Harsh Patel

73 of 100, ranked 6 of 14 people with a score, against a cohort median of 72. The bands are cut from this period's own distribution, not fixed thresholds: ≤ 66 is behind, ≥ 73 is leading, and this person is in the middle 40% of people with a score this period. With every component at the cohort median this person would score 73; they score 73. That difference is carried by craft (66 against a cohort median of 97, worth -9.9 points) and review (0 against a cohort median of 15, worth -1.5 points). 1 of 6 components (completed work) could not be measured at all, so the remaining weights were rescaled — this score rests on 78% of the intended evidence.

A band compares this person to the 14 people who have a score in this window — nothing more. It is not a rating, it does not persist, and it moves when the cohort or the window moves.

Effect is a counterfactual, not a share: how many points this person’s score would move if that one component dropped to the cohort median for it. The effects sum to 73 − 73 = 0, where 73 is what this person would score with every one of their components at the cohort median. Negative rows are the reason the total sits where it does. A component with no data is withheld and the remaining weights are rescaled — it is never scored as zero, and it never counts against anyone.

Measured but deliberately not scored: 526.9 hours tracked, 272 commits, 626,833 lines changed, 10.7% activity, first-pass CI 22.2%. The first four measure motion rather than delivery and are trivially inflated. CI is excluded because it exists on a minority of repositories, so scoring it would rank people by which repository they work in. This score cannot see architecture, planning, mentoring, incident response or infrastructure work at all.

• ⚑ 626,833 lines changed in this window

  Rule: fires above 500,000 lines added + deleted — measured 626,833.

  What it does not mean: Churn is never scored, precisely because it is this easy to inflate. The flag exists to stop a large number being read as a large contribution. Check for a vendored directory or a bulk import.

What they actually did

• ec2 DeleteSecurityGroup×181 · 543 pts
• autoscaling SuspendProcesses×173 · 346 pts
• elasticbeanstalk UpdateEnvironment×85 · 340 pts
• cloudformation UpdateStack×76 · 304 pts
• autoscaling UpdateAutoScalingGroup×132 · 264 pts
• autoscaling TerminateInstanceInAutoScalingGroup×48 · 240 pts
• route53 ChangeResourceRecordSets×119 · 238 pts
• ec2 AuthorizeSecurityGroupIngress×75 · 225 pts

CloudTrail records the API call, not the intent. It cannot tell a considered change from one corrected a minute later, and it retains 90 days — so an empty earlier period means “not retained”, never “nothing happened”.