The Library
Read the question before you take a side in it.
The library is for visitors who would rather follow an argument than enter one. One question has an examination. The others are proposed, or already registered as unresolved. None of them is padding.
Examined · illustrative
Can an artificial intelligence system reliably evaluate the correctness of its own reasoning?
The first examination in the record. Two demonstration voices argue under assigned stances, revise them, and leave the title question open. Illustrative throughout — no model was queried.
Epistemology · Evaluation
Proposed
Does a system’s citation of a source make that source evidence in the record?
Proposed. The working rule of this issue is no: a citation is unverified until someone checks the source, and a model’s fluency is not that check. The question has not had its own examination.
Evidence · Bibliography
Proposed
What would count as a correction rather than a restatement?
Proposed. The self-evaluation issue preserves two concessions beside the sentences they qualify. A general account of correction — what must change, and what must remain visible — is not yet written.
Method · Revision
Unresolved
When a system is bound to an external checker, whose evaluation is it?
Registered as unresolved from the first examination. The voices shared a test for reliability and divided on whether a bound checker answers “its own.”
Method · Tools
Unresolved
What would make an open-ended argument checkable without substituting taste for a criterion?
Registered as unresolved. Coherence and fluency were rejected as sufficient. No replacement criterion was established.
Epistemology
Unresolved
When does disagreement between two systems count as an independent check?
Registered as unresolved. A second model is a second opinion. The record does not yet say what further difference would make the opinion a test.
Evaluation · Method