Skip to content

Experts ​

This lens answers: who should I ask about this package? That isn't always the person with the most commits. Someone who wrote most of it five years ago and left knows less today than someone who changed it last month, so Experts counts recent work for more and suggests only people who are still committing.

Its biggest limit is that it sees nothing but commits, so "knows it" is an estimate from who changed the code and when. Reviewers, designers and anyone who reads the code without changing it don't show. A high share is a good place to start asking, not proof that someone understands the package.

Unlike most lenses, this one reads the repository's whole history, whatever --since says. See What --since changes.

Reading the numbers ​

What you seeWhat it means
"49% of what's known"This person's share of the package's changes, where each change counts half as much every six months. The shares of everyone who ever changed the package add up to 100%, including people who have left. It compares people, and doesn't say how well anyone knows the code.
A filled dotAn active person holds a share. The bigger the dot, the bigger the share.
A hollow dot, or "(quiet)"This person has made no commit anywhere in the repository in the 90 days before the newest commit analyzed. What they did still counts toward what's known, but they aren't suggested. See Who counts as quiet.
"Ask Rao 49%, Okoye 28%"The active people to ask, most knowledge first. Only people with 3% or more can be named.
"Most commits: Kowal (quiet)"Who made the most commits to the package over all of history, with no fading. It appears in a package's tooltip only when that isn't the person to ask.
"408 commits over all of history"The package's commits since the repository began, not just in the analyzed period, so it's usually higher than the counts on other lenses.
"1 package nobody active knows well"No active person holds 15% of what's known. This is the lens's badge, and one such package is enough to fire it.
"Ask Wei 7%" next to "Nobody active knows it well (15%+)"Not a contradiction. Wei is active and is the best there is, but holds too little to count as knowing it well.
"Nobody active to ask"No active person is among the people listed for the package: those with 3% or more, the top 8 at most. Someone active may have changed it a little, below that line.
"15%" on a package flagged as nobody knowing it wellShares are rounded for display. 14.7% shows as 15% and is still under the line.
"1 of 3 packages to ask someone other than the top committer about"The lens card's headline when nothing is flagged: packages where the first person to ask isn't the person with the most commits.
"1 person to ask, across all 3 packages"The card's headline when neither of the above applies. It's how many different people come first across the packages.
"Left out: 1 untouched for a year (Age's dormant packages) and the root and dot-folders (2)."Packages with enough history to rate that the lens doesn't show. The list isn't the same as Age's dormant packages: Age looks at what's in the code now and needs 5 or more source files, while this counts any rated package with no change in a year.
"The history before this window couldn't be read, so knowledge here covers the window only."Reading the older history failed, so every share is from the analyzed period alone, and the lens card says "this window only." Tooltips still say "over all of history," so trust this warning over them. Run it again with --fresh, and a longer --git-timeout if the read timed out.

The table's tiers describe who you can ask. They aren't scores:

TierMeaning
One to askOne active person holds 3% or more, and they have the most commits.
Several to askMore than one active person does, and the first has the most commits.
Not the top committerThe person to ask isn't the person with the most commits.
Nobody knows it wellNo active person holds 15%. The name is short for "nobody active": people who have gone quiet may know it very well.

Four example repositories ​

These repositories are invented, but each result is what GitRelic reports for commits like these.

A Go backend whose longest-serving author moved on ​

1 of 3 packages to ask someone other than the top committer about

In 1 of 3 packages the person to ask isn't the top committer: for services/billing, ask Deepak Rao (49%), not Marta Kowal, who has gone quiet.

Deepak Rao is first to ask about 2 packages, more than anyone.

The table's row for that package:

PackageCommitsAskTheir share %AlsoMost commits
services/billing408Deepak Rao49Okoye 28%Marta Kowal (quiet)

What it means: Marta made 310 of the package's 408 commits, 76%, but Marta's last was 14 months ago. Fading leaves Marta with 23% of what's known. Deepak has 70 commits, and Sam Okoye 28, all recent enough to count for more.

What to look at: select the package. Its ridgeline says "ask Rao, Okoye · most commits Kowal (quiet)" and shows Marta's ridge falling while the others rise. If Marta is still at the company, they may be the right person to ask for the reasons behind old decisions. Quiet only means they stopped committing here.

A Python repo that recently lost the people who knew a package ​

1 package nobody active knows well

services/ingest has nobody active who knows it well (15%+ of what's known): whoever changes it next learns it alone.

In 1 of 3 packages the person to ask isn't the top committer: for services/ingest, ask Chen Wei (7%), not Ana Lindqvist, who has gone quiet.

Chen Wei is first to ask about 2 packages, more than anyone.

Left out: 1 untouched for a year (Age's dormant packages) and the root and dot-folders (2).

What it means: two people held 93% of what's known, and both stopped committing in the last several months. Chen made 3 small commits recently, which is why the second line names them. Being first to ask at 7% means they're the only option, not an expert. The first line is the one to act on.

What to look at: the package is still changing, or it would have been left out. Decide who takes it on while Ana and Ben may still be reachable, and check Handoffs to see whether anyone has taken the package over.

