Add managed content and keyboard usage insights
CI / verify (push) Has been cancelled
CI / publish (push) Has been cancelled

Introduce versioned official content workflows and privacy-safe keyboard analytics, while preventing repeat DeviceCheck sign-ins from incorrectly restricting eligible accounts.
This commit is contained in:
Rocky
2026-08-21 13:34:03 +08:00
parent b5212dcdc2
commit d0abe27623
55 changed files with 4690 additions and 111 deletions
+45
View File
@@ -69,6 +69,51 @@ Managed client success events are excluded from this total.
`successful AI requests / distinct value-active users` for the selected period.
## Keyboard input usage
Keyboard input metrics use finalized UTC-day summaries produced on-device.
They describe manually committed OSGKeyboard text only and are independent
from AI value events, billing and referral qualification.
### Keyboard input active users
Distinct account identities, falling back to pseudonymous installations, with
at least one accepted keyboard usage summary in the selected UTC-date window.
### Activation-to-input conversion
Keyboard input active users divided by distinct identities with either a
`KEYBOARD_ACTIVATED` event or an accepted keyboard usage summary in the same
UTC-date window. Including summary-only identities prevents missing activation
telemetry from producing rates above 100%.
### Chinese, English and bilingual active users
- Chinese active: at least one committed Han-script character.
- English active: at least one committed Latin letter.
- Bilingual active: both Chinese and English counts are non-zero.
These populations overlap and must not be summed.
### Character volume and language share
Character volume is the sum of client-classified Chinese, English and other
committed characters. Chinese and English share use only classified language
characters as the denominator:
- Chinese share: `Chinese / (Chinese + English)`.
- English share: `English / (Chinese + English)`.
Both shares are unavailable when the denominator is zero. Other characters
remain visible in total volume but do not dilute the language split.
### Input sessions
An input session is a keyboard activation containing at least one manually
committed character. Chinese-only, English-only, mixed-language and other-only
session counts form a complete partition. Average characters per input session
is `total committed characters / input sessions`.
## Retention
The cohort date is the UTC date of a user's first successful AI feature.