Reading a container
Journals, books and series, the three pages that hold works rather than being one.
A work is one thing you can read. A container is the thing it was published in, and this site has three of them:
spr journal 10994 # a journal, by its Springer id
spr book 978-3-031-28170-9 # a book, by isbn
spr series 558 # a series, by its Springer id
Each is its own record with its own fields, because a journal has an impact factor and no price, a book has four ISBNs and no volumes, and a series has neither. One record with a kind field would have been mostly empty on every page it was used for.
The one thing to know first
A work page announces itself. Google Scholar reads the Highwire meta tags on an article, so citation_title, citation_doi and citation_author are there, correct, and quickly fixed when they are not.
A container page announces nothing. There is no citation_journal_title on a journal home page, no citation_isbn on a book page, and the schema.org block on a series is a breadcrumb list. Every field on these three pages comes from rung 3 or rung 4, which means Springer's own data-test region names or, failing that, a css class and a printed English label.
spr extraction --record journal prints exactly which, and the why column says why each row sits where it does:
$ spr extraction --record series
RECORD FIELD RUNG SOURCE WHY
series series_id region datalayer.Book Series Id arrives as a JSON number here and as a string on a journal, so the reader accepts both
series title region datalayer.Book Series Title the thinnest payload on the site states this and the id and nothing else
series editors region [data-test=editor-links-1] the same dl per role shape a journal uses, with Series Editor in place of Editor-in-Chief
series latest_titles region [data-test=latest-titles] .c-card five books out of many thousands, which is why the record names them latest rather than titles
series indexed_in region [data-test=series-abstract-and-index-services-list] the same claim a journal makes about itself, in the same words
Rung 3 is not a worse answer than rung 1. It is a less durable one, and the record says which you got so that a field breaking after a redesign is a thing you can see rather than a thing you discover months later.
A journal
$ spr journal 10994
title Machine Learning
id 10994
url https://link.springer.com/journal/10994
issn 1573-0565, 0885-6125
publisher Springer, Springer
model Hybrid Access
publishing continuous, articles appear before the issue closes
open access 664 articles
metric Journal Impact Factor 4.9 (2025)
metric 5-year Journal Impact Factor 8.0 (2025)
metric Downloads 2.4M (2025)
subjects Machine Learning, Control, Robotics, Automation, Artificial Intelligence, ...
editors (1)
1 Michelangelo Ceci Editor-in-Chief
indexed in (29)
ACM Digital Library, ANVUR, BFI List, Baidu, CLOCKSS, CNKI, CNPIEC, ...
volumes 0 held, more at https://link.springer.com/journal/10994/volumes-and-issues
Two ISSNs, electronic first, and they are kept apart rather than folded into one field, because a citation that gives you 0885-6125 and a record that only knows 1573-0565 do not match and should.
Every editor keeps their role. Editor-in-Chief, Associate Editor and Editorial Board Member are different jobs, and a flat list of names throws that away for no gain.
A metric is only emitted with the year it was measured in. The journal prints one line with no year, "Submission to first decision (median), 5 days", and it goes to missed with that sentence attached rather than being emitted as a metric that nothing can be compared against:
$ spr journal 10994 --envelope
...
missed metrics: the page printed "Submission to first decision (median), 5 days" with no year, and a metric with no year is not comparable with anything
2.4M is printed as 2.4M. The record carries 2400000 beside it, but the human form prints the publisher's own number, because that is the one you can go and check.
The volumes are a separate page
The last line of a journal is a pointer, not a list:
volumes 0 held, more at https://link.springer.com/journal/10994/volumes-and-issues
0 held and a url is a different fact from an empty list. It says the collection exists, this record does not have it, and here is where it is. --volumes makes the second request and fills it in:
$ spr journal 10994 --volumes
...
volumes 348 of 348 held
volumes (114), 348 issues
volume 115 January - August 2026
Issue 8 2026-08
Issue 7 2026-07
...
volume 105 January - December 2016
Issue 3 2016-12 Special Issue of the ECML PKDD 2016 Journal Track
114 volumes and 348 issues, 86 of which carry a themed collection title. The volume year comes from the volume's own printed span rather than from its issues, because volume 115 runs January to August 2026 and a volume that crosses a year boundary would otherwise take the year of whichever issue happened to be listed first.
A book
A book is addressable two ways and both work as given:
spr book 10.1007/978-3-031-28170-9 # by doi
spr book 978-3-031-28170-9 # by isbn
It is the richest page on the site, and it is read from four sources at once: eight meta names, three JSON-LD blocks, the analytics payload, and the printed bibliographic table at the foot of the page.
