Cull operational/diagnostic events from desktop analytics (~9% of desktop-era volume)

[mdmohsin7]

Context

Original scope was just the Update Check Started/Not Found/Failed triplet with a ~22% of desktop / ~$140/month claim sourced from short PostHog/Mixpanel windows. Pulled the full 2-year Mixpanel raw-event export (92,242,888 events, May 2024 → May 2026) and re-counted against the actual bundle. Numbers below replace the originals.

The corrected picture: the Update Check triplet is real but smaller than claimed, and it's only one of ~10 operational/diagnostic events firing at high cadence from desktop. Total cull is ~8.66% of the April-2026 desktop-era month, not 22%.

Two scopes shown — most of these events shipped recently on desktop, so lifetime % undersells the current noise; April 2026 is what "going forward" would look like.

Counts (from the export bundle)

Event	Lifetime	%92M	April 2026	%20M	Code reasoning
Update Check Started	584,698	0.63%	427,441	2.08%	Sparkle SUScheduledCheckInterval=600 → fires every 10 min while app is running. Operational. Update Available / Update Installed carry the actual signal.
Update Not Found	467,464	0.51%	338,031	1.65%	Pairs with each Started call. Same.
Update Check Failed	89,299	0.10%	68,841	0.34%	Sparkle network failures. Sentry territory.
App Became Active	377,493	0.41%	259,817	1.27%	NSApplication.didBecomeActiveNotification — fires on every cmd-tab back. Pure focus noise.
App Resigned Active	354,490	0.38%	245,560	1.20%	NSApplication.willResignActiveNotification — every cmd-tab away. Same.
Display Info	85,896	0.09%	60,725	0.30%	Screen-dimensions diagnostic. No product signal.
Launch At Login Status	85,238	0.09%	60,634	0.30%	Polled config-state read, not a user action.
Settings State	226,471	0.25%	171,557	0.84%	Periodic 3-boolean settings snapshot. State, not action.
App Startup Timing	69,739	0.08%	49,187	0.24%	Cold-start timing diagnostic. Belongs in perf telemetry, not analytics.
Notification Settings Checked	133,324	0.14%	93,035	0.45%	Gated on change in AppState.swift, but still fires often. State, not action.

Keep: All Settings State (29,355 / 21,622) — code comment says "fired on app launch, at most once per day". Bounded, gives a daily snapshot worth having.

Totals

Scope	Total events	Cull volume	% cull
Lifetime (2 yr)	92,242,888	~2,473,612	2.68%
April 2026 alone	20,514,441	~1,775,267	8.66%

April 2026 is the desktop-on-Mixpanel month before the Apr 29 PostHog cutover (PR #7086). It's the most representative window for "what desktop noise looks like at steady state."

Recommended cull list

1. Update Check Started, Update Not Found, Update Check Failed — drop all three. Keep Update Available / Update Installed only. Originating call sites: desktop/Desktop/Sources/UpdaterViewModel.swift:47,88,129 → AnalyticsManager.updateCheck* → PostHogManager.swift:611-639.
2. App Became Active, App Resigned Active — drop entirely; not actionable. Call sites: desktop/Desktop/Sources/OmiApp.swift:1254,1260 → PostHogManager.swift:335,339.
3. Display Info, Launch At Login Status, Settings State — drop, or fold into the existing daily All Settings State snapshot. Call sites: OmiApp.swift:439, AnalyticsManager.swift:956, AnalyticsManager.swift:719.
4. App Startup Timing — move to perf telemetry (Sentry) instead of PostHog analytics. Call site: AnalyticsManager.swift:268.
5. Notification Settings Checked — keep only the change-detected case; AppState.swift:805-820 already gates it. The 93k/month is presumably already real changes, but worth a property-distribution check before deciding.

What we'd lose

• Per-poll telemetry on the updater, focus changes, settings reads, screen geometry. All recoverable from local logs / Sentry / server-side state if ever needed.

Owner

Desktop. ~30 lines deleted across PostHogManager.swift, AnalyticsManager.swift, UpdaterViewModel.swift, OmiApp.swift. Single PR.

Methodology

Counts come from data.mixpanel.com/api/2.0/export for project 3314908 (Based Hardware), windowed 2024-05-01 → 2026-05-05, exported via scripts/migration/mixpanel_export.sh (175 daily/monthly chunks, 0 failures, 4.18 GB gz, 92,242,888 lines). Per-event counts via zgrep -c '\"event\":\"<NAME>\"' chunks/*.jsonl.gz.

Refs

• PostHog project omi-macos (302298)
• Mixpanel Based Hardware (3314908)
• Desktop PostHog cutover: PR Remove Mixpanel from desktop app #7086 (2026-04-29)
• Flutter PostHog cutover: PR feat(app): migrate Flutter analytics from Mixpanel to PostHog with event parity #7111 (just merged)