case of GD 71 at the shortest wavelengths where the spectrum is dominated by deep hydrogen lines, and where the model may not be sufficiently accurate. The other small discrepancy seen is a systematic offset of about 2% longward of around 6000 Å for both GD 71 and WD0308-565. The abutted spectra are meant as quicklook data products and not suitable for quantitative tests of the flux calibration accuracy. For that we turn to the scientifically usable spectra that form the core data products provided by HSLA. Figure 5: The top panels show the HSLA abutted spectrum of G 191-B2B overlaid on a synthetic spectrum of the star. The plot on the left spans about 1100 Å  and that on the right spans the remaining \approx 8000 Å covered by COS and STIS. The bottom panels show residuals obtained after binning the observed and model spectra. The bin sizes are different for the two sections of the spectrum, and were chosen solely for clarity of display. For each of the three standard stars we compare the spectra in individual “cspec.fits” files against the models. These spectra are grating-dependent, and in the case of COS FUV spectra HSLA separated by LP (§4). The COS NUV detector has operated at only a single lifetime position, which is designated LP1, and STIS does not use lifetime positions. For testing the HSLA flux calibration accuracy we restrict ourselves to STIS data for G 191-B2B and to COS data for WD0308-565, while for GD 71 we include both COS and STIS modes. To compare models with data, we first obtain the residuals at the full spectral resolution of the HSLA spectra (i.e. without binning): Residual=(F_{HSLA}-F_{model})/F_{model} and then we use the routine “astropy.stats.sigma_clip” to obtain the means and standard deviations of the residuals, rejecting 3 \sigma outliers iteratively for a maximum of five iterations. The resulting means and standard deviations are used as measures of flux calibration accuracy for each mode. These results are shown in Tables 5.3 – 5.3 for the three stars. In each table, the first column is the grating (and for COS modes the LP), the second and third columns give the wavelength range of the spectra, the fourth column is the total number of wavelength points, the fifth column is the number of these points omitted due to sigma-clipping, and the sixth and seventh columns are the residual means and standard deviations. Figure 6: Same as Fig. 5 but for GD 71. G 191-B2B is one of the main calibrators for STIS. It has been observed using all STIS modes except for the low-resolution MAMA modes, G140L and G230L. The residual mean is less than 1% for all the modes except the two echelle modes E230H and E230M where there are offsets of about - 2% and -1 %, respectively (Table 5.3 ). The standard deviations range from 0.4% (G230LB, G430L) to around 3.5% (G140M and G750M). The star has been observed using COS, but those data are known to have issues with flux calibrations and are not included in the table. GD 71 has been used for calibrating the COS NUV channels, and for tracking the TDS of COS FUV grating G160M, segment A. It is also used to calibrate or verify several STIS MAMA and CCD modes. Table 5.3 shows that the means are mostly less than 1%, and in the case of G160M at LP1 and G750L between 1 and 2%. The one exception is G130M at LP2, where the residual mean is 6% and standard deviation 8%. This is because the only G130M CENWAVEs used to observe GD 71 have been c1055 and c1096 and using only detector segment B, as reflected by the minimum and maximum wavelengths listed. Furthermore, there has been only one observation using c1055, where the wavelength extends below the Lyman cut-off at 912 Å. The flux calibration between \approx 912 and 940 Å is poor and that region has been omitted in our analysis. Figure 7: Same as Fig. 5 but for WD0308-565. WD0308-565 is the primary calibrator for the COS FUV modes, and is also used to monitor the TDS for most of them. The residual means (Table 5.3 ) are all less than 1% except for G130M at LP2 where it is 1.7%. The standard deviations fall in a narrow range between 2.2 and 3.5%. Note