{ "metadata": { "title": "Dark Factory Wiki Browser", "subtitle": "Offline human browser for the Markdown Dark Factory LLM wiki.", "schema_version": "1.0", "markdown_root": "reports/wiki", "canonical_index": "reports/wiki/index.md", "html_artifact": "reports/wiki/index.html", "json_artifact": "reports/wiki/wiki-data.json", "page_count": 28, "source_url_count": 11, "beads_issue": "dark-factory-e68" }, "counts": { "pages": 28, "types": { "comparison": 4, "concept": 6, "index": 1, "log": 1, "query": 3, "schema": 1, "source": 6, "system": 6 }, "graph_edges": 214, "open_tensions": 42, "source_urls": 11 }, "pages": [ { "title": "Dark Factory LLM Wiki", "slug": "index", "type": "index", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "index", "dark-factory", "llm-wiki" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_links": [], "related": [ "SCHEMA", "log", "dark-factory", "agentic-pipeline-runner" ], "wikilinks": [ "dark-factory", "what-is-the-durable-artifact-in-a-dark-factory", "dot-pipeline", "agentic-pipeline-runner", "code-review-vs-validation", "agent-orchestrator", "SCHEMA", "index", "log", "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo", "software-factory", "llm-wiki-pattern", "nlspec", "attractor-pattern", "strongdm-attractor", "digital-twin-universe", "cxdb", "healer", "llm-wiki-vs-rag", "tool-nodes-vs-llm-nodes", "specs-vs-code", "how-do-dark-factories-validate-without-reading-code", "how-do-dot-pipelines-map-to-agent-workflows" ], "tensions": [ "This wiki is a curated synthesis; source cards remain the provenance layer for checking exact claims." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "This is a durable Markdown wiki for the dark factory source set. It compiles article, spec, and repository sources into reusable pages so future agents can start from the synthesized map before returning to raw sources.", "headings": [ { "level": "1", "title": "Dark Factory LLM Wiki" }, { "level": "2", "title": "Best Starting Path" }, { "level": "2", "title": "Infrastructure" }, { "level": "2", "title": "Source Cards" }, { "level": "2", "title": "Concepts" }, { "level": "2", "title": "Systems" }, { "level": "2", "title": "Comparisons" }, { "level": "2", "title": "Queries" } ], "markdown_path": "index.md", "html_id": "page-index", "body_text": "Dark Factory LLM Wiki This is a durable Markdown wiki for the dark factory source set. It compiles article, spec, and repository sources into reusable pages so future agents can start from the synthesized map before returning to raw sources. ## Best Starting Path 1. Read dark-factory for the operating model. 2. Read what-is-the-durable-artifact-in-a-dark-factory for the durable-artifact synthesis. 3. Read dot-pipeline and agentic-pipeline-runner for the executable workflow layer. 4. Read code-review-vs-validation before making claims about not reading generated code. 5. Read agent-orchestrator when work involves worker/session/PR orchestration. ## Infrastructure - SCHEMA: frontmatter, page-type, provenance, and wikilink rules. - index: this navigation page. - log: ingest and maintenance history. ## Source Cards - dan-shapiro-you-dont-write-code-2026-02-13: primary Shapiro article on not writing or reading generated code, StrongDM's validation posture, CXDB, and Healer. - 2389-dark-factory-dot-file-2026-03-09: primary 2389 post on DOT pipelines as the dark factory artifact and the attractor pattern. - strongdm-attractor-nlspec: StrongDM Attractor NLSpec repository for unified LLM, coding loop, and DOT runner specs. - karpathy-llm-wiki-gist-2026-04-04: LLM wiki pattern source that motivates persistent Markdown synthesis. - jleechanorg-dark-factory-repo: concrete Python DOT pipeline runner repository with CXDB, Healer, backends, benchmarks, and holdout isolation. - jleechanorg-agent-orchestrator-repo: AO fork repository for fleet/session/worktree/PR orchestration and evidence/skeptic automation. ## Concepts - dark-factory: software production model where humans specify intent and validation policy while agents generate and validate code. - software-factory: broader repeatable system for turning intent into software artifacts. - llm-wiki-pattern: persistent Markdown synthesis pattern for compounding agent knowledge. - nlspec: natural-language specification intended for coding agents to implement. - dot-pipeline: Graphviz-directed process graph used as an executable workflow definition. - attractor-pattern: three-layer convergence around LLM client, coding loop, and pipeline engine. ## Systems - strongdm-attractor: specification corpus for Attractor-style software factory layers. - digital-twin-universe: high-fidelity local test universe for validating generated systems. - cxdb: event-log and observability layer for pipeline runs. - healer: diagnosis layer that clusters factory failures and supports repair loops. - agentic-pipeline-runner: runner that parses DOT graphs, invokes agents/tools/gates, and records outcomes. - agent-orchestrator: worker/session/worktree/PR orchestration layer that can serve as a factory backend or adjacent substrate. ## Comparisons - llm-wiki-vs-rag: maintained synthesis versus query-time retrieval over raw sources. - code-review-vs-validation: human diff inspection versus layered outcome evidence. - tool-nodes-vs-llm-nodes: deterministic tool execution versus model-backed reasoning nodes. - specs-vs-code: durable intent/process artifacts versus generated executable