Skip to main content

Value points

A value point is the platform's unit for delivered software. It is a score of finished, shipped work, normalized so that a point represents comparable delivered value whoever wrote it and whatever language, editor, or AI tool produced it — differences in raw size or verbosity between languages and tools do not change what a point means. That is what lets you compare output across people and tools, and divide it by what it cost.

  • Deployment: SaaS, On-premises

As a rough calibration — the scale the model is anchored to, not a benchmark to hold a team to — an average engineer using traditional development delivers one to two value points in a week. Where AI analysis is unavailable — no AI analysis provider connected, or content analysis off — the product reports output points instead: the same unit and scale, but computed from size and static signals rather than from the content of the change, and less accurate as a result. Reports label which one you are seeing (VP/$ versus OP/$). On SaaS, content analysis is not enabled by default, so output points are what a new organization sees first.

A value point is read from delivered work, not from effort​

Value points are scored per pull request, on code that shipped and stayed shipped. Drafts and abandoned work earn nothing. This is deliberately different from the counts most teams already have: lines of code measure how much got typed, pull requests measure how the work was chopped up, and story points are a guess made before the work started. Value points measure what the finished work was worth.

Every score is inspectable. The per-branch scores behind any headline number can be read in developer traces, and previous scoring policy versions are preserved when you save a change, so historical scores stay traceable (see Tune value points).

Human and AI work are split by evidence​

Every kept line is tagged at capture time as AI-written or human-written, based on session evidence from the AITrax tracker rather than self-reporting. A pull request's value points are split between agentic work — done through an agent, CLI, or MCP tool — and in-editor work — human-directed work in an IDE, with or without inline AI — in proportion to who wrote the kept lines. The proportion only attributes the pull request's already-scored value between the two sides; the score itself comes from the analysis of the finished change, not from counting lines.

Value per dollar divides output by what it cost​

Value points are the numerator of the platform's efficiency metrics. Value per dollar (VP/$) divides value points by the cost of producing them, and the platform computes it at two scopes. The overall figure divides all value points by combined AI and engineering labor cost — the headline return read. The per-side figures divide the value points credited to each side of the work split by the cost attributed to that side; the AI adoption report's acceleration factor is agentic VP/$ divided by in-editor VP/$. Both scopes answer the same question: whether the AI investment is paying off.