Closing the graphs, and re-verifying every separation over them

The governing conclusion, recorded first:

No semantic transformation can be verified over an open representation graph.

This is the rendering analogue of the projection ledger's requirement that every information-preserving claim name its observable object. An open graph is not incomplete coverage — it is a verification precondition that has not been satisfied. A separation measured over an open graph is unverifiable in principle, not merely unverified in practice.

---

Step 1 — closing the remaining graphs

Openness had exactly two causes. They resolved differently.

Animation — CLOSED

link.flow drives a useFrame loop; a single-frame snapshot could not see it. @react-three/test-renderer exposes advanceFrames, which makes it observable with no new dependency.

Measured: after advanceFrames(30, 16) a traffic mesh has moved from the origin to (11, 0.5, 13) — the link endpoint. Now a permanent test.

Html — NOT CLOSABLE with any instrument available

Measured, not assumed:

• Under the test renderer, Html renders nothing. The jsdom body.innerHTML is length 0; the label text appears in neither the scene graph nor the DOM. drei's Html portals into a canvas parent the test renderer never creates.
• The browser instrument cannot reach it either: the page rendering WorldScene is session-gated (getSession in app/page.tsx), and authenticating as the operator is outside what I may do.

What would close it, in order of honesty:

1. A browser-based instrument driven by the operator, holding a real session — the standing constraint makes session-gated probes the user's action.
2. Extracting the label-colour decision into an exported pure function, with the call site verified by parser. This is structural closure, not observational: it proves the function is correct and is called, not that the pixel is right. It should be labelled as the weaker grade, not counted as closure.

Closure has grades, and conflating them would repeat the substitution defect:

Grade	Meaning
Observational	the rendered output is measured
Structural	the carrier derives from a tested function whose call site is verified
Open	neither

---

Correction — the two instruments disagree, and the parser under-counts

The source parser reported status with one carrier (wireframe). Measured at the render, status has three: position.y, emissiveIntensity, wireframe.

The parser missed two because they are computed into variables rather than written as inline attribute expressions referencing .status. The render is authoritative for carrier counts; the parser is a lower bound.

Every carrier count in semantic-locality.md is therefore a floor, not a figure. The closure verdicts survive — all three status carriers are in the scene graph, so it remains closed — but the counts do not.

Corrected closure table

Fact	Carriers (measured)	Observable	Closure
activity	2	2/2	closed
status	3 (was 1)	3/3	closed
flow	2	2/2 (animation now instrumented)	closed
identity	≥4	3/4	open — span.style
health	≥3	2/3	open — span.style
blocked	≥2	1/2	open — span.style

Three of six closed, up from two. Every remaining open graph has the same single cause: span.style.

---

Steps 2 and 3 — re-running every separation over its graph

Separation	Graph	Verdict
unstaffed vs idle	status — closed	VERIFIED
identity vs alert (emissive)	identity — open	remains provisional
uncertainty (health=unknown → opacity)	health — open	remains provisional

The one that survives

Re-measured over the closed status graph:

steady      y=3.20  int=1.5   wf=false
idle        y=2.20  int=0.35  wf=false
unstaffed   y=2.20  int=0.35  wf=true    ← distinguished by wireframe alone

unstaffed and idle share two carriers and are separated by the third. Since all three are observable, this separation is verified, not provisional — the first semantic separation in this renderer to reach that standing.

The ones that do not

Both remaining separations touch facts whose graphs are open at span.style. Their earlier "success" was measured over the instrumented subset only, and the measurement could not have detected relocation into the uninstrumented carrier — which is precisely what happened to identity/alert.

They are not reported as failures. They are reported as unverifiable, and that distinction is the point: the evidence does not exist either way.

An unrelated collision, still standing

active and steady differ in 0 slots. This is the collision the reachability correction flagged: it matters only if status = active ∧ blocked = 0 is reachable, which remains unresolved and is not a closure question.

---

Status against the sequence

1. Close every representation graph — animation done; span.style blocked on an instrument I cannot build.
2. Re-run separations over closed graphs — done. 1 verified, 2 unverifiable.
3. Record which survive — done, above.
4. Legend amendments — not yet. Two of three separations remain unverifiable, and the Legend must ratify semantics that survived verification over closed graphs.

The single blocker for steps 1–3 is now one carrier, and closing it requires either an operator-run browser instrument or an explicit decision to accept structural closure as sufficient for that carrier. Both are the user's call, not mine.