Gaps, open questions and honest tensions
The build order for this site was published unresolved, and so is this. A reference site for a primitive should be easiest to argue with at the points where its own authors are least sure, which is what this page is for.
1. Six gaps: what this site needs built
| Item | Why it matters | |
|---|---|---|
| G1 | The naming ruling done | The corpus flagged the collision and left it unresolved. The site could not use the word until this page existed. Ruled → |
| G2 | A second working twin | One proves the interface claim. The pattern claim needs two. The AWS IAM twin is closest: finish it and publish it |
| G3 | The coverage measure, computed once | Provisions-without-twins over the AI Act is fully specified and has never been run. One number makes the discipline page concrete, and standards.sgit.ai wants the same number |
| G4 | The 2FA demo | The designed milestone for actors, the fractal build and the change pipeline. Everything downstream labels itself against whether it exists |
| G5 | A twin schema | "properties, behaviours, functions, inputs and outputs" is stated everywhere and typed nowhere. One Schema__Twin in the house style moves the definition from prose to code |
| G6 | The name-watch check done | A grep for new senses of the word on every build. Cheap, and the collision already happened once. How it runs → |
Two of the six are closed by this site's first release, and both are the cheap ones. G2 and G3 are the two that would change what the site can honestly claim, and G3 is the cheapest high-value artefact available anywhere in this material.
2. Five open questions
Q1. How connected is a twin, really?
The agent-twin brief carries the objection itself: "the standing question of how connected the twin actually is to reality."
Where it stands. Connected-or-not is binary in the model, and that is precisely what makes coverage computable. Freshness, latency and fidelity are not binary and are not captured. A twin updated "manually once a week" is connected, and on a Friday it is six days stale. The discipline needs a staleness dimension and does not have one, which means today's coverage number would treat a live API and a weekly spreadsheet as identical.
Q2. Who may update a twin?
Twins act, and twins are updated, and nothing states who is authorised to change a twin's definition. That definition is now the graph's contact with reality: change it and every measure grounded on it changes meaning without any measure changing.
Where it stands. The broker composition suggests the answer, in that a change to a twin could itself require a mandate and produce a receipt. Nothing states it.
Q3. Does the discipline of reality survive simulation?
The world model runs counterfactuals, which are attacks that have not happened. But "everything has to exist."
Where it stands. The likely answer is that simulated states are marked as projections from real twins, never as facts, and can never be cited as evidence by a measure. That rule has not been written, and until it is, a world model built on this graph is a mechanism for the exact pollution the discipline exists to prevent. The tension in full →
Q4. Is the broker's concentration risk acceptable?
It "becomes the highest-value target in the estate because it must hold usable credentials, which inverts the catastrophic failure property the rest of the architecture depends on."
Where it stands. The corpus states the cost and does not decide, and neither does this site. This is a real architectural fork: the whole estate is built on never concentrating credentials, and the broker deliberately concentrates them because that is the only way to keep them out of the agent's hands. Somebody who owns the estate has to choose, in the open. The cost, stated →
Q5. When is a twin done?
"even one specific use case can become massive once you capture all the elements", and the brief that says so offers no stopping rule.
Where it stands. The grounding ladder's stopping test probably applies: "the last node where going deeper would neither improve observability nor change a decision." Somebody should say so explicitly, because without a stopping rule the discipline of reality has no natural boundary and a twin can absorb an unbounded amount of work.
3. Five honest tensions
- One working twin, nine applications. The primitive is load-bearing across five sites and demonstrated once. Publishing that ratio is the credible move, and the S3 container carries a great deal of weight until G2 lands.
- The discipline of reality against the simulation ambition. The most distinctive rule in the cluster and its most ambitious design pull in opposite directions. Q3 is where they meet.
- "Connected or not" is the powerful claim and the oversimplified one. Binary connectedness is what makes coverage computable. Q1's staleness dimension is what it hides.
- The broker inverts the estate's core property, on purpose. Publishing that as a decision to be taken is more honest than presenting the design unqualified.
- The best material arrived last and least verified. The 19 August pair is 8,488 words, converted from an external session, with market research "as supplied". Strong ideas, weakest provenance in the cluster, labelled accordingly wherever they appear.
4. Loose ends worth an hour each
- Verify the S3 container's current state in the sg-compute repository. The vault specs shipped; confirm the S3 face specifically.
- Check whether the "digital twins already built" that the AWS engine reuses are on disk, and where.
- Run the name-grep across the whole estate once, not just this site. There may already be a third sense of the word out there. The ruling this would test →
- Confirm the 2FA scenario's public framing stays clear of the February governance records.
- Ask whether the broker market research's vendor claims were ever checked. Nothing from that file is quoted here until they are.
5. The build order, published as given
The pack that commissioned this site specified an order, and instructed that it be published unresolved rather than presented as completed work. It ran: what is a twin, then the discipline, then built, then actors, then naming, then simulation, then shipped. That is the order this site was written in, and the order the navigation follows.