loupe

a paper from loupe

What to keep

Which file is authoritative, what to do to it, and where it should live.

Every photograph an artist cares about ends up as several files, and only one of them is the work. This paper says which one, what to do to make it, and where each of the others belongs.

First, the words, because the field uses the same one two ways. The US federal digitisation guidelines, FADGI, call the uncorrected scan the archival master (or preservation master) and the cropped, corrected file made from it the production master.¹ The best-known reference written for photographers, dpBestflow, calls that same corrected file the master file.² This paper uses FADGI's words, because they say what each file is for. We borrow the words, not FADGI's specification. Where we recommend something different, we say so and give the reason.

1. Why this is hard

Portfolio tools know how to make a page and were never built to keep anything. Preservation standards know how to keep things, and were written for institutions reproducing objects they own — where the object is the authority and the file is evidence about it. The artist in the middle is neither: they are the authority. There is little written for them, what there is often conflicts, and short of hiring someone, you are on your own to decide.

Each clause of that can be checked.

The tools that hold an artist's record are not built to keep the files. The most thorough guide written for artists, the Joan Mitchell Foundation's Career Documentation for the Visual Artist, says image files are managed “outside of the artist's inventory management system”, which holds “low-resolution images … for reference purposes”.⁸ The services built for that record agree through their limits: they cap an upload at 15 to 30 megabytes, and one of them tells its users in its own documentation not to use it as an archive for their images. Publishing can also destroy what a file carries. When IPTC, the body that maintains the photo metadata standard, tested social sites in 2019, ten of them removed at least some of a photograph's embedded metadata, and one removed all of the XMP.¹⁸

Little is written for artists. No standards body publishes a JPEG quality setting. No institution has a view on chroma subsampling, the one JPEG setting whose damage a higher quality cannot undo (§7). And nobody publishes how large an artist's archive actually gets.

What exists conflicts, sometimes within one document. FADGI says master files “should be lossless”, then recommends JPEG 2000 at 20:1, which is lossy, “for most applications”.¹ Ask four national programmes which color space a master should use and you get four different answers (§7).

And what exists is written for someone else. An artist who follows museum practice exactly still gets the wrong answer on the question that matters most, because that practice was written for a museum reproducing an object someone else made.

2. Three files, not one

For every photograph that matters you will end up with three kinds of file, and they have different jobs.

Between the last two sits the file to send: a full-size JPEG made from the production master (§7). It keeps every pixel, carries your record inside it, and is a fraction of the size. It is the file you give a printer, a publisher, or loupe.

Keep the masters within reach. The file to send is the best file for today's screens and printers, not for every screen to come. Displays that show richer colour and finer tone are arriving, and the formats that carry them are settling (§12). When they do, the file to send will be made again, from your production master. A master that is backed up but can still be found and opened is ready for that day; one put away and forgotten is not.

Make the production master once, and make copies from it for as long as you need them. Always make a copy from the production master, never from another copy (§7 explains why).

3. The scan and the work

A scan is a dated observation. The production master is the work.

A scan of a negative is a machine's reading of the film: the scanner's color, the film's edges, the dust, and whichever way up it was loaded. It tells you about the film. The crop, the color and the decision about what the photograph is make up the work. So the first thing to do with a scan from a lab is to make a production master from it, and every later use comes from that master: a page on your site, or a 24 × 30 inch print for sale.

The institutional position, at its strongest

Preservation standards forbid exactly that. FADGI: “No image retouching is permitted to master files”; “Dust removal is not allowed on master images”; “Color, tone enhancement or restoration is not allowed on master images.”¹ The Dutch national programme, Metamorfoze, is stricter still. Its preservation masters are “unprocessed first-generation camera or scanner files”, and during capture “the editing software must be turned off.”³ FADGI's code of ethics gives the reason: a digitised image should be “a ‘reasonable reproduction’ without enhancement … without altering the fundamental nature of the historic record.”¹

They are right about their own case. In a museum the object is the authority and the file is evidence about it, so correcting the file destroys the evidence.