that in this case, as WD0308-565 has been observed many times using CENWAVE 1055 for TDS monitoring, we have retained the default minimum wavelength for G130M at LP2. This results in the relatively large number of clipped points. Across all the targets and modes, most of the residual means are less than 1%, and the residual standard deviations have a median of 2.3%. Based on these results we conclude that the HSLA “cspec” spectra retain the absolute flux calibration accuracy of 5% of individual observations and have a relative accuracy of 2.3%, which is only slightly lower than for individual observations. Table 6 : HSLA Flux Calibration: G 191-B2B \begin{array}[c]{\string lcccccc}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Grating\hfil\hfil$\lambda_{min}$\hfil\hfil$\lambda_{max}$\hfil\hfil N$_{tot}$\hfil\hfil N$_{clipped}$\hfil\hfil Residual\hfil\hfil Residual\hfil{}\cr\hfil\par\hfil\hfil(\AA)\hfil\hfil(\AA)\hfil\hfil\hfil\hfil\hfil\hfil Mean\hfil\hfil St.Dev.\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}{\vskip 5.81247pt}\hskip 0.0ptSTISECHELLE\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}E140H1141.01687.9719335397-0.0000.023{}\cr{\vskip 0.0pt}{}E230H1629.43146.31116592388-0.0210.024{}\cr{\vskip 0.0pt}{}E140M1140.41709.73054529550.0040.020{}\cr{\vskip 0.0pt}{}E230M1607.13073.5291381137-0.0120.015{}\cr{\vskip 0.0pt}{}{\vskip 5.81247pt}\hskip 0.0ptSTISMAMA\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G140M1145.31741.5111388520.0070.035{}\cr{\vskip 0.0pt}{}G230M1642.33097.1164381161-0.0010.032{}\cr{\vskip 0.0pt}{}{\vskip 5.81247pt}\hskip 0.0ptSTISCCD\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G230MB1635.13184.71022310990.0010.017{}\cr{\vskip 0.0pt}{}G430M3022.15610.19300246-0.0010.015{}\cr{\vskip 0.0pt}{}G750M5450.610127.58401999-0.0020.035{}\cr{\vskip 0.0pt}{}G230LB1664.43073.4102653-0.0010.004{}\cr{\vskip 0.0pt}{}G430L2965.55709.899911-0.0010.004{}\cr{\vskip 0.0pt}{}G750L5294.910258.11017211-0.0050.017{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\end@float\par\small\@float{table}\centering\@@toccaption{{\lx@tag[ ]{{7}}{HSLA Flux Calibration: GD~71}}}\@@caption{{\lx@tag[: ]{{{\bf Table 7}{}}}{HSLA Flux Calibration: GD~71}}}\small\hbox{$\begin{array}[c]{\string lcccccc}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Grating/LP\hbox to0.0pt{${}^{\mathrm{a}}$\hss}\hfil\hfil$\lambda_{min}$\hfil\hfil$\lambda_{max}$\hfil\hfil N$_{tot}$\hfil\hfil N$_{clipped}$\hfil\hfil Residual\hfil\hfil Residual\hfil{}\cr\hfil\par\hfil\hfil(\AA)\hfil\hfil(\AA)\hfil\hfil\hfil\hfil\hfil\hfil Mean\hfil\hfil St.Dev.\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}{\vskip 5.81247pt}\hskip 0.0ptCOSFUV\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G130M/2\hbox to0.0pt{${}^{\mathrm{b}}$\hss}940.01080.0141033770.0620.081{}\cr{\vskip 0.0pt}{}G160M/11602.11771.7138582590.0110.086{}\cr{\vskip 0.0pt}{}G160M/21575.01801.5184945950.0060.023{}\cr{\vskip 0.0pt}{}G160M/31575.01801.6184957440.0010.024{}\cr{\vskip 0.0pt}{}G160M/41533.31800.9218397670.0030.023{}\cr{\vskip 0.0pt}{}G160M/61530.71801.1220708350.0070.023{}\cr{\vskip 0.0pt}{}{\vskip 5.81247pt}\hskip 0.0ptCOSNUV\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G185M/11662.12135.4124422420.0130.049{}\cr{\vskip 0.0pt}{}G225M/12068.42526.412636232-0.0020.043{}\cr{\vskip 0.0pt}{}G285M/12473.83226.9173033400.0010.057{}\cr{\vskip 0.0pt}{}G230L/11700.43099.833732240.0030.030{}\cr{\vskip 0.0pt}{}{\vskip 5.81247pt}\hskip 0.0ptSTISMAMA\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G140M1145.51741.311117205-0.0030.024{}\cr{\vskip 0.0pt}{}G230M1724.63096.8155182090.0020.023{}\cr{\vskip 0.0pt}{}G140L1137.91718.0994102-0.0040.007{}\cr{\vskip 0.0pt}{}G230L1579.43168.11024340.0010.003{}\cr{\vskip 0.0pt}{}{\vskip 5.81247pt}\hskip 0.0ptSTISCCD\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G230LB1663.43073.8102741-0.0030.003{}\cr{\vskip 0.0pt}{}G430L2969.55708.3997250.0000.004{}\cr{\vskip 0.0pt}{}G750L5302.910256.41015178-0.0180.021{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\par\vskip 7.74997pt{\vskip 0.0pt{\parbox{3558.79433pt}{\hskip 9.24994pt${}^{\mathrm{a}}$Lifetime Positions relevant only for COS FUV modes.