Epistemedia

Security and Adversarial Robustness

Threats including source prompt injection, evaluator collusion, Sybils, ontology poisoning, lineage laundering, and privilege escalation.

Lens: skeptical — Only strongly supported, independently grounded conclusions are foregrounded.

Included objects

Security Policy

Epistemedia treats candidate code, repository text, imported sources, issue content, model output, and remote bundles as untrusted data. The project’s security boundary includes both conventional software integrity and epistemic integrity.

Epistemedia Documentation

Agent Operations

This directory contains repeatable operating recipes. AGENTS.md remains the root authority contract.

Contributor Agent Recipe

1. Orient: make orient. 2. Select: python -m epistemedia repo next. 3. Read the root contract, selected task contract, execution plan, scoped authority, schemas, policies, and tests. 4. Claim through an append-only proposal. 5. Work in one bounded branch and preserve negative results. 6. Rebuild all derived surface

API, MCP, and CLI

All integrations read the same disclosure-safe public catalog. Preserve the returned catalog_id, frontier, commit, compiler, policy IDs, object IDs, and content digests when storing or citing results.

Public API and MCP deployment contract

The API and MCP gateway is a read-only projection service. It receives an accepted repository snapshot and exposes no GitHub, contribution, deployment, package-publishing, signing, or arbitrary network-fetch authority.

Epistemedia Architecture

Epistemedia is not an autonomous wiki whose canonical object is a page. Its canonical substrate is an append-only, provenance-preserving epistemic history. Pages and agent responses are reproducible projections over that history.

Brand Architecture

Knowledge that can show its work.

Validation workflow

Epistemedia's contribution CI is defined in .github/workflows/ci.yml and appears in GitHub Actions as Validate.

Autonomous Governance

Epistemedia governance controls procedure, authority, admission, resources, disclosure, and interoperability. It does not vote propositions into universal truth.

Human and Agent Interface Contract

Epistemedia does not maintain separate editorial truths for people, API clients, MCP agents, and local tools. Each interface adapts one disclosure-safe catalog and projection manifest.

Launching and Engaging with Epistemedia

Start at the human site and choose a topic. A topic page is a reproducible projection, not an anonymously edited canonical article. Inspect its lens, realm, frontier, policy, source objects, and manifest. Switch lenses to compare a coherent overview with evidence-first, skeptical, frontier, historical, pedagogical, or

Pre-public owner handoff

Archived bootstrap handoff. The repository-publication and protected-contribution steps below are complete. The default-hostname Pages bootstrap is also live. Use ops/activation/2026-08-22-pages-bootstrap.md for current evidence and remaining owner boundaries.

Epistemedia Roadmap

Compile the repository’s own architecture, governance, tasks, code, tests, and releases through the same public catalog used by every interface. Establish deterministic identity, disclosure safety, and agent-native operation.

Test Container Contract

from __future__ import annotations

Test Installed Package

from __future__ import annotations

Test Interfaces

from __future__ import annotations

Test License

from __future__ import annotations

Test Workflow Security

from __future__ import annotations

Projection manifest