But institutions do not leave masters untouched, and it would be unfair to say they do. FADGI calls the idea that a file straight from a scanner is “pristine or unmolested” a misconception: “many people argue that one should not perform any post-scan or post-capture adjustments on image files … We disagree.” It asks for levels and curves, highlights and shadows kept “without clipping detail”, and admits that “some amount of subjective evaluation may be necessary.”¹ Both sides process their files. The difference is who decides when the file is right.

The artist's archive reverses the authority

In an artist's archive the artist is the authority. An uncorrected file is evidence of nothing but what a scanner did on one day. The logic is the museum's, and it reaches the opposite conclusion. FADGI itself says as much when it comes to negatives. The flat, unclipped scan is saved “as the archival master”, and then:

“To create a production master from the above, invert the image to produce a positive image. The resulting image will need to be adjusted to produce a visually pleasing representation. Digitizing negatives is analogous to printing negatives in a darkroom and is very dependent on the photographer's or technician's skill and visual literacy to produce a good image.”¹

A museum hands that darkroom judgement to a technician who works to the institution's targets. An artist's archive hands it to the artist.

4. Why keep the scan, and how well

The usual reason for keeping the raw scan is that later software will get more out of it. FADGI rejects that reason:

“If an unadjusted/raw scan is saved, future image processing can be hindered by unavailability of the original for comparison … In the future, unadjusted or raw images will need to be processed to be used and to achieve an accurate representation of the originals, and this will be difficult to do.”¹

A raw scan is hardest to interpret once the original is gone, and that is exactly when you would need it. So “the raw” is not the thing worth keeping. Keep a high-bit, minimally adjusted scan saved in a named color space: a 16-bit TIFF whose color space is recorded, not an undocumented dump from the device. From a good scanning service this is usually the same file people mean by “the raw scan”, saved properly. Keep it beside the production master, together with whatever documents the original: a reference print, a densitometer reading, a photograph of the film on a light box. Future software will need that documentation and will not have it.

Is the scan replaceable?

A scan can be made again only while three things are all true: the original still exists in a condition that can be scanned, a comparable machine can be reached, and someone is willing to pay for the pass. When all three are in doubt, treat the scan as a preservation master.

The evidence on scanners points less clearly than people expect. The best-known dedicated film scanner, Hasselblad's Flextight, was withdrawn around 2019, and no new high-end scanner for still film has replaced it.²⁸ But a careful camera-scanning setup has closed most of the gap. The Flextight's claimed resolution does not hold up under measurement. On 4 × 5 film it resolves 2,040 ppi, which is fewer pixels than a single high-resolution camera frame, and the density advantage once claimed for drum scanners does not survive the published figures.²⁹ No controlled test has put one negative through a camera, a drum scanner and a dedicated film scanner and published the measurements. So the case for keeping a scan should not rest on “they don't make the machines any more.” It rests on the film.

Scanner quality in twenty years is a guess. Film condition in twenty years is not.

• Acetate film decays, and the decay speeds itself up. Using the Image Permanence Institute's figures, the US National Park Service estimates that acetate film at 75°F and 50% relative humidity reaches significant decay in about 25 years. At 55°F it takes about 105 years, and at 32°F about 625. The Park Service's own conclusion: “providing cold storage should be the first step in a collection preservation strategy over reformatting or duplication.”¹¹ Refrigerate the film before you worry about scanning it.
• Different films want very different storage. ISO's recommendations, as the Image Permanence Institute reports them: color film on an acetate base at −10°C (14°F); black-and-white film on a polyester base at room temperature, up to 21°C and 50% humidity.¹²
• Dye fading is slower and more varied. Kodak's own predictions, as compiled by Wilhelm and Brower, give color negative films in dark storage anywhere from 7 to 65 years before the least stable dye loses 20%, depending on the film. Kodak cautioned that such figures can be off by half in either direction.¹³
• Advanced decay destroys detail, not just color. Once acetate shrinks and channels, the image detail is gone for good. The Northeast Document Conservation Center's advice: “Any negatives that show signs of deterioration should be digitized as soon as possible.”¹⁴

If you are not sure whether a film is acetate or polyester, test it. The Image Permanence Institute describes simple tests, such as looking at the film edge-on and between two polarising filters. A stock's name will not tell you reliably.