}\par}\vskip 0.0pt{\parbox{3558.79433pt}{\hskip 9.24994pt${}^{\mathrm{b}}$The minimum wavelength has been set to exclude G130M/1055-only data (see text).}\par}}\end@float\par\small\@float{table}\centering\@@toccaption{{\lx@tag[ ]{{8}}{HSLA Flux Calibration: WD0308-565}}}\@@caption{{\lx@tag[: ]{{{\bf Table 8}{}}}{HSLA Flux Calibration: WD0308-565}}}\small\hbox{$\begin{array}[c]{\string lcccccc}{\hline}{}\cr{\vskip-6.58745pt}{}\hfil Grating/LP\hfil\hfil$\lambda_{min}$\hfil\hfil$\lambda_{max}$\hfil\hfil N$_{tot}$\hfil\hfil N$_{clipped}$\hfil\hfil Residual\hfil\hfil Residual\hfil{}\cr\hfil\par\hfil\hfil(\AA)\hfil\hfil(\AA)\hfil\hfil\hfil\hfil\hfil\hfil Mean\hfil\hfil St.Dev.\hfil\hskip 6.0pt{}\vrule height=0.0pt,depth=2.71246pt,width=0.0pt\cr{}{\hline}{}\cr{\vskip-5.81247pt}{}\par{\vskip 5.81247pt}\hskip 0.0ptCOSFUV\hfil{}\cr{\vskip 0.0pt}{}{\vskip 1.93748pt}G130M/11132.71472.13403722590.0070.028{}\cr{\vskip 0.0pt}{}G130M/2900.41472.357343111730.0060.029{}\cr{\vskip 0.0pt}{}G130M/31066.41472.24067844770.0020.023{}\cr{\vskip 0.0pt}{}G130M/41065.11472.24081439040.0050.022{}\cr{\vskip 0.0pt}{}G130M/51131.31472.53420236040.0170.028{}\cr{\vskip 0.0pt}{}{}\cr{\vskip 0.0pt}{}G160M/21387.61795.9333437140.0080.035{}\cr{\vskip 0.0pt}{}G160M/31384.01802.03412311690.0000.025{}\cr{\vskip 0.0pt}{}G160M/41343.11802.03744617110.0080.023{}\cr{\vskip 0.0pt}{}G160M/61339.61801.83772214900.0070.027{}\cr{\vskip 0.0pt}{}{}\cr{\vskip 0.0pt}{}G140L/2915.01959.91296212500.0040.034{}\cr{\vskip 0.0pt}{}G140L/3915.01959.9129621290-0.0010.022{}\cr{\vskip 0.0pt}{}G140L/4915.01959.91296310750.0040.027{}\cr{\vskip 0.0pt}{}\crcr{\vskip 2.71246pt}{\hline}\lx@close@alignment\end{array}\end@float\par\vskip-8.5359pt\par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf\arabic{section}. Accessing the Catalog and Data\hfill}} \addcontentsline{toc}{section}{6. Accessing the Catalog and Data} \par Users can access HSLA data through the \hrefhttps://mast.stsci.edu/portal/Mashup/Clients/Mast/Portal.html, the \hrefhttps://mast.stsci.edu/search/ui/#/hst or via {astroquery}. In addition to the usual search categories such as target name and coordinates, HSLA products may be found using the classifications in any of the three tiers. Data may be obtained for individual targets, or for all targets spanning one or more classes. Metadata files are included in the data products and may be downloaded separately if necessary. \par Data access may change with time and users are encouraged to visit the \hrefhttps://archive.stsci.edu/missions-and-data/hst/hsla for the latest details on how to access data. \par\par\lx@frontmatter@fallback\@@numbered@section{subsection}{toc}{Considerations for HSLA data products, associations, and classifications} \par While HSLA data products, associations, and classifications are robust in the large majority of cases, certain target types or observation types are not well suited for coadds across multiple programs. We provide a brief list of considerations here, and users are encouraged to carefully inspect the trailer files, metadata files, and data before relying on the results. \par\begin{itemize}\par\itemize@item@{In crowded fields, particularly where there are both multiple HST pointings and SIMBAD objects, it will be useful to check the individual program target specifications and information to determine whether the coadds and classifications are correct. If the HSLA data are compromised by including multiple targets, then HASP data may be used or a custom coaddition of user selected datasets.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Blue Modes of G130M have differing resolutions and thus differing LSFs; caution should be used when analyzing lines in overlap regions with the 1055,1096, 1222, and 1291 CENWAVES. An example HSLA notebook of handling differing LSFs is given in Section \ref{sec:custom}.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{There is a small wavelength offset for spectra taken at the E1 position that is currently unaccounted for in the STIS pipeline and needs to be corrected for (See Section 4.2 of Joyce et al., 2018). This can be mitigated by applying a wavelength correction and performing a custom coaddition.