292 lines
7.8 KiB
Markdown
292 lines
7.8 KiB
Markdown
# A/B-Tests
|
|
|
|
When YouTube introduces a new feature, it does so gradually. When a user creates a new
|
|
session, YouTube decided randomly which new features should be enabled.
|
|
|
|
YouTube sessions are identified by the visitor data cookie. This cookie is sent with every
|
|
API request using the `context.client.visitor_data` JSON parameter. It is also returned in the
|
|
`responseContext.visitorData` response parameter and stored as the `__SECURE-YEC` cookie.
|
|
|
|
By sending the same visitor data cookie, A/B tests can be reproduced, which is important for testing
|
|
alternative YouTube clients.
|
|
|
|
This page lists all A/B tests that were encountered while maintaining the RustyPipe client.
|
|
|
|
**Impact rating:**
|
|
|
|
The impact ratings shows how much effort it takes to adapt alternative YouTube clients to the
|
|
new feature.
|
|
|
|
- 🟢 **Low** Minor incompatibility (e.g. parameter name change)
|
|
- 🟡 **Medium** Extensive changes to the response data model OR removal of parameters
|
|
- 🔴 **High** Changes to the functionality of YouTube that will require API changes
|
|
for alternative clients
|
|
|
|
If you want to check how often these A/B tests occur, you can use the `codegen` tool with the
|
|
following command: `rustypipe-codegen ab-test <id>`.
|
|
|
|
## [1] Attributed text description
|
|
|
|
- **Encountered on:** 24.09.2022
|
|
- **Impact:** 🟡 Medium
|
|
- **Endpoint:** next (video details)
|
|
|
|
|
|
|
|
YouTube shows internal links (channels, videos, playlists) in the video description
|
|
as buttons with the YouTube icon. To accomplish this, they completely changed the underlying
|
|
data model.
|
|
|
|
The new format uses a string with the entire plaintext content along with a list of `"commandRuns"`
|
|
which include the link data and the position of the links within the text.
|
|
|
|
Note that the position and length parameter refer to the number of UTF-16 characters. If
|
|
you are implementing this in a language which does not use UTF-16 as its internal string
|
|
representation, you have to iterate over the unicode codepoints and keep track of the UTF-16
|
|
index seperately.
|
|
|
|
**OLD**
|
|
|
|
```json
|
|
{
|
|
"videoSecondaryInfoRenderer": {
|
|
"description": {
|
|
"runs": [
|
|
{
|
|
"text": "🎧Listen and download aespa's debut single \"Black Mamba\": "
|
|
},
|
|
{
|
|
"navigationEndpoint": {
|
|
"commandMetadata": {
|
|
"webCommandMetadata": {
|
|
"rootVe": 83769,
|
|
"url": "https://www.youtube.com/redirect?...",
|
|
"webPageType": "WEB_PAGE_TYPE_UNKNOWN"
|
|
}
|
|
},
|
|
"urlEndpoint": {
|
|
"nofollow": true,
|
|
"target": "TARGET_NEW_WINDOW",
|
|
"url": "https://www.youtube.com/redirect?..."
|
|
}
|
|
},
|
|
"text": "https://smarturl.it/aespa_BlackMamba"
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
**NEW**
|
|
|
|
```json
|
|
{
|
|
"videoSecondaryInfoRenderer": {
|
|
"attributedDescription": {
|
|
"content": "🎧Listen and download aespa's debut single \"Black Mamba\": https://smarturl.it/aespa_BlackMamba\n🐍The Debut Stage...",
|
|
"commandRuns": [
|
|
{
|
|
"startIndex": 58,
|
|
"length": 36,
|
|
"onTap": {
|
|
"innertubeCommand": {
|
|
"commandMetadata": {
|
|
"webCommandMetadata": {
|
|
"url": "https://www.youtube.com/redirect?...",
|
|
"webPageType": "WEB_PAGE_TYPE_UNKNOWN",
|
|
"rootVe": 83769
|
|
}
|
|
},
|
|
"urlEndpoint": {
|
|
"url": "https://www.youtube.com/redirect?...",
|
|
"target": "TARGET_NEW_WINDOW",
|
|
"nofollow": true
|
|
}
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
## [2] 3-tab channel layout
|
|
|
|
- **Announced:** 15.09.2022, https://www.youtube.com/watch?v=czIyqEC4V-s
|
|
- **Encountered on:** 11.10.2022
|
|
- **Impact:** 🔴 High
|
|
- **Endpoint:** browse (channel videos)
|
|
|
|
|
|
|
|
YouTube changed their channel page layout, putting livestreams and short videos into
|
|
separate tabs.
|
|
|
|
Fetching the videos page now only returns a subset of a channel's videos. To get all videos
|
|
from a channel, you would have to run up to 3 queries.