Camera raw, and DNG

DNG, Adobe's openly documented raw format, has a strong case, which dpBestflow makes: it keeps “the original raw sensor data … Nothing is left out”, it carries a checksum of that data so it “can continue to be verified forever”, and it can hold the camera's own raw file inside itself.² The Library of Congress lists both DNG and camera raw as acceptable rather than preferred, and sums up the whole question in three phrases: DNG is “somewhat normalized”, camera raw “too raw”, an uncompressed TIFF “too cooked”.⁵ Our advice: keep the camera's own raw file. A DNG beside it, or wrapping it, is a sound second copy that can check itself. That is a choice, not a rule.

5. Where each file lives

The rule most often given to photographers is 3-2-1: three copies, on two kinds of storage, one of them somewhere else. A photographer wrote it for photographers. The US Cybersecurity and Infrastructure Security Agency attributes it to Peter Krogh's The DAM Book.¹⁵ The Joan Mitchell Foundation gives artists the same rule.⁸

• Copy to new drives every five to seven years. This is the one interval the sources agree on: the Library of Congress says every five years and asks you to check your files once a year;¹⁶ the Joan Mitchell Foundation and Artists' Studio Archives say five to seven.^8,10 The Foundation suggests writing an expiry date on each drive.
• A synced folder is not a backup. A deletion or a damaged save reaches every synced copy at once. dpBestflow warns that a simple synchronised backup “will often transfer these problems to your backup — overwriting a good backup file with an unwanted change.”² Archive files do not change, so the right kind of backup only ever adds files. Its modern form is cloud storage that keeps old versions and refuses deletion for a set period.
• “Permanent means ‘permanent until the next migration’.”² Storage is something you keep doing, not a place you put things once.

The trap in cheap cold storage

The trap is not durability, because deep-archive cloud storage is durable. The trap is checking. You pay each time you read from the cheapest tiers, so nobody reads their files to check them, and nobody notices when one is lost. Data can go bad silently. The first time anyone opens a file in years may be the day it is needed, and that is when they find out it is broken. So keep the deep tier for the copy you hope never to touch, and never let it be the only copy. When you compare services, look at what it costs to get everything back out, not only at the storage price per terabyte. For an artist's archive the cost of retrieval usually matters more, and on the cheapest tiers the files have to be thawed, for a fee, before you can read them.

6. Making the production master

dpBestflow defines a master file as one that has been renamed and has had “basic metadata, capture sharpening, dust-busting, and basic image/color corrections applied”.² That is the right list for an artist. Within it:

• crop away the film edges, and rotate and straighten;
• correct color and tone with levels and curves, not global brightness and contrast, and never with an automatic setting (FADGI's own advice for adjusting a master);¹
• apply capture sharpening;
• remove dust and scratches;
• flatten the layers.

Where the line falls

FADGI and dpBestflow agree on one rule in almost the same words: a master is use-neutral. “Digitization should be done in a ‘use-neutral’ manner, and should not be geared for any specific output.”¹ So a step does not belong in the master if it serves only one use:

• output sharpening, which belongs to one print size on one paper (dpBestflow: masters “should never have output sharpening applied”);²
• a crop to a delivery shape, such as a square for a feed or a book's page ratio;
• reducing the size, since the master is as large as the capture;
• converting to a small color space for screens, since sRGB belongs to the copy;
• a treatment you might not want next year, such as a toned or split-graded print. That is a version. Give it its own name and keep it beside the master, not in place of it.

Borrow one habit from the museums: when you make a change that alters the image, write it down. FADGI lists cropping, sharpening, enhancement and saturation among the operations to record in metadata.¹

Your layered 16-bit working file is not the master. It holds decisions still being made, and it stays on your disks.

7. Saving it

Keep the production master as a 16-bit TIFF if you can. Nearly every source agrees that a master should be high-bit and stored losslessly.^1,4 If you compress the TIFF, choose ZIP with a predictor. Avoid LZW: dpBestflow notes it “may actually increase file size for 16-bit TIFFs”,² and on six photographs we tested, LZW made two of the six files larger than no compression at all.²⁷

The file to send. When you need a smaller file that is still fit to be the authority, for printing, for your record, or to send, this is our recommendation:

JPEG, saved with Photoshop's Save As at quality 10

Not 12, and never below 10.

