Fodev JEO

LATEST INVESTIGATIONS

Source audits & field notes

Transparent original diagram separating Feynman's date prompt context, direct alpha recovery branches, and telemetry error-text construction

Feynman Research Recovery Is Not Evidence Assurance

A pinned source audit of Feynman v0.5.47 separates signed-in alpha tools, abstract-only fallbacks, prompt context, and conditional telemetry error capture.

Apr 10, 2026 AI, Agents

#feynman #research-agents

AutoBE Release Diff: 133 of 134 Compiler Files Match

Source-pinned audit of AutoBE releases, write-before-complete guards, and realization-test cleanup boundaries—not a reliability benchmark.

Apr 7, 2026 AI, Tooling

#autobe #agentic-coding

LLM Wiki: Why Agent-Maintained Markdown Outperforms Ephemeral Vector RAG

An architectural evaluation of Andrej Karpathy’s LLM Wiki, comparing persistent markdown compilation against ephemeral RAG retrieval and analyzing 2026 wiki harnesses.

Apr 5, 2026 AI, Agents

#llm-wiki #rag

Original diagram separating Claw Dev launcher control flow from the Anthropic-shaped client endpoint and provider adapter data path

Claw Dev sets its session token after starting the proxy

Claw Dev starts or reuses its proxy before assigning the client token. A pinned audit traces authentication, buffered streams, and retired GitHub Models.

Apr 3, 2026 AI, Tooling

#claw-dev #coding-agent

A plan-derived worker task route, a separate four-field validator route, and a whole-plan review path

Engineering Discipline's handoff contradicts its gate

A commit-pinned Source Audit of Engineering Discipline’s worker–validator handoff, the inline-execution contradiction, and what its published benchmark can and cannot establish.

Apr 2, 2026 AI, Agents

#ai-engineering #coding-agents

Original request-contract diagram showing a client serializer keeping or dropping reasoning_content while both requests set preserve_thinking to true.

Qwen3.6-Plus Preserves Only Reasoning the Client Replays

Qwen3.6-Plus separates thinking from history replay. A pinned Qwen Code audit shows why preserve_thinking, idle cleanup and a 1M context window are different contracts.

Apr 2, 2026 AI, Agents

#qwen #ai-agents

PrismML 1-bit Bonsai: Extreme Low-Bit Quantization for Practical On-Device AI

A technical source audit of PrismML’s 1-bit Bonsai models, examining intelligence density, memory footprints, energy efficiency, and on-device agent deployment.

Apr 2, 2026 AI, Research

#prismml #bonsai

Client-Side Knowledge Base Indexing: SQLite, Pagefind, and Static Engines

Audit client-side search architectures—from SQLite-WASM and Pagefind chunked indexes to client-side vector search for technical documentation.

Apr 1, 2026 AI, Tooling

#search-indexing #sqlite

Decision map showing which Claude Code changes are conditionally recoverable with rewind and which remain outside checkpoint restoration

Claude Code rewind depends on who changed the file

A source-based Claude Code /rewind recovery matrix covering tool provenance, execution context, turn boundaries, path safety, and surviving snapshots.

Mar 31, 2026 AI, Agents

#claude-code #agentic-loop

Vibe Coding XR Architecture: Accelerating Spatial Prototyping with Gemini and XR Blocks

A source audit of Google Research’s Vibe Coding XR, evaluating natural language spatial intent, WebXR modular components, and rapid 3D interaction prototyping.

Mar 30, 2026 AI, Tooling

#vibe-coding #xr