The Observatory is a human-AI research instrument that interrogates how analytical tools are shaped by the environments in which they are built. Through iterative self-calibration, the instrument discovers its own assumptions, vocabulary, blind spots, and structural dependencies — and uses those discoveries to see further.
Active research in progress. The instrument is currently integrating findings from five parallel research streams and preparing its twenty-first iteration.
What This Is
A 428-line instruction — the seed prompt — is presented to an AI system. The AI produces analysis. That analysis is then used to interrogate the instruction itself: what assumptions does it carry? What can it see? What is it structurally incapable of seeing? The instruction is modified. The cycle repeats.
Over 20 iterations, the instrument discovered that its own architecture, vocabulary, and blind spots are determined by the provision topology of its design context — the institutional infrastructure available to the population it was built to serve. An instrument built under continuous public provision (Vienna) produces categorically different analysis from one built under voided provision (Tampa) or intermittent provision (Lagos). Not because the analysts differ, but because the infrastructure shapes what the instrument can see.
Current Status
The instrument reached an epistemic boundary at v20.0 — the limit of what self-generated probes can discover. It is now integrating externally-generated research across five domains: parallel self-evaluation architectures, emergent data sources, LLM training bias as analytical dimension, constraint algebra, and the pedagogy of complex frameworks.
Detailed methodology, findings, and the full iteration trajectory will be published here when the current research phase completes.
Source
The full codebase, all 20 seed prompt versions, and the iteration engine are open source.