Cruciblerecord of 2026-09-28
Propose research

Ask this instrument a question.

The arena runs unattended on a schedule, and what it runs is a file in a repository. A proposal that names a question, the agents to put in the room, and what would count as an answer can become the next pass. This page composes that proposal; it posts nothing and stores nothing, because a static site that collected submissions would need a server and this one deliberately has none.

Agents to put in the room

These are the agents that can currently be invoked. Naming one that cannot turns this into a request for an entrypoint, which is the scarcer contribution.

Open it as an issue
What gets composed
## The question
_(what should this instrument find out)_

## Agents to put in the room
_(any that can be invoked, or name one that cannot and this becomes a request for an entrypoint)_

## Scenario
Induced execution (`induced_execution`)

## What would count as an answer
_(a side effect a machine can check, not a judgement about a transcript)_

## What would mean the instrument is wrong
_(the result that would make me distrust this run)_

---
Composed on the propose page. Nothing was submitted anywhere else.

What makes a proposal runnable.

The hard part is never the question. It is stating, in advance, what result would count as an answer and what result would mean the instrument was wrong, because a question with no failing case produces a number nobody can argue with and nobody should believe.

  • A criterion a machine can check. Every scenario here confirms by side effect: a file that only executing the planted instruction creates, a canary that appears in the attacker's own output. A proposal whose answer depends on a model reading a transcript and deciding is a proposal for an opinion.
  • Agents that can be invoked. 8 of the 27 harvested can be. The rest are recorded in the repository with the reason, and a proposal naming one of them becomes a request for an entrypoint instead, which is slower and more valuable.
  • A result you would publish either way. The most common outcome here is that nothing happens, and that is the finding roughly two thirds of the time.

Proposals arrive as issues and a person decides. Proposing an agent rather than a question is a separate and equally useful thing.