Project statistics

“Human-equivalent” estimates the effort a competent human team might need to deliver the observed changes, including implementation, review, debugging, and validation. It is not actual time worked, a measure of individual performance, or a claim that lines of code equal value.

The base estimate is 88461.39 hours, with a 44230.66–176922.78 hour uncertainty range. The current dataset uses heuristic. Generated, vendored, lock, and binary files are excluded from effective churn.

Those estimates measure churn — everything written, including what a later week rewrote. As an independent check, an offline census of the code that actually survived counts 992,042 lines across the tracked repositories, which at 10 finished lines per competent-engineer hour is 47.69 person-years of work. The weekly estimates come to 0.89× that anchor. The census is never folded into the estimate; it exists so a figure that drifts away from the artifact is visible rather than silent.

The chart covers the last six months and distinguishes time before the first recorded workflow from time after it. Uptime is proven two ways: by matching observations of the same engine process, and by records the daemon itself wrote — every gateway request, run event, worker spawn, and run start or finish is proof it was alive at that instant, and records closer than fifteen minutes are joined into one interval. Two consecutive absent observations prove downtime. What is left is silence, and silence is read by its length: a gap of twenty-four hours or more after the first workflow counts as likely downtime, a gap of eight hours or more as possible downtime, and anything shorter is treated as idle. Workflow starts remain visible even when a run interval is unusable; long spans and clustered recovery completion timestamps are rejected.

Last updated 2026-08-25T18:48:52Z · schema 4 · source ecf6a49c289ecaefb431