Environment: Azure Container Apps Job (D32-benchmark, 32 vCPU / 64 GiB), westus3
ClickHouse Cloud Production tier (3 replicas × 16 vCPU × 64 GiB = 48 vCPU / 192 GiB)
Connection: HTTPS, Compression=true, set_async_insert=1, set_wait_for_async_insert=0
PBP sink: none (--no-pbp)
Merge parallelism cap: 16 (was 4 on smaller tier — raised in commit ba0bb337)
Job execution: oc-exp-1k-p32-7c1jm5n

Backend: clickhouse, Scale: 100000 worlds, Environment: local
Streaming mode: totalWorlds=100000, chunkSize=500 [PBP DISABLED] [parallel=32]
PBP sink: local at ./pbp-parquet

Streaming Run Report
--------------------
Total wall time:         1886.50 s
Simulation time:         1412.06 s
OC write time:            327.67 s
PBP write time:             0.00 s
Merge time:               136.50 s
Peak working set:        18962.6 MB
Chunks completed:            200
OC rows (merged):        244,800
PBP rows (written):            0

Notes:
- 100K OC-only rebaseline on Production 3×16 (48 vCPU aggregate).
- Vs prior Production Scale (2×4) baseline of 2,991s: total dropped 37%
  (2,991s → 1,887s), with the entire improvement coming from merge:
  1,210s → 137s, an 8.9× speedup.
- Merge cap raised from 4 to 16 to exploit the new per-replica RAM budget
  (was constrained to 4 on 14.4 GiB/replica; 64 GiB/replica comfortably
  handles 16 × ~680 MB concurrent arrayFlatten working sets).
- Sim phase unchanged (client-bound, prototype). Now 75% of wall time.