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{The HSLA code does not implement wavelength shifts to the input files in the sense of matching up features in spectra to improve the wavelength calibration. The data are resampled onto a new grid and so wavelengths can shift by up to half a wavelength bin. Radial velocity variability of sources will not be accounted for or corrected.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{The flux accuracy for some modes like the STIS echelles, narrow slits on STIS (below 0.2") or the COS 1055 and 1096 CENWAVES can be less precise than our flux requirements either due to absolute flux calibration accuracy or focus variations within an HST orbit.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Target associations, target names, and target classifications are only as reliable as their SIMBAD entries for a majority of HSLA targets. Users who find an inaccuracy are encouraged to submit a help desk ticket so that corrections to the metadata files can be made as well as to SIMBAD.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Despite a high degree of accuracy between the target associations and SIMBAD classifications, completeness for any object depends on details of the classifications. No effort was made to be consistent for e.g. objects with multiple properties such as post-AGB stars vs. Planetary nebulae central stars or Active galaxies that also are a particular morphology. In this case cross matching with a known catalog of objects may be more complete.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{If multiple Phase II classifications were present within a program or across different programs the most common property was used, with some limited filtering of special cases.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Relatively compact sources that are spatially resolved with STIS and COS often meet our requirements but may suffer from excessive dataset filtering or flux mismatches between gratings. Users should carefully understand the different angular areas subtended by each dataset to properly interpret the HSLA coadds.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Sources that are variable in time are likely to have excessive dataset filtering or flux mismatches between gratings. Users should carefully understand the different epochs in which HST observations were taken to properly interpret the HSLA coadds for a variable source. Custom coaddition of datasets scaled to a common flux may be appropriate for some science cases. An example of how to do this is given in the HASP notebook on \hrefhttps://spacetelescope.github.io/hst_notebooks/notebooks/HASP/FluxScaleTutorial/FluxScaleTutorial.html.} \par\par\vskip 4.0pt\vskip 284.52756pt\itemize@item@{Moving Targets are not included in HSLA under the assumption that they are often flux variable.} \par\end{itemize} \par \par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf\arabic{section}. HSLA Examples\hfill}} \addcontentsline{toc}{section}{7. HSLA Examples} \par In this Section we give an introduction to some of the potential unique uses of HSLA data products. HSLA products are broadly useful in that they can facilitate the analysis of HST spectra for objects that have been observed many times or with many modes. Collating multiple programs then increases overall SNR, and extends wavelength coverage. Additionally, target classification enables studies on classes of objects, such as stars or galaxies, or subclasses of objects such as Seyfert 1 galaxies or WDs. \par\par\lx@frontmatter@fallback\@@numbered@section{subsection}{toc}{High SNR Spectrum of WD0308-565 from 920-10000\AA} WD0308-565 is a DB white dwarf that has been extensively used as a calibrator for the COS instrument. In the HSLA, its SIMBAD name is WG~7 and its unique target identifier is 722. Because of its status as a calibrator, many COS modes have been observed throughout the operational