output. ## Queries - what-is-the-durable-artifact-in-a-dark-factory: answer to whether DOT, specs, validation, or code is the durable layer. - how-do-dark-factories-validate-without-reading-code: validation stack for no-diff-review workflows. - how-do-dot-pipelines-map-to-agent-workflows: mapping of graph nodes to agent, tool, gate, join, and human workflow roles.", "search_text": "Dark Factory LLM Wiki index index high index dark-factory llm-wiki dan-shapiro-you-dont-write-code-2026-02-13 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec karpathy-llm-wiki-gist-2026-04-04 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo SCHEMA log dark-factory agentic-pipeline-runner dark-factory what-is-the-durable-artifact-in-a-dark-factory dot-pipeline agentic-pipeline-runner code-review-vs-validation agent-orchestrator SCHEMA index log dan-shapiro-you-dont-write-code-2026-02-13 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec karpathy-llm-wiki-gist-2026-04-04 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo software-factory llm-wiki-pattern nlspec attractor-pattern strongdm-attractor digital-twin-universe cxdb healer llm-wiki-vs-rag tool-nodes-vs-llm-nodes specs-vs-code how-do-dark-factories-validate-without-reading-code how-do-dot-pipelines-map-to-agent-workflows This is a durable Markdown wiki for the dark factory source set. It compiles article, spec, and repository sources into reusable pages so future agents can start from the synthesized map before returning to raw sources. Dark Factory LLM Wiki This is a durable Markdown wiki for the dark factory source set. It compiles article, spec, and repository sources into reusable pages so future agents can start from the synthesized map before returning to raw sources. ## Best Starting Path 1. Read dark-factory for the operating model. 2. Read what-is-the-durable-artifact-in-a-dark-factory for the durable-artifact synthesis. 3. Read dot-pipeline and agentic-pipeline-runner for the executable workflow layer. 4. Read code-review-vs-validation before making claims about not reading generated code. 5. Read agent-orchestrator when work involves worker/session/PR orchestration. ## Infrastructure - SCHEMA: frontmatter, page-type, provenance, and wikilink rules. - index: this navigation page. - log: ingest and maintenance history. ## Source Cards - dan-shapiro-you-dont-write-code-2026-02-13: primary Shapiro article on not writing or reading generated code, StrongDM's validation posture, CXDB, and Healer. - 2389-dark-factory-dot-file-2026-03-09: primary 2389 post on DOT pipelines as the dark factory artifact and the attractor pattern. - strongdm-attractor-nlspec: StrongDM Attractor NLSpec repository for unified LLM, coding loop, and DOT runner specs. - karpathy-llm-wiki-gist-2026-04-04: LLM wiki pattern source that motivates persistent Markdown synthesis. - jleechanorg-dark-factory-repo: concrete Python DOT pipeline runner repository with CXDB, Healer, backends, benchmarks, and holdout isolation. - jleechanorg-agent-orchestrator-repo: AO fork repository for fleet/session/worktree/PR orchestration and evidence/skeptic automation. ## Concepts - dark-factory: software production model where humans specify intent and validation policy while agents generate and validate code. - software-factory: broader repeatable system for turning intent into software artifacts. - llm-wiki-pattern: persistent Markdown synthesis pattern for compounding agent knowledge. - nlspec: natural-language specification intended for coding agents to implement. - dot-pipeline: Graphviz-directed process graph used as an executable workflow definition. - attractor-pattern: three-layer convergence around LLM client, coding loop, and pipeline engine. ## Systems - strongdm-attractor: specification corpus for Attractor-style software factory layers. - digital-twin-universe: high-fidelity local test universe for validating generated systems. - cxdb: event-log and observability layer for pipeline runs. - healer: diagnosis layer that clusters factory failures and supports repair loops. - agentic-pipeline-runner: runner that parses DOT graphs, invokes agents/tools/gates, and records outcomes. - agent-orchestrator: worker/session/worktree/PR orchestration layer that can serve as a factory backend or adjacent substrate. ## Comparisons - llm-wiki-vs-rag: maintained synthesis versus query-time retrieval over raw sources. - code-review-vs-validation: human diff inspection versus layered outcome evidence. - tool-nodes-vs-llm-nodes: deterministic tool execution versus model-backed reasoning nodes. - specs-vs-code: durable intent/process artifacts versus generated executable output. ## Queries - what-is-the-durable-artifact-in-a-dark-factory: answer to whether DOT, specs, validation, or code is the durable layer. - how-do-dark-factories-validate-without-reading-code: validation stack for no-diff-review workflows. - how-do-dot-pipelines-map-to-agent-workflows: mapping of graph nodes to agent, tool, gate, join, and human workflow roles.", "backlinks": [ { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "related", "wikilink" ] }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "kinds": [ "related" ] }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "kinds": [ "wikilink" ] } ] }, { "title": "Dark Factory Wiki