I have been sketching the architecture for what I am tentatively calling Tessera. Named for the small tiles that compose a mosaic. Individual decisions are small. Together, they form a picture of how a mind works.
The critical design insight came this week: the goal is not to recreate my voice or my personality. The goal is to preserve judgment.
What Judgment Means in This Context
Judgment is how I recognize patterns. How I weigh risk. How I decide what matters and what does not. How I escalate, when I slow down, when I move decisively, and how I correct course when new evidence appears. Judgment is the layer above domain expertise, where tradeoffs are reconciled across domains.
Most AI systems trained on personal data try to replicate language patterns. They learn how you write. That is almost worthless for what I need. I do not need a system that sounds like me. I need a system that chooses the same class of move I would choose, given the same constraints.
Extracting Decisions, Not Text
The architecture must treat every artifact in my corpus not as text to be learned, but as decisions to be reconstructed. Every email argued a position. Every journal entry weighed a tradeoff. Every recorded call navigated constraints. Every escalation embodied a judgment about severity, urgency, and accountability.
From those artifacts, I need to extract: where I committed, where I deferred, where I escalated, where I reversed myself, and where I held a line despite pressure. Each decision anchored to signals, constraints, stakeholders, risk tolerances, and outcomes.
Over time, this should form a multi-dimensional decision lattice that spans domains but preserves causality. Not a timeline. A lattice. That is the architectural bet. If I am right, this becomes something genuinely new. If I am wrong, I will have documented a lot of interesting thinking on the way to finding out.