github profile for fresher resume — the 12-minute audit, with the 7 signals recruiters actually open
recruiters spend 12 minutes on a github profile before they decide to shortlist. 7 signals drive 90% of the shortlist decision. here is the audit we run on every dreamclerk applicant's github, and what it tells us to fix.
in cohort 2 (n=287), we ran a 12-minute github audit on every applicant who put a github link on their resume. 213 of 287 had a link. the audit covered 7 signals. we correlated each signal with the binary shortlist decision.
the 7 signals, in order of effect size:
- at least 1 merged PR to a public repo (effect size: 0.34). the artifact is a force-push, a review thread, a maintainer's "lgtm", and a merge commit. without the artifact, the github profile is "code i wrote alone." with the artifact, it is "code a stranger thought was good enough to ship."
- a pinned repo with a 3-paragraph readme (0.29). the readme says: what the repo does, how to run it, and what the author would do differently now. the third paragraph is the part most readmes skip. it is also the part the rubric scores.
- a profile readme (0.21). 41% of cohort 2 applicants with a github link had a profile readme. of those, 67% passed the shortlist. of those without, 49% passed. the readme is a 6-minute investment that moves the needle.
- the most-recent commit date on the default branch is <90 days old (0.18). recency is a signal of momentum. an old default branch with a recent PR open on a side branch scores lower.
- commit messages are imperative and specific (0.14). "fix: index choice for the search endpoint" beats "wip" or "update". the audit reads the last 20 commit messages on the default branch.
- no abandoned first-attempt repos on the first page (0.11). archive the repos with 1 commit and no readme. the recruiter sees the first 6 repos in the grid; if 4 of them are empty first-attempts, the profile reads as a learner, not a finisher.
- the github username matches the resume name or li handle (0.06). a small signal, but a real one. a handle that looks like "darkcoder99" or "anonymous-2023" reads as throwaway.
the 7 signals together explain 90% of the shortlist decision. the remaining 10% is a noise floor (recruiter mood, time of day, JD pressure). the audit is what you can move.
the 12-minute audit, step by step
if you have 12 minutes, this is the protocol.
minute 0–2: pin the 4 right repos. the shape is: 1 capstone, 2 small-but-finished, 1 fork-with-a-merged-pr. the capstone is the biggest one. the 2 small-but-finished are the ones with tests, a readme, and a deployment link. the fork-with-a-merged-pr is a public repo you have a merged PR in. archive the rest.
minute 2–5: write the 3-paragraph readme. the first paragraph is "what this is." the second paragraph is "how to run it" with a one-command setup. the third paragraph is "what i would do differently now" — and this is the load-bearing paragraph. it shows the rubric's revisit dimension.
minute 5–8: write the profile readme. 6 short sections: who you are, what you are building, what you are looking for, what you have shipped, what you are reading, how to reach you. no emoji, no "passionate about", no "quick learner". the profile readme is the most-skimmed doc on your github; write it like a 1-page cv.
minute 8–10: clean the commit messages on the most-recent branch. if your last 5 commits are "wip", "fix", "update", "wip2", "final" — fix them. git rebase -i HEAD~10 lets you rewrite them. do not rewrite history on a public branch; do it on a feature branch and merge.
minute 10–12: check the most-recent commit date. if the default branch is >90 days stale, ship a 1-line PR — even a typo fix in a readme. the recency signal is the cheapest of the 7 to move.
what the audit does not cover
the audit does not look at:
- the contribution graph. nobody looks at the graph.
- the streak count. the streak count is a vanity metric.
- the number of followers. the follower count is noise.
- the number of stars. the stars are a popularity signal, not a quality signal.
- the "highlighted" line on the profile. the highlighted line is a distraction; remove it.
the audit is a 7-signal rubric, scored in 12 minutes. the rubric is published. the audit is reproducible. if you do not pass the audit, the fix is mechanical and takes 12 minutes.
what we got wrong in cohort 1
in cohort 1, we did not run the audit before the shortlist. we ran it after. of the 31% who passed the resume screen, 8% had a github profile that would have failed the audit. those 8% would not have been shortlisted in cohort 2. the audit moved the shortlist decision by 11pp, in our data, and the cost was 12 minutes per applicant.
the audit is now a pre-shortlist gate. if your github does not pass the audit, we send you the protocol and ask you to ship 1 PR. the median time-to-fix is 4 days. the median callback lift is +11pp.
the 4-repo shape, in detail
the 4-repo shape is the single highest-leverage move on the audit. the shape:
- the capstone — a real, end-to-end project with a deployment link, a test suite, a readme, and a 3-paragraph "what i would do differently now." the capstone is the repo the recruiter opens first. if the capstone is a tutorial-clone (a todo app, a weather app, a chat app), the recruiter closes the tab.
- the small-but-finished — 2 repos that are <500 lines, have a clear scope, a clear readme, and a clear "what i would do differently now." these are the repos that prove the capstone is not a one-off. a finisher has 2 small-but-finished. a sprinter has 1 capstone and 9 abandoned first-attempts.
- the fork-with-a-merged-pr — a public repo you have a merged PR in. the recruiter reads the PR thread. the thread is the artifact. a maintainer's "lgtm" is the most credible signal on the profile.
the 4-repo shape takes 30–60 hours of work to build from zero, and 12 minutes to audit. the audit is free. the building is the work.
— ananya subramanian, chennai, july 2026