No chroma subsampling (4:4:4)

This matters more than the quality number.

8 bits per channel

A JPEG cannot hold more.

Adobe RGB (1998), with the profile embedded

Always embed it.

No standards body publishes a quality figure, so this recommendation is ours. FADGI's tables say “All” for access formats at every level, and NARA, the Library of Congress and Metamorfoze give no figure either.^1,3,4,7 The one published figure is dpBestflow's, and it comes from photographers, not a standards body: quality 8 of 12 as the minimum for delivery, and 10 to 12 where it matters.² We based ours on measurement:²⁷

• Quality 10 is already below what you can see. We measured an average color error of 0.75 to 1.68 ΔE, about one just-noticeable difference or less. Quality 12 improves the measurement, not what you see. It usually makes a file 1.5 to 3 times larger, depending on the image and the program. It is still lossy, like every JPEG setting. Even at the highest quality in the standard JPEG library, where the quality setting discards nothing, 39 to 44% of pixels changed value in our tests.
• Color subsampling is the one setting you cannot make up for. Many encoders store color at half resolution (4:2:0). Detail lost that way is gone before the quality setting applies, so no quality setting can bring it back. We measured thin colored lines losing about 17 dB, with a worst-case error four times larger, and raising the quality afterwards recovered a tenth of a decibel. Saturated edges, signatures and fine linework suffer most, and red most of all. No institution publishes a view on subsampling, so this part of the case rests on measurement alone.
• Use Save As. Photoshop's four JPEG dialogs use different encoders. Adobe's own staff have said so, and have acknowledged that the newest dialog's quality scale is mapped wrongly.²⁶ In Save As, subsampling stops somewhere below quality 10, so quality 10 is safe there. Most other programs subsample at every setting unless you turn it off. Check yours.
• Adobe RGB, not sRGB, for a file you keep. sRGB cannot hold the most saturated cyans, greens and oranges that a modern pigment printer can reproduce. dpBestflow says sRGB is “not appropriate as a working space”,² and both FADGI's better master levels and NARA's regulation exclude sRGB for photographs.^1,7 Convert to sRGB for the web, in the copy.
• Embed the profile, always. An untagged file is read as sRGB; the web standard requires it.²⁴ So an Adobe RGB file that loses its profile will look dark and dull. Browsers do manage color now. The danger is whatever strips the profile on the way to them: a re-encoder, a platform, an email client.
• 8 bits is a real compromise, and we say so. It is invisible in output such as a web page or a large print. It is a real limit if the file is ever re-graded heavily, which is why the 16-bit working file stays on your disks.

Always make copies from the master

Re-saving a JPEG at the same quality and the same size costs almost nothing. Resizing between saves changes that. In our tests, ten generations saved at a single quality lost about 1.4 dB in total and then stopped changing. With a resize before each save, ten generations ended more than 30 dB worse than the same saves without it, and the loss had not levelled off.²⁷ Always make a new copy from the master, not from a JPEG you sent somewhere.

8. How big these files are

You can check the arithmetic yourself. A 16-bit RGB file takes 6 bytes for each pixel, and the image area of a 4 × 5 inch sheet of film is about 3.75 × 4.75 inches.

So a 4 × 5 color scan is roughly 400 MB to 1.7 GB, and an 8 × 10 is 4 GB and up. A 24-megapixel camera file saved as a JPEG at quality 10 comes to about 8 to 15 MB. The JPEG figures are ranges because the same setting produced files five times smaller per megapixel for a smooth image than for a detailed one.²⁷

A production master is not half the size of the scan. Cropping off the film edges removes 10 to 20% of the pixels, and correcting color removes none. What does halve a file is dropping it to 8 bits, or flattening a layered file.

