All posts
Developer tracking7 min read

Developer performance metrics that hold up: the ones read from merged commits

For a CTO or a CEO choosing what to measure: the per-developer metrics that hold up come from merged commits (delivered, reviewed, tested, defects, code score), and each one has limits to state.

The DeployIt Team

We build DeployIt, an MCP server that lets a whole company ask its code from Claude and track each developer's work.

The developer performance metrics that hold up are the ones read from merged commits: what each developer delivered, whether a colleague reviewed it, whether it came with tests, which defects it caused, and a code score from 1 to 5 that is always read with its number of commits.

You are a CTO or a CEO, and sooner or later you have to say something precise about each developer's work. Hours, closed tickets and lines of code are easy to count and easy to dispute. A metric holds up when it passes three tests: it counts finished work, the person measured can trace it back to their own commits, and it states its own limits.

In your Claude · example

Over the half-year, what did each developer deliver, and how was it reviewed, tested and scored?

Here is the half-year for the three tracked developers. Each score comes with its number of commits; defects from the last 60 days are not mature yet.

Half-yearCamille R.Idriss B.Léa M.
Merged commits251214186
Reviewed by a colleague71%60%66%
Tested52%31%40%
Confirmed defects372
Code score3.9 (251 commits)3.2 (214 commits)4.1 (186 commits)
Drawn from your commits651 merged commits · sample data

Which developer performance metrics hold up?

Five. All of them are written from the same event: a commit merged into the product, credited to its author.

MetricHow it is countedWhat it saysWhat it does not say
DeliveredCommits merged on each repository's landing branch, counted when they landWhat reached the product, by developer and by periodHow large or how hard the work was
Reviewed by a colleagueThe pull request was reviewed by someone who is neither its author nor a botThat a second person read the changeHow deep the review went
TestedThe change comes with tests, or notWhether tests travel with the codeWhether those tests are enough
DefectsA fix points back to the commit it repairs: “to confirm”, then “confirmed”Where the period's fixes come fromAnything final about the last 60 days
Code scoreA language model reads each commit: 1 to 5, on five criteriaAn average reading of qualityAnything, without its number of commits

Read them together. One column alone settles nothing.

What counts as delivered?

Only work that reaches each repository's landing branch counts, and it counts when it arrives. Example: a commit dated in October but merged on 3 November counts for November. Work still waiting on a branch is not counted.

Two consequences. A month can look thin because a long piece of work landed just after it: read the quarter before concluding. And the count is about people: bots are excluded, detected coding agents are counted separately, and an undeclared agent counts as human work.

Reviewed and tested: what do the two shares mean?

“Reviewed by a colleague” has a strict meaning: someone who is neither the author nor a bot. For each pull request, DeployIt keeps who opened it, who reviewed it and how long it took; for each review, who gave it and its verdict, approved or changes requested (on GitLab, only approvals are known). A developer is therefore read in both directions: the reviews they receive, and the reviews they give.

“Tested” is a yes or a no on each change. Over a period, both become shares. Example: 71% of Camille R.'s merged pull requests reviewed by a colleague, 52% of their commits tested.

Neither share is a grade. Each is a question to bring to the conversation, and what counts as a code review is a subject in itself.

Why are defects “to confirm” before they are “confirmed”?

DeployIt counts a defect when a fix leads back to the commit it repairs. Counted by the machine alone, it is “to confirm”. An administrator, or an automated investigation that rereads the code, can mark it “confirmed”; only an administrator can dismiss it.

Two reading rules follow. Use “confirmed” when the number carries weight, and treat “to confirm” as a list to look at. Then wait: defects show up after the fact, the last 60 days are not mature yet, and the count can still rise. The mechanism is described in how a fix finds the commit it repairs.

How do you read a code score from 1 to 5?

Each merged commit is read by a language model and scored on five criteria: correctness, maintainability, robustness, tests and security. The figure you see is an average, which is why it always comes with the number of commits it covers.

Example: 4.6 over 3 commits and 3.9 over 251 commits are not the same kind of information. The score is read by month, quarter, half-year or year; a wider period puts more commits behind it. More in what a code grade per developer says, and what it doesn't.

What do these metrics not say?

