Before you arrive

• Order the boxes in advance, always; same-day retrieval is a myth in about
  half the places I have worked
• Read the finding aid on the train, not at the desk
• Charge everything, because the plug sockets are theoretical

The first hour

Do not read. Survey. Open every folder in the first box, note what is in it,
close it, move on. You are building an index of the day.

The second to fourth hours

Now read, in the order the survey suggested, which is almost never the order
the boxes are in.

Photograph everything you touch

1. A shot of the folder label, first, every single time
2. Then the item
3. Then, if it is more than one page, a shot of your own hand holding up the
   page number

Step one is the one people skip, and it is the one that makes the photos
usable six months later. A folder of 400 undated images with no context is not
research, it is a burden you handed to yourself.

The last hour

Stop reading at three. Spend the last hour writing down what you found, in
prose, while the boxes are still in front of you.

The note I write at the end of every day

Three lines: what I found, what I expected and did not find, and the one box I
would open first if I came back tomorrow. That third line has paid for itself
more often than the other two together.

What drives the choice

• Demand: the collections people already ask for get scanned first, which
  makes them get asked for more
• Condition: too fragile to handle is a reason to scan and a reason not to
• Rights: clear copyright is a green light, uncertain copyright is a wall
• Funding: a grant names a collection, and that collection gets scanned
  whether or not it was next

The compounding effect

A researcher searching from home sees the scanned tenth. They cite it. The
citations become evidence of demand. The demand justifies the next round of
scanning, on the same tenth.

This is not a conspiracy, it is a ratchet, and it runs on its own.

What a good project publishes alongside

1. The selection criteria, in writing
2. What was excluded and why
3. The capture specification, so a later project knows whether to rescan
4. A stable identifier scheme, so the links survive the next migration

That last one is where most projects fail. A URL like
https://archive.example/viewer?id=4471&session=99 is not an identifier, it
is a coincidence, and it will not resolve in five years.

The one that works

Persistent identifiers, resolved centrally. The Handle System
and DOIs both do this, and both work because the resolution is someone’s job
rather than a side effect of a web server’s routing table.

What I want in every viewer

A visible, copyable citation string. That is it. Half the linkrot in my
bibliography exists because I copied what was in the address bar.

Reading a gap

A chain that runs 1890, 1912, then jumps to 1958 has a gap that covers the
years that matter most in European collections, and everyone in the field
knows it. The Washington Principles exist for
precisely that shape of gap.

What the gap can mean

1. The dealer’s records burned, which happened constantly
2. The owner was private and asked to stay private
3. Nobody wrote it down, because until the 1990s nobody had to
4. Something happened that someone did not want written down

Only the fourth is the one people mean. The other three are more common.

The research path

• Auction catalogues, which are often the only dated evidence
• Dealer stock books, where they survive
• Exhibition catalogues, which place an object in a room on a date
• Insurance schedules, the most reliable and the hardest to see

Links out of a post like this

Every outbound link in this post carries rel="nofollow", which is a small
statement: I am pointing at this, not vouching for it. In a field where
pointing at a dealer’s site is not an endorsement of the dealer, that
distinction earns its keep.

An internal link is different. The catalogue is not the collection
is mine, and pointing at it is vouching for it.

The honest conclusion

Most gaps are boring. The discipline is treating every one as a question until
it is answered, without pretending a question is an accusation.

What gets lost

• Anything the cataloguer did not think was the point
• Anything in a language they did not read
• Everything on the back of everything

Why the search fails

You search for a word. The record contains a description of the word. If the
cataloguer wrote correspondence and you searched for letters, the
collection is invisible to you and neither of you did anything wrong.

The fix is controlled vocabulary, which is why
Library of Congress Subject Headings
exist and why nobody enjoys them.

Searching around the record

1. Find one item you know is in the collection
2. Look at the subject headings on its record
3. Search those headings, not your own words
4. Repeat with whatever new headings turn up

This is citation chaining pointed at metadata instead of bibliography.

The underscore problem

Archive systems love identifiers like MS_1904_series_b and
coll_finding_aid_v2. Written in running prose without code formatting, a
naive emphasis rule eats the middle: MS_1904_series_b would come out with
1904 in italics if underscores delimited emphasis anywhere. They do not,
only at word boundaries, which is why this is emphasis and that is not.

What to do about it

Write to the archivist. Every collection has one, most of them answer, and a
five-line email saves a wasted day in a reading room three hundred miles away.

The four sections

Every aid has the same skeleton, whatever the institution calls it. The
Society of American Archivists glossary
is the reference I keep open, and the
ISAD(G) standard is the reason the sections
are the same everywhere.

Note that the label there contains brackets of its own. A link label is
matched by counting brackets, not by finding the first ], which is exactly
what a title like ISAD(G) or [Sic] needs.

Scope and content

The paragraph that tells you what is actually in the collection, as opposed
to what the collection is named after. Read this twice.

Arrangement

Series, subseries, boxes, folders. The unit you can request is usually the
box, so the folder list is a preview, not a menu.

Restrictions

The section people skip. It is where you learn that the interesting series is
closed until 2041.

Citing what you find

The house style at most reading rooms is collection, series, box, folder,
institution. A citation marker like [1] in running prose is not a link and
must never become one; it is a bracket, a digit and a bracket, and any
renderer that turns it into an anchor has broken the citation.

For the full form see the Chicago manual’s archives
chapter, which is behind a paywall
and worth exactly what your institution pays for it.