Incremental Opportunities for Public Projects — 2026-08-08
Scope: this window's new Architecture/Fundamentals/ notes (idempotency-keys, load-balancing-algorithms) checked against Projects/public-projects.md. The 2026-07-31 pass covered MCP statelessness, Aurora DSQL, and Docker sandboxing against earlier AIDigest content — this pass covers two Fundamentals notes not yet checked against the repo.
Idempotency Keys gives three existing MongoDB projects a vendor-neutral reference they're currently missing
mongodb_distributed_lock (TTL-based distributed lock), mongodb_to_avoid_duplicate (CacheManager combining caching, dedup, and locking), and real_time_fraud_detector (API Gateway/SQS/Lambda/ElastiCache velocity checking) are all already solving instances of the exact problem this window's idempotency-keys note formalizes — but none of their descriptions currently reference the three-part structure the note lays out (stable per-operation key, payload fingerprint to catch key-reused-with-different-payload, and an atomic claim to prevent two concurrent duplicates both passing a read-then-write check). Worth a documentation pass on mongodb_to_avoid_duplicate specifically: the note's "Common pitfall" section — storing a key without a payload fingerprint, which looks correct until a client bug reuses a key across two genuinely different requests — is a concrete correctness check worth running against that project's actual CacheManager implementation, not just a naming exercise.
Load Balancing Algorithms gives the three loadbalance_* demo projects a decision framework to document, not just a working config
loadbalance_caddy, loadbalance_custom_network, and loadbalance_roundrobin_config all demonstrate round-robin load balancing mechanically, but none currently explains why round robin was the right algorithm choice for that demo versus the alternatives. This window's load-balancing-algorithms note's central framing — the real fork is between load-optimizing algorithms (round robin, least connections) and affinity-preserving consistent hashing, and picking the wrong family for a given tier "doesn't fail loudly, it just quietly degrades" — is a one-paragraph addition each of these three READMEs is currently missing. It would also set up a natural fourth demo: a consistent-hashing variant showing what breaks (cache hit rate collapsing) when round robin is pointed at a stateful cache tier instead of a stateless one — the exact failure scenario the note's own "Reel Script" walks through.