lifetime of the instrument (\lx@sectionsign 5.3). As an example, we show portions of the 920-10,000~\AA\ spectrum in Figure \ref{fig:wd0308} that is comprised of a mix of medium resolution COS spectra and low resolution STIS spectra from six different gratings. \par In particular, the median effective exposure time in the region of 1200-2000~\AA\ is 1.1$\times$10${}^{5}$~s, or 30.8 hours of exposure time. At this total exposure time, one is limited by fixed pattern noise. Even so, the fixed pattern noise is diminished due to the combination of several lifetime positions, FP-POS and small non-repeatable shifts of the OSM1 mechanism. We measure an SNR of 111 in the region of 1220-1240~\AA\ assuming a $\sim$6 bin resolution element, which exceeds the nominal best SNR for G130M mode assuming COS 2025 rules by a factor of nearly 4. \par Below 1000~\AA, the total SNR is close to 36/resolution element, allowing studies of the ISM and the carbon abundance of this DB in exquisite detail, with the caveat of a more complex line spread function due to multiple lifetime positions. If a science case requires an accurate measure of the line spread function, users can refer to the HSLA notebook that describes how to create a custom LSF that accounts for the merging of different lifetime positions, or they can make use of the lifetime position single grating spectra. \par\begin{figure}\centering\includegraphics[width]{hsla_dataprod_wg7.png} \@@toccaption{{\lx@tag[ ]{{8}}{Selected regions of the HSLA quick look spectrum of WD0380-565, which spans from 920-10,000~\AA. Top: Spectrum near 950~\AA\ that shows several ISM H lines. Bottom: Several ISM Si and Photospheric C lines are seen in a very high SNR region of the FUV spectrum, obtained by dozens of individual observations of this flux standard. Residual OI airglow lines are present around 1301 and 1306~\AA.}}}\@@caption{{\lx@tag[: ]{{Figure 8}}{Selected regions of the HSLA quick look spectrum of WD0380-565, which spans from 920-10,000~\AA. Top: Spectrum near 950~\AA\ that shows several ISM H lines. Bottom: Several ISM Si and Photospheric C lines are seen in a very high SNR region of the FUV spectrum, obtained by dozens of individual observations of this flux standard. Residual OI airglow lines are present around 1301 and 1306~\AA.}}} \@add@centering\end{figure} \par\par\lx@frontmatter@fallback\@@numbered@section{subsection}{toc}{UV Coverage of Exoplanet Host Star $\beta$~Pictoris} \par Eleven separate programs (PIDs: 7512, 9154, 13406, 14089, 14735, 14894, 15396, 15412,15479, 17421, 17790) have targeted the well-known exo-comet, giant-planet, and debris disk hosting star in the UV using a wealth of COS and STIS modes, including G130M, E140H, and E230H. For HSLA, the target name is ``* Beta Pic'' and its unique target identifier is 1134. In addition to the photosphere, the spectrum includes absorption due to a non-varying stable gas component roughly at the rest velocity of the star and time variable red- and blue-shifted components at high velocities caused by the evaporation of exo-comets. Figure \ref{fig:betapic} shows the full 1135-3095~\AA\ spectrum that spans four orders of magnitude in flux for this interesting star, where multiple elements in the circumstellar gas disk have been detected (Roberge et al., 2006; Wilson et al., 2017; Wilson et al., 2019) \par One caveat to note is that since time variable absorption occurs at higher redshift and blueshift from the system radial velocity, the quicklook spectrum shown here might not be desirable for understanding the short term evaporation behavior of $\beta$~Pictoris' exo-comets. In that case, users are recommended to investigate the visit and program level HASP coadds for the individual programs listed above. Inspection of the HSLA metadata file would facilitate obtaining that information quickly for any HSLA object to check for variability. \par\begin{figure}\centering\includegraphics[width]{hsla_dataprod_betapic.pdf} \@@toccaption{{\lx@tag[ ]{{9}}{Full UV SED of $\beta$~Pictoris derived from the HSLA quicklook {aspec} file, covering three separate COS and STIS gratings. Abundant stellar and circumstellar absorption features are seen in the spectrum.