Tracking

Null for the same reason recipients[].email is: shared bytes, unknowable reader.

human looked like somebody reading. proxy is an image proxy, Gmail's above all: a genuine reading by a reader we cannot see, counted once and then cached out of our sight. machine is a scanner or Apple Mail Privacy Protection, which fetches on delivery and means only that the message arrived.

Whether this hit moved the numbers. False for every machine hit, and false again for a hit landing within thirty seconds of the last counted one on the same copy. A preview pane redrawing is the same reading rather than a second one. includeMachine filters on this field and not on kind, so it is what hides those collapsed repeats as well.

Read out of the User-Agent, which is a claim rather than a fact and is absent altogether on plenty of hits.

What the edge already knew about the request, so this is as precise as the location will ever be, and on a proxied hit it is the proxy's country rather than the reader's. There is no ip field on this resource, but the address is not discarded: it is written into the event log for the message, readable at GET /emails/{id}/events, and into the email.opened and email.clicked webhook payloads.

The destination that was followed.

True when another page follows. Pass nextCursor back as cursor to read it.

The id of the last row on this page, or null on the last page.

Null for the same reason recipients[].email is: shared bytes, unknowable reader.

human looked like somebody reading. proxy is an image proxy, Gmail's above all: a genuine reading by a reader we cannot see, counted once and then cached out of our sight. machine is a scanner or Apple Mail Privacy Protection, which fetches on delivery and means only that the message arrived.

Whether this hit moved the numbers. False for every machine hit, and false again for a hit landing within thirty seconds of the last counted one on the same copy. A preview pane redrawing is the same reading rather than a second one. includeMachine filters on this field and not on kind, so it is what hides those collapsed repeats as well.

Read out of the User-Agent, which is a claim rather than a fact and is absent altogether on plenty of hits.

What the edge already knew about the request, so this is as precise as the location will ever be, and on a proxied hit it is the proxy's country rather than the reader's. There is no ip field on this resource, but the address is not discarded: it is written into the event log for the message, readable at GET /emails/{id}/events, and into the email.opened and email.clicked webhook payloads.

True when another page follows. Pass nextCursor back as cursor to read it.

The id of the last row on this page, or null on the last page.

Present when the report is the whole response body. Absent under an email's tracking field, which is part of that email rather than a resource in its own right.

The tracking record, tmsg_ + 24 hex. Not the message id and not the send id.

The msg_ this went out as, when the send service handled it. Null for mail the mailbox agent sent on its own behalf, which is most composer, MCP and assistant traffic. Those messages do get a send record, but nothing links this tracking row to it. Tracking covers the mailbox rather than only the traffic that came through this API.

RFC 5322 Message-ID. Not a correlation key. The header is rewritten on the way out. Correlate on id.

What was APPLIED to this message, resolved when it was sent from the setting of the address it was sent from (its own, else its domain catch-all's, else off, while a broadcast copy is on unless the broadcast or its address turned it off) and any per-send override, not what is switched on now. Turning tracking on today does not make yesterday's mail start reporting, and a report that implied otherwise would read as "nobody opened it".

As opens, for link rewriting. The two are independent switches.

Whether every reading on this message can be pinned to a named recipient. False as soon as an unattributed copy has activity of its own, which is what happens whenever one body went to the whole list rather than a separate one per person. This is the flag that decides whether "Bob has not opened it" is a sentence a client is entitled to write, or whether all it may say is that somebody did. recipients carries the same fact one row at a time, and one row at a time is where it gets missed.

Opens that looked like a person, with repeat fetches within thirty seconds collapsed. A preview pane redrawing is not a second reading. Through Gmail this is a floor and not a total: its proxy fetches the image once and caches it, so later readings never reach us.

Counted clicks. Stronger evidence than an open, and worth weighting as such: images are blocked far more often than links go unfollowed, so a message with clicks and no opens was certainly read.

Every open hit, the automated ones included. openCountRaw - openCount is everything that was filtered out: Apple Mail Privacy Protection and corporate link scanners, which fetch on delivery whether or not a person ever looks, and alongside them the repeat fetches collapsed by the thirty-second window. Both are recorded and neither is counted, because discarding them outright would leave a gap in the log that nothing could explain. Do not read the difference as a machine count on its own. A message reopened twice in a minute lands in it too.

As openCountRaw, for clicks.

One entry per tracked copy, which is not always one entry per person. Absent only from the trimmed form on a GET /emails row; every /tracking response carries it, list included.

The rewritten links, in the order they appeared in the message. Only links in the new part of the body are here: the quoted history under a reply belongs to whoever wrote it, and routing their URLs through our redirector would both rewrite their message and record the recipient "clicking" something we did not put there. Repeated destinations share one entry, because a campaign page linked from a header image, a button and a footer is one question asked three times. Absent only from the trimmed form on a GET /emails row.

True when another page follows. Pass nextCursor back as cursor to read it.

The id of the last row on this page, or null on the last page.

Messages in the window that carried a pixel or a rewritten link.

Of those, how many a person opened at least once. Distinct MESSAGES, not hits. A message opened five times is one opened message, and conflating the two is how open rates above 100% get published.

The denominator of openRate. Tracking is two switches rather than one, so this is the messages that actually carried a pixel, not every tracked message.

The denominator of clickRate, and the reason it is published. A message with no links is never click-tracked, so counting it against the click rate makes that rate a measure of how much of your mail contains a link. Most mail is a reply with no links, so the difference from tracked is usually large.

A percentage of trackedForOpens, deliberately not of everything sent. A workspace that tracks one message in ten has an open rate for those ten; dividing by all its mail produces a number that falls every time somebody sends an untracked reply, which is not a fact about how anyone is reading.

A percentage of trackedForClicks, as openRate is of trackedForOpens.

Hits rather than messages. Labelled as a total because it is the easy one to misread.

Hits excluded from every number above: the sum of openCountRaw - openCount, so Apple Mail Privacy Protection and corporate scanners together with the repeats the thirty-second window collapsed. Reported rather than hidden, because a reader who cannot see how much of the traffic was machinery has no way to judge the rest of the panel. Gmail's image proxy is not in this number: that fetch is a real person displaying the message, and it is counted.

Median seconds from send to first counted open, over the messages that were opened at all. Null when none were. A median over an empty set is not zero.

SPARSE. A day on which nothing was sent has no entry rather than a row of zeroes, so a chart has to fill the gaps itself. Days break at offsetMinutes east of UTC, so that they break where the reader's day does rather than where the database's does.

Mail clients by counted opens, as far as a User-Agent can be trusted to name one.