Daily active (DAU)
Attempts started / day
New accounts / day
Weekly active (WAU)
Monthly active (MAU)
Funnel — registration → payment
Language (attempts)
Gender (warm-up)
Age range
Occupation
Taking it for whom
Sign-in doors
Registered via
Repeat usage
“Active” = answered ≥1 test item (page views aren’t tracked). DAU/WAU/MAU always cover the
last 1/7/30 days regardless of the window above. Sign-in doors count linked identities, not people —
one person can have several.
Revenue / day (so'm)
Refunds & extras
Provider × status
Declines by provider code
Checkout states (window)
Comeback offer (10-min discount)
Latest 15 checkout attempts
Promo sales (window)
Promo code inventory
Promo codes
Revenue counts real provider transactions only (Atmos card rail + Click); the old free
placeholder unlocks are shown as “test unlocks” and never as money. Reversed = refunded — those rows leave revenue
automatically. Promo: 100% codes unlock for free (never counted as revenue); smaller discounts are real card/Click
payments at the reduced price, so they appear in revenue at what was actually paid. The code table always shows
every code regardless of the date filter. Times shown in Tashkent (UTC+5).
Gemini totals — estimated
By model
AI est. cost / day (USD)
SMS totals — actual
SMS by template & delivery
Delivery verdicts
SMS cost / day (so'm)
AI costs are estimates: exact token counts × the configured price table (thinking tokens
billed as output). SMS costs are the actual so'm Eskiz charged, reconciled after each send — fresh sends are
legitimately unpriced for a few minutes. SMS rows aren’t linked to attempts, so demographic filters never apply to them.
Why people ask for money back
Transfer state
Requests + feedback (latest 100 in window)
A request auto-accepts for the user (3-day promise); the money moves by hand
(scripts/refund_payment.py). waiting transfer is the to-do list — it clears when the payment
row flips to reversed. The two feedback columns are the product's rawest signal; read them whole.
Answering activity by hour — Tashkent time (UTC+5)
0006121823
Pace
Where abandoners stop (items answered)
Medians are computed over the newest 2000 attempts. “Abandoners” = attempts that never produced
a result; the buckets show how many items they answered before walking away.
Result quality
Quality flags seen
AI pipeline
Top result titles
Top talents (scales)
Top suggested careers
Top historical figures
Title/talent/career/figure breakdowns scan the newest 2000 results. “Skipped” stories belong to
INVALID-tier results, which never get a narrative by design.
Checks
Health checks look at right now — they ignore the date window and any filters on purpose.
Hit Refresh to re-check. If “AI stories delivered” goes red, those users paid and are waiting: that’s the one to act on first.