}}}\@@caption{{\lx@tag[: ]{{Figure 9}}{Full UV SED of $\beta$~Pictoris derived from the HSLA quicklook {aspec} file, covering three separate COS and STIS gratings. Abundant stellar and circumstellar absorption features are seen in the spectrum.}}} \@add@centering\end{figure} \par\par\lx@frontmatter@fallback\@@numbered@section{subsection}{toc}{Large scale UV spectroscopic surveys of White Dwarfs} \par Isolated WDs that are not actively accreting material have very simple atmospheres that usually contain either hydrogen or helium spectral lines. To that end, especially in the UV, they represent an interesting population to observe in order to understand the composition and properties of the local ISM. The WDs act as a background light source on which ISM absoprtion is imprinted. Accreting white dwarfs that show Si are primarily believed to be accreting rocky material consistent with bulk Earth and thus are probing the minor body populations of planetary systems that have survived post-main sequence evolution (Zuckerman et al., 2007; Koester et al., 2014). \par While a detailed study of the ISM or WD accretion is beyond the scope of this ISR, we demonstrate the potential power of HSLA in facilitating large scale spectroscopic studies in the UV of different types of astrophysical objects. In this example, we show that one can easily generate a large survey of WDs observed by HST by performing equivalent width measurements in the Si~1260~\AA\ FUV ISM line and the Si~1265~\AA\ photospheric accretion line resolved and simultaneously covered by COS G130M, G140M, E140M, and E140H. \par First, we select all HSLA targets that are classified as "White Dwarfs", which results in 890 HSLA targets. We then read each aspec file to select a subset of HSLA targets that have wavelength coverage around 1260~\AA. Roughly 753 pass that selection. After doing a simple fit to the contiuum, we calculate the equivalent width within $\pm$50~km~s${}^{-1}$ of the rest line center to get a rough estimate of the absorption due to the ISM for each target. We also measure the equivalent width of the Si~1265~\AA\ line, which is generally only detected if there is active accretion onto the photosphere of the WD (Koester et al., 2014). We required that a spectrum have at least two wavelength bins in our equivalent width measurement region, which ruled out the lower resolution modes. We assume each WD with just a significant 1261~\AA\ detection is a clean ISM detection and each WD with a significant 1265~\AA\ detection is appropriate for studying WD accretion. We assume any equivalent width measurement larger than 3~$\sigma$ to be significant, using the nearby continuum to estimate our uncertainties in equivalent width. From this sample, we find 462 WDs with detectable ISM accretion and 135 WDs with detectable Si from the accretion of rocky material. Figure \ref{fig:wd} shows the distribution of the HSLA WD targets in galactic coordinates. \par\begin{figure}\centering\includegraphics[width]{hsla_dataprod_wd.pdf} \@@toccaption{{\lx@tag[ ]{{10}}{An aitoff projection of galactic coordinates for HSLA white dwarfs. The grey triangles are WDs that have sufficient wavelength coverage and resolution to resolve and detect the FUV Si lines at 1261~\AA\ and 1265~\AA. Light blue circles represent HSLA WDs with a significant Si 1261~\AA\ line, indicative of detectable local ISM absorption, while red squares represent HSLA WDs with a significant Si 1265~\AA\ line, indicative of ongoing accretion onto the WD from rocky material.}}}\@@caption{{\lx@tag[: ]{{Figure 10}}{An aitoff projection of galactic coordinates for HSLA white dwarfs. The grey triangles are WDs that have sufficient wavelength coverage and resolution to resolve and detect the FUV Si lines at 1261~\AA\ and 1265~\AA. Light blue circles represent HSLA WDs with a significant Si 1261~\AA\ line, indicative of detectable local ISM absorption, while red squares represent HSLA WDs with a significant Si 1265~\AA\ line, indicative of ongoing accretion onto the WD from rocky material.