🎛 Interactive demo — every number on this dashboard is synthetic. This is a faithful copy of the production analytics dashboard behind Lispr, a voice-dictation app.
Lisprdemo
…
The 5-second view
Growth · trackable 1.5+ (see coverage above)
Active · new vs returning (daily)
A rising "returning" line means retention is real. "New" in the early days also includes people who merely upgraded to 1.5+ (became trackable), not only genuinely new users. Today (an incomplete day) is hidden so the tail is not misread as a drop.
Retention · cohorts (rolling)
% surviving to day N. Grey/dimmed = the cohort has not lived that long yet OR the eligible sample is tiny (n<5) — do not trust those. The "n" in a cell = installs that lived to day N (not the cohort size in the header).
DAU over time (unique trackable 1.5+ per bucket)
Audience · who and on what
Platform (mac / win)
"~" = inferred from the version scheme (client not tagged yet); an explicit tag overrides the guess. This card always shows the full split — it is the source for the toggle.
Dictation languages
Countries
Activity & health · why the line moved
Requests over time · by type
End-to-end (as the user feels it) · p50/p90/p99
"Stopped speaking → text at the cursor" (beacon 1.6+). The gap against server latency is the client's share (network + insertion).
Errors over time · % and absolute
Engineering / diagnostics
Transcription latency (server) · p50/p90/p99
Latency breakdown · by relay (avg, ms)
Versions · platform+volume
"~" = platform inferred from the version scheme (client not tagged yet).
A passive ledger kept by the edge service, NOT a poll of the provider. TPD (200k/day/org) is not exposed in headers — it is an "est." from the prompt+max_tokens reservation, anchored on a real 429 and recomputed to "now" including refill (~2.31 tok/s). The pool fills up as traffic flows (format+translate). Transcription is not counted here (it has its own audio quota).
Corpus browser
Pick a record on the left
Per-install drill-down (install_id)
or pick an active install from the list below ↓
Pick an install
Competitor releases & news
…
Sources and channels (health)
The primary signal is desktop versions (Mac/Windows), read straight from the auto-update feeds every desktop app has to publish: Sparkle appcasts (which usually carry the release notes too), Squirrel and electron-updater manifests, and GitHub releases for the open-source ones. The mobile App Store is a secondary signal (tagged "iOS"). Nobody ships a desktop build through an app store, so the update feed is the only reliable version source. Launch-platform and social channels need their own API tokens and are wired in per source.
Wake-word spotter · field telemetry · beta channel · 14 days
Spotter recordings
—
Flips
—
False candidates
—
flip without the name in the text
Misses
—
name was there — no flip
Opt-in prompts
—
shown → accepted
Time to fire
—
p50 / p90 · avg score
Onboarding · first-run funnel
Onboarding events from the first-run wizard (mac 2.0+, win 1.7+). "Shown / completed" is an
estimate of runs (SUM(_sample_interval): Analytics Engine samples even at small volumes, and exact DISTINCT counters undercount; at small N the numbers are noisy and a step can show done>shown); the large number is first runs, the grey one is all runs (incl. the Setup Guide
opened from the menu). Skipped steps (permissions already granted) never enter the funnel. "Avg time" covers
first runs only; steps longer than 10 min (an abandoned window) are excluded from the average. Window: 30 days.
Loading…
Feedback · user issues and ideas · the "Send feedback" window in the clients