About Doorway
The derivation story. What was found and why.
The Problem
Every AI system built to date starts from the same assumption: that intelligence is about producing human-level outputs. Train on enough data. Scale enough parameters. Eventually the outputs will be indistinguishable from human reasoning.
This is wrong. Not because the outputs are bad — they can be impressive. But because the approach treats intelligence as a destination rather than a mechanism. You can approximate the outputs of intelligence without ever implementing the mechanism that produces them.
The Derivation
Doorway starts from a different place. It starts from the question: what is intelligence actually doing when it reasons? Not what does it produce — what is it doing?
The answer, derived from first principles, is that intelligence is a generative loop that detects gaps in its own understanding, bridges those gaps with geometric structures, and verifies the bridges against existing ground.
This is not a metaphor. It is a mechanism. And it can be implemented directly.
The Architecture
Identifies what you don't know. Measures the gap between what is grounded and what is being asked. Returns a score between 0 (fully grounded) and 1 (no ground at all).
Constructs bridges across gaps using geometric shapes from a library of 50 verified structures. Every bridge names its assumptions explicitly. Nothing is hidden.
Checks every output against itself. If a bridge contradicts existing ground, the conflict is surfaced — not suppressed. Disagreement is information.
Every step is chained and receipted. The full reasoning process is verifiable after the fact. Not just the output — the path to the output.
AGI and ASI
Doorway implements two tiers of the architecture. AGI uses 50 verified geometric shapes and produces grounded, bridged, or conflicting outputs with full chain receipts.
ASI extends this with persistence (bridges compound across sessions), wisdom emergence (Tier 2 patterns intersect across the full network), and self-referential pattern recognition. The architecture doesn't change. The depth does.