They are drawn from commits, pull requests and reviews. DeployIt does not see a screen, messages or the time spent on a task. Design work, mentoring or an incident handled at night appear only if they end in a commit, a pull request or a review.

Context comes from what the company declares: absences and working time, so that a quiet month is not read as a drop.

No single figure is enough. The authors of the SPACE framework say it of developer productivity in general: one metric or one dimension cannot measure it (Forsgren et al., ACM Queue, February 2021; abstract on Microsoft Research, checked on 17 November 2026). Five facts read side by side are better than one. They remain facts about what was merged.

And a model can be wrong. A code score is an indication.

What framework goes with tracking each developer?

DeployIt tracks each developer, commit by commit. That is its purpose, and it comes with rules written in the terms of service:

  • the company informs the people it tracks before connecting its repositories, or turns statistics off;
  • a score set by a model is an indication, never on its own the reason for a decision;
  • the human decision prevails over a defect counted by the machine.

The statistics are read by the people to whom an administrator gave the Leadership profile, by the data tool the company connects, and by its administrators. The company can export everything DeployIt knows about a tracked person, and erase it. What a tracked developer can expect is on the privacy page.

Where do you read these metrics?

In Claude. A person with the Leadership profile asks in plain words; Claude calls stats_schema, then stats_query, a structured query and not free-form SQL. For the team over a period, the team_report prompt returns the same table every time: see the monthly developer report, asked of Claude. The datasets and the reading rules are in the docs.

What DeployIt does here

DeployIt is an MCP server. Your company gets one link, https://mcp.deployit.ai/<company>/<code>; each person adds it to Claude (on the web, on the desktop or in Claude Code) and signs up with a name and an email, which is not verified; their questions are logged under that name. Everyone starts with the Everyone profile: questions about the product, answered from the code; a filter removes the code, the file paths and the names of team members it recognises from the answers, without proving their absence, and a leak test tells you what still gets through. An administrator gives the Developers profile (excerpts) or the Leadership profile (developer statistics) to whoever needs it. Repositories are connected from GitHub or GitLab; DeployIt makes no commit, branch or pull request, and only adds one webhook per repository.

A working copy of the code is kept while a repository is connected, hosted by DigitalOcean in London (UK); at indexing, each source file is sent to the model to be summarised, then the passages needed for an answer and the diffs of the commits that are scored; OpenAI's models by default, or the provider whose key you bring.

The whole product is described in DeployIt: ask your code and follow each developer's work from Claude.

Next step: tell your developers, open the trial, connect your repositories and ask Claude for last quarter, developer by developer. Tracking starts at connection, with 6 months of history.

Price. Free plan: 15 questions a week for the whole company. A 30-day trial, no card. Plans from €290 a month excluding tax for up to 25 tracked developers: see pricing.

Ask your code from Claude

30-day trial, no card. Or try the demo first: add https://mcp.deployit.ai/demo to your Claude connectors.

Connect my code

Frequently asked questions

What are the best developer performance metrics?

The ones that count finished work and can be traced back to it. Read from merged commits, that gives five: what each developer delivered, the share reviewed by a colleague, the share that came with tests, the defects traced to their commits, and a code score from 1 to 5. Each is read with its limits, and the code score always with its number of commits.

Is the number of commits a good measure of a developer?

On its own, it only says what landed, not what it was worth. It becomes useful when only merged commits are counted, on each repository's landing branch and when they land, and when it is read next to review, tests, defects and the code score. In DeployIt, bots are excluded and detected coding agents are counted separately.

Can a code quality score be used to evaluate a developer?

As an indication. In DeployIt, each commit is scored from 1 to 5 by a language model on five criteria (correctness, maintainability, robustness, tests, security), and the average is always given with the number of commits it covers. The terms of service state that such a score is never, on its own, the reason for a decision about a person.

Do developers have to be told they are being measured?

DeployIt's terms of service say so: the company informs the people it tracks before connecting its repositories, or turns statistics off. It is for the company to determine the basis for this tracking under its own law. DeployIt provides the list of what is measured, and the export and erasure of everything it knows about a person.

Over which period should developer metrics be read?

By month, quarter, half-year or year. A month with few commits says little: read the quarter. Defects show up after the fact, so the last 60 days are not mature yet, and a closed period can still move when a commit lands late or a defect is dismissed.

Continue reading

Connect my code