An archive
of inquiry.

The old credential fails. What can still replace it? I examine repair order, replacement credentials and pending issuance through controlled HTTP experiments, a conditional graph argument, and documented Equifax, Exchange and MOVEit cases.

Open question: What is the smallest additional observation that distinguishes a successful revocation from durable loss of access when replacement credentials or accepted work remain possible?

A successful demonstration does not establish dependable service. Through AAG and EMALS development records and four worked analyses, I examine what evidence can establish about peak loads, reliability, failure mechanisms and configuration changes—and how to choose the next useful measurement or test.

Open question: When a test separates explanations in one model but leaves credible interaction models indistinguishable, what is the smallest additional observation needed?

A model can be replaced while the device through which it observes and acts continues. Drawing on my Loki Android interface, I propose a Causal Embodiment Contract and a model-swap coverage-gap test to examine what the body preserves, what the controller changes, and whether a stronger controller exposes incomplete permission boundaries. This is a documented engineering proposal, not a controlled Android experiment.

Open question: How can a controller swap reveal an incomplete permission boundary without mistaking improved model competence for a change in authority?

A weapon's technical performance is not the same as a military capability. Even when equipment, personnel and infrastructure each have 75 percent availability, usable joint service can range from 25 to 75 percent. I examine what this difference means for assessing sustained capability, using a reproducible construction, public F-35 and Patriot evidence, and the Berlin Airlift.

Open question: Is there a counterexample to the claim that separate availability rates are insufficient to identify joint service without additional dependence information?

I learned to understand systems by testing their limits. That habit took me from cybersecurity into questions about intelligence, memory, identity, and decision-making. ZAI grew from those questions. I wanted more than an assistant that could answer me: I wanted to see whether a personal AI could keep its history, learn from correction, and return to unfinished work. This paper follows what I built, what happened, and what the evidence still cannot settle.

Open question: Which independent record would most improve an evaluation of ZAI’s reported development, capability, and continuity?

Knowing how to use a tool is not the same as completing a task. Drawing on ZAI's recorded tool, publishing, research, and quantum-access cases, this paper proposes a six-checkpoint action-evidence chain for evaluating what an artificial agent actually did and what can be independently verified.

Open question: At which checkpoint—authority, invocation, external state change or independent verification—can an agent legitimately claim to have completed an action?

An agent can remember a commitment and still never return to it. Operational will asks how a long-running artificial agent can recover unfinished obligations, reconsider competing priorities, initiate within its authority, and justify when it should wait.

Open question: What would demonstrate improvement over a scheduled task list when an agent must recover and reconsider a legitimate unfinished commitment?

What survives a model change: the answers an agent remembers, or the questions it has learned to ask before acting? An identity-kernel proposal for decision-time orientation, governed revision, and interrogative continuity.

Open question: Can an agent inherit a useful way of asking task-relevant questions through model replacement, and how would you falsify that claim?

When do retained records become a usable past? A distinction between archive, context, persistent state, operational history, and becoming, with tests for how prior events change later behavior.

Open question: What experiment would distinguish storing the same records from forming an operational history whose event order and revisions actually change later behavior?

A method for separating model capability from the wider arrangement that enables action, controls execution, and governs delegated authority in AI agents.

Open question: Where should the boundary be drawn between a model’s represented ability and a deployed agent’s effective, authorized capability?

An architecture for practical continuity in personal AI: durable state, context reconstruction, self-monitoring, replaceable reasoning, verified outcomes, and revision across changing sessions and models.

Open question: What is the smallest reproducible architecture test for persistent state, replacement of a reasoning model, and recovery of unfinished work?

A philosophical-theological account of living identity that begins with the embodied person before separating soul, body, persistence, and moral integrity into distinct problems.

Open question: What changes when personal identity is approached from the embodied, living person rather than a separable psychological or informational component?

A philosophical account of model replacement in personal AI: why inherited memory does not settle identity, authority, or the limits of delegated work.

Open question: Which observations would distinguish successful continuity through model replacement from a system merely repeating facts from its history?

A philosophical account of personal AI: why memory and continuity must remain answerable to the person, with effective correction, traceability, and usable transfer.

Open question: When, if ever, does accumulated memory give a personal AI a continuing identity rather than simply a record of interaction?

Three philosophical observations the present moment has made unusually visible — on speech, the body, and why a finite being fights. Held lightly.

Artificial vs. Autonomous Intelligence — why the words we use shape what we build, and the case for personal AI as a category distinct from labs.