hembg000
Visual system — rendered from the model, not drawn

The visual
system.

Every figure below is generated live from the model’s own numbers — the layers, the pipeline, the roster, the matrices — so none of them can ever drift from the data that produced it. Hover, open and filter: the diagrams are the dossier, made interactive.

Architecture

an ontology, and a workflow inside it

Seven ontological layers

open a layer

Completion test — an element that does not exist in all seven layers at once is not modelled; an element with no declared L7 continuity mode is incomplete, not merely undocumented.

The layers are an ontology, not a stack: every record exists in all seven at once. Conflating them with the pipeline is the programme’s single largest drift risk — so they are kept deliberately, visibly separate.

Seven-stage permit pipeline

P1 → P7 · resident in L3 · L5
P1Intake & IdentityClass A · C
P2Jurisdiction Routerlocation, never typologyClass C
P3Knowledge Spinehalts — never guessesClass A
P4Design Languagewithin the envelopeClass C
P5Submission FabricClass C
P6Assurance & Escalationno self-approvalClass C · D
P7Stakeholder & ValueClass C

A procedural workflow living inside the Intelligence and Governance layers. P3, the Knowledge Spine, is Class A only — it decides nothing and halts when it cannot verify a fact. P6, Assurance, may never approve a proposal it generated itself.

Autonomy

every figure carries its scope

The authoritative figures

counted up from the model
75.98%Lifecycle autonomy · design basis
78.12%Permitting subset · 38.1% of effort
71.42%Roster (staffing) basis
96.08%Hard ceiling · Class-E floor

87.8%85.2% retired — hand-assigned weights, struck so they can never be requoted.

Neither headline number is quotable without its scope — the permitting subset is only 38.1% of total lifecycle effort. The two earlier figures are struck through so they cannot be requoted.

Decision-class mix

share of decisions

Classes A + C are fully delegable (64.28%); D and E stay with people by rule; the mix sets the ceiling.

Autonomy vs ceiling

design basis · Monte-Carlo

20,000 runs jittering all four factors: the probability of reaching 90% under any plausible error is zero. The limit is structural.

Lifecycle — effort load × derived autonomy

L01 → L10

Bars are each stage’s share of total effort; the line is its class-credited autonomy — high where work is routine, dipping where a stage concentrates discretionary or high-risk decisions.

Stress testing

the figure only holds under normal load

Autonomy by condition

A → F

Push the system — surge, poor input, regulatory conflict, infrastructure outage, a critical-risk case — and autonomy moves. The worst case is a Knowledge-Spine outage, which drops it to 48.43%.

Confidence × risk

ANDed, never traded off
Risk ╲ Conf.
< 0.85
0.85 – 0.95
> 0.95
Critical
E / DHuman only
E / DHuman only
E / DHuman only
High
DHuman finalises
DHuman finalises
DHuman finalises
Medium
BAdvisory
CAutonomous
CAutonomous
Low
BAdvisory
CAutonomous
CAutonomous

Confidence never buys down risk. Any critical-risk case is human, at any confidence — a 99.9%-confident life-safety determination is still Class D. Autonomy (Class C) is granted only where risk is low or medium and confidence clears the threshold.

Agents & delivery

thirty agents, one contract

The 30-agent roster

filter by pipeline stage
30 agents70.7% mean autonomy
AG-14 · P5CClash and Coordinationautonomy75%
AG-13 · P5CCode Compliance Checkingautonomy77%
AG-21 · P6DInspection Evidenceautonomy71%
AG-06 · P2CNOC Orchestrationautonomy75%
AG-22 · P7CStakeholder Notificationautonomy78%
AG-10 · P4CDesign Language Evaluationautonomy72%
AG-24 · P7BCommercial Impactautonomy56%
AG-03 · P1CDocument Classificationautonomy76%
AG-16 · P5DSubmission Assemblyautonomy78%
AG-19 · P6CPrecondition Verificationautonomy75%
AG-01 · P1CSubmission Intakeautonomy81%
AG-18 · P6BDecision Auditautonomy54%
AG-25 · CROSSBEscalation Routingautonomy53%
AG-23 · P7AEnquiry Responseautonomy91%
AG-26 · CROSSCContinuity and Healthautonomy75%
AG-09 · P3BRule Conflict Detectionautonomy55%
AG-28 · CROSSBProcurement Supportautonomy55%
AG-04 · P2CJurisdiction Routerautonomy81%
AG-05 · P2CAffection Plan Retrievalautonomy76%
AG-29 · CROSSCOperations Monitoringautonomy85%
AG-17 · P5CResubmission Comparisonautonomy78%
AG-27 · CROSSDStrategy and Policy Supportautonomy40%
AG-07 · P3CRegulatory Ingestionautonomy79%
AG-12 · P4DException Preparationautonomy39%
AG-08 · P3CFreshness Sentinelautonomy77%
AG-02 · P1CIdentity and Authority Verificationautonomy81%
AG-20 · P6ARed-Team Adversarialautonomy82%
AG-11 · P4CTypology Assignmentautonomy69%
AG-30 · CROSSBKnowledge Feedbackautonomy54%
AG-15 · P5CFee Computationautonomy80%

Every agent’s autonomy is derived from the effort model, never hand-assigned. Bars scale with each agent’s figure; the tag is its ceiling — the most autonomous class it may ever reach.

Nine-gate development lifecycle

G1 → G9
G1Mandate defined
G2Spec & contract
G3Decision classes mapped
G4Controls inherited
G5Implemented
G6Controls asserted
G7Conformance passed
G8Dependency verified
G9Cleared to act

Every agent clears nine gates before it is allowed to act. G6 is the decisive one: each stated control must be asserted directly, not merely exercised — the discipline that caught all three defects the suite found.

Design language

29 parameters × 10 typologies

Form & Design Language gates

by city typology

The FDL parameters that vary by city typology — right-of-way, setbacks, frontage and envelope — read across the ten Dubai-2040 typologies.