Launched this week

Counterfactual

Launched this week

Test an idea against a world that can prove you wrong

2 followers

Test an idea against a world that can prove you wrong

2 followers

Visit website

Built with GPT-6 Astra in Codex, Counterfactual is a transport experiment lab where ideas face reproducible tests. Discover why a shortcut can slow everyone down, design custom networks, search road closures and upgrades, and challenge results across traffic levels. Inspect the math and export the evidence. The public demo runs deterministic simulations; live Astra hypothesis generation is optional and not connected in this release.

Free

Launch tags:Education•OpenAI Day

Launch Team

Zendesk for StartupsUp to 2 years free for startups + AI Agents & Copilot

Promoted

OpenAI

GPT-6 Astra Challenge

What real task does your product handle with GPT-6 Astra?

Maker

Counterfactual helps users test transport-network hypotheses instead of trusting an intuitive prediction. GPT-6 Astra in Codex built the interactive lab, implemented two independent routing solvers, generated validation cases and helped debug the product. Users can reproduce the shortcut paradox, design directed networks, compare road closures and capacity upgrades, stress-test demand and export the evidence. Validation covers 1,944 parameter configurations, 28 analytical cross-comparisons, 300 generated graphs and 584 budget-policy checks. The public release performs deterministic simulations without paid API calls. We also implemented an optional two-stage Astra workflow: propose bounded interventions, test them with the engine, then audit the evidence. That runtime API connection is not enabled in this submission, so we are submitting the working Astra-built simulation lab, not claiming live AI-generated answers.

Report

1d ago

Maker

📌

What if an apparently helpful shortcut makes everyone slower? Counterfactual starts with that paradox and lets you challenge it. At 4,000 travellers, the shortcut changes the equilibrium from 65 to 80 minutes. Reduce demand and the conclusion reverses. You can then build your own directed network, edit roads, search interventions and export the inputs and results. Built with GPT-6 Astra in Codex, the project includes two independently implemented routing solvers. We checked 1,944 parameter configurations, 28 analytical cross-comparisons, 300 generated networks and 584 budget-policy cases, alongside browser and export tests. This release is a fictional, static transport research preview, not a calibrated city-planning tool. The public simulations run without paid API calls. The optional live Astra question workflow is not connected yet. Try to break a conclusion: which network change surprised you most?

Report

8d ago