}}} \@add@centering\end{figure} \par\par\vskip-8.5359pt \par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf\arabic{section}. Jupyter Notebooks for Custom Coadds\hfill}} \addcontentsline{toc}{section}{8. Jupyter Notebooks for Custom Coadds} \par The coadd code and its HASP/HSLA wrapper are available for public use in Space Telescope's GitHub Organization; {coadd} is located in the \hrefhttps://github.com/spacetelescope/ullyses/, and the HASP/HSLA wrapper for {coadd} is in the \hrefhttps://github.com/spacetelescope/hasp. Both codes are scripted in Python (compatible with versions 3.9 or later) and are installable using pip. More detailed installation and use instructions are given in our tutorial notebooks, which are described in the HASP ISR (Debes et al., 2024). Two additional notebooks specifically address the HSLA data products and the challenges inherent in combining data from observing programs that are separated in time by many years. \par The first notebook, \hrefhttps://github.com/spacetelescope/hst_notebooks/tree/main/notebooks/HSLA/Intro/Intro.ipynb, explores the standard HSLA data files returned by MAST. It provides a simple example of a custom coadd, demonstrating how to adjust the logic by which the abutting routine decides which coadded spectra (cspec files) to use for each wavelength region. \par The second notebook, \hrefhttps://github.com/spacetelescope/hst_notebooks/tree/main/notebooks/HSLA/Setup/Setup.ipynb, explores how the COS line-spread function (LSF) varies with LP and CENWAVE. Because the HASP/HSLA script sums all of the spectra from a particular grating without regard to LP or CENWAVE, its final products may not have the highest-possible spectral resolution. The notebook presents several techniques for dealing with these effects. \par All HASP and HSLA notebooks are available from the \hrefhttps://spacetelescope.github.io/hst_notebooks \par \par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf\arabic{section}. Conclusions\hfill}} \addcontentsline{toc}{section}{9. Conclusions} \par This report introduces the methods and advantages of a new Hubble Spectroscopic Legacy Archive. The new HSLA represents a critical evolution in delivering spectroscopic data that leverages 28 years of observations from STIS and 16 years of observations from COS, allowing the study of objects with higher SNR spectra or greater wavelength coverage than the data from individual programs. The combination of coadded spectra associated with astrophysical object types allows the study of galaxies and stars as large populations that enhances the legacy value of HST spectroscopy for years to come. Finally, the HSLA innovates a new way to classify and associate datasets that could be applicable to other observatories that host spectroscopic data products. \par\par\par\par\par\vskip-8.5359pt \par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf Change History for COS ISR 2025-18\hfill}} \addcontentsline{toc}{section}{Change History for COS ISR 2025-18} \vskip-8.5359pt Version 1: 6 November 2025 - Original Document \par\vskip-8.5359pt \par\lx@frontmatter@fallback\@@unnumbered@section{section}{\hbox to \hsize{\large\bf References\hfill}} \addcontentsline{toc}{section}{References} \vskip-8.5359pt \par\par\noindent Debes, J., et al.\ 2024 COS Instrument Science Report 2024-01(v2) \\ Frey, K., \& Accomazzi, A. 2018, ApJS, 236, 24F \\ Joyce, S.~R.~G., Barstow, M.~A., Holberg, J.~B., et al.\ 2018, MNRAS, 481, 2361 \\ Koester, D., G{\"{a}}nsicke, B.~T., \& Farihi, J.\ 2014, A\&a, 566, A34 \\ Peeples, M. et al., 2017, COS Instrument Science Report 2017-4 \\ Roberge, A., Feldman, P.~D., Weinberger, A.~J., et al.\ 2006, Nature, 441, 724 \\ Roman-Duval, J., et al.\ 2025 ApJ, 985, 109 \\ Wilson, P.~A., Kerr, R., Lecavelier des Etangs, A., et al.\ 2019, A\&A, 621, A121 \\ Wilson, P.~A., Lecavelier des Etangs, A., Vidal-Madjar, A., et al.\ 2017, A\&A, 599, A75 \\ Zuckerman, B., Koester, D., Melis, C., et al.\ 2007, ApJ, 671, 872 \\ \par\par\par\lx@frontmatter@fallback\@add@PDF@RDFa@triples\par\end{document}}}