Comfy-Org / Comfy-Org/Comfy-Desktop
Desktop->browser handoff telemetry needs a join key, or #1358/#14457 still cannot produce a rate
- Dominant language
- TypeScript
- Stars
- 458
- Forks
- 59
- Avg merge
- 22h 18m
- Merged PRs (30d)
- 45
Description
#1358 (Desktop) and Comfy-Org/ComfyUI_frontend#14457 (frontend) add events on both sides of the Desktop -> browser handoff. As drafted they will not produce a usable rate, so the gap they were written to close stays open.
## Problem
The arrival event carries no properties, and nothing correlates the two halves. So the handoff count and the arrival count cannot be divided by each other: there is no way to tell whether a given arrival corresponds to a given handoff, and the two numbers come from different populations (one Desktop-side via `posthog-node`, one browser-side via `web`).
We would land two events and still not be able to answer the actual question, which is what fraction of browsers that Desktop opens ever arrive at the redemption page.
## Why it matters
Today 41% of people who start a Desktop sign-in end up signed in nowhere, and we cannot subdivide that number at all. There is no event of any kind between "Desktop called `openExternal`" and "the code was redeemed", so a browser that never opened is indistinguishable from a user who walked away, which is indistinguishable from a redeem that failed silently.
That black box is why the 8 August regression ran for two weeks without anyone being able to localise it from telemetry alone.
## Suggested fix
Either:
1. Thread an attempt id through both sides. Desktop generates it at code creation, includes it on `sign_in_started` and in the URL it opens, and the frontend echoes it on the arrival event. Straightforward but requires both PRs to move together.
2. Or emit arrival server-side from the redeem endpoint, which already receives the installation id at code creation and can tie the two together without a frontend change at all.
Option 2 is less coupling and less client code, and it also survives the user closing the tab before any browser-side event flushes. Worth considering before landing the two PRs as drafted.
Whichever is chosen, the acceptance criterion is that we can compute "browsers opened -> browsers arrived -> codes redeemed" as three numbers over the same population.
## Related
- #1358 Make Desktop browser handoff observable
- Comfy-Org/ComfyUI_frontend#14457 Track Desktop login browser arrival
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.