docs(status): r4 soak PASS closes the multi-hour gate + prod NetworkPolicy checklist (audit G-1, G-3)

[brownjuly2003-code]

Closes audit findings G-1 (P2, doc integrity) and G-3 (P3, ops default) from the 2026-07-20 deep audit.

G-1: docs/STATUS.md still said "Multi-hour (4 h+) sustained >= 100 is still open" while the tracked evidence doc docs/perf/throughput-realpath-paced100-4h-r4-2026-07-19.md records r4 PASS (100.0 produce / 99.9 flink / 99.9 apply eps, 1 440 000 events exactly once, dup=0, fail=0, lag 0->0, zero Flink restarts). STATUS now: header refreshed, r4 row added to the throughput ladder, narrative rewritten (including the honest r1/r3 failure pointer), Next list = P2-6 packaging + flink-runtime bump to retire the pyarrow safety ignore.

G-3: docs/deployment.md Helm operator checklist gains the missing line: networkPolicy.enabled=true before any multi-replica production rollout + confirm the cluster enforces NetworkPolicy (chart default stays false by design).

Docs-only, no code. test_docs_single_source_of_truth.py 9 passed locally.

🤖 Generated with Claude Code

[github-actions]

DORA Metrics

• Window: last 30 days
• Branch: main
• Deployment frequency: 117 total / 27.3 per week
• Lead time for changes: avg 0.84h / median 0.0h
• Change failure rate: 77.78% (91/117)
• MTTR: 20.13h across 5 incident(s)