← All Dashboards
Sprint 65 Sprint 66 Sprint 67 Sprint 68 Sprint 69

Sprint 68 Dashboard

CareerGateway — Mar 16 – Mar 22, 2026

LIVE — pulls from Google Sheet on refresh
Loading data...

Capacity vs Planned vs Tracked

Workload Distribution

Time Tracked by Ticket

Sprint 68 vs Historical Averages

Retrospective — We Should...

▶ Start

  • Incident log for production failures — When a critical bug like module generation failure surfaces in beta, immediately open a doc tracking root cause, impact, and resolution steps so everyone stays aligned without repeated sync calls.
  • Queue-based approach from the start — Design new async workflows (video gen, module gen) with a queuing system from the beginning, not as a retrofit — avoids the reliability issues hit in G20B-760.
  • Early HeyGen reliability monitoring — Set up alerts for HeyGen API failures and add retry logic with user-visible progress feedback before shipping video generation features.
  • Beta user check-in cadence — Schedule structured weekly touchpoints with beta customers to surface friction early, especially around core flows like journey creation.

■ Stop

  • Shipping async features without queuing — Module generation and video avatar creation depend on long-running processes — always design the queuing layer first, not after bugs surface in production.
  • Underestimating reliability work — Stability and bug-fix tickets (G20B-760, G20B-753) consumed significantly more hours than estimated. Stop treating reliability work as lower-weight SP tasks.
  • Leaving third-party dependency failures unmonitored — HeyGen failures were silent to the user. Stop shipping third-party integrations without fallback states and error visibility.

✓ Keep

  • Pairing on complex reliability tickets — Gab and Claire pairing on the module gen overhaul (G20B-760) got more ground covered than solo work would have.
  • PM creating bug tickets immediately — Continued pattern of logging issues to Jira instantly on discovery keeps the backlog accurate and triageable.
  • Cross-functional awareness of AI costs — GCP spend dropped significantly this sprint — keep monitoring and attributing cost changes to specific engineering decisions.
  • Slack-first for async unblocking — Daily blockers surfaced quickly via #app-development chat; keep this as the default escalation path before calling a sync.
  • Claude Code + Perplexity research workflow — Continue using Perplexity for scoping third-party integrations and Claude Code for implementation and review cycles.
  • Overestimating time for complex unknowns — Sprint 68's highest-effort ticket was appropriately scoped with buffer; keep 2–3× estimation on new async/infrastructure work.