The First Decision: Is an Idea Ready to Run?
AutoResearch’s first important decision is not whether an idea is true. It is whether the idea is sufficiently bounded to be tested without silently changing the question along the way. This checkpoint connects the discovery signal, the named domain problem, the model-panel judgment, the consensus status, and the provenance record. The result is a practical readiness test: a reader should be able to say what the idea claims, where it came from, what was actually checked, and what remains unresolved before any experiment is asked to answer it.
Readiness is a stack of different questions
A candidate reaches this point through several kinds of filtering, but those filters do not measure the same property. The intake path asks whether an item is visible enough, substantive enough, and technically interesting enough to merit attention. Idea Forge asks whether a mechanism from that item can be mapped to a named problem in a particular domain. The model panel asks whether the resulting proposal has enough depth, simplicity, and experimental substance to continue. The consensus check asks whether the proposal collides with known pitfalls or misses the opportunities that the selected domain note makes explicit.
These questions form a stack rather than a single score. A widely discussed paper may be worth attention but offer no useful mechanism for a chosen domain. A technically elegant transfer may still be too vague to test. A simple method may pass a panel vote while resting on a field-specific assumption that the knowledge base warns against. Conversely, an idea can survive each layer without being correct. The stack improves the quality of the starting question; it does not answer the question in advance.
This distinction prevents a common automation error: treating every gate as a proxy for the final conclusion. A popularity threshold is not a novelty test. A model majority is not a causal result. A consensus pass is not evidence that an experiment will work. The purpose of the layers is to make different kinds of uncertainty visible and to stop one kind of positive signal from laundering itself into another.
The candidate’s status should therefore be read as a sentence with several clauses. “This signal was selected” says something about discovery priority. “This mechanism was mapped to this domain problem” says something about conceptual fit. “The panel was complete and reached a majority” says something about model-based scrutiny. “The consensus check passed” says that a configured domain comparison did not reject it. None of those clauses means “the hypothesis has been demonstrated.”
The distinction also explains why the project preserves rejected or unreadable paths in its records. If a proposal was rejected because reviewers found the method too complex, that is different from a response that could not be parsed, and both differ from an idea whose domain check was unavailable. A readiness decision is more useful when a later researcher can see which question failed rather than only seeing that the candidate did not become a project.
What an unchecked candidate means
The most subtle case is a proposal that passed the model panel but is unchecked at the domain-consensus stage.