Schema", "slug": "SCHEMA", "type": "schema", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "schema", "llm-wiki", "dark-factory" ], "sources": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_refs": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_links": [], "related": [ "index", "log", "llm-wiki-pattern" ], "wikilinks": [ "llm-wiki-pattern", "dark-factory" ], "tensions": [], "bead_refs": [ "dark-factory-rr4" ], "summary": "This wiki is a persistent Markdown knowledge base for the dark factory, software factory, DOT pipeline, NLSpec, and Attractor source set. It follows the llm-wiki-pattern: source cards preserve provenance, compiled pages preserve reusable...", "headings": [ { "level": "1", "title": "Dark Factory Wiki Schema" }, { "level": "2", "title": "Frontmatter" }, { "level": "2", "title": "Page Types" }, { "level": "2", "title": "Claim Handling" }, { "level": "2", "title": "Link Conventions" } ], "markdown_path": "SCHEMA.md", "html_id": "page-schema", "body_text": "Dark Factory Wiki Schema This wiki is a persistent Markdown knowledge base for the dark factory, software factory, DOT pipeline, NLSpec, and Attractor source set. It follows the llm-wiki-pattern: source cards preserve provenance, compiled pages preserve reusable synthesis, and query pages preserve answers future agents are likely to need again. ## Frontmatter Every Markdown page uses the same fields: - title: human-readable title. - slug: stable page ID. Use lowercase kebab case except infrastructure pages such as SCHEMA. - type: one of index, schema, log, source, concept, system, comparison, or query. - created: date first added to this wiki. - updated: date last materially changed. - status: active, seed, or needs-review. - tags: compact retrieval hints. - sources: source card slugs or raw URLs that support the page. - related: internal page slugs that should be read next. - confidence: high, medium-high, medium, or low. - contradictions: preserved tensions, disagreements, or caveats. - beadrefs: Beads issue IDs connected to the work. ## Page Types Source pages store provenance, capture dates, compact takeaways, and explicit caveats. Concept pages store reusable vocabulary and cross-source claims. System pages describe concrete systems or subsystems. Comparison pages sharpen boundaries. Query pages answer likely future questions with reusable synthesis. ## Claim Handling Do not paste full article text into the wiki. Use source cards and outbound source URLs as provenance. When a claim combines multiple sources, prefix it with inference: so future agents can distinguish synthesis from direct source reporting. ## Link Conventions Use wikilinks such as dark-factory for local pages. The link text must match a local page slug or infrastructure page. Use normal Markdown links only for source URLs and external references.", "search_text": "Dark Factory Wiki Schema SCHEMA schema high schema llm-wiki dark-factory karpathy-llm-wiki-gist-2026-04-04 index log llm-wiki-pattern llm-wiki-pattern dark-factory This wiki is a persistent Markdown knowledge base for the dark factory, software factory, DOT pipeline, NLSpec, and Attractor source set. It follows the llm-wiki-pattern: source cards preserve provenance, compiled pages preserve reusable... Dark Factory Wiki Schema This wiki is a persistent Markdown knowledge base for the dark factory, software factory, DOT pipeline, NLSpec, and Attractor source set. It follows the llm-wiki-pattern: source cards preserve provenance, compiled pages preserve reusable synthesis, and query pages preserve answers future agents are likely to need again. ## Frontmatter Every Markdown page uses the same fields: - title: human-readable title. - slug: stable page ID. Use lowercase kebab case except infrastructure pages such as SCHEMA. - type: one of index, schema, log, source, concept, system, comparison, or query. - created: date first added to this wiki. - updated: date last materially changed. - status: active, seed, or needs-review. - tags: compact retrieval hints. - sources: source card slugs or raw URLs that support the page. - related: internal page slugs that should be read next. - confidence: high, medium-high, medium, or low. - contradictions: preserved tensions, disagreements, or caveats. - beadrefs: Beads issue IDs connected to the work. ## Page Types Source pages store provenance, capture dates, compact takeaways, and explicit caveats. Concept pages store reusable vocabulary and cross-source claims. System pages describe concrete systems or subsystems. Comparison pages sharpen boundaries. Query pages answer likely future questions with reusable synthesis. ## Claim Handling Do not paste full article text into the wiki. Use source cards and outbound source URLs as provenance. When a claim combines multiple sources, prefix it with inference: so future agents can distinguish synthesis from direct source reporting. ## Link Conventions Use wikilinks such as dark-factory for local pages. The link text must match a local page slug or infrastructure page. Use normal Markdown links only for source URLs and external references.