Measure safe-read latency

Summarize observed samples without inventing a performance guarantee.

Type: Build Languages: Go Stage: 1 of 4 Prerequisites: Basic functions, collections, JSON and the previous stages when applicable. Time: ~2 hours

What you build

Summarize nonnegative finite latency samples using nearest-rank percentiles. Keep the sample count and maximum so a tiny experiment does not masquerade as a stable distribution. The library adapter performs GET requests only.

Worked example

Samples [2,4,10,20,80] have P50=10, P95=80, maximum=80 and sample count 5. Sorting must not mutate the caller's array.

Implement the contract

Summarize([]float64)(Distribution,error) returns samples, p50Ms, p95Ms and maxMs. Reject empty, negative, NaN and infinite values. SafeURL(string)error accepts HTTP(S) URLs without credentials or fragments.

Reject malformed input before producing a partial artifact. The tests include held-out records separate from the demonstration fixture.

Run your work

python3 scripts/project_test.py hedged-request-lab --init /tmp/hedged-request-lab-work
python3 scripts/project_test.py hedged-request-lab --stage 1 --path /tmp/hedged-request-lab-work --strict

Initialization copies cumulative scaffolding without replacing your work. Implement the TODO functions in the first-stage starter; later-stage starter notes describe the new behavior. Tests import the learner workspace. Run the command in the README against your own input after passing the stage.

Inspect and extend

Collect repeated controlled runs. Nearest-rank P95 on five points is simply the maximum; it is not a tail estimate for production.

Go standard library and Node standard library. All fixtures and implementations are original.