Leaderboard & Counting Service · View 04 of 21 · People and journeys
The finding
- The sentiment drops at "find me", not at "open". A list that loads in 40 ms and does not contain the member's last action reads as broken, while a list that is 20 seconds stale and does contain it reads as fine.
- That asymmetry is why freshness is spent on one member's own contribution and budgeted for everyone else's view of the ranking.
- The second trough is a rank that moves between two reads in one session. Pinning the projection version per session costs nothing and removes it.
What the platform owes here
- Own value and rank band within 1 second of an accepted event (assumption).
- Top-N p99 ≤ 40 ms on cache hit; rank-of-member p99 ≤ 120 ms (assumptions).
- An as-of timestamp on every response, so the product can say "as of" rather than imply live.
Risks
- The overlay is a second code path for the same number. If it ever disagrees with the projection by more than the member's own delta, the platform has two answers and no way to choose.
- Pinning a version per session means a long-lived session can read a projection the platform would rather retire.