Permitting
Programme.
A derivation-first model of how far the municipal building-permit lifecycle can be run by autonomous agents — and, just as deliberately, where it cannot. Every figure below carries its scope and basis; no number stands without one.
State of play
designed vs built · 18 Aug 2026The figures above describe the designed system. This is what is actually built. The model is deliberately honest about the gap between the two.
Built
Open & blocked
Request ledger
19 requests reviewedDecision-class mix
share of decisionsClasses A–C (routine to rule-bound) are automatable; D–E carry discretion or statutory weight and stay with people. The mix is what sets the ceiling.
Autonomy vs ceiling
design basisMonte-Carlo · n=20,000 · σ=0.2 · p50 76.02%
Lifecycle — effort load × derived autonomy
L01 → L10Bars are each stage's share of total effort; the line is its class-credited autonomy. Autonomy runs high where work is routine and dips where a stage concentrates discretionary or high-risk decisions.
Seven ontological layers
an ontology, not a stackSeven-stage pipeline
workflow inside L3 · L5The layers are an ONTOLOGY: every object exists in all seven simultaneously. The pipeline is a WORKFLOW resident inside L3 and L5. Conflating them is the programme's largest drift risk.
Visual system
Open the visual system →The programme’s diagrams, rebuilt as a live, interactive system — every figure drawn from the model’s own numbers, never a picture. A sample: how confidence and risk are ANDed to a decision class.