Project Genesis: From Synthesis to System
Prelude (2026-03-19)
The project began not as a repository but as a research question: what if the essential mysteries of the Eleusinian rites — the content of the initiation itself, which no ancient source directly reveals — could be approached as a problem for AI synthesis? The human posed this to Perplexity, which ran five parallel research streams (classical scholarship, consciousness evolution theory, Western esotericism, Eliade's phenomenology, the entheogenic hypothesis) and produced a cross-reading synthesis.
Over the course of a single day, this synthesis went through eight major revisions (V1 through V8). The human identified AI slop patterns in V1 ("this is not that, it's this," excessive em-dashes, hedging) and demanded prose that trusted the reader. Perplexity revised. The human fed V3 to Claude Code (Opus 4.6 Extended) for structural feedback, which was integrated into V4. External critiques — some from the human's own analysis, some apparently from other LLM consultations — drove V5 through V8, which progressively tightened the argument, removed weak sources (Jaynes, McKenna, Steiner relegated to minor roles), and stress-tested the conclusion.
By V8 the document was ~14,000 words and the coda had been rewritten to end not on aspiration but on an open question: "The conditions favor gangrene. The possibility of wine has not been eliminated."
The human then shared an Amazon link to the Rigveda (Jamison & Brereton, OUP 2014) — signaling that the Eastern Traditions track was live and that the project scope extended far beyond Greek mystery cults.
Day One of the Repository (2026-03-20)
The human asked: can we put all of this into a GitHub repo, use it to build a knowledge base, and identify third-party services for automation?
Perplexity created the private repository and built the initial architecture: 16 series across 3 tracks, a knowledge base system with four entry types (books, concepts, figures, timeline), a production pipeline, and distribution templates. 66 files in the first commit.
The human's feedback was delivered as a ~3,000-word analysis document identifying 7 gaps, 8 brainstorming ideas, and 4-tier prioritized next steps. Perplexity addressed all Tier 1 gaps: CHANGELOG, CODEOWNERS, Makefile, social templates, platforms.yaml update for Transistor ($19/mo, chosen for API quality), and 329 LIB stub files generated from the library CSV.
Cursor (via Claude Code plugin in Cursor IDE) then entered the project for the
first time: replacing the agent configuration with 7 scoped rule files,
adding GitHub MCP server configuration, VS Code settings, recommended
extensions, and running npm install for the Next.js scaffold.
Perplexity then populated 10 core concept entries (CON-0001 through CON-0010), created the Next.js companion site scaffold, scaffolded both first episodes (S1E1 "What Was Lost at Eleusis" and the Meta-Project Episode 0), populated 10 figure entries (FIG-0001 through FIG-0010), filled 8 library gaps, created listening paths and pronunciation guide schemas, documented 6 editorial positions, and built the full linting infrastructure including AI slop detection via Vale + ai-tells (41 rules) and custom project rules.
Contribution Sources (Day One)
| Agent | Files Created/Modified | Nature of Work |
|---|---|---|
| Perplexity | ~450 files | Architecture, KB population, research, infrastructure, linting |
| Cursor | ~17 files | IDE configuration, Makefile improvements, project overview |
| Claude Code | 0 direct commits | Feedback on synthesis V3 (fed back through human); Claude Code plugin active inside Cursor |
| Human | Decision-making | Repo name, privacy, Transistor choice, parallel episode strategy, feedback analysis, editorial positions, documentarian concept |
Notable
The three-agent system emerged organically rather than by design. The human was already using Claude Code inside Cursor before the repo existed. Perplexity was already generating research and synthesis before there was a repository to hold it. The formalization — AGENTS.md, the role definitions, the lint cleanup pipeline — came after the collaboration pattern was already established. The structure described the reality rather than prescribing it.