What Our Commit History Says About Team Health

The Pattern Before the Conversation

Most teams find out someone is unhappy when they hand in their notice. By then the decision was made weeks ago, maybe months. The signs were there earlier — not in Slack messages or stand-ups, but in the work itself.

Commit history is not a performance metric. It is a pulse. And like any pulse, the interesting signal is not the individual beat but the rhythm over time.

Burst and Rest Is Normal. Burst and Silence Is Not.

Small teams ship in waves. A hard push before a release, a quieter stretch after. This is healthy. Human energy is not constant, and pretending otherwise leads to steady-state mediocrity that looks productive on a graph and feels dead in a room.

What matters is the shape of the cycle. A team that bursts, rests, then re-engages is recovering. A team that bursts, goes silent, and stays silent is withdrawing. The difference is hard to spot in a single week. Over a month, it is obvious.

Watch for the gap between the last hard push and the next voluntary commit. When that gap stretches — when the rest period stops looking like rest and starts looking like avoidance — something has changed. Not in the code. In the person writing it.

PR Size Creep Tells You About Trust

Early in a team's life, pull requests are small. People want feedback. They want to prove their thinking. They submit tight, focused changes and wait for review.

Over time, PR sizes drift upward. Some of this is natural — confidence grows, context deepens, and a developer who once split work into five pieces can hold the full problem in their head and ship it as one. Fine.

But there is a version of PR size creep that signals something else: a person who has stopped expecting useful review. They batch more work because they assume nobody will read it carefully. Or they avoid the friction of back-and-forth by presenting finished work that is hard to argue with. The PR grows not because the author is more capable, but because the review process has become a rubber stamp.

When you see someone who used to submit small, frequent changes start shipping large, infrequent ones — ask why. Not about the code. About how the process feels to them.

The Friday Afternoon Commit

One specific commit pattern should catch your attention: regular work at odd hours, especially late Friday or over weekends, by someone who did not used to work that way.

Sometimes this means a person is excited. They found a problem they care about and lost track of time. That kind of energy is unmistakable — the commits are creative, exploratory, often a little messy.

Other times, it means a person is compensating. They feel behind. They do not feel safe admitting it during work hours. So they quietly try to close the gap on their own time, hoping no one notices the timestamps.

Same data. Opposite meaning. The commit history shows you when. Only a conversation tells you which.

Silence Is the Loudest Signal

The pattern we have seen most reliably before someone leaves: a slow fade. Not a dramatic stop. A gradual reduction in initiative. Fewer comments on other people's work. Fewer self-directed improvements. The commits that remain are responsive — assigned tasks, reviewed tickets, requested fixes. The proactive work disappears first.

This is hard to see if you only look at throughput. Output can stay constant while ownership evaporates. A person can ship the same number of changes and be completely checked out, because the changes they stopped making were the ones that came from caring about where the product is going.

By the time the commit history goes fully quiet, the departure is already decided. The useful window was earlier — when the optional work started falling away and nobody asked about it.

What This Means in Practice

We do not measure commits per day. We do not gamify contribution graphs. Treating commit history as a performance metric destroys the signal, because people optimize for the metric instead of the work.

What we do is pay attention to rhythm changes. Not the absolute level of activity, but the delta. A person who shipped five things last month and two this month is telling you something. A person whose review comments went from thoughtful to perfunctory is telling you something. A person who stopped starting things and only finishes what is assigned is telling you something.

The cadence of work tells a story before anyone speaks up. Your job is not to surveil the data. It is to notice the pattern and have the conversation early enough that it matters.

The best retention tool is not a comp adjustment or a title change. It is the question asked six weeks before someone starts interviewing: "You seem different lately. What is going on?"

That question is easy to ask. Knowing when to ask it is the hard part. The work itself will tell you, if you are paying attention.