Problem

Reports had grown around separate web and mobile implementations, feature-specific refresh behavior, and two competing sources of state. A self-reinforcing React-to-Redux reconciliation loop could repeatedly regenerate reports, while each new product surface had to rebuild core behavior such as filtering, saved state, exports, printing, and drilldowns.

Decision & implementation

Designed one report lifecycle with typed cross-platform contracts, Redux-owned state, configuration-driven consumers, saved-report behavior, synchronized charts and virtualized tables, and explicit regeneration across eight financial-data domains. The migration was staged so existing reports could move onto the platform without a single high-risk cutover.

Tradeoffs

A staged migration temporarily preserved adapters and parallel paths, but it kept rollback practical and let each report prove parity before cutover. The platform standardized lifecycle and state ownership without forcing every report into the same presentation or financial model.

Outcome

In September 2026, 44,171 report users generated 191,078 opens across 14 product experiences. The reconciliation failure fell from 1,874 occurrences across 176 users in April to four occurrences across two users in September. In matched four-month windows around the transition, new Reports bugs declined 48% and the open backlog declined 41%; those broader quality trends reflect the platform work alongside ongoing stabilization by the team.

Problem

The legacy chart layer struggled with production-sized datasets and forced the product to carry two visualization architectures during migration. In one 12-month daily report, drawing roughly 365 bars was the primary cause of an extended page freeze; grid autosizing and quadratic row de-duplication added to the cost. The product needed a faster library and a migration plan that would not destabilize live reports.

Decision & implementation

Ported representative report charts to seven candidate libraries, then measured mount, paint, update, and unmount behavior under production-shaped load. The resulting proposal covered technical fit, licensing, accessibility, rollout, validation, and rollback. I then built shared typed configuration and interaction patterns so individual charts could migrate behind flags while preserving product behavior.

Tradeoffs

Canvas rendering solved the dominant high-density bottleneck but required deliberate work around semantics, interaction parity, and test coverage that SVG had exposed through the DOM. Feature flags and dual implementations added short-term complexity in exchange for measurable, reversible production rollout.

Outcome

Completed the conversion of more than 35 visualizations and removed amCharts, 14 rollout flags, 24 dual implementations, and 2,274 net lines of transitional architecture. In an equivalent production flow, first-open time improved from 1,601 to 715 milliseconds and blocking chart setup fell from about 880 to 12.5 milliseconds. Reports of freezes tied to high-density amCharts rendering stopped after rollout.

Problem

A common Spending Plan interaction took about two seconds before users could see the effect on their monthly plan. The delay sat in a high-frequency financial workflow where repeated waiting made the product feel unresponsive and increased the risk of masking state problems behind loading behavior.

Decision & implementation

Profiled the interaction and redesigned the state flow so the client could derive the result from financial data it already owned instead of waiting for the slower path. Added regression coverage for the calculations and coordinated a cross-team test plan across current, future, and prior-month behavior.

Tradeoffs

The faster path was limited to financial state the client already owned; it did not replace server authority where fresh data was required. That boundary kept the optimization deterministic and made the new calculation practical to cover with regression tests.

Outcome

Reduced the interaction to less than 50 milliseconds without hiding the delay behind a spinner. The change preserved the financial rules and established test coverage around a previously fragile workflow.