Email reports, annotations, and remembering why

A dashboard answers questions you remember to ask. This note is about three features for the questions you don't: scheduled email reports, annotations, and a weekly team health digest. They're all, in different ways, tools against forgetting.

Email reports

Every hour the scheduler runs reports:send, which walks the active email reports and sends the ones that are due — daily reports once per day, weekly after seven days, monthly after thirty. Each report belongs to a site, takes a comma- or whitespace-separated list of recipients, and is composed of sections you toggle per report: the summary numbers, top pages, top sources, top events, goals, week-over-week comparison, and anomalies. The defaults lean editorial — summary, top pages, top sources, week-over-week, and anomalies on; events and goals off until you ask.

The anomalies section is the one I'd call load-bearing. A weekly email that says "here are your numbers" gets skimmed; one that says "here are your numbers, and Tuesday was 80% above a typical Tuesday" gets read. The anomaly engine I wrote about a couple of weeks ago feeds the same findings into these digests, so the push channel carries the judgment, not just the totals.

Annotations

Annotations are dated notes attached to a site: "shipped the redesign," "newsletter went out," "got linked from that podcast." Each one has a date, up to 5,000 characters of text, and optionally a specific page path — so a note can be about the whole site or pinned to /pricing. The composer offers quick-pick dates for the last two weeks (most notes are about something that just happened), and the page picker suggests your top pages from the last 30 days with a search box.

The payoff comes later, in two places. On the annotations timeline itself, each note is shown with its traffic context — the views on that day, the change versus the prior 7-day average, and a small sparkline of the surrounding two weeks — so a note reads as "shipped the redesign, and here's what traffic did." And on the anomalies view, Tracely surfaces any annotations on or within a day of each anomaly right next to it — the "oh right, we deployed that morning" moment, delivered automatically instead of excavated from a Slack scrollback. Six weeks after the fact, nobody remembers why traffic doubled on a Wednesday. The annotation remembers.

This is also why annotations are part of the site export bundle alongside goals and funnels: they're analytical context, not decoration, and they should survive a migration.

The team health digest

The third piece is the quietest. Once a week — Monday morning, if you've enabled it — tracely:team-health-digest computes a health score per team over the trailing seven days from a few blunt signals: pageview volume, JavaScript errors (tracked as $javascript_error events), and average largest contentful paint from the performance metrics. It logs one line per team, labeled with the ISO week, and can optionally email a short Markdown digest to each team member.

It's off by default, and the email is a second opt-in on top of that. I have strong feelings about products that treat your inbox as a growth channel; a digest should exist because someone decided they want a Monday pulse, not because I want a weekly impression. But for a team running several sites, one line per week — traffic moving, errors flat, pages fast — is a genuinely useful ambient signal, the kind of thing you don't study but would notice going red.

Push versus pull

The common shape here: dashboards are pull, and pull depends on the discipline to keep looking. These three are push and memory — the report arrives whether or not you remembered to check, the annotation holds the explanation until you need it, the digest keeps a floor under "is everything roughly fine." None of it is novel, and all of it is the difference between analytics you have and analytics you use. If you're evaluating what's included, these are the features I'd actually weigh.