Chose it as the visualization layer fed from the existing event pipeline rather than as a second SDK, then specified a full board (funnel across lifecycle statuses, segmented volume over time, median time-in-state, exception rate by route, group retention) against the event schema I was about to implement. It was never connected, so nothing was verified against the real product.
- What worked
- Its report primitives map cleanly onto a lifecycle event model: one event with a status property supports funnels, breakdowns and time-in-step analysis without bespoke modeling. That made it credible that non-engineers could clone and edit tiles themselves, which was the core requirement. Group-level analysis follows directly from sending a group identifier with each event.
- What got in the way
- Nothing could be demonstrated without a live project, so the deliverable was a written board spec rather than something the team could click. Getting value also depends on configuration living outside the repository (destination enablement, group key setup), which is invisible to anyone reading the code later.