Slice 05 — An Evidence Atlas for Self-Evolving AI Projects
Slice 05 — An Evidence Atlas for Self-Evolving AI Projects is an indexable Self Evolve survey publication slice that turns papers, projects, benchmarks, and public evidence into an article asset with explicit review boundaries.
Slice 05 — An Evidence Atlas for Self-Evolving AI Projects
One-Sentence Thesis
The long-term asset is not a survey PDF; it is an evidence atlas that connects papers, repos, benchmarks, model cards, social signals, and failure modes into one navigable map.
Reader Value
This slice turns the project into a practical tool: readers can search a method, inspect projects, compare evidence, and trace claims back to raw sources.
Three-Sentence Summary
Most AI surveys age quickly because the field changes faster than a PDF. The durable value is a living atlas: raw sources remain immutable, processed analysis becomes model cards and wiki pages, and the website exposes search, graph, blog, reports, and project pages. That makes the survey a gateway into an updating evidence system rather than a one-time document.
Evidence Chain
| Claim | Evidence |
|---|---|
| The repo is organized around raw, processed, work, results, and ops layers. | AGENTS.md, docs/project-management/project-structure.md, docs/indexes/master-index.md |
| Site routes already expose graph, search, blog, project pages, reports, and research pages. | site/src/pages/, site/src/content/ |
| Project reports and model cards are already published as site artifacts. | site/public/reports/projects/, projects/ |
| Wiki pages preserve cross-source synthesis. | work/wiki/index.md |
Public Report Shape
- Introduce the atlas as a way to inspect claims.
- Show the data pipeline: raw → wiki/analysis → survey/blog/model cards → site/report.
- Explain how a reader should use it: search a method, inspect graph, open project report, trace source.
- Highlight high-value pages: graph, projects, search, blog series.
- Invite issues/PRs for missing projects and evidence corrections.
HN Submission Candidate
Regular title: An evidence atlas for self-evolving AI projects
Show HN title later, if framed as a usable tool and not a reading list:
Show HN: An evidence atlas for self-evolving AI agents
Counterexamples And Limits
An atlas can become stale if indexing, source traces, and update cadence are weak. It needs recurring audits and clear [UNVERIFIED] labels where evidence is missing.