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "related", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "related", "wikilink" ] }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "kinds": [ "related" ] }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "kinds": [ "related" ] }, { "slug": "llm-wiki-vs-rag", "title": "LLM Wiki vs RAG", "type": "comparison", "kinds": [ "related" ] } ] }, { "title": "Dark Factory Wiki Log", "slug": "log", "type": "log", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "log", "ingest", "dark-factory" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_links": [ { "label": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "url": "https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/", "origin": "body" }, { "label": "2389: The Dark Factory Is a .dot file", "url": "https://2389.ai/posts/the-dark-factory-is-a-dot-file/", "origin": "body" }, { "label": "strongdm/attractor", "url": "https://github.com/strongdm/attractor", "origin": "body" }, { "label": "Karpathy LLM Wiki gist", "url": "https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f", "origin": "body" }, { "label": "jleechanorg/dark-factory", "url": "https://github.com/jleechanorg/dark-factory", "origin": "body" }, { "label": "jleechanorg/agent-orchestrator", "url": "https://github.com/jleechanorg/agent-orchestrator", "origin": "body" } ], "related": [ "index", "SCHEMA" ], "wikilinks": [ "SCHEMA", "index", "log", "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "karpathy-llm-wiki-gist-2026-04-04", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo", "dark-factory", "software-factory", "llm-wiki-pattern", "nlspec", "dot-pipeline", "attractor-pattern", "strongdm-attractor", "digital-twin-universe", "cxdb", "healer", "agentic-pipeline-runner", "agent-orchestrator", "llm-wiki-vs-rag", "code-review-vs-validation", "tool-nodes-vs-llm-nodes", "specs-vs-code", "what-is-the-durable-artifact-in-a-dark-factory", "how-do-dark-factories-validate-without-reading-code", "how-do-dot-pipelines-map-to-agent-workflows" ], "tensions": [ "The initial ingest compiles point-in-time repository snapshots; repository claims should be refreshed before current operational decisions." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Added a standalone offline browsing layer for humans:", "headings": [ { "level": "1", "title": "Dark Factory Wiki Log" }, { "level": "2", "title": "2026-06-17 Visual Browsing Layer" }, { "level": "2", "title": "2026-06-17 Initial Ingest" }, { "level": "2", "title": "Sources Captured" }, { "level": "2", "title": "Pages Created" }, { "level": "2", "title": "Synthesis Notes" } ], "markdown_path": "log.md", "html_id": "page-log", "body_text": "Dark Factory Wiki Log ## 2026-06-17 Visual Browsing Layer Added a standalone offline browsing layer for humans: - index.html: generated light-theme wiki browser with sticky navigation, search, type filters, page cards, backlinks, source/provenance chips, open tensions, maintenance links, and an inline factory graph. - wiki-data.json: generated structured sidecar with every Markdown page, unique slugs, page types, tags, source URLs, related links, backlinks, tensions, and graph edges. - buildviewer.py: Python standard-library generator and --check validator. Canonical knowledge remains in the Markdown wiki pages. Regenerate with python3 reports/wiki/buildviewer.py and verify with python3 reports/wiki/buildviewer.py --check. ## 2026-06-17 Initial Ingest Created reports/wiki/ as a persistent LLM wiki for the dark factory source set. ## Sources Captured - Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either. https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/ - 2389: The Dark Factory Is a .dot file https://2389.ai/posts/the-dark-factory-is-a-dot-file/ - strongdm/attractor https://github.com/strongdm/attractor - Karpathy LLM Wiki gist https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f - jleechanorg/dark-factory https://github.com/jleechanorg/dark-factory, observed main HEAD 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae - jleechanorg/agent-orchestrator https://github.com/jleechanorg/agent-orchestrator, observed main HEAD 3923e3cce815b541af59c2756945d184fde4ac25 ## Pages Created - Infrastructure: SCHEMA, index, log. - Sources: dan-shapiro-you-dont-write-code-2026-02-13, 2389-dark-factory-dot-file-2026-03-09, strongdm-attractor-nlspec, karpathy-llm-wiki-gist-2026-04-04, jleechanorg-dark-factory-repo, jleechanorg-agent-orchestrator-repo. - Concepts: dark-factory, software-factory, llm-wiki-pattern, nlspec, dot-pipeline, attractor-pattern. - Systems: strongdm-attractor, digital-twin-universe, cxdb, healer, agentic-pipeline-runner, agent-orchestrator. - Comparisons: llm-wiki-vs-rag, code-review-vs-validation, tool-nodes-vs-llm-nodes, specs-vs-code. - Queries: what-is-the-durable-artifact-in-a-dark-factory, how-do-dark-factories-validate-without-reading-code, how-do-dot-pipelines-map-to-agent-workflows. ## Synthesis Notes - Preserved the main tension: dark factory aspirations reduce human code reading, but validation design and human-owned policy become more important. - Added repository sources requested after initial implementation started: jleechanorg-dark-factory-repo and jleechanorg-agent-orchestrator-repo. - Distinguished agentic-pipeline-runner from agent-orchestrator: the runner owns workflow graph execution; the orchestrator owns worker/session/worktree/PR lifecycle.", "search_text": "Dark Factory Wiki Log log log high log ingest dark-factory dan-shapiro-you-dont-write-code-2026-02-13 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec karpathy-llm-wiki-gist-2026-04-04 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo index SCHEMA SCHEMA index log dan-shapiro-you-dont-write-code-2026-02-13 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec karpathy-llm-wiki-gist-2026-04-04 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo dark-factory software-factory llm-wiki-pattern nlspec dot-pipeline attractor-pattern strongdm-attractor digital-twin-universe cxdb healer agentic-pipeline-runner agent-orchestrator llm-wiki-vs-rag code-review-vs-validation tool-nodes-vs-llm-nodes specs-vs-code what-is-the-durable-artifact-in-a-dark-factory how-do-dark-factories-validate-without-reading-code how-do-dot-pipelines-map-to-agent-workflows Added a standalone offline browsing layer for humans: Dark Factory Wiki Log ## 2026-06-17 Visual Browsing Layer Added a standalone offline browsing layer for humans: - index.html: generated light-theme wiki browser with sticky navigation, search, type filters, page cards, backlinks, source/provenance chips, open tensions, maintenance links, and an inline factory graph. - wiki-data.json: generated structured sidecar with every Markdown page, unique slugs, page types, tags, source URLs, related links, backlinks, tensions, and graph edges. - buildviewer.py: Python standard-library generator and --check validator. Canonical knowledge remains in the Markdown wiki pages. Regenerate with python3 reports/wiki/buildviewer.py and verify with python3 reports/wiki/buildviewer.py --check. ## 2026-06-17 Initial Ingest Created reports/wiki/ as a persistent LLM wiki for the dark factory source set. ## Sources Captured - Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either. https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/ - 2389: The Dark Factory Is a .dot file https://2389.ai/posts/the-dark-factory-is-a-dot-file/ - strongdm/attractor https://github.com/strongdm/attractor - Karpathy LLM Wiki gist https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f - jleechanorg/dark-factory https://github.com/jleechanorg/dark-factory, observed main HEAD 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae - jleechanorg/agent-orchestrator https://github.com/jleechanorg/agent-orchestrator, observed main HEAD 3923e3cce815b541af59c2756945d184fde4ac25 ## Pages Created - Infrastructure: SCHEMA, index, log. - Sources: dan-shapiro-you-dont-write-code-2026-02-13, 2389-dark-factory-dot-file-2026-03-09, strongdm-attractor-nlspec, karpathy-llm-wiki-gist-2026-04-04, jleechanorg-dark-factory-repo, jleechanorg-agent-orchestrator-repo. - Concepts: dark-factory, software-factory, llm-wiki-pattern, nlspec, dot-pipeline, attractor-pattern. - Systems: strongdm-attractor, digital-twin-universe, cxdb, healer, agentic-pipeline-runner, agent-orchestrator. - Comparisons: llm-wiki-vs-rag, code-review-vs-validation, tool-nodes-vs-llm-nodes, specs-vs-code. - Queries: what-is-the-durable-artifact-in-a-dark-factory, how-do-dark-factories-validate-without-reading-code, how-do-dot-pipelines-map-to-agent-workflows. ## Synthesis Notes - Preserved the main tension: dark factory aspirations reduce human code reading, but validation design and human-owned policy become more important. - Added repository sources requested after initial implementation started: jleechanorg-dark-factory-repo and jleechanorg-agent-orchestrator-repo. - Distinguished agentic-pipeline-runner from agent-orchestrator: the runner owns workflow graph execution; the orchestrator owns worker/session/worktree/PR lifecycle.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "related", "wikilink" ] }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "kinds": [ "related" ] }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "kinds": [ "related", "wikilink" ] } ] }, { "title": "Attractor Pattern", "slug": "attractor-pattern", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "attractor", "architecture", "convergence" ], "sources": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_refs": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "strongdm-attractor", "agentic-pipeline-runner", "dot-pipeline" ], "wikilinks": [ "strongdm-attractor", "agentic-pipeline-runner", "agent-orchestrator" ], "tensions": [ "Independent convergence is plausible from the sources, but the evidence is a small cluster of related implementations and specs." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "The attractor pattern is the repeated three-layer architecture that appears in StrongDM Attractor-style systems:", "headings": [ { "level": "1", "title": "Attractor Pattern" }, { "level": "2", "title": "Why It Converges" }, { "level": "2", "title": "Local Examples" } ], "markdown_path": "concepts/attractor-pattern.md", "html_id": "page-attractor-pattern", "body_text": "Attractor Pattern The attractor pattern is the repeated three-layer architecture that appears in StrongDM Attractor-style systems: - Unified LLM/client layer. - Coding agent loop or worker/session layer. - DOT pipeline engine that orchestrates multi-step workflows. ## Why It Converges inference: Systems that try to automate software work repeatedly need provider abstraction, an execution loop for agent/tool interaction, and a higher-level process graph. Those pressures make the three-layer shape an architectural attractor. ## Local Examples - strongdm-attractor specifies the layers as NLSpecs. - agentic-pipeline-runner implements a DOT pipeline runner and integrates multiple agent backends. - agent-orchestrator handles fleet/session/worktree orchestration and can sit under a factory runner as a worker substrate.", "search_text": "Attractor Pattern attractor-pattern concept medium-high attractor architecture convergence 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec jleechanorg-dark-factory-repo strongdm-attractor agentic-pipeline-runner dot-pipeline strongdm-attractor agentic-pipeline-runner agent-orchestrator The attractor pattern is the repeated three-layer architecture that appears in StrongDM Attractor-style systems: Attractor Pattern The attractor pattern is the repeated three-layer architecture that appears in StrongDM Attractor-style systems: - Unified LLM/client layer. - Coding agent loop or worker/session layer. - DOT pipeline engine that orchestrates multi-step workflows. ## Why It Converges inference: Systems that try to automate software work repeatedly need provider abstraction, an execution loop for agent/tool interaction, and a higher-level process graph. Those pressures make the three-layer shape an architectural attractor. ## Local Examples - strongdm-attractor specifies the layers as NLSpecs. - agentic-pipeline-runner implements a DOT pipeline runner and integrates multiple agent backends. - agent-orchestrator handles fleet/session/worktree orchestration and can sit under a factory runner as a worker substrate.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "related" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "related", "wikilink" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "wikilink" ] } ] }, { "title": "Dark Factory", "slug": "dark-factory", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "dark-factory", "automation", "validation", "software-engineering" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "software-factory", "dot-pipeline", "code-review-vs-validation", "digital-twin-universe" ], "wikilinks": [ "digital-twin-universe", "dot-pipeline", "healer" ], "tensions": [ "Dark factory rhetoric minimizes human code reading, but practical systems still need humans to define intent, gates, threat models, and acceptable risk.", "Human gates may be necessary, but too many gates recreate the manual bottleneck.", "Disposable generated code sounds freeing, but the runner, tests, and observability stack become critical infrastructure.", "\"Do not read code\" creates a security problem unless validation can catch unsafe behavior and supply-chain drift." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "A dark factory is a software production model where humans specify intent and validation policy while automated agents generate, test, repair, and ship code with minimal direct human reading of generated diffs.", "headings": [ { "level": "1", "title": "Dark Factory" }, { "level": "2", "title": "Core Claims" }, { "level": "2", "title": "Practical Definition" }, { "level": "2", "title": "Open Tensions" } ], "markdown_path": "concepts/dark-factory.md", "html_id": "page-dark-factory", "body_text": "Dark Factory A dark factory is a software production model where humans specify intent and validation policy while automated agents generate, test, repair, and ship code with minimal direct human reading of generated diffs. ## Core Claims - The bottleneck shifts from writing code to designing a system that can safely create and validate code. - inference: The durable work is not the emitted code alone; it is the factory definition: specs, process graphs, validation contracts, holdouts, evidence, and observability. - Code review becomes less central than outcome validation, independent review, sandboxing, and telemetry. - The strongest versions require realistic evaluation environments such as digital-twin-universe, durable process artifacts such as dot-pipeline, and repair loops such as healer. ## Practical Definition A dark factory is not \"LLM writes a patch.\" It is a production pipeline with explicit inputs, agent roles, gates, isolated evaluation, failure logging, and policy for retries or human escalation. ## Open Tensions - Human gates may be necessary, but too many gates recreate the manual bottleneck. - Disposable generated code sounds freeing, but the runner, tests, and observability stack become critical infrastructure. - \"Do not read code\" creates a security problem unless validation can catch unsafe behavior and supply-chain drift.", "search_text": "Dark Factory dark-factory concept medium-high dark-factory automation validation software-engineering dan-shapiro-you-dont-write-code-2026-02-13 2389-dark-factory-dot-file-2026-03-09 jleechanorg-dark-factory-repo software-factory dot-pipeline code-review-vs-validation digital-twin-universe digital-twin-universe dot-pipeline healer A dark factory is a software production model where humans specify intent and validation policy while automated agents generate, test, repair, and ship code with minimal direct human reading of generated diffs. Dark Factory A dark factory is a software production model where humans specify intent and validation policy while automated agents generate, test, repair, and ship code with minimal direct human reading of generated diffs. ## Core Claims - The bottleneck shifts from writing code to designing a system that can safely create and validate code. - inference: The durable work is not the emitted code alone; it is the factory definition: specs, process graphs, validation contracts, holdouts, evidence, and observability. - Code review becomes less central than outcome validation, independent review, sandboxing, and telemetry. - The strongest versions require realistic evaluation environments such as digital-twin-universe, durable process artifacts such as dot-pipeline, and repair loops such as healer. ## Practical Definition A dark factory is not \"LLM writes a patch.