How DevGhost estimates effort
Last updated June 6, 2026
DevGhost estimates the cognitive effort behind code changes — how hard the work was, not how many lines changed or how long someone sat at a keyboard. The estimate is expressed in the hours a mid-level developer (3–4 years) who knows the codebase and works without AI would need: writing the code, testing it by hand, and fixing it in review. It deliberately excludes meetings, planning, and waiting for review.
A pipeline, not one AI call
Effort is not the output of a single model call. First, a language model reads the actual change — what was added, removed, and restructured — and judges its difficulty for the reference developer, rather than counting lines or commits. On top of that runs a deterministic, rule-based layer that keeps any single model guess from swinging the result.
The deterministic layer
The rule layer does the work a careful reviewer would:
- Classifies the nature of each change and recognizes high-stakes work — infrastructure, data migrations, security — separately.
- Filters out mechanical and generated changes: mass find-and-replace, generated or moved code, and formatting.
- Applies sets of correction rules and guardrails so one model guess can’t swing the result.
- Breaks large and combined commits down in more detail instead of scoring them as one lump.
Spreading effort over time
A single commit's effort is spread across up to five working days and capped at five productive hours per day, so one large merge doesn't distort a single day's picture — and the daily figures that feed Ghost% stay realistic.
Consistency and calibration
The same standard is applied to everyone automatically. Each commit is evaluated once and the result is fixed, which is what makes the numbers comparable and reproducible.
The algorithmic layer encodes empirical patterns from real-world enterprise development — which changes cost more than they look, and which are cheap despite their size — and those rules are checked against reference estimates. The system behaves less like a line counter and more like an experienced tech lead assessing the work.
From effort to Ghost%
Daily effort is then compared to the Ghost norm — three productive hours per working day — to produce Ghost%, the team's output relative to a pre-AI baseline. The estimates are probabilistic and are not a measurement of any individual.