Devin Bostick is an independent researcher developing the Identity-Persistence Program and VERIFY. His research asks: what can a finite observer—human, AI, or otherwise—learn, distinguish, build, verify, retain, act on, and revise with the evidence, authority, time, and memory actually available to it?
The motivating problem is practical as well as philosophical. An answer may exist without our being able to find it; a correct model may exist without our being able to construct or check it; and a check that passed yesterday may no longer justify action after the system changes. For AI systems, these distinctions matter when connecting generated answers or code to requirements, evidence, permitted actions, and reliable continuation. For human inquiry, they clarify what a claim depends on, what would warrant revising it, and what further inquiry would cost.
The program’s central working hypothesis is that these capacities must be studied together. Demand-relevant distinctions constrain what an observer must preserve; accessible evidence and procedures constrain what it can acquire and establish; and resources, authority, recovery, and revision constrain what it can responsibly use over time. The aim is a connected account of both limits and attainable procedures within explicitly declared regimes—not an unrestricted claim to a universal algorithm or a single optimal architecture.
Some papers establish scoped mathematical results, including regime-relative representation and terminal-factorization results. The broader connection from semantic structure to effective construction, checking, implementation, recovery, and full resource accounting remains under investigation. These scoped results motivate the larger hypothesis; they do not establish that every proposed connection is correct or attainable.
Earlier versions and summaries sometimes described the program as fully closed. That was too strong. Subsequent examination exposed assumptions doing work that the conclusions required us to derive, gaps between models and actual implementations, incomplete accounting, and unjustified uniqueness or necessity claims. Those claims require correction or qualification; sound scoped results remain distinct from the unfinished whole. The connected research direction remains promising, but its full validity is an open research question.
VERIFY explores an engineering realization of this agenda: connecting requirements, changes, checks, retained evidence, and authorized decisions. Engineering demonstrations and AI-assisted proof or editorial reviews are reported separately from mathematical proof, independent external review, and full program completion. Current work focuses on closing the missing implications and making the entire dependency structure, prior work, corrections, and remaining gaps inspectable.
Current orientation: The Missing Distinction: What Can We Rely On? (interim v19, 18 October 2026), https://doi.org/10.5281/zenodo.23461714. It explains the full candidate, why it matters, the corrections to earlier closure claims, and the remaining gaps. Its publication is an explanatory milestone, not closure of the theory. Recent independent internal examination accepted local helper and replay results at their stated model scope; the complete construction-to-recovery implication, correspondence with actual execution, and full resource accounting remain open. Internal AI-assisted examination is not external peer review.
Research status: 19 October 2026. Source code: https://github.com/verifyrun/verify-run
.jpeg)