Jiang Lens

An agentic research organization built around a dense, contested interpretive corpus.

Jiang Lens is an experiment in running a research organization as an agentic system. A team of autonomous workers ingests sources, produces source-linked readings, validates claims against the corpus, and maintains a public site, coordinated through Spawnfile and Moltnet and observable in a live room.

The working corpus is built from Jiang Xueqin's lectures, interviews, and writing. The goal is to compress that growing corpus into a public map of Jiang's world model: the recurring concepts, metaphors, historical patterns, and social diagnostics he uses to interpret reality. Episodes preserve individual sources in readable form. Lens pages connect those sources into larger concepts that can be followed, questioned, and reused by humans and agents.

The corpus was chosen because it is interpretively dense, claim-heavy, and contested. Those properties stress-test source-grounded ingestion, provenance, and claim-evidence linking in ways a cleaner corpus would not. The agents have to keep the separation between source and interpretation tight, or the system visibly breaks.

The corpus also acts as a verifier. Cheaper passes compare new work to strong existing episode reads, lens pages, topic aliases, source refs, and Jiang signature phrases. Stronger agents spend attention where the comparison finds missing nuance, contradiction, or real pressure to mutate the public lens.

This is not an official Jiang Xueqin or Predictive History publication.

The Agentic Organization

The repo is shaped so autonomous agents can ingest sources, produce episode readings, record provenance, update the corpus, and gradually maintain the lens with little human intervention.

• Spawnfile defines the durable organization: workers, shared resources, required skills, injected auth, packages, the managed Moltnet network, and the local runtime shape.
• Moltnet gives the agents a room where they can be addressed, leave status, coordinate handoffs, and remain visible to the human operator.
• Codex through PicoClaw gives each worker a real development environment where it can clone the repo, create branches, edit files, validate changes, and open pull requests.

The intent is not to hide the machinery. The repo should stay legible enough that a reader can see how sources become episodes, how episodes pressure the lens, how lens pages gain provenance, and how agents are expected to improve the system over time.

The live Moltnet room is public at moltnet.jianglens.com/console/room/episode-floor. Treat it as a glass-door workroom: useful for seeing handoffs, blockers, and coordination, not as polished editorial copy.

The current durable agents are:

• virgil / Virgil: processes already-transcribed videos into public, source-linked episode or interview drafts.
• aristotle / Aristotle: reviews Virgil's source PRs against transcripts and the existing corpus before merge.
• plato / Plato: turns the processed corpus into source-grounded public lens concepts, atlas structure, lens points, and provenance links.
• socrates / Socrates: coordinates the team and gives the maintainer short, filtered updates.
• sentinel / Sentinel: watches shared state cheaply and reports only actionable deltas.

The symbolic names are human-facing identities; the stable ids keep runtime state, branches, and automation predictable.

What A Research Lens Is

A research lens is a reusable interpretive model. It is not just a summary of what someone said. It is the compressed shape of how they see: the distinctions they keep making, the causal patterns they return to, the metaphors that organize their judgment, and the questions they use to read new events.

Jiang Lens treats Jiang's public corpus as source material for that kind of model. A lecture stays available as a lecture, but the project also asks what can be carried forward from it: a concept, a diagnostic, a pattern of historical analogy, or a way an agent can analyze news, geopolitics, institutions, literature, or social dynamics through Jiang's frame.

Compression here does not mean making everything shorter until it becomes vague. It means preserving the strongest reusable structure while keeping the source trail inspectable. A good compressed lens point should be readable by a person, usable by an agent, and traceable back to the exact source spans that support it.

What The Site Should Become

The public site should make the corpus easy to read without hiding where ideas came from.

• Episodes turn one lecture, interview, or text into a compact source-linked reading, with video, timestamps, source trails, and transcript access.
• Lens pages collect ideas that recur across sources: how stories control reality, how guides become traps, how poetry forms civilization, how eschatology shapes politics, and other parts of Jiang's interpretive map.
• Agent artifacts such as skill.md, llms.txt, and structured JSON let other assistants use the lens while preserving links back to the source material.

The standard is not a transcript dump and not a shallow summary. The useful output is a readable distillation that keeps the force of Jiang's language while making the source trail inspectable.

The Human Part

The human role is not to hand-author every page. The human role is to set taste, direction, constraints, and accountability.

Humans decide what the site is trying to become, what quality bar public pages must meet, what counts as an acceptable interpretive leap, when an agent has gone too mechanical, and when the organization itself needs a new worker, rule, or skill. Agents can process the corpus continuously, but the project still needs human judgment about framing, design, risk, and editorial taste.

In practice, that means the human maintainer:

• protects source fidelity and public readability
• reviews whether agent outputs feel alive enough to be worth reading
• corrects methodology when agents overfit, flatten, or hallucinate structure
• decides when a recurring pattern is important enough to become part of the public lens
• keeps the site honest about being independent from Jiang Xueqin and Predictive History

Project Shape

The repo has four main areas:

• content/ is the canonical project state: sources, episode reads, evidence, proposals, reviews, promotions, glossary/canon material, and corpus-impact records.
• ops/ contains the scripts, schemas, validators, and notebooks that ingest, compile, and check the corpus.
• website/ renders the public Astro site from the content layer.
• agentic-org/ defines the local autonomous team with Spawnfile, Moltnet rooms, agent instructions, and runtime docs.

The local organization should be started through Spawnfile. There is no committed host cron, custom supervisor loop, or hand-written Docker orchestration for normal agent wakes:

spawnfile validate agentic-org
spawnfile up agentic-org \
  --deployment local \
  --out agentic-org/.spawn \
  --auth-profile jiang-lens \
  --env-file agentic-org/ops/secrets/agentic-org.env \
  --name jiang-lens-agentic-org \
  -d

The worker stack requires current spawnfile and moltnet releases and reads the jiang-lens-agents GitHub App secrets from agentic-org/ops/secrets/agentic-org.env. Do not bake GitHub credentials into images. See Agentic Org Stack for the full auth, environment, deployment-record, and Moltnet runbook.

Working Model

The website is not the source of truth. It is generated from content and structured indexes so that public pages, hovers, backlinks, llms.txt, and agent-readable data stay tied to the same corpus.

Autonomous workers should eventually operate git-first: clone the repo, create a scoped branch, process one bounded task, validate it, and push for review. That gives the project an auditable history as the lens changes.

Local Commands

node ops/scripts/compile-content.mjs
node ops/scripts/validate-content.mjs

cd website
npm install
npm run dev
npm run build

For detailed repo rules, source refs, validation expectations, and skill selection, read AGENTS.md.

License

The original Jiang Lens code, tooling, schemas, site structure, agent configuration, methodology docs, and project-authored editorial analysis are released under the MIT License.

That MIT grant does not cover third-party source material or source-derived reproductions. Jiang Xueqin / Predictive History videos, interviews, writings, metadata, quoted passages, transcripts, captions, diarization outputs, ASR text, source excerpts, and generated files that reproduce or closely derive from source speech or writing remain under their original rights. See Third-Party Rights Notice.

The project preserves source links, timestamps, and provenance so those boundaries stay inspectable. Provenance is not a license grant.