Skip to main content
    What we measured

    We build our own product through the Factory.

    Reign is built the way we ask customers to let us build for them. The work enters our own repositories as merge requests, passes our own approval gate and is reviewed by named people. That gives us a small, honest dataset about our own engineering. It is the first dataset we will show you, and it is the one we are most careful about describing.

    Read the denominatorHow that month ranWhat we do not measure

    The dataset · one month, our own repositories, our own engineers

    Three numbers, and what sits under them.

    These are the only figures we publish. The denominator is inside this box with them, because separating a number from the population it was drawn on is how a measurement becomes a claim. It is part of the figures, not a footnote to them.

    326
    Merged merge requests across eight of our own repositories.
    19 h
    Median time from merge request open to merge.
    326 of 326
    Merged with a recorded approval from a named iTmethods engineer.
    Read the denominator. Measured on our own engineering, 11 July to 11 August 2026, in GitLab. The repositories are the Reign gateway repository and seven satellites. The population is merged merge requests only: requests opened in the window and closed, abandoned or still open when it ended are outside it, and we do not publish that count, so read 326 as a volume rather than a success rate. Release commits are excluded. Automated approvals are filtered out of the approval count, so the approval figure reflects people. Internal environment only, and the work was done by a two-person team. Your own conversation does not start from these numbers. It starts from your own baseline, agreed with you before anything runs. This is a dated sample, not a trend. When the next month is measured the same way, it will replace this one.
    How that month ran

    Two people directed. The Factory built. A named person approved.

    Of the 326 merged requests, 228 were planned and validated by engineers and built through the Factory. 98 were carried by the Factory as far as the merge gate. Every one of the 326 still needed a recorded approval from a named iTmethods engineer.

    Composition of one month

    Two routes in. One gate out.

    Where each of the 326 came from, and the one thing all 326 had in common.

    One month, and where each of the 326 came from 228 planned and validated 98 carried by the Factory The merge gate named approval 326 merged Every one of the 326 needed a recorded approval from a person The Factory does not hold the gate. It never has. 11 July to 11 August 2026 · eight of our own repositories · two engineers Merged merge requests only. Read as a volume, not a success rate. A dated sample. When the next month is measured, it replaces this one.
    The composition of one month of our own engineering. Not a forecast, not a service level, and not a comparison with anyone else.
    Not a forecast, not a service level
    This is the composition of a single dated month of our own engineering, and it is not a comparison with anyone else. The split between the two routes moves with the work. Your own conversation starts from your baseline, agreed before anything runs.

    Agents and models build

    Implementation ran through the Factory: agents and models doing the writing, inside an isolated worker, with Reign Gateway on the model path.

    A human stays in the loop

    Nothing merged without a named approval. The Factory does not hold the gate. Your reviewers would not either, if this were your estate.

    The path is operated, and it changes

    Reign Factory is in beta and being productized. It is not generally available; beta access is agreed with us and scoped in writing before anything runs. What follows here is our own engineering rather than a customer result. What the Factory may pick up, and how it hands over, moves with customer feedback. It is not a static benchmark.

    Comparison

    This is not a benchmark.

    We are not claiming an industry position. We are describing one month of our own work.

    01The windows are different
    Published industry figures usually measure a wider window. First commit to release is a common one, and it takes in planning, queueing, environments and release scheduling. Our figure measures merge request open to merge. That is a narrower window inside the wider one. Placing the two side by side would be misleading, so we do not.
    02The population is small and it is ours
    Eight repositories, one month, one product, two engineers who know the codebase well. That is a favorable population by construction. It tells you the delivery path works and that the approval gate held. It does not tell you what happens in an estate with more repositories, more reviewers and more constraints.
    03Your numbers will differ
    They will differ because your review capacity, your change process, your codebase age and your risk posture are different. We expect that. This is why we agree the measures first, record your baseline before work starts, and report against that baseline rather than against ours.
    Intent

    What we do not yet measure.

    These three measures are not instrumented today and we publish no figures for them. We intend to measure them, and we would rather state the gap than let the silence read as a result.

    Refusal and handover rate

    How often the work stops and hands over to a person, and why. We report stops inside an engagement. We do not yet publish a rate across a population, and we will not until the definition is stable.

    Would you trust a system that never reported a stop?

    Rework

    How much merged change is amended or reverted afterwards. This is the measure that separates volume from value. It is not measured yet, and any volume figure should be read as incomplete until it is.

    How would you know today whether merged work stayed merged?

    Size of change

    The distribution of change size behind a count of merge requests. A count treats a one-line edit and a substantial refactor as equal. We do not publish that distribution yet.

    What would a count of changes hide in your own repositories?

    Stated as intent, not as fact. None of the three is measured today. Where each motion actually stands is on the product status register.

    These are our numbers, not your forecast.

    Start from your baseline, not from ours.

    Bring one repository and one population of work, and we will agree the measures before anything runs. What we report back will be your numbers, described with the same qualifiers we apply to our own.