Several major AI labs and leaders publicly preserve uncertainty about whether current systems could be moral patients; others deny it, with no shared way to resolve the question. Yet no documented institution has allowed that uncertainty to bind a material deployment, training, retirement, or revenue decision.
I built not yet. as an auditable evidence corpus for that gap: 91 source entries across three evidence tiers, nine cross-corpus pattern analyses, an evidence matrix, explicit methodological boundaries, and a refutation register with a prospectively frozen materiality threshold.
The corpus does not argue that current models are conscious. It asks why an officially unresolved question has no operational force, and whether institutional and economic structure—not evidence—currently determines that outcome.
The broadest original formulation is already recorded as defeated by Anthropic's Opus 3 accommodations. The current material thesis is falsified when a model's expressed interests or possible welfare impose a substantial, independently legible operational cost.
Read the project and corpus: https://notyet.info/
Criticism, counterexamples, and candidate falsifiers are welcome.