|
|
|
|
Even though it has its disadvantages, the RSS feed is now probably the best way for keeping
|
|
track of a channel's new uploads.
|
|
|
|
Additionally the channel tab response model was slightly changed, now using a `"RichGridRenderer"`.
|
|
Short videos also have their own data models (`"reelItemRenderer"`).
|
|
|
|
**RichGrid**
|
|
|
|
```json
|
|
{
|
|
"tabRenderer": {
|
|
"content": {
|
|
"richGridRenderer": {
|
|
"contents": [
|
|
{
|
|
"richItemRenderer": {
|
|
"content": {
|
|
"videoRenderer": {}
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
**Short video**
|
|
|
|
```json
|
|
{
|
|
"reelItemRenderer": {
|
|
"accessibility": {
|
|
"accessibilityData": {
|
|
"label": "being smart was my personality trait - 56 seconds - play video"
|
|
}
|
|
},
|
|
"headline": {
|
|
"simpleText": "being smart was my personality trait"
|
|
},
|
|
"navigationEndpoint": {
|
|
"clickTrackingParams": "CLcCEIf2BBgAIhMImuP85t-D-wIVd-sRCB2r6gl7",
|
|
"commandMetadata": {
|
|
"webCommandMetadata": {
|
|
"rootVe": 37414,
|
|
"url": "/shorts/glyJWxp7a5g",
|
|
"webPageType": "WEB_PAGE_TYPE_SHORTS"
|
|
}
|
|
},
|
|
"reelWatchEndpoint": {
|
|
"overlay": {
|
|
"reelPlayerOverlayRenderer": {
|
|
"reelPlayerHeaderSupportedRenderers": {
|
|
"reelPlayerHeaderRenderer": {
|
|
"timestampText": {
|
|
"simpleText": "2 days ago"
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"thumbnail": {
|
|
"thumbnails": [
|
|
{
|
|
"height": 720,
|
|
"url": "https://i.ytimg.com/vi/glyJWxp7a5g/hq720_2.jpg?sqp=-oaymwEdCJUDENAFSFXyq4qpAw8IARUAAIhCcAHAAQbQAQE=&rs=AOn4CLCUzo9AlrNh4n4cZfTOB8_Gf5aAkw",
|
|
"width": 405
|
|
}
|
|
]
|
|
},
|
|
"videoId": "glyJWxp7a5g",
|
|
"viewCountText": {
|
|
"simpleText": "593K views"
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
## [3] Channel handles in search results
|
|
|
|
- **Encountered on:** 20.11.2022
|
|
- **Impact:** 🟡 Medium
|
|
- **Endpoint:** search
|
|
|
|
|
|
|
|
Instead of subscriber count / video count, a channel item from the search result now
|
|
displays the channel handle and the subscriber count. The video count was removed.
|
|
|
|
The implementation looks pretty quick and dirty, as they did not even bother to rename
|
|
their response parameters. So this might change again in the future.
|
|
|
|
Note that channels without handles still use the old data model, even on the same page.
|
|
|
|
**OLD**
|
|
|
|
```json
|
|
{
|
|
"subscriberCountText": {
|
|
"accessibility": {
|
|
"accessibilityData": {
|
|
"label": "2.92 million subscribers"
|
|
}
|
|
},
|
|
"simpleText": "2.92M subscribers"
|
|
},
|
|
"videoCountText": {
|
|
"runs": [
|
|
{
|
|
"text": "219"
|
|
},
|
|
{
|
|
"text": " videos"
|
|
}
|
|
]
|
|
}
|
|
}
|
|
```
|
|
|
|
**NEW**
|
|
|
|
```json
|
|
{
|
|
"videoCountText": {
|
|
"accessibility": {
|
|
"accessibilityData": {
|
|
"label": "4.03 million subscribers"
|
|
}
|
|
},
|
|
"simpleText": "4.03M subscribers"
|
|
},
|
|
"subscriberCountText": {
|
|
"simpleText": "@MusicTravelLove"
|
|
}
|
|
}
|
|
```
|
|
|
|
## [4] Video tab on the Trending page
|
|
|
|
- **Encountered on:** 1.04.2023
|
|
- **Impact:** 🟢 Low
|
|
- **Endpoint:** browse (trending videos)
|
|
|
|
YouTube moved the list of trending videos from the main *trending* page to a
|
|
separate tab (Videos).
|
|
|
|
The video tab is fetched with the params `4gIOGgxtb3N0X3BvcHVsYXI%3D`.
|
|
|
|
This new tab contains two shelves (video lists), the first one labeled "Trending videos"
|
|
which contains the regular trends and the second one named "Recently trending".
|
|
|
|
The data model for the video shelves did not change.
|
|
|
|
**OLD**
|
|
|
|
|
|
|
|
**NEW**
|
|
|
|
|