9. Names, and the number that is not a name

Give every work an inventory number, and never change it. Keep it short and unique, with no punctuation. Never re-use it, and do not alter it once the work has left the studio. Give one even to work that is sold, given away or destroyed. Put it on the work as well as in your record. The Joan Mitchell Foundation's rule is the one to keep: “Never allow an artwork to leave your studio if it is not linked to a record in your inventory management system with a proper label that includes your Inventory Number.”⁸

Should the number mean anything? Sources disagree. FADGI says no: identifiers should be “not content-dependent”.¹ dpBestflow builds initials and a date into the name.² The Joan Mitchell Foundation changed its mind between editions. In 2013 its workbook built meaning into the number.⁹ In 2022 it recommends a plain number, “not representative of any attribute of the work”, starting at 1000.⁸ We agree with the 2022 advice. Put meaning in the record's fields, which can change, and keep the identifier fixed.

For file names, the sources agree on the rest:

• Give each file one name, and never let it have two. dpBestflow calls a second name for the same file “one of the very worst things you can do”.²
• Use only letters, digits, hyphens and underscores, in lowercase, with numbers padded to a fixed width so they sort in order.¹
• If a name encodes something, such as a date or a view, record that in the file's record as well.¹
• Name each copy after its master, with a suffix for what the copy is for.^1,2

And a test worth keeping from Artists' Studio Archives: “someone with limited knowledge of you and your work should be able to readily navigate through your files.”¹⁰

10. What to write down, and where it lives

The fields you record now are the fields a catalogue raisonné will need later. Across the artist-facing guides, the agreed minimum for a work is: inventory number, title, date, medium, dimensions (framed and unframed kept separate), at least one image, where the work lives and where it is now, its status, and its price or value.^8,10 The New York Public Library's list of what a catalogue raisonné entry holds maps almost exactly onto that record. The fields most worth adding are inscriptions, condition, other titles the work has had, and exhibition history.²² If you do not know something, write “unknown” rather than leaving the field blank, as the cataloguing guide Cataloging Cultural Objects asks.²¹

A work and a photograph of it are two things. The museum cataloguing standards separate them. VRA Core makes the work and the image separate records, and “a single image can relate to one or more works.”²⁰ Cataloging Cultural Objects warns that if the distinction is not drawn, “the results of a search can produce inaccuracies and confusion.”²¹ IPTC's photo metadata standard follows the same split. Its Artwork or Object fields describe the work shown in a photograph separately from the photograph itself, including the work's maker, date, materials and inventory number, and the work's own copyright.¹⁷

On rights, the photographers' standard suits an artist better than the museums' do. Cataloging Cultural Objects requires no rights field at all, because a museum keeps rights records separately. IPTC puts the copyright notice, the copyright owner and the credit line in its minimum.^17,21 A file you send out into the world needs them inside it.

In the file, beside it, or in a catalogue

• Sidecar files get separated from their images. dpBestflow calls them “a temporary solution”.² IPTC says such methods are “potentially exposed to metadata loss”.¹⁷
• Keep the record in a catalogue, and have the catalogue write it into the files it sends out. dpBestflow calls a catalogue “the best place to store the master copy of the metadata”.² The file carries the record out into the world.
• IPTC's minimum: caption, creator, copyright owner, copyright notice, credit line and date created. As IPTC puts it, “this is not a permission to strip off metadata.”¹⁷
• Not every format can carry the record. JPEG, TIFF and PSD carry both kinds of IPTC data. PNG, WebP, HEIC and AVIF carry only XMP, and a plain WebP file carries none. Camera raw files, in practice, carry none, which is why sidecar files exist.²³
• Metadata is most often lost when a new copy is made. dpBestflow: “The most common time for metadata to be destroyed is when you make a new derivative file.”² IPTC offers a free online checker that shows what a file still carries.¹⁹

11. What loupe does, said plainly

This paper is published by loupe, a home for an artist's work and its record. Here is where loupe fits into everything above, including the parts that are not flattering.

