blushneybusiness-glitch/roblox-omniskills/rbx-analytics-reading

rbx-analytics-reading

When to load

Any dashboard interpretation, any "what do I change" question after numbers exist, before any ad spend decision, and automatically under rbx-retention-cadence and rbx-wf-weekly — read the curve before prescribing. If the user can't access the dashboard this session, emit a fetch prompt via rbx-handoff-dashboard rather than guessing.

Green targets

Roblox colors your metrics against similar-genre games; these are the working bars when the dashboard's own comparison is unavailable. Benchmarks drift — for a high-stakes call, have rbx-psych-research verify current values.

• Session length is the most genre-relative number here: simulators/RPGs run healthy at

15–25 min; obby/minigame burst genres run short sessions and compensate with sessions-per-user-per-day — judge the pair, never session length alone.

• First-session length is an acquisition metric, not just retention: qualified

play-throughs feed the discovery algorithm, so a weak first session throttles home recommendations before it ever shows up in D1.

• The gate before ad spend: D1 ≥ 20% and median session ≥ 8 min. Paying for traffic into

a red-D1 game buys churn.

Symptom → change

A red metric indicts the checklist that owns its timeframe. Fix where the timeframe lives.

Reading rules

• Compare within genre, never globally. A horror game's D30 looks terrible next to a

simulator's by the nature of the loop, not by quality.

• Cohort before celebrating. Post-update numbers mix veterans with new players; filter

to the new-user cohort acquired after the ship date for any FTUE claim. Retention-tab cohort curves show updates as breaks at their date — that break is the honest signal.

• Sample honesty: under ~1,000 new users/day the daily numbers are noise; read weekly

aggregates and never redesign off a Tuesday dip.

• Two-timeframe rule: pushing D1, D7, and D30 green simultaneously usually means

compromising all three. Pick the two the genre supports; keep the third above its floor.

• Every change names its metric before shipping — target metric + expected direction,

reviewed after 7 days of cohort data. No hypothesis = it's a vibe; label it as one.

Sharp edges

• CRITICAL — Cohort mixing: judging an FTUE update by all-player D1 buries the effect

under veterans who never saw it. Symptom: "the update did nothing." Fix: new-user cohort filtered by acquisition date, 7-day window.

• HIGH — Acting on the dashboard's 24-hour-delayed charts as if they were live: a fix

shipped this morning cannot show today. Use "View Events" for did-it-fire checks and the dashboard only for trends.

• HIGH — Ad spend before the D1 ≥ 20% / session ≥ 8 min gate: sponsored traffic

converts worse than organic, so red metrics get redder while the budget burns.

• MEDIUM — Reading averages where distributions matter: mean session length hides the

bimodal split between bounced players (< 1 min) and engaged ones; a rising average can mean losing the casual half, not improving the game.

Load with

rbx-analytics-events (the numbers exist because it ran), rbx-analytics-funnels (when the symptom localizes to a flow), rbx-retention-cadence + rbx-retention-daily (the D7/D30 fixes), rbx-wf-firstsession (the D1 fix), rbx-handoff-dashboard (when the user must fetch auth-gated charts).