Pointer compression shrinks V8 heap references so the same object graph fits in less RSS, but it does not fix retention. After enabling it, re-baseline heapUsed and rss under a fixed warm-up, then compare retained size by constructor against the pre-change snapshots. Watch for external buffers and native addons that sit outside the compressed heap — those can dominate RSS even when the JS heap looks half as large.

Ops prove-out on a canary that stays up across a realistic traffic window: confirm the post-GC slope of heapUsed flattens or drops versus the prior build under the same mix, and that p99 latency does not regress from denser GC. If rss barely moves while heapUsed halves, inventory Buffer and external memory. A smaller heap number alone is not the win if the process still climbs across hours on reused instances.