A TypeScript project with one author ​

1 person to ask, across all 3 packages

Noor Haddad is the one to ask about every package.

What it means: every package reads 100%, and nothing is flagged. That doesn't make it safe: the lens is saying there's nobody else. Ownership's bus factor is the place to read that risk.

A Java repo where one person reformatted everything ​

2 of 5 packages to ask someone other than the top committer about

In 2 of 5 packages the person to ask isn't the top committer: for libs/search, ask Lars Dahl (36%), not Paolo Ricci, who has gone quiet.

Lars Dahl is first to ask about 2 packages, more than anyone.

Tooltips on the other three packages read "Ask Bello 78%, Dahl 22%" and similar.

What it means: Lars made 6 commits in the last month, each touching all five packages for formatting. That's 6 of each package's 86 commits, yet Lars holds 22% to 36%, because a commit counts the same however small it is, and those are the newest. In libs/search and libs/shipping, whose owner went quiet, Lars comes out first to ask.

What to look at: a column of dots on every row is a hint. Select one of the packages and Lars's ridge appears as a late spike. GitRelic can't tell a reformat from real work, so check those commits before asking Lars. The useful answer here may be that libs/search has no real expert, which the lens can't see, since Lars clears the 15% line.

How it's measured ​

GitRelic takes every commit on the analyzed branch, skipping merge commits and bots, and works out each package on its own.

How older work fades ​

Each change counts half as much every 180 days, measured back from the newest commit analyzed, not from today.

A change madeCounts for
Just now1
6 months ago0.5
1 year agoabout 0.25
2 years agoabout 0.06
3 years agoabout 0.015
5 years agoabout 0.001

For example, take a package with three people:

PersonChangesWhenCounts forShare
Mina1002 years agoabout 646%
Bo106 months ago538%
Cy2just now215%

Two things follow. Mina made 89% of the commits and has under half of what's known. And a share only compares people: everyone fades at the same rate, so a package nobody has touched for months still adds up to 100%.

What counts ​

  • A commit counts once per package, whether it changes one line or a thousand. A commit touching two packages counts once in each.
  • The author gets the credit. Co-authored-by: trailers don't count here, so people credited only as co-authors get nothing.
  • Bots never count. AI agents' own commits do, and an agent can be named as the person to ask, such as "ask copilot-swe-agent[bot] (24%)." See what counts as a bot and an AI agent. A person's commit with an AI trailer counts as that person's.
  • Files that aren't anyone's work are left out: fixtures and generated files, unless you pass --include-fixtures, and project plumbing such as manifests, lockfiles, changelogs, licenses, CI and tool configuration, always. Tests count.
  • Identities are merged across all of history, so someone who changed email is one person, as long as GitRelic can tell (see Contributors). --no-merge-identities turns it off.

What counts as a package ​

  • Packages are the repository's declared workspaces, else packages/*, apps/*, libs/* or services/*, else the top-level folder.
  • A package needs 20 or more commits over its history to be rated, counted after the rules above.
  • It's shown while it has changed in the last year, measured back from the newest commit analyzed. A package nobody has touched in a year has nobody active by definition, and that's a question for Age.
  • The root and top-level dot-folders are left out. Those are the root's own files and folders like .github.

The layout is today's, and history is by path. Moves aren't followed, so old commits count toward wherever their files were at the time:

  • A package that moved starts again at its new path. Its old commits stay under the old folder, which can read as a separate package until a year after its last change. Until then the new path may have too few commits to be rated. See Renames.
  • A package that has been deleted still appears until a year after its last change, because nothing here checks that it still exists. Its deletion counts as a change. It can be flagged as nobody knowing it well, and "whoever changes it next" won't apply.

Who counts as quiet ​

Quiet here has one rule: no commit in the 90 days before the newest commit analyzed. Exactly 90 days is still active, and 91 is quiet.

  • Any commit counts as activity, anywhere in the repository, in any file. Someone who works in another package, or who only bumped a lockfile last week, is active, and can be suggested for a package they last touched years ago, as long as their share is 3% or more.
  • The clock is the repository's, not today's. It runs from the newest commit analyzed, which can be a bot's. On a repository whose last commit was two years ago, the people who worked in its last 90 days read as active. If only a bot still commits, everyone reads as quiet.
  • There's no "gone" on this lens. Someone quiet for 91 days and someone who left six years ago look the same: a hollow dot.

Quiet means something else on Contributors. Don't read one as the other:

On this lensOn Contributors
CutoffAlways 90 daysStretches with the analyzed period: 90 days at least, and a quarter of its length on long ones
StatesActive or quietActive, quiet or gone
Quiet meansNo commit in 90 days, however long ago the last one wasBetween the active and gone cutoffs, so "recent, but not lately"

Ownership and Handoffs use the same 90-day rule as this lens.

The two can disagree on the same screen. The person card uses the Contributors words, since it's shared across lenses. On a long --since, someone with a last commit 8 months ago can read "Active" on their card while this lens calls them quiet. Their card can also read "Gone" while the lens says "quiet." For who to ask, the lens's rule is the one that applied.

