fastmon Docs

Observations

Changes fastmon found on its own and recorded with a start and an end. Sudden shifts and slow trends. Beta.

An observation is a change fastmon noticed by itself and wrote down with a start, an end, and the numbers behind it. You configure nothing. An observation opens and ends on its own, and your team can mark it as seen.

Beta

Observations are in beta. Thresholds, texts, and the layout of the page may still change.

Two kinds

KindWhat it catchesCompared againstVisible after
Shifta sudden change within a quarter hour: traffic, TTFB, LCP, JavaScript errors, botsthe same quarter hour on the 7 days before30 minutes for a degradation, 2 hours for an improvement or a note
Trenda slow change over weeks: LCP, INP, CLS, TTFB, error sharethe 14 days before the last 7the next midnight UTC

Every observation has a direction. A degradation is a problem and an improvement a highlight. A note is information nobody can act on: a traffic rise, or any change of bot traffic.

States

StateMeaning
going onthe change is measurable right now
opena recurring shift between two occurrences, or a trend waiting for its next midnight
pauseda trend with too little data in the last 7 days. It neither ends nor moves
acknowledgeda team member marked it as seen. It leaves the open cards and the sidebar count, and still ends on its own
endedthe change is gone, or it became the new usual level

Shifts

A shift is a quarter hour far from what is usual for that quarter hour. Usual is the median of the same quarter hour on the 7 days before, per metric. It needs 4 of those days with data, so a new application opens nothing in its first 4 days.

The diagram shows the two ways a shift opens, and how it ends.

Life of a shift: opens after 2 quarter hours above 3 times the usual value or 8 above twice, ends under 1.5 timesWay 1: 2 quarter hours in a row over 3× (or under 1/3)visible after the 2nd quarter hour3× usual1.5× usualusualstartends: back under 1.5×Way 2: 8 quarter hours in a row over 2×, degradations onlyvisible after the 8th quarter hour, 2 hours2× usual1.5× usualusualstartends: back under 1.5×quarter hours, compared with the same quarter hour on the 7 days before

Opens in one of two ways. The shift starts at the first of these quarter hours.

  • Two quarter hours in a row at 3 times the usual value, or a third of it.
  • Eight quarter hours in a row, 2 hours, at twice the usual value in the worse direction.

Each of these quarter hours needs at least 150 page views, 500 for bots. A degradation at 3 times is written after its second quarter hour. Everything else once it has lasted 2 hours.

Ends when a quarter hour is back under 1.5 times the usual value, or above two thirds of it after a drop. The observation says why:

ReasonMeaning
back to normalthe metric returned to its usual level
new levelthe shift lasted 4 days or more and became the usual level. The problem did not go away, fastmon just no longer reports it
no data4 days in a row without a reference, as a dead tracker leaves

Recurs when a shift of the same metric and direction starts within 2 hours of the same time of day, no more than 7 days after the last one ended. It stays one observation, labelled "Shift, recurring", lists every occurrence, and ends 7 days after its last end. Bot shifts recur at any time of day and in either direction.

MetricMeasured per quarter hourDirection
Traffichuman page viewsa drop is a degradation, a rise a note
Server response time (TTFB)p75 in msslower is a degradation, faster an improvement
Load time (LCP)p75 in msslower is a degradation, faster an improvement
JavaScript errorsshare of page views with a first-party errormore is a degradation, fewer an improvement
Botspage views that are not humana surge or a collapse is a note

A degradation and every traffic or bot shift also list what carried the change: the countries, page types, devices, and domains with the largest share of it, and the releases around its start.

A trend is a slow change over weeks, judged once a day at midnight UTC over whole days. The measure is the share of page views above the poor threshold of LCP, INP, CLS, or TTFB, or the share with a first-party error. The diagram shows the reference, the line where a trend opens, and the line where it ends.

Life of a trend: opens when the last 7 days reach 1.3 times the 14 days before, ends under 1.1 times1.3× opens1.1× endsreferencereference: the 14 days before, frozen at the openingthe last 7 dayswhole days (UTC), judged once a nightshare of slow page views
  • Reference is the pooled share of the 14 days before the last 7. It is frozen when the trend opens.
  • Opens when the last 7 days reach 1.3 times the reference and at least 2 percentage points more. An improvement opens at 1/1.3 of the reference and 2 points less.
  • Ends at the first midnight whose last 7 days are under 1.1 times the frozen reference. A metric that settles at a new level keeps its trend open for months.
  • Pauses while one of the last 7 days is missing or the week holds fewer than 1,000 page views with the metric.

The first verdict needs 14 whole days of data. An open trend shows since when the metric has been at its new level.

On the Observations page

Observations in the sidebar shows what is open now as cards, the selected time range as a timeline per area, and every observation of the range as a table. Each row opens a detail page with the title, the state, the period, and the change against the usual value. A shift adds its drivers and the releases around its start.

Acknowledge at the top right marks an observation as seen, with a note of up to 1,000 characters for the team. Owners and members can acknowledge, viewers see it. A new occurrence of a recurring shift keeps the acknowledgement, and the note stays after the observation ended.

Observations send no notification. The weekly overview mentions trends: the open degradations and the improvements of the week.

On this page