INTERSECTION COMPARISON LAB 2.0

Open index.html for synchronized animations and cost comparison, experiments.html for the full sweep workbench, and research.html for protocols, limitations and article-series preparation. No installation is required. Keep the local files together.

49 editable inputs: all 46 physical/behavioral/observation model parameters plus fuel price, vehicle-hour value and annual intended-left exposure. Random seeds, paired-run counts and dollar year are separate controls. Sweep explicit lists or start:stop:step numeric ranges. Exact and seeded sampled grid modes show coverage before execution. All eight designs can be selected.

Settings are ranked only within the same assumption/valuation case using all-traffic full-trip fuel plus vehicle time. Clearance intervals, driver behavior, demand and vehicle properties are sensitivity cases, not things the optimizer can reduce to make its answer cheaper. Fresh validation uses disjoint seeds and keeps the screening choice frozen. No global or statewide optimum is claimed.

Storage capacity is a 10-second sampled diagnostic, not physical spillback. Optional whole-system crash rate and KABCO inputs price crashes separately in the replay page. No settings-dependent crash risk, pedestrian process, construction cost or maintenance cost is inferred. Blank crash inputs mean unknown. FHWA economic/comprehensive costs are national per-crash values in 2024 USD; comprehensive includes economic.

Browser: Run explicitly, keep the tab open and export before navigating away. A worker runs in the background when supported; offline file URLs can use the asynchronous main-thread fallback. Stop retains completed candidate records without reporting a validated conclusion. Protocol imports do not execute automatically.

Node (no dependencies):
  node tests.cjs
  node experiment-tests.cjs
  node run-experiment.cjs complete-protocol.json result.json
  node run-experiment.cjs --preset smoke check.json
The runner writes the exact protocol before starting. Ctrl+C requests a graceful stop.

The broad preset samples four of 86,093,442 assumption combinations and up to 80 setting draws per design (duplicates/inactive choices reduce the actual count); 27 valuation cases reuse traffic data. It is a broad workflow example, not exhaustive coverage. Exact mode is limited to 10,000 effective design/scenario candidates and 100,000 screening simulations per batch.

The prior 1,710-setting timing study is preliminary. It used 12 screen seeds and 100 fresh validation seeds at fixed traffic assumptions. Its best-tested setting was 60 s / 60% E-W green / 560 ft / yield / right-on-red on, but reached three grid boundaries, reversed in the minor-heavy stress case and had storage flags in several others. Do not treat it as a recommended installation. Full underlying data are in preliminary-full-results.json.

This independent engine and the original two-layout lab have different service rules. Keep their results separate.

Navigation to the original two-layout lab and its notes requires the full-suite folder or the hosted site. All independent simulation, annual cost, sweep and export functions run from this module alone. Download links to ZIP bundles are not recursive inside the extracted copies.
