READ → IMPLEMENT → OBSERVE

Experiments

Each experiment takes a chapter's implementation and pushes it somewhere the theory predicts a specific, observable outcome.

1.
Percentile convergence under load
Feed the reservoir tracker increasing volumes of synthetic request durations and observe how quickly p50 and p99 estimates converge and diverge in accuracy.
2.
Document vs. graph traversal cost
Run the same 'friends of friends' query against the document store (manual joins) and the graph traversal, and compare the number of lookups each requires.
3.
Sequential vs. random write throughput
Compare append-only log writes against in-place updates at random offsets on the same amount of data, and measure the throughput gap directly.
5.
Observing replication lag
Introduce an artificial delay on the follower's log-shipping channel and read from the follower immediately after a leader write to reproduce a stale read.
7.
Reproducing write skew
Run two concurrent transactions under snapshot isolation that each check an invariant and then violate it together, showing why snapshot isolation alone doesn't prevent write skew.
8.
Timeout vs. real failure
Configure the unreliable network to add random delay without ever dropping a node, and show that a fixed timeout still produces false-positive failure detections.
11.
Consumer lag under a traffic spike
Burst-produce messages into the partitioned log faster than the consumer processes them and graph the growing gap between the latest offset and the consumer's committed offset.