Perceptual information budget — and two corrections

Measurement. Classifies every apparently-invisible fact by whether it is genuinely independent, then measures the remaining carrier bandwidth and the resulting deficit.

---

First: two errors in perceptual-state-space.md

Both are the same class — the sweep varied field values that cannot occur — and both inflated the deficit.

Error 1 — building.blocked is not a fact

Building has fourteen fields: id label shortLabel kind districtId position footprint height accent activity health capacity evidence detail. There is no blocked. The fixture added one, the renderer ignored it, and it was recorded as an invisible fact. It is not invisible; it does not exist.

(blocked exists on CityNode, the HQ view model consumed by CityMap — a different type in a different layer.)

Error 2 — agent status was varied independently of its own inputs

Measured in build.ts:

status = unstaffed  if placement.kind === "unstaffed"
       | active     if blocked > 0
       | steady     if queueDepth > 0
       | idle       otherwise

status is fully derived. Varying status = active while holding blocked = 0 produced a state composeWorld can never emit, and the resulting "active ≡ steady" collision measured nothing real.

Re-measured over reachable states only — 4 distinct of 6:

Reachable state	Distinguishable
idle (b0 q0)	✓
steady (b0 q3)	✓
steady (b0 q9)	✗ collides with q3
active (b5 q9)	✓
active (b1 q1)	✗ collides with b5
unstaffed	✓

The status vocabulary is fully distinguishable. All four categories render distinctly. The real remaining gap is magnitude within a category: a viewer sees blocked but not how blocked.

The lesson generalises: a state space must be enumerated over reachable states, not over the Cartesian product of field values. Half the deficit I reported was unreachable states colliding with each other.

---

Classification of the genuinely invisible

Fact	Category	Evidence
agent status	Derived	computed from placement + blocked + queueDepth; all three are categorically visible, so it needs no carrier of its own
agent load	Redundant	a priority-weighted normalisation of the same queue that produces queueDepth. Two carriers would spend two channels on one underlying fact
agent queueDepth magnitude	Independent	q3 and q9 are byte-identical; nothing visible distinguishes them
agent blocked magnitude	Independent	b1 and b5 are byte-identical
building health = warn	Independent	read from incident facts; not reconstructible from height, activity, footprint or identity
building capacity	Independent	read from capacity facts; and three states collide — 8/10, 10/10 and unmeasured
district health	Independent	keyed separately on district id (index.health.get(d.id)), not aggregated from member buildings, so it cannot be inferred from them
district revenue	Independent	declared in the estate; nothing visible implies it
focus, shortLabel	Observer-dependent	text carried by Html overlays. Requires reading, not perceiving — and is outside this instrument

A fifth category was forced by measurement: not a fact (building.blocked). That is a correction to the map, not an ontology extension — the vocabulary already distinguishes "a fact" from "a field I invented."

---

The carrier budget

Measured across the reachable sweep. Capacity is perceptual — how many states an observer can reliably separate — and is stated as an estimate where it cannot be measured without observers.

Carrier	Est. capacity	Current use	Remaining	Collision risk
position.y	high (continuous)	building height; agent idle-vs-not	low — already dual	high: adding a third meaning makes elevation unreadable
geometry params	high	footprint; height	low	medium
opacity	~3 levels	lot (0.35); unknown (0.5); district activity	~1 level	high — already carries two distinct meanings
color	high (hue space)	building kind; agent identity	low	high — identity must stay stable
emissive	high	identity only (cleaned)	none — reserved	would re-break identity
emissiveIntensity	~3 levels	activity; lot; agent status	~0	high
wireframe	2 states	unstaffed	0 — binary, spent	n/a
object presence (beacon)	additive	health=fail; agent blocked	medium — more marker types possible	medium: markers compete visually
scale	high	none	full	none
labels (Html)	high, low-salience	shortLabel, focus	unknown	low, but requires reading
animation	~2–3 rates	link flow	unknown	unmeasured by this instrument

One high-bandwidth carrier is entirely free: scale. Everything else is either spent, binary, or reserved for identity.

---

The deficit

Independent facts requiring a carrier, and the states each needs:

Fact	States needed	States rendered	Deficit
building health	4 (ok / warn / unknown / fail)	3	1
building capacity	3 (headroom / at-limit / unmeasured)	1	2
district health	4	1	3
district revenue	≥2 (declared / none)	1	1
agent queueDepth magnitude	≥2 bands	1	1
agent blocked magnitude	≥2 bands	1	1

Perceptual deficit = 9 distinguishable states missing.

Down from the impression left by the previous map, which counted eight invisible facts — of which one did not exist, one was derived, and one was redundant.

---

Decision importance — reasoned, not measured

This ranking is judgement, and is marked as such. No measurement here establishes what an executive needs; that requires the observer study. Ranked by the decision that becomes wrong if the fact is not perceived:

Rank	Fact	Decision at risk if unperceived
1	building capacity at-limit vs unmeasured	Capacity planning. Acting on "we have headroom" when the truth is "we never measured" is the failure mode with no recovery — it is the missing-vs-negative error, and it is currently a three-way collision.
2	building health = warn	Preventive intervention. A warning is the only state that is actionable before failure; invisible warnings convert prevention into response.
3	district health	Portfolio triage. A whole product line can be failing while every building in it looks ordinary.
4	agent blocked magnitude	Escalation sizing. Blocked is visible; severely blocked is not, so triage between two blocked agents has no visual basis.
5	agent queueDepth magnitude	Load balancing. Lower consequence — reassignment is reversible.
6	district revenue	Investment. Important to the business, but slow-moving and available in text; the render is not where this decision is made.

Note the inversion against implementation difficulty: the top-ranked item (capacity) needs three states from a carrier that has roughly one level left, while the lowest-ranked (revenue) could take scale tomorrow. Ranking by consequence rather than convenience is what surfaces that.

---

Status

Do not propose a Legend amendment. The deficit is 9, one high-bandwidth carrier remains, and two candidate carriers — labels and animation — are still unmeasurable by this instrument. Extending it to Html and animation is the prerequisite to any encoding decision, because two of the eleven carriers cannot currently be budgeted at all.