Skip to content

t1k:game-analytics

FieldValue
Modulet1k-insights
Version2.14.25
Effortmedium
ToolsGlob, Grep, Read

Keywords: ads per dau, build comparison, drop rate, event taxonomy, feature adoption, game analytics, kill criteria, level funnel, measurement pack

/t1k:game-analytics

Playbook for the questions a mobile-game studio asks about a release: which measurement pack answers it, which events the game must emit, and how to compare build N with build N-1 without fooling yourself. It never drives a dashboard. Tool driving goes to t1k:bytebrew (and the Firebase marketing skill when it is installed).

QuestionPack (references/pack-<name>.md)Where the read happens
Is build N as stable and fast as N-1? Crash-free, ANR, FPS, load timerelease-health (references/pack-release-health.md)Crashlytics for crashes and ANRs; performance events in the analytics tool
Do players play as long and come back? DAU, sessions, playtime, D1/D7engagement-retention (references/pack-engagement-retention.md)analytics (../t1k-bytebrew/references/analytics.md)
Does build N earn more per player? ARPDAU, LTV, payer conversionmonetization (references/pack-monetization.md)monetization (../t1k-bytebrew/references/monetization.md)
How many ads of each type per version, and how often do they fail?ads-by-type (references/pack-ads-by-type.md)custom-workspace (../t1k-bytebrew/references/custom-workspace.md) breakdowns
Where do players leave, and which level or day spikes?level-funnel (references/pack-level-funnel.md)custom-workspace (../t1k-bytebrew/references/custom-workspace.md) funnels
Is the new feature good: keep, iterate or kill?feature-evaluation (references/pack-feature-evaluation.md)live-ops (../t1k-bytebrew/references/live-ops.md) A/B tests and breakdowns
Does the first session convert; do sinks balance sources? (optional)onboarding-economy (references/pack-onboarding-economy.md)custom-workspace (../t1k-bytebrew/references/custom-workspace.md) funnels

| Dated progress report by game version and blueprint window: did the changes go the right way? | progress-report.md (references/progress-report.md) | t1k-bytebrew (../t1k-bytebrew/SKILL.md) (primary reads), then the Firebase blueprint timeline below | | Which blueprint_version was live when? (Firebase Remote Config history) | firebase-remote-config-history.md (references/firebase-remote-config-history.md) with scripts/rc-blueprint-history.cjs | Firebase CLI, pinned through bun | | Analytics counts from Firebase (GA4 Data API or BigQuery), only when the owner has chosen one | firebase-analytics-numbers.md (references/firebase-analytics-numbers.md) | GA4 Data API or the BigQuery export | | How to log in to the Firebase CLI without a localhost callback | firebase-login.md (references/firebase-login.md) | Firebase CLI |

Any question that matches no row: read references/event-taxonomy.md first, then references/version-comparison.md, and say which pack you chose and why.

  • Event names come from the taxonomy. Packs copy names from references/event-taxonomy.md and never define events.
  • The comparison method lives once. Every build comparison follows references/version-comparison.md. Packs add only their own deltas.
  • Report gaps, never substitute. A missing or renamed event is a gap: name it and the nearest variant, and never swap a near match in silently. Rule: t1k-bytebrew SKILL (../t1k-bytebrew/SKILL.md).
  • Not applicable is not a gap. A system the game does not have (no shop, no economy) reads n/a by design.
  • No tool driving here. This skill never opens a browser or dashboard. Delegate the read to t1k:bytebrew.
  • No product identifiers. Use placeholders <appId>, <bundleId>, <adUnit>, <gameName> in every output and file.
  • Invented numbers are labelled. Every threshold this skill sets reads default, calibrate per title. Vendor figures are context, not rules, and carry their URL and retrieval date.
  1. Pick the pack from the Decision Tree. If two rows match, run both, each with one primary metric named before you look.
  2. Audit the events against references/event-taxonomy.md. Check the SDK layer as well as game code. Map aliases (## Observed aliases), list every variant separately before any sum, and list each missing event under gaps:.
  3. Fix the cohort and the window per references/version-comparison.md: install cohorts, equal-length windows in UTC, at least 7 full days (default, calibrate per title), no release day, no partial day. The install-date filter equals the window in every query. Run the arm viability check before reading any rate.
  4. Delegate the read to t1k:bytebrew with the exact event names, the UTC date range and the build dimension. Quote a dashboard limit only from the live page.
  5. Give the verdict in the four-word vocabulary: improved, regressed, inconclusive or not-measurable. Write n per arm, the difference with its interval, the window, the direction and the gaps: line. A verdict without n is not a verdict.

scripts/rc-blueprint-history.cjs is mechanical: it lists Remote Config versions and extracts blueprint_version per version, with no judgment. The cheap-tier agent t1k-data-verifier runs it and returns its JSON rows unchanged. The invoking session chooses the window, applies the sample guards and writes the verdict in references/progress-report.md.

  • Event names, vendor bindings and limits: references/event-taxonomy.md (each vendor URL there is dated 2026-11-11).
  • Build comparison method and interval formulas: references/version-comparison.md.
  • Crash-free definition and the window caveat: https://firebase.google.com/docs/crashlytics/crash-free-metrics [retrieved 2026-11-11]
  • Tool driving: t1k:bytebrew (../t1k-bytebrew/SKILL.md).