$ spr book 978-3-031-28170-9
title The Economics of Family Taxation. Optimal Tax Issues from a Household Economics Perspective
doi 10.1007/978-3-031-28170-9
kind book, Monograph
isbn ebook 978-3-031-28170-9
isbn print 978-3-031-28172-3
isbn hard 978-3-031-28169-3
isbn soft 978-3-031-28172-3
publisher Springer Cham
edition 1
pages XI, 102
series Population Economics 1431-6978
published 2023-04-26
hardcover 2023-04-27
softcover 2024-04-28
access not free to read, access=No
metrics 1169 accesses, 22 citations
prices
eBook EUR 85.59 ebook
Softcover Book EUR 99.99 book
Hardcover Book EUR 99.99 book
chapters (7 of 9 rows, front and back matter included)
Four ISBNs, four fields. One book has an electronic ISBN, a hardcover ISBN, a softcover ISBN and a print ISBN, and on this book the softcover and the print ISBN happen to be the same string while the hardcover is different. Collapsing them into one isbn field picks one edition and silently discards two, and which one you get depends on which source the reader happened to look at first. The doi resolves to the electronic edition, so that is the one JSON-LD names, and it is not the one the page prices most prominently.
Three publication dates. The ebook came out 2023-04-26, the hardcover the day after, and the softcover a year later. A book with one published field is a book you cannot cite the softcover of.
The price kind is not the printed label. eBook, Softcover Book and Hardcover Book are prose that changes with the locale. The kind column beside each price is the value the publisher's own order form posts to its cart, which is the stable one. Prices are localized by requesting IP, so what you see depends on where you fetch from, and the printed string is kept beside the parsed amount for that reason.
Front and back matter are rows. --chapters prints all nine rows and names the two that are not chapters:
$ spr book 10.1007/978-3-031-28170-9 --chapters
...
chapters (7 of 9 rows, front and back matter included)
i-xi Front Matter
1-14 Standard Optimal Taxation with Single Agents: What It Is and What to Use in Its Place
10.1007/978-3-031-28170-9_1
15-30 Optimal Taxation in the Presence of Household Production
10.1007/978-3-031-28170-9_2
...
99-102 Back Matter
Front matter has a page range and no DOI. That is why a chapter is its own type here rather than the generic reference a linked work usually gets, since a reference has no room for a page range and no way to say that a row is deliberately without an identifier.
A book behind a subscription is read rather than refused. Everything above comes off a page with access=No, and the exit code is 4 to say so.
A series
$ spr series 558
title Lecture Notes in Computer Science
id 558
issn 1611-3349, 0302-9743
editors (4)
1 Elisa Bertino Series Editor
2 Wen Gao Series Editor
3 Bernhard Steffen Series Editor
4 Moti Yung Series Editor
latest titles (5)
2027 Superposition for Higher-Order Logic
2027 Consolidated Ada 2022 Reference Manual. Volume 1 - Core Language
...
2027 Vibe Coding open access
titles 5 held, more at https://link.springer.com/series/558/books
The five books are the five the page shows, and LNCS has many thousands. The field is called latest_titles for that reason, and the pointer under it says 5 held and where the rest are, in the same shape a journal points at its volumes.
Each card credits either authors or editors, and the two are kept apart. The page's own itemprop says editor on both, so the printed Authors: or Editors: label is what this reads instead, because a monograph with one author and a proceedings volume with four editors are different books and the page is careful to say which.
Where a conference goes
There is no conference page on this site. /conference/aaai answers 404 with a zero byte body, so no conference url is ever constructed.
What exists is a conference named inside a proceedings title, and it is parsed out of that title only when the title actually says so:
Computer Vision - ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part I
name 16th European Conference
acronym ECCV
year 2020
The location is never read, even though it is sitting right there in that string, because "Glasgow, UK" and "Lecture Notes in Computer Science, Munich" are the same shape and only one of them is a place the conference happened. A title that names no conference produces no conference at all rather than a guess. A wrong conference year is worse than no conference year, since a reader has no way to tell it was invented.
The envelope, on every one of them
--envelope works the same on all three:
$ spr series 558 --envelope
...
envelope html, ok, 144018 bytes, 3 redirects, fetched 2026-08-18 10:57:40 UTC
8 fields answered, 0 missed, 24 regions unread
24 regions unread on a series, 43 on a journal and a book. Those are named components this tool saw and did not read, listed rather than dropped, so that the day one of them starts carrying something worth having it is already written down.
What comes next
A work has three subpages of its own and counts and assets covers them: how often it was read, its figures at full size, and the rows of its tables, which are not on the article page at all.