Wiki Schema

Domain

LLM Wiki / Agent Memory / Context Compression / Knowledge Integration research and implementation framework.

Conventions

• File names: lowercase, hyphens, no spaces.
• Stable document structure is preserved across iterations.
• Every wiki page starts with YAML frontmatter.
• Use explicit source citations for every non-trivial claim.
• Distinguish: fact, inference, speculation, community view.
• Do not treat Twitter/X, Reddit, or Hacker News discussion as verified fact unless cross-validated by papers, docs, repos, or working implementations.
• If evidence is missing, write: unknown / insufficient evidence / speculative.
• Raw sources are immutable; corrections go into wiki pages.
• Raw sources must preserve parsed original/source text whenever accessible, not only source_url and extracted claims. A raw file may include analysis notes, but those must appear after the preserved source text and be clearly labeled.
• If extraction is blocked, truncated, or only summarized by tooling, mark raw_preservation and extraction_status explicitly and schedule a later full-text/PDF/API retrieval pass.
• Every update must append to log.md.
• Every file change must be synchronized through the wiki git repository: check git status, stage relevant files, commit with a clear message, and push if a remote is configured/available. If push is not possible, report the local commit hash and reason.

Required Research Document Sections

Executive Summary Core Thesis Key Concepts Karpathy Gist Analysis Architecture Patterns Existing Projects Community Consensus Major Debates Failure Cases Engineering Constraints Practical Integration Blueprint MVP Plan Recommended Stack Open Problems Research Questions Personal Developer Opportunities Source Map

Frontmatter

---
title: Page Title
created: YYYY-MM-DD
updated: YYYY-MM-DD
type: entity | concept | comparison | query | summary
tags: [from taxonomy below]
sources: [raw/...]
confidence: high | medium | low
contested: true | false
---

Tag Taxonomy

• topic: llm-wiki, agent-memory, rag, context-engineering, knowledge-integration, personal-ai-os
• architecture: memory-architecture, retrieval, compression, indexing, graph-memory, vector-memory, symbolic-memory
• evidence: paper, blog, github, hn, reddit, tweet, product-docs
• implementation: prototype, mvp, scalable-architecture, local-first, cloud-first
• analysis: tradeoff, failure-case, opportunity, debate

Page Thresholds

• Create or update a concept page when a claim affects architecture, engineering tradeoffs, or implementation direction.
• Avoid pages for passing mentions unless they are required by the stable research structure.
• Mark source quality and reliability explicitly.

Reliability Scale

• high: primary source, paper, official docs, reproducible repo, or direct author statement.
• medium: engineering blog, HN thread with author participation, community discussion with technical detail.
• low: search-result snippet, secondary summary, inaccessible Reddit/Twitter content, marketing copy, unverified claims.

Update Policy

When new evidence conflicts with existing content:

1. Keep the old view under a dated correction note.
2. Add the new evidence with source and date.
3. Mark contested: true if unresolved.
4. Update Source Map notes.