\" It is a production pipeline with explicit inputs, agent roles, gates, isolated evaluation, failure logging, and policy for retries or human escalation. ## Open Tensions - Human gates may be necessary, but too many gates recreate the manual bottleneck. - Disposable generated code sounds freeing, but the runner, tests, and observability stack become critical infrastructure. - \"Do not read code\" creates a security problem unless validation can catch unsafe behavior and supply-chain drift.", "backlinks": [ { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "wikilink" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "related", "wikilink" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "related", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "kinds": [ "wikilink" ] }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "kinds": [ "related" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "wikilink" ] }, { "slug": "healer", "title": "Healer", "type": "system", "kinds": [ "related" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "related", "wikilink" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "related" ] } ] }, { "title": "DOT Pipeline", "slug": "dot-pipeline", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "dot", "pipeline", "graphviz", "process-graph" ], "sources": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_refs": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "agentic-pipeline-runner", "tool-nodes-vs-llm-nodes", "how-do-dot-pipelines-map-to-agent-workflows" ], "wikilinks": [ "dark-factory", "jleechanorg-dark-factory-repo", "strongdm-attractor-nlspec" ], "tensions": [ "DOT is simple and inspectable, but complex software lifecycle graphs can still become hard to reason about without schema validation and visual conventions." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "A DOT pipeline is a Graphviz directed graph used as an executable workflow definition. Nodes represent work, tools, gates, human approvals, joins, exits, or model calls. Edges represent routing, dependencies, retries, success/failure...", "headings": [ { "level": "1", "title": "DOT Pipeline" }, { "level": "2", "title": "Why DOT" }, { "level": "2", "title": "Factory Role" }, { "level": "2", "title": "Implementation Notes" } ], "markdown_path": "concepts/dot-pipeline.md", "html_id": "page-dot-pipeline", "body_text": "DOT Pipeline A DOT pipeline is a Graphviz directed graph used as an executable workflow definition. Nodes represent work, tools, gates, human approvals, joins, exits, or model calls. Edges represent routing, dependencies, retries, success/failure branches, and joins. ## Why DOT - It is text, versionable, diffable, and visually renderable. - It can encode both linear and branching workflows. - It separates the process blueprint from a specific runner implementation. ## Factory Role In dark-factory, a DOT pipeline can be the durable artifact that records how intent becomes validated software. The runner may change; the graph preserves the development process. ## Implementation Notes The jleechanorg-dark-factory-repo uses DOT graphs for task-sized lanes, minimal feature loops, PR review, bug-fix TDD, gates, and benchmarks. The strongdm-attractor-nlspec defines a generalized DOT runner with typed attributes, handlers, checkpointing, human gates, model stylesheets, validation, and extensibility.", "search_text": "DOT Pipeline dot-pipeline concept high dot pipeline graphviz process-graph 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec jleechanorg-dark-factory-repo agentic-pipeline-runner tool-nodes-vs-llm-nodes how-do-dot-pipelines-map-to-agent-workflows dark-factory jleechanorg-dark-factory-repo strongdm-attractor-nlspec A DOT pipeline is a Graphviz directed graph used as an executable workflow definition. Nodes represent work, tools, gates, human approvals, joins, exits, or model calls. Edges represent routing, dependencies, retries, success/failure... DOT Pipeline A DOT pipeline is a Graphviz directed graph used as an executable workflow definition. Nodes represent work, tools, gates, human approvals, joins, exits, or model calls. Edges represent routing, dependencies, retries, success/failure branches, and joins. ## Why DOT - It is text, versionable, diffable, and visually renderable. - It can encode both linear and branching workflows. - It separates the process blueprint from a specific runner implementation. ## Factory Role In dark-factory, a DOT pipeline can be the durable artifact that records how intent becomes validated software. The runner may change; the graph preserves the development process. ## Implementation Notes The jleechanorg-dark-factory-repo uses DOT graphs for task-sized lanes, minimal feature loops, PR review, bug-fix TDD, gates, and benchmarks. The strongdm-attractor-nlspec defines a generalized DOT runner with typed attributes, handlers, checkpointing, human gates, model stylesheets, validation, and extensibility.