Scores tended to be compressed. Most work landed in a narrow band in the middle, so a trivial change and a hard one looked closer than they actually were. The new model fine tunes our parameters to widen things out.
How effort is scored#
Three inputs feed the score. The first is structural — how big the change actually is. Lines, files touched, how nested the code is, how spread out the work is across the codebase. File types are weighted differently too: docs count less than application code, for example.
The second is cognitive — how hard the code is to follow once you sit down with it. Nested loops, branching, decision points. Part of this read is done by a model that looks at the actual code rather than just counting symbols.
The third is semantic. A bug fix and a chore and a new feature aren't worth the same, so we apply a small adjustment based on what kind of change it is.
What to expect#
- Low-effort work scores lower than before. Small mechanical changes read as small.
- High-effort work can exceed the old ceiling. Genuinely hard changes aren't capped anymore.
- The middle thins out, so the score is a clearer signal of how hard something actually was.
- All effort metrics after June 1st 2026 are now using the new calculation, this means we recalculated the effort of the first 5 days of June.
If you want more detail around these changes or how we handle effort feel free to reach out to support@mesmer.co.