What --since changes ​

Almost nothing in this lens's answer. --since "3 months ago", the default 12 months and --since all normally give the same shares, the same people to ask and the same packages. GitRelic reads the history before the period for this lens too, as it does for Handoffs. The clock doesn't move either, because the newest commit is in every period.

There are two exceptions. A commit made at exactly the start of the period is counted twice, so a count can be one too high. And when a commit's date is older than the period's start but commits made before it are newer, such as after a rewrite that backdated a commit, git can stop reading at that commit on a run with a period, so the commits behind it can be missed, which can shift shares. It's rare and usually a few commits, and --since all reads them all.

What it means when you're reading the dashboard:

  • Someone with no commits in the period can still appear here. Selecting them shows a card with "Commits: none in this window" and their last commit date.
  • Numbers beside this lens stay period-bound. The time strip, a package card's File changes, and a person card's Commits cover the analyzed period only. So a package can read "408 commits over all of history" here and much less elsewhere. Neither is wrong.
  • "Most commits" here isn't always Ownership's top contributor, which comes from the analyzed period only.
  • You can't narrow this lens to recent history. Its fading already favors recent work. For the analyzed period alone, use Ownership, which follows --since.

"All of history" means what your copy has. A shallow clone has only the commits it fetched, and only commits on the analyzed branch count.

In the dashboard ​

Find this lens under Ownership & Knowledge.

  • The matrix is the main picture. Packages run down the left, by their last folder, with the full path in the tooltip. People run across the top. A dot's area follows the person's share. Rows are grouped by whom to ask first, so people who know the same code form blocks. Active people come first, and quiet ones sit past a line, marked "quiet: knew it, can't ask." A package's own dot in the accent color means nobody active knows it well.
  • What the matrix leaves out. It has the 24 packages with the most commits. A person gets a column by holding 10% or more of at least one of those, and then their smaller shares down to 3% show too. So a row can look emptier than its tooltip, which names people without a column. The table has every package.
  • Hovering a package shows its commits, whom to ask, the top committer when that's someone else, and the flag if it has one. A dot shows the person's share and whether they're "First to ask," "2nd to ask" or "Quiet 90+ days: knew it, can't ask." A person's name shows how many of these 24 packages they're first to ask about and their last commit date.
  • Select a person to light their column, which is everything they know among these rows. Someone without a column dims the whole matrix.
  • Select a package (its name, its row, a dot, a table row or search) to see its ridgeline, which replaces the matrix. There's one ridge per person, for up to 8 people who ever knew the most, drawn by quarter from the package's first commit to the newest commit analyzed.
  • Reading a ridge. Its height is how much that person's changes count for at that time, on one scale for the whole package. It climbs while they work and falls by half every six months after. It's an amount, not a share, so a falling ridge doesn't always mean a falling share. Quiet people have faded ridges and hollow dots, and the first three people to ask are named in the accent. The title says, for example, "ask Rao, Okoye · most commits Kowal (quiet)." Hovering says "Knew it most in 2025 Q2" and either "23% of what's known now" or, under 3% or outside the top 8, "Next to nothing known now."
  • The lens card, with nothing selected, leads with the headline number, then the findings quoted above. They're the warnings first, then where to ask someone else, then who is asked most, and what was left out.
  • The table (T) lists every package shown, most commits first: commits, whom to ask, their share, up to two more people, and who has the most commits. Packages with nobody active say "nobody active."
  • Anywhere in the dashboard, a package's card has a Who to ask row with the top two, such as "Rao 49%, Okoye 28%," in a warning color when nobody active knows it well. A person's card has a First to ask row with how many packages, when they're first on any.

When no package qualifies, the picture says "No package changed in the last year has 20 or more commits in its history." That's normal for a young or small repository. A report saved before this lens existed says "Rerun gitrelic to see who knows what."

What it can't tell you ​

  • Who understands the code. Changing code isn't the only way to know it, and not everyone who changes code understands it. Reviewers, whoever wrote the spec and people on call won't show.
  • How big anyone's work was. A typo fix and a rewrite each count as one change. Because recent changes count for so much, a repo-wide reformat or rename can put one person on every row.
  • Whether someone can be asked. Active and quiet come from commits alone. A quiet person may be down the hall, working in another repository. An active person may have left last week.
  • Whether someone active knows this package now. Being active is about the whole repository, and their share here can rest on work from years ago, if everyone else's is older still.
  • Who did the work behind a squash merge. A squash credits its one author for everyone's work, and merge commits aren't counted, so people who only merge get nothing.
  • Whether six months is the right pace. It's a fixed choice that applies to every package. A slower pace would favor long-time authors, and a faster one whoever touched it last. Stable code that nobody needs to change fades as fast as anything else.
  • Ownership — who made the most commits in the analyzed period, how much rests on one person, and the bus factor.
  • Handoffs — who took each package over from whom, also over all of history.
  • Contributors — who is active, quiet or gone by that lens's rule, and how identities are merged.
  • Age — the code nobody has changed in a year, which this lens leaves out.
  • Renames — what a move does to each lens, since this one doesn't follow files that move.

Released under the MIT License.