This post shows how to use your git history to write a self-review with facts.
In the performance series I said team metrics do not belong in a review. This is different: your own numbers, used by you.
The problem
In a yearly review, people write about recent work and forget the first months. Without data, the manager fills the gaps with impressions.
A brag document fixes this. Julia Evans wrote about it. Most people stop writing it in February. Git already has the data: every PR has a title, a date, a size and a reviewer.
What a good claim looks like
A good claim has three parts:
- Claim. What you did, in one sentence.
- Number. A fact anyone can check.
- Story. Why it mattered.
Weak: "I helped a lot with code reviews."
Strong: "I reviewed 246 PRs from other people, up from 160 last year. Most were in payments-api. My median first review came 2.6 hours after a PR opened."
Numbers are not a score. Use them to find your stories.
Step 1: compare this year with last year
I use ShowOffer. It scans your GitHub activity and builds the tables below. The screenshots use sample data for a made-up engineer, Jordan Vale.
Look for the biggest changes. Here:
- Reviews given: +54%. Why? New people joined, or Jordan became a lead.
- Time to merge: -29%. PRs merge faster. Why?
- PR size: -21%. Probably the reason for the row above.
Each big change is a paragraph in your review. A drop is not always bad: fewer PRs can mean you wrote a design doc. Say so.
Step 2: look at each month
A total hides time. 112 PRs can be a steady year or one busy month.
- Peaks. What shipped in June? Open the PR list for that month.
- Dips. Holidays, on-call, onboarding. Write the reason down. A dip with a reason is fine.
Here this year is above last year almost every month.
Step 3: pick your stories
List your biggest PRs. Important work often leaves a big PR.
For each one, answer three questions:
- Before. What was the problem?
- After. What changed?
- Result. What can you measure? Take this number from your dashboards, not from git.
Lockfiles and generated files inflate line counts. Skip those rows.
Step 4: count the quiet work
Two numbers work well together:
- 2.2 reviews per PR. Jordan reviewed twice as many PRs as they merged.
- 2.6 hours review response. Median time to first review.
Write it as one sentence: "For each PR I merged, I reviewed two from teammates, usually within three hours."
"Reviews given" counts PRs, not comments. One PR with five rounds counts once.
What git cannot show
Git sees code. Keep a short list during the year for the rest:
- Design docs and RFCs
- Mentoring and onboarding
- Incidents and postmortems
- Glue work: planning, unblocking other teams, docs. See Tanya Reilly's talk on glue work.
Also remember:
- Private repos need a token that can see them.
- Squash merges and direct pushes change commit counts. PR counts are more stable.
- Pair work often lands under one name. Say so.
One page for the review
Attach one page with the key numbers. Write the stories in the form.
With ShowOffer:
- Open the report generator. Paste a read-only GitHub token, or click "Try with sample data".
- Pick the review period. Turn on compare with the previous period.
- Pick your orgs and repos.
- Add the widgets you need. Click "Fit chosen widgets".
- Download a PDF or PNG.
It runs in your browser. The token stays in memory and goes only to api.github.com. It reads titles, dates and counts, never your code.
You can do it by hand too. This GitHub search gives the PR count: is:pr author:YOU merged:2025-11-01..2026-10-31.
Checklist
- [ ] Every claim has a number or a link.
- [ ] You compared this period with the last one.
- [ ] You explained the dips.
- [ ] Each big PR has a before, an after and a result.
- [ ] Reviews and other quiet work are in.
- [ ] Non-code work has its own section.
Note
Jordan Vale and all numbers in the images are made up. I built ShowOffer, so I am not neutral. The method works without it.
Next week, part 2: the 1:1.
GitHub · LinkedIn · Substack · 4thwithme.dev/blog
May the --force be with you.







Top comments (0)