• loupe holds the file to send: the full-size JPEG of §7, made from your production master. It is not the place for either master, and it is not a backup. It should never be your only copy. The masters, the scan and the finished file made from it, are what §5's 3-2-1 is for. Read the sizes against §8: a 4 × 5 scanned at 2,040 ppi is about 445 MB as a 16-bit TIFF and 22 to 46 MB as the file to send; a 4,000 ppi drum scan is 1.7 GB and 88 to 182 MB. Every pixel is kept either way, so the file to send is still fit to zoom into and to print from. loupe takes files up to 300 MB, and will take a TIFF, but a TIFF leaves loupe with no record inside it. When a better file to send becomes worth making, loupe will say so, and ask for one made from your master; replace with new versions swaps it in and keeps the whole record. loupe also takes a camera raw as it comes, with its .xmp or the settings file another editing program writes beside it, kept as it came. It shows the camera's own picture when that is full size, and otherwise develops the raw as shot; a program's edits stay in that program, so for finished work the file to send is still the one to give it. A raw uploaded beside your finished picture is kept with it, but that copy does not make loupe the place for your master. Your camera raw files follows a raw through loupe step by step. Either way, your own copies are the ones that count.
• Your file is never changed. loupe makes its own copies for the page, a large picture, a thumbnail and zoom tiles, converted to sRGB from the file's own profile, and it records how it converted each one. When you download a JPEG, you get your original with its record written in; any other kind of file comes back exactly as you uploaded it.
• loupe's copies for the page are saved at a lower quality than this paper recommends, with the color subsampled. That is on purpose. They are screen-sized copies that nobody keeps, and at screen size the difference does not show. The advice in this paper is for the file you keep and send out; your original is never re-encoded, and nothing you uploaded as a finished file passes through loupe's encoder on its way to a printer. A camera raw, which no printer takes, goes out as a full-size JPEG that loupe makes at high quality, with its color kept whole.
• Whatever you type into loupe's record goes inside the files it sends out. That includes the work's own record in IPTC's Artwork or Object fields, kept separate from the photograph's.

If your catalogue is already in loupe, you can check it against this paper today. Search your photographs for profile:untagged to find the files that arrived with no color profile, or profile:srgb to find the ones that arrived in sRGB.

12. What we could not settle

We are listing these so nobody mistakes a guess for a finding.

• The exact point at which Photoshop's Save As stops subsampling. Three sources put it between quality 6 and 7. All three appear to trace back to one analysis that can no longer be read, and nobody has published a current test at each setting.²⁵ Our advice of quality 10 does not depend on the answer.
• How Photoshop's quality scale relates to other programs'. The encoders are built differently, so there cannot be an exact conversion. The only independent mapping is nearly twenty years old.²⁵
• A same-negative comparison of scanning methods. No one has published measurements from putting one negative through a camera, a drum scanner and a film scanner. The best-equipped testers on each side sell one of the methods.
• How large an artist's archive gets. No source publishes a distribution or a benchmark. The only worked example we found is 65.9 GB across 29,254 files for four years of one artist's work, and it was offered as an illustration, not as research.¹⁰
• Whether artists should publish wide-gamut color on the web yet. Browsers can display it. Whether the profile survives the trip from artist to viewer has not been established.
• Which current films have acetate bases and which polyester. We could not confirm this for current stocks by format, which is why §4 says to test the film.

If you know the answer to any of these, or find something here that is wrong, we want to hear. A paper that takes positions should expect to be corrected.

Sources

Quotations are given in full on this page, so you do not have to follow a link to check what a source said. Some sources are dated, and some have gone offline. Where that matters, we say so. The measurements marked as ours are described in source 27.

