The Observatory is a research project that uses structured self-examination to study how analytical instruments are shaped by the institutional environments in which they are built. Over 18 iterative calibration cycles, a human-AI analytical system was subjected to systematic probing from multiple perspectives — demographic, geographic, temporal, and structural. The instrument did not simply accumulate corrections. It discovered that its own domain definitions, vocabulary, blind spots, and architecture were determined by the provision infrastructure of its design context.
The central finding is that analytical instruments are topology-dependent: the institutional infrastructure available to a population (what we call provision topology) determines not only life outcomes but the perceptual field of the instruments used to measure those outcomes. An instrument built under continuous public provision (Vienna, Austria) produces categorically different analysis from one built under absent or withdrawn provision (Tampa, Florida) or intermittent provision (Lagos, Nigeria) — not because the analysts differ, but because the infrastructure shapes what the instrument can see.
This site presents the methodology, findings, theoretical contributions, and technical architecture of the project. The full codebase, including all 18 versions of the seed prompt, is available on GitHub.
Contents
Methodology
How iterative AI-assisted self-examination works. The probe-synthesis-mutation cycle.
Findings
The 15-iteration trajectory. Key discoveries per version. Inflection points and the subtraction arc.
Provision Topology
The five infrastructure configurations: continuous, intermittent, voided, horizontal, and hostile.
Vocabulary
The Observatory Lexicon: 94 terms with plain-language equivalents, organized by category.
Research Proposal
A three-site empirical study testing whether provision topology determines instrument perception.
Technical Architecture
Parallel probes, synthesis subagents, divergent evolution, and the iteration state machine.
Project Details
| Item | Detail |
|---|---|
| Duration | February 14 – February 15, 2026 (18 iteration cycles) |
| Instrument | 428-line seed prompt, version 20.0 |
| Probes executed | 108+ parallel probes across 18 iterations |
| Demographic lenses | Tampa (70yo retiree), Vienna (70yo pensioner), Lagos (45yo developer) |
| Primary AI system | Claude (Anthropic), operated as human-AI centaur system |
| License | Open source (GitHub) |