Superlog's findings gate stops short of resolution proof
A source audit of Superlog’s findings gate, outcome validation and declared server dispatch separates payload acceptance from incident-resolution proof.
Fodev JEO
LATEST INVESTIGATIONS
Source audits & field notes
DeepSeek Harness: What If the Agent Loop Itself Were Just Another Plugin?
DeepSeek open-sourced its agent harness with no privileged core — the model adapter, tool registry, session log, and the agent loop are all replaceable plugins. A look at what that architecture buy...
Self-Improving Agent Systems Require External State and Verifier Loops
A survey of autonomous agent compounding: why model weights stay frozen while state files, verifier sub-agents, and dynamic workflows accumulate improvements.
Obsidian's Rewrite Approval Setting Stops at Ingest
A pinned Obsidian Second Brain audit finds that ingest asks before rewriting by default, but reconcile instructions and an MCP write helper do not share that gate.
GraphRAG, Agent Memory, and Business Ontology Do Different Jobs
HyGRAG retrieves evidence, an LLM Wiki preserves synthesized knowledge, and Genie Ontology supplies governed business semantics; their update and trust boundaries differ.
LLM Serving Repositories Target Different Layers
A pinned-source audit finds that fifteen repositories span serving engines, runtimes, frameworks, kernels, training, and speech recognition—not one comparable speed ranking.
SuperGemma4's Perfect Code Score Misses a Syntax Error
A pinned audit of SuperGemma4’s 20-prompt quick bench finds one 10/10 code answer Python cannot parse, while baseline gains remain unreproducible.
Sentence Transformers v6.1 Makes Input Order Part of Your Index
Sentence Transformers keeps familiar calls, but v6.1 changes structured-message ordering. A pinned audit separates index compatibility, Qwen prompts and reranker scores.
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.
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.
인기 태그
Search article titles, categories, and tags. Full text is searched when needed.
Type to search articles.