Day 12 — Align One Incident Timeline

Today Signal Room answers a deceptively hard question: did the latency spike happen before or after the deployment and the on-call observation?

Run the alignment artifact

cd rust/rust-ai-engineering
cargo run -q -p mosaic-evidence --example alignment_pipeline

Represent time deliberately

Use a unit-bearing integer type:

pub struct Micros(pub i64);

pub struct TimeRange {
    pub start: Micros,
    pub end: Micros,
}

Do not pass bare f64 values where one subsystem means seconds and another means milliseconds. Checked constructors enforce start <= end and checked arithmetic prevents wraparound.

Each source clock needs an alignment transform:

pub struct ClockMap {
    pub source_origin: Micros,
    pub incident_origin: Micros,
    pub rate_numerator: i64,
    pub rate_denominator: i64,
}

Audio and video clocks may drift. Dashboard captures may know only wall time. A human voice note may say “about two minutes earlier.” Store uncertainty rather than manufacturing precision.

Build the timeline as data

Every event contains incident time, uncertainty, evidence IDs, and a short observation. Sort using a stable rule and preserve ties. The rendered Markdown report is a view; JSON is the replayable source.

Break it deliberately

• shift one source clock by one timezone;
• use milliseconds as microseconds;
• create two events with equal time;
• map a range beyond the source duration;
• remove the transform required to restore a citation.

Ask which invariant catches each mutation. A model prompt cannot repair a unit error hidden in the timeline.

Completion proof

Produce a five-event timeline containing a deployment, trace anomaly, dashboard observation, voice-note statement, and recovery. Every event must map back to its original source locator.

Deep reference: Timeline and geometry alignment.

Next: Day 13 — Hypotheses and Evals →.