Infographic 05 · D3
Cell colour is what an employer weights. The bar on the right is how strong the evidence actually is. The interesting rows are the ones where those two disagree.
The last two rows are the honest gaps: distributed consensus is consumed rather than implemented, and every deployment is single-region. Both are weighted highly by the employers being targeted, and both are stated rather than padded.
| Skill | Evidence | Spotify | Bank | Depth | |
|---|---|---|---|---|---|
| Event logs & consumer groups | BillboardActivityService · Kafka, 3 groups | HIGH | HIGH | HIGH | 5 / 5 |
| Idempotency & delivery semantics | three sinks, three mechanisms | HIGH | HIGH | CRIT | 5 / 5 |
| Partitioning & shard-key design | MatchService · (service_region, geohash4) | HIGH | HIGH | HIGH | 5 / 5 |
| Concurrency correctness in Go | fatal map race G-07 · calling_race_test.go | CRIT | HIGH | HIGH | 5 / 5 |
| Realtime transport & presence | GlobalChatService · WebSocket, TTL presence | MED | HIGH | MED | 5 / 5 |
| Contract discipline in CI | OpenAPI + AsyncAPI + 24 protobufs | HIGH | MED | CRIT | 5 / 5 |
| Operational writing & postmortems | 54 manual chapters · gap registers | CRIT | MED | CRIT | 5 / 5 |
| Kubernetes & autoscaling | Solomon · k3s, Helm, HPA, NetworkPolicy | CRIT | HIGH | MED | 4 / 5 |
| Auth, tokens & key handling | AuthGateway · RS256, ACL, mode-sync | MED | MED | CRIT | 4 / 5 |
| Large-file & media pipelines | R2 multipart, IngestDO, TUS, range streaming | MED | CRIT | LOW | 4 / 5 |
| Audio DSP & fingerprinting | vDSP FFT on-device · kissfft.wasm at the edge | LOW | CRIT | NONE | 4 / 5 |
| ML platform & release gating | Solomon · 9 promotion-blocking gates | HIGH | MED | HIGH | 4 / 5 |
| Distributed consensus & replication | consumed, not implemented | CRIT | MED | HIGH | 2 / 5 |
| Multi-region & failover | single-region deployments | HIGH | HIGH | CRIT | 2 / 5 |