Two instance cumulative-bucket staircases merge into one fleet staircase before estimating p99. Matching le bucket identity is retained.

Prometheus histograms retain a bucketed distribution

A latency mean can conceal a slow subset of requests. A histogram retains more distribution information by counting observations into buckets, but it still discards detail within each bucket. Prometheus can estimate quantiles from those counts without storing every observation. The estimate depends on bucket boundaries and aggregation; it is not a recovery of the complete original latency distribution. ...

January 6, 2027 · 6 min · XB Field Notes
Successful-request mean is 89.47 ms, compared with the five fast failures at a 5 ms mean. Proportional bars show why adding failures can lower the combined mean without improving useful service.

The golden signals of health

Latency, traffic, errors, and saturation are four complementary ways to ask whether a service is serving users well. They are a starting set of questions, not four universal counters that prove health by themselves. Their usefulness depends on where they are measured, which work they describe, and how their time windows align. A fast response can still be a failed operation. ...

November 29, 2026 · 5 min · XB Field Notes
A complete synthetic population has 100 requests and four failures, shown as a 10-by-10 grid with four red cells. A deliberate selection keeps five records, including two failures: 4/100 is 4%, while 2/5 is 40%. The selected fraction is not service-wide.

Metrics and logs need an explicit population

Metrics and logs are different representations of observed behavior, not competing definitions of truth. A metric usually summarizes measurements under a set of dimensions. A log record describes an event and can carry detailed context. Both can answer aggregate questions if their coverage and semantics support the calculation. Neither becomes authoritative merely because its dashboard looks cleaner or its records contain more fields. ...

October 16, 2026 · 5 min · XB Field Notes
A scrape-failure alert points to three investigation destinations: the target error, a comparison of metrics and public API paths, and logs with changes. These are evidence to read, not a claim that the API failed or a release caused it.

Alerts tell you what to read

An alert should give its reader a useful first investigation, not merely a red indicator. Name the observed condition, affected scope, persistence window, responsible team, and next evidence to inspect. These details are part of the rule’s behavior: test them with the trigger instead of treating notification text as decoration added after the incident. ...

September 22, 2026 · 5 min · XB Field Notes
A wide one-to-five-second interval and a narrow one-to-two-second interval give different p99 estimates for the same raw two-second p99: 4.2 and 1.8 seconds.

Histograms preserve the latency shape

A mean can make an uncomfortable service look fast. If ninety-five requests take 50 milliseconds and five take two seconds, the average is only 147.5 milliseconds. A latency histogram retains more information about that slow minority without storing every request. What it preserves is a bucketed approximation of the distribution, not the complete original observations or an automatically exact percentile. ...

September 20, 2026 · 5 min · XB Field Notes
GET and HEAD use matched URL, request headers, and release. Both response cards show type, validator, coding, and length categories. Only GET has content and checksum evidence. HEAD length can be omitted; if provided it describes GET.

HEAD should describe the same representation as GET

HEAD is a metadata view of GET; status and representation headers should tell the same story.

June 22, 2026 · 2 min · XB Field Notes
Expiry evidence consists of notAfter and a chosen threshold. Separate verification asks about name, chain, and trust. SNI selects the virtual host and does not validate its certificate.

Certificate expiry is a date, not a dashboard color

Use SNI, inspect the public certificate, and alert before renewal becomes an incident.

June 10, 2026 · 2 min · XB Field Notes