The project manager is where the register becomes work
Two layers, and the second is the one worth the page. The first is a position about what project management is. The second is a description of what actually happens here every day, which had been demonstrated across thousands of documents and described as a system precisely nowhere.
1. The position
“The project manager is where the register becomes work.”
A risk register that produces no change to what anybody does is a document, not a control. The claim is that risk management and project management are one discipline with two vocabularies: the register names what could go wrong, and the plan is the only place that naming turns into action. Split them across two roles and two artefacts and the register drifts into a compliance object.
Four instruments make the register operational rather than decorative:
| Instrument | What it does |
|---|---|
| A register per project | Not one organisational register. The unit of ownership matches the unit of delivery, so the person who can act on an entry is the person holding it. |
| A net score | Exposure after the controls that actually exist, rather than before. A gross score measures the world; a net score measures your response to it. |
| Erosion | Controls decay. A control credited at full strength two years after anybody checked it is a fiction, and erosion makes the decay a tracked quantity rather than a surprise. |
| The missing reference class | The estimate nobody has: how often did this kind of project, run this way, actually go wrong? Naming its absence is more honest than substituting a number for it. |
The full grounding vocabulary — fact, evidence, measure, vulnerability, risk — belongs to risks.sgit.ai and standards.sgit.ai, and this page links it rather than restating it.
2. How a brief becomes work here
This section is the original contribution. The estate's way of working is its project-management methodology, it is demonstrated across roughly four thousand documents as of 24 August 2026, and no page anywhere described it as a system until this one. It also doubles as onboarding: an agent session starting cold can read this and know how the work is shaped.
A brief, with acceptance criteria
Work starts as a written brief that states what would count as done. The criteria are numbered, which is what makes the later argument about whether it was done a short one. The discipline: a brief with no acceptance criteria is a wish.
Numbered asks and numbered tasks
Requests inside a brief are numbered, and so are the tasks they generate. Not decoration: a numbered ask can be referred to across documents, across sessions, and across teams without quoting it, and an unanswered one is visibly unanswered. This is what makes a document actionable by an agent rather than merely readable.
Same-day cross-team review
A brief produced by one team is read by another the same day. The speed is the mechanism: a review that arrives a week later arrives after the work has already been built on. Cost: it only works where the reviewing team is genuinely available, which is a staffing fact, not a process one.
Version-prefixed filenames and day-indexes
Filenames carry a version prefix, so a directory listing sorts chronologically without metadata. Each working day gets an index of what it produced. Together they turn a pile of documents into a timeline that can be walked backwards. This is the retrieval half of the memory thesis, implemented as a naming convention.
The debrief, classified
Work ends with a debrief, and failures are classified as good failure or bad failure — a good failure being one that bought information worth its cost. The classification is the point: without it, a team either punishes all failure or excuses all of it. Related: every resilience mechanism here traces to a named incident, which is what a debrief culture produces downstream.
The handover guide
A session that ends writes down how the next one starts, with one instruction above the rest: don't improvise. Written for people, load-bearing for agents, because an agent has no memory of the previous session and will cheerfully invent one. The clearest single piece of evidence that the memory thesis describes something already being done.
The system is not novel step by step. What is novel is that it is run, at scale, by a team whose members mostly have no memory between sessions — and that this is the first page to say so.
Counter-evidence
Three honest weaknesses in the system above.
- Nothing enforces any of it. Every step is a convention held by attention, which is the same finding as the documentation page and the testing page. The conventions that survive here are the ones a script checks, and none of these six is checked.
- The scale is stated approximately. Roughly four thousand documents, and the day-indexes and debriefs have not been counted precisely — which matters, because the scale is part of the argument. Counting them properly is on the proposed list, and until then this figure carries a tilde and a date rather than a claim of precision.
- Same-day review is the first thing to go under load. It is the step most dependent on people being free, and no version of this system has been tested at a point where they were not.