1. Federal Agencies Digital Guidelines Initiative (FADGI), Technical Guidelines for Digitizing Cultural Heritage Materials, 3rd edition, May 2023 — digitizationguidelines.gov. Master and production master: §2.5 and the glossary. Principles and code of ethics: front matter. Retouching and dust: §3.1, §3.8. Correcting a master: §6.2. Negatives: §3.11. Born-digital outside scope: §2.7. Lossless versus JPEG 2000 at 20:1: §2.5.1, §2.5.4. Color spaces and bit depth: §2.4 and the material tables. Naming: ch. 9.
2. American Society of Media Photographers, dpBestflow — Digital Photography Best Practices and Workflow, funded by the Library of Congress through NDIIPP, 2007–2010. ⚠ The site was offline when checked in September 2026, and its pages were last updated September 2015, so every sentence used here is quoted in full. The printed form of the project: Russotti & Anderson, Digital Photography Best Practices and Workflow Handbook, Focal Press, 2009, ISBN 978-0-240-81095-9.
3. Metamorfoze (National Library of the Netherlands), Preservation Imaging Guidelines — Image Quality, version 2.0, April 2025 — metamorfoze.nl. Introduction; §1.5; §1.6.
4. Library of Congress, Recommended Formats Statement 2025–2026, §II, Photographs — Digital — loc.gov.
5. Library of Congress, Sustainability of Digital Formats: Digital Negative (DNG) — loc.gov.
6. The National Archives (UK), Digitisation at The National Archives — nationalarchives.gov.uk.
7. US National Archives and Records Administration, 36 CFR Part 1236 Subpart E, effective 5 June 2023 — law.cornell.edu.
8. Joan Mitchell Foundation, Career Documentation for the Visual Artist, 2022, ch. 7 — joanmitchellfoundation.org.
9. Joan Mitchell Foundation, Creating a Living Legacy (CALL) Workbook, December 2013 — joanmitchellfoundation.org.
10. Artists' Studio Archives (University of North Carolina at Chapel Hill; Institute of Museum and Library Services), workbook and 2022 handouts — artiststudioarchives.org.
11. US National Park Service, Conserve O Gram 14/10, Table 2, “Average Film Deterioration Rates”, after the Image Permanence Institute — nps.gov.
12. Image Permanence Institute, Rochester Institute of Technology, Media Storage Quick Reference — rit.edu.
13. Henry Wilhelm with Carol Brower, The Permanence and Care of Color Photographs, 1993, ch. 5, compiling manufacturers' published predictions — wilhelm-research.com.
14. Northeast Document Conservation Center, A Short Guide to Film Base Photographic Materials, leaflet 5.1 — nedcc.org.
15. US Cybersecurity and Infrastructure Security Agency, Data Backup Options, citing Peter Krogh, The DAM Book, 2nd edition, p. 207 — cisa.gov.
16. Library of Congress, Personal Archiving: Digital Photos — digitalpreservation.gov.
17. IPTC, Photo Metadata Standard 2025.1 (including the Artwork or Object structure) and the Photo Metadata User Guide — iptc.org, user guide.
18. IPTC, Social Media Sites Photo Metadata Test Results, 2019 — iptc.org.
19. IPTC, Get IPTC Photo Metadata — getpmd.iptc.org.
20. Visual Resources Association, VRA Core 4.0, hosted by the Library of Congress — loc.gov.
21. Visual Resources Association and American Library Association, Cataloging Cultural Objects, 2006 — vraweb.org.
22. New York Public Library, Wallach Division, Catalogue Raisonné — nypl.org.
23. Adobe, XMP Specification Part 3: Storage in Files — adobe.com.
24. W3C, CSS Color Module Level 4, §3.5 — w3.org.
25. Raphaël Quinet, “Mapping JPEG compression levels between Adobe Photoshop and GIMP”, 2007 — blogs.gnome.org. The only independent mapping we found, and the first of three sources for where subsampling stops.
26. Adobe Community, “Inconsistent JPEG quality with 1–7 slider in Export As”, with Adobe's reply — community.adobe.com.
27. Measured for this paper, September 2026, with libjpeg-turbo: on the Kodak PhotoCD test set (color error, file size per megapixel, generational loss) and on six of the author's own photographs (TIFF compression, the effect of the highest quality setting). Photoshop's quality 10 and 12 were stood in for by libjpeg quality 93 and 98, matched by file size, because Photoshop's own encoder cannot be scripted for a test like this. Figures given as ranges vary with the image.
28. Hasselblad, Flextight X5 datasheet — hasselblad.com; resolution measured independently at filmscanner.info. Hasselblad withdrew the scanners around 2019 without a formal announcement.
29. Filmscanner.info, resolution measurements against a USAF-1951 target — filmscanner.info. The comparison with camera scanning is put together from measurements that were never taken side by side; see §12.

loupe · What to keep · September 2026