INDEPENDENT MODULE / 01
One traffic sample.
Eight ways through.
Run known intersection designs together. Follow the same drivers. Measure the whole trip—including detours, second stops, and fuel.
01 / SYNCHRONIZED REPLAY
Same clock. Same destinations.
Animation shows the first replication; results below use every completed paired run.
● Intended left turn ● Through ● Right turn Hollow ring = selected driver. Routes and point queues are schematic; card metrics are full-run means.
Center light ↔ crossover light
What the evidence says ↗02 / FULL-TRIP RESULTS
Compare the cost of getting through.
| Design | Wait (s) | Trip (s) | Fuel / 1k trips (US gal) | Stops / trip | Δ wait vs reference (s, 95% CI) | Exit flow (veh/hour) |
|---|
Exit flow counts all trips leaving during the observation hour, including warmup entrants. It remains an all-movement measure when “left turns only” is selected.
| Fuel components / 1,000 selected trips | Driving distance | Idling | Speed recovery | Route distance (m) | U-turn + return stops (%) | Red turns / 1k trips |
|---|
03 / ANNUAL COST COMPARISON
One exposure. Every selected design.
Full-trip time includes detours, queues and acceleration losses; total fuel includes each modeled restart. No component is added twice. All designs use the selected reference above and identical intended-left exposure. This does not scale these designs to all Michigan intersections.
| Design | Fuel + time / 1k trips | Annual operating cost | Δ vs reference ± 95% seed interval | Implied external visits/year | Model flags |
|---|
These are operating costs, not total social costs. Crash valuation below is separate; construction, maintenance, emissions, pedestrians and physical spillback are not priced. The original two-layout engine and this independent engine have different service rules; their outputs are not pooled.
04 / EXPERIMENT WORKBENCH
Sweep settings. Separate assumptions. Validate the result.
49 model and valuation inputs, exact or seeded sampled grids, all eight designs, independent validation, and downloadable protocols. The earlier 1,710-setting search is retained as a preliminary timing study in the research dossier.
Crash assumptions: separate from operational rankings
This module has no calibrated crash prediction model or collision engine. Blank means unavailable, not zero risk. Enter a site-specific assumed rate to calculate exposure × rate; these expected counts do not inject incidents or change traffic. The two-layout lab retains the separate crash-process experiment.
Supply each design's overall crash severity mix below: K fatal, A serious, B minor, C possible injury, O property damage only. All five percentages must sum to 100%; blank is unknown. No type mix or Michigan-left rate is transferred to the other designs. These values never affect operating rankings or the settings search.
| Design | K % | A % | B % | C % | O % | Expected crashes/year | Separate annual crash cost |
|---|
National per-crash costs at maximum severity; comprehensive already includes economic costs. No occupant multiplier. Source audit · FHWA Table 1. Assumed rates must cover comparable systems, periods, definitions and unique external-trip exposure. No combined total is inferred.
Rates: crashes per million unique entering trips, covering the whole studied system. Do not transfer a junction-only rate to a crossover network without matching boundaries, period and crash definition.