Discarded work

What it measures

The weekly share of human PRs that reached an end state and were closed without merging, out of all human PRs that ended that week (merged + closed-unmerged). Weeks with no ending PRs show a gap, not a zero.

Why it matters

Work that never shipped is invested effort that leaves no trace in throughput. A persistently high share usually points at unclear scope, feedback arriving late, or duplicated work — all fixable at the team level, and all invisible without this view.

How to read it

• Read multi-week patterns, not single weeks — one closed PR in a quiet two-PR week reads 50%.
• The first week of the window is partial, so its point is not comparable.
• A nonzero floor is normal and healthy.
• Weeks with no ending PRs show a gap, not a zero.
• The drilldown lists the closed-unmerged PRs, longest-open first.

Anti-blame

Common misreads

• "Closed = wasted" — spikes, prototypes, supersession, duplicates found late, and dependency-blocked work all close healthily.
• A low share is not automatically good — it can mean nothing is ever abandoned, even when it should be.
• This counts PRs, not effort — a one-line PR and a two-week PR weigh the same.
• A closed PR can be reopened and later merged — when that happens it leaves this view, so an old week's number can change.
• Not every close is a decision: deleting a base branch auto-closes every PR targeting it, and work rebased or pushed outside the PR also records as closed-unmerged. Both inflate the share without anyone abandoning anything.