CTV device ID feed | Pixalate Docs

The Connected TV (CTV) Device ID Data Feed lists CTV device IDs linked to invalid traffic (IVT). Use it to avoid bidding on or serving ads to those devices.

The feed is tuned for CTV inventory. CTV generates lower volumes than mobile traffic, so Pixalate segments CTV signals separately to build CTV baselines for inclusion. The time horizon over which signals are observed, and the statistical significance thresholds, are also set for CTV.

What’s in the feed

IVT types. The feed covers every IVT type in the Pixalate dashboard except those that integration issues can trigger, such as duplicateImpressions and duplicateClicks. See Reported Invalid Traffic (IVT) Types. It also uses this CTV-specific type:

IVT type Definition
continuousPlay Devices that appear to be on for the vast majority of the day.
Proxy traffic. Proxy traffic in this feed means fake server-side ad insertion (SSAI) or other fraudulent proxy traffic. Pixalate uses its own framework to detect fake SSAI proxies, and adds only the devices linked to that activity.

IVT type	Definition
`continuousPlay`	Devices that appear to be on for the vast majority of the day.

At a glance


Update cadence	Hourly
File format	CSV
File name	`OttDeviceIdBlocklist_YYYYMMDD.csv`
Columns	`deviceId`, `fraudType`, `os`, `ifaType`, `deviceName`, `probability`
File size	Up to 2 GB

The file name carries a date only, not a time or version, and a newer file can reuse an earlier file’s name and replace it. Don’t use the file name to find the newest file; check its last-modified time instead. See Delivery and integration.

Schema

Column	Type	Description
`deviceId`	string	The CTV device ID linked to invalid traffic.
`fraudType`	string	The name of the IVT type linked to the device ID.
`os`	string	The operating system linked to the device ID.
`ifaType`	string	The type of identifier for advertising (IFA). See IFA types.
`deviceName`	string	The device brand or platform. See Device names.
`probability`	float	How likely it is that the device ID is linked to this fraud type, from 0.5 to 1. Higher means riskier.

Example rows (made-up IDs and values):

00000000-0000-4000-8000-00000000000a,continuousPlay,<os>,RIDA,Roku,1.0
00000000-0000-4000-8000-00000000000b,<fraudType>,<os>,AFAI,Amazon,0.84

IFA types

These ifaType values are in the feed:

`ifaType`	Description
`AAID`	Standard Android device identifier
`AAID_GENERIC`	Generic (non-standard) Android device identifier
`AAID_MD5`	MD5 Android device identifier
`AAID_SHA1`	SHA1 Android device identifier
`AFAI`	Standard Amazon device identifier
`AFAI_GENERIC`	Generic (non-standard) Amazon device identifier
`AFAI_MD5`	MD5 Amazon device identifier
`AFAI_SHA1`	SHA1 Amazon device identifier
`GENERIC`	Generic (non-standard) identifier that doesn’t belong to a major vendor and has no standard format
`IDFA`	Standard Apple device identifier
`IDFA_GENERIC`	Generic (non-standard) Apple device identifier
`IDFA_MD5`	MD5 Apple device identifier
`IDFA_SHA1`	SHA1 Apple device identifier
`IFA_GENERIC`	Generic identifier with the standard IFA format
`LGAID`	Standard LG device identifier
`LGAID_GENERIC`	Generic (non-standard) LG device identifier
`LGAID_SHA1`	SHA1 LG device identifier
`MD5_GENERIC`	Generic identifier with MD5 format
`MSAI`	Standard Microsoft device identifier
`MSAI_GENERIC`	Generic (non-standard) Microsoft device identifier
`MSAI_MD5`	MD5 Microsoft device identifier
`MSAI_SHA1`	SHA1 Microsoft device identifier
`RIDA`	Standard Roku device identifier
`RIDA_GENERIC`	Generic (non-standard) Roku device identifier
`RIDA_MD5`	MD5 Roku device identifier
`RIDA_SHA1`	SHA1 Roku device identifier
`SHA1_GENERIC`	Generic identifier with SHA1 format

Device names

The major deviceName values include, but aren’t limited to: Amazon, Apple, Bravia, Chromecast, HbbTV, Insignia, JadooTV, LG, Microsoft, Nvidia, Opera TV, Roku, Samsung, Sony, Toshiba and Vizio.

Choose a probability threshold

Every listed device ID has a probability of at least 0.5, calculated by Pixalate’s machine learning models. You decide where to set your threshold, based on the quality and scale of your supply. As a general guide:

Threshold	What you filter out
Probability = 1	Only the worst offenders (deterministic).
Probability ≥ 0.90	Users that are fraudulent beyond reasonable doubt.
0.75 ≤ probability < 0.90	Users with clear and convincing evidence of fraud.
0.5 ≤ probability < 0.75	Users that are more likely than not to be fraudulent (preponderance of the evidence).

Lowering the threshold can reduce the number of impressions you deliver. Each time you change it, check the effect on impression delivery.

Apply it pre-bid

Load the latest file on the servers that make your bid or serving decisions.
For each CTV request, look up the device ID.
If it’s listed at or above your threshold, don’t bid or serve.

Check for new files

The feed updates hourly. Poll at least once an hour, and use the file’s last-modified time, not its name, to tell when a new file has arrived. Hourly feeds follow this pattern: