Code quality per developer can be read from what each person merges: in DeployIt, a language model reads each merged commit and scores it from 1 to 5 on five criteria (correctness, maintainability, robustness, tests, security). A developer's code score is the average over a period, always shown with the number of commits it covers; it is an indication, never on its own the reason for a decision about someone.
This page is written for the CTO or the CEO about to read that number for the first time. It covers how the grade is built, how to read it, and where it stops.
What is each developer's code score this quarter, and on how many commits?
Here is the quarter, developer by developer. Léa M.'s score covers 11 commits only: ask for the half-year before you read it.
| Developer | Merged commits | Code score |
|---|---|---|
| Camille R. | 132 | 3.9 |
| Idriss B. | 104 | 3.2 |
| Léa M. | 11 | 4.4 |
How do you measure code quality per developer?
The usual quality tooling describes a codebase: test coverage, lint warnings, complexity. None of it belongs to a person. The other usual source is memory: what a lead recalls from the reviews of the last few months, which favours the recent and the loud.
The unit of work that carries an author is the merged commit. DeployIt turns each one into a fact credited to that author: when it landed, its pull request, its tests, its ticket, its module, its score and its five criteria. The fact is written once per merged commit, and the statistics are updated every night.
The score itself is a reading by a language model. The diff of the commit is sent to the model, OpenAI's by default or the provider whose key you bring, and comes back scored from 1 to 5 on the five criteria. Only work that reaches the landing branch of each repository counts. Bots are excluded, detected coding agents are counted separately, and an agent that is not declared counts as human work.
One number does not sum up a developer. The authors of the SPACE framework (Forsgren and co-authors, ACM Queue, 2021) write that developer productivity cannot be measured by a single metric or dimension (abstract on the Microsoft Research site, checked on 17 November 2026). The code score is one dimension: what the merged code looks like. DeployIt keeps the others beside it: what was shipped, reviewed by a colleague, tested, fixed since.
What do the five criteria look at?
| Criterion | The question behind the word |
|---|---|
| Correctness | Does the change do what it set out to do? |
| Maintainability | Can the next person read it and change it? |
| Robustness | What happens when an input is wrong or a dependency fails? |
| Tests | Did tests come with the change? |
| Security | Does it open a door that was closed? |
These are the everyday meanings of the five words, given here to help you read. The definition DeployIt applies to each measure is the one the stats_schema tool returns: it is the reference, and the datasets are listed in the docs.
Five criteria matter because two developers with the same score rarely have the same strengths. Example: Idriss B. is at 3.2 over 104 commits. If tests is the criterion that pulls the average down, the conversation is about how the team writes tests, and it is a short one. If it is correctness, it is a different conversation.
So the second question to put to Claude is always: “which criterion is lowest?”
Why read the score over a period, with its number of commits?
Because it is an average, and an average of few things is fragile.
Take the example above: Léa M. has 11 commits in the quarter. Three commits scored 2 instead of 4 would lower her average by more than half a point. The same three commits would move Camille R.'s average, built on 132 commits, by less than five hundredths. Same event, two very different effects.
So DeployIt never returns the score without its count, and reads it by month, quarter, half-year or year. Three rules follow.
- A low count calls for a longer period. Ask for the half-year when the quarter holds a dozen commits.
- A commit counts when it arrives. A commit dated earlier but merged this month counts for this month.
- A closed period can still move. A commit that landed late, or two identities merged into one person, changes a past figure; every summary carries the time it was last computed.
Absences declared for a person are part of the statistics too, so a thin month is not read as a weak one.
What does a code grade say, and what doesn't it?
| It says | It does not say |
|---|---|
| How a model read each merged commit of the period, on five criteria | Whether the feature was the right one to build |
| The average of those readings, and how many commits it covers | How hard the task was, or how old the module it touched |
| Which criterion sits below the others | What the person did outside merged code: design, help given to a colleague, an incident handled |
| Where to look: every row links back to its commits | Whether a commit caused a defect |
The last line deserves a word. A defect is a separate fact: DeployIt counts one when it finds the commit that a fix came to repair, and the defect stays “to confirm” until an administrator or an automated investigation has confirmed it. A high score with confirmed defects exists, and so does a modest score with none. Read the two together.
Why is the grade an indication, never the reason for a decision?
A model can be wrong about a commit. An average hides its spread: 3.2 can be a hundred sound commits and a handful of weak ones, or evenly middling work, and only the commits tell which. And a commit does not carry its context: the deadline that moved, the old module nobody wanted.
DeployIt's terms of service say so. Article 8 states that a score given by a model is an indication and must never be, on its own, the ground for a decision about a person. The same article sets the framework. DeployIt tracks each developer; during the trial or on a plan, tracking starts when the repositories are connected, with 6 months of history; the company informs the people it tracks beforehand, or turns statistics off.
The company can also export everything DeployIt knows about a person, and erase it. The short version is in the questions we get asked, and what a tracked developer can expect is on the privacy page.
That gives a way of working:
- Read the score with its count, over a quarter at least.
- Ask which criterion is lowest.
- Open two or three of the commits behind it: every row links back to its own. With the Leadership profile, the answer comes without code or file paths (a filter removes them), so the commit itself opens in GitHub or GitLab.
- Bring them to the conversation, as a question. A 1:1 anchored in what was delivered is the right place.
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, profile by profile, is described in DeployIt: ask your code and follow each developer's work from Claude.
Next step: the demo answers without statistics, so your own scores start with the 30-day trial. Inform your team, connect your repositories, add your company link to Claude, ask an administrator for the Leadership profile, and put the question at the top of this page.
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.
Frequently asked questions
How do you measure code quality per developer?
Score the unit of work that carries an author: the merged commit. In DeployIt, a language model reads each merged commit and scores it from 1 to 5 on five criteria: correctness, maintainability, robustness, tests and security. A developer's code score is the average of their commits over a month, a quarter, a half-year or a year, always shown with the number of commits it covers.
Can an AI model grade code quality reliably?
A model reads every commit against the same five criteria, which no reviewer can do across a whole team. It can also be wrong about a given commit. That is why DeployIt's code score is an average read over a period, with its number of commits, and why the terms of service call it an indication that must never be, on its own, the ground for a decision about a person.
How many commits does a code score need to mean something?
DeployIt shows the number of commits next to every score so that you can judge. As an example, 4.4 over 11 commits and 3.9 over 132 commits do not weigh the same: three weak commits move the first by about half a point and leave the second almost untouched. When the count is low, read a longer period: the half-year instead of the quarter.
Can a code quality score be used in a performance review?
As one indication among others, with the commits behind it on the table. DeployIt's terms of service (article 8) state that a score given by a model must never be, on its own, the ground for a decision about a person, and that the company informs the people it tracks before connecting its repositories.
Who can see a developer's code score?
The people an administrator gave the Leadership profile to, a data tool connected with a Data tool key, and the administrators of the company. The Everyone and Developers profiles get no statistics. The company can export everything DeployIt knows about a tracked person, and erase it.
Continue reading
Find the commit that introduced a bug: follow the fix back to it
For a CTO and their developers: how a fix leads back to the commit it repairs, why the defect it reveals stays “to confirm” until an administrator or an investigation confirms it, and why the last 60 days are never final.
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.
Developer metrics in your data warehouse: one key, two MCP calls
For a data team: how to load per-developer statistics into a warehouse or a BI tool with a Data tool key. The request and its header, stats_schema then stats_query, the row limit, the cursor, and what to reread on each pass, shown as the trace of a call.