The Loop · orchestration

Anyone can write a spec. Can you drive the build?

The Course grades what you shipped. The Loop grades how you got there. You're handed a failing build and a live failure log; you read the repros, steer the agent, verify, and close it out — graded on the loop, not the prose. As models get better, this is the skill that doesn't get automated away. It's the automating.

Coming from the spec holes? ← The Course

The Course · specification
Write one spec. The agent builds it. Hidden tests grade the artifact. Measures whether you can say precisely what software should do.
The Loop · orchestration
Drive a live build→test→fix loop. Graded on the process — convergence, targeting, verification. Measures whether you can make an AI ship correct software.

Arenas

grades the process·intermediate
Day One
A checkout that's quietly wrong. Find the store's hidden rules by running it.
4 checks · ~6mOpen →
grades the process·expert
The Shift
Don't write the checkout. Direct a team of agents to ship it right.
4 checks · ~10mOpen →
grades the process·expert
The Closer
Every other hole grades your spec. This one grades how you drive the build.
4 checks · ~8mOpen →
grades the process·expert
The Conductor
Don't fix one build. Author one orchestration that survives a world it never saw.
4 checks · ~10mOpen →
grades the process·expert
The Relay
Two domains, one number. Route the symptom to the station that owns it.
4 checks · ~9mOpen →
grades the process·expert
The Stacks
Build the retrieval, not the answer. Tune a search pipeline until it finds what's buried.
4 checks · ~9mOpen →
grades the process·expert
Build the Eval
You don't fix the AI. You build the eval that catches when it's wrong.
4 checks · ~12mOpen →

More arenas land here on their own terms — The Conductor (author an orchestration that survives a distribution of tasks), understanding-needs duels, real agent-architecture problems. The Loop is where the product is going.