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "wikilink" ] }, { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "related", "wikilink" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "related" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "related", "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "related" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "related" ] }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "kinds": [ "related" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "related" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "related" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "related" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "related" ] } ] }, { "title": "LLM Wiki Pattern", "slug": "llm-wiki-pattern", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "llm-wiki", "knowledge-base", "markdown" ], "sources": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_refs": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_links": [], "related": [ "llm-wiki-vs-rag", "SCHEMA", "log" ], "wikilinks": [ "index", "log" ], "tensions": [ "LLM-maintained synthesis compounds learning, but can also compound stale or mistaken synthesis unless source cards and logs stay current." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "The LLM wiki pattern turns repeated research into a durable Markdown knowledge base. Raw sources remain provenance; the wiki stores curated source cards, compiled topic pages, comparison pages, query notes, backlinks, and update logs.", "headings": [ { "level": "1", "title": "LLM Wiki Pattern" }, { "level": "2", "title": "Why It Matters Here" }, { "level": "2", "title": "Maintenance Rules" } ], "markdown_path": "concepts/llm-wiki-pattern.md", "html_id": "page-llm-wiki-pattern", "body_text": "LLM Wiki Pattern The LLM wiki pattern turns repeated research into a durable Markdown knowledge base. Raw sources remain provenance; the wiki stores curated source cards, compiled topic pages, comparison pages, query notes, backlinks, and update logs. ## Why It Matters Here Future agents should be able to read this wiki before returning to the Shapiro, 2389, StrongDM, Karpathy, and repository sources. The goal is not to hide sources; it is to avoid re-deriving the same concept map every session. ## Maintenance Rules - Add a source card for every major source. - Update concept pages when new sources change the synthesis. - Preserve contradictions instead of smoothing them away. - Keep index and log current.", "search_text": "LLM Wiki Pattern llm-wiki-pattern concept high llm-wiki knowledge-base markdown karpathy-llm-wiki-gist-2026-04-04 llm-wiki-vs-rag SCHEMA log index log The LLM wiki pattern turns repeated research into a durable Markdown knowledge base. Raw sources remain provenance; the wiki stores curated source cards, compiled topic pages, comparison pages, query notes, backlinks, and update logs. LLM Wiki Pattern The LLM wiki pattern turns repeated research into a durable Markdown knowledge base. Raw sources remain provenance; the wiki stores curated source cards, compiled topic pages, comparison pages, query notes, backlinks, and update logs. ## Why It Matters Here Future agents should be able to read this wiki before returning to the Shapiro, 2389, StrongDM, Karpathy, and repository sources. The goal is not to hide sources; it is to avoid re-deriving the same concept map every session. ## Maintenance Rules - Add a source card for every major source. - Update concept pages when new sources change the synthesis. - Preserve contradictions instead of smoothing them away. - Keep index and log current.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "kinds": [ "related", "wikilink" ] }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "llm-wiki-vs-rag", "title": "LLM Wiki vs RAG", "type": "comparison", "kinds": [ "related" ] } ] }, { "title": "NLSpec", "slug": "nlspec", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "nlspec", "specification", "agent-readable" ], "sources": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09" ], "source_refs": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09" ], "source_links": [], "related": [ "strongdm-attractor", "specs-vs-code", "dot-pipeline" ], "wikilinks": [], "tensions": [ "Natural language specs are accessible to agents and humans, but ambiguity must be controlled with validation and examples." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "An NLSpec is a natural-language specification written to be implemented by coding agents. In the StrongDM Attractor repository, NLSpecs define desired behavior for a unified LLM client, a coding agent loop, and a DOT pipeline engine.", "headings": [ { "level": "1", "title": "NLSpec" }, { "level": "2", "title": "Role In The Factory" }, { "level": "2", "title": "Risk" } ], "markdown_path": "concepts/nlspec.md", "html_id": "page-nlspec", "body_text": "NLSpec An NLSpec is a natural-language specification written to be implemented by coding agents. In the StrongDM Attractor repository, NLSpecs define desired behavior for a unified LLM client, a coding agent loop, and a DOT pipeline engine. ## Role In The Factory - It is the contract between human intent and generated implementation. - It is durable enough to rebuild implementation code from scratch. - It can define interfaces, validation behavior, error cases, tool contracts, and extension points. ## Risk inference: NLSpecs are stronger than casual prompts but weaker than executable tests. A dark factory needs NLSpecs plus validation: conformance tests, holdouts, static checks, and observability.", "search_text": "NLSpec nlspec concept high nlspec specification agent-readable strongdm-attractor-nlspec 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor specs-vs-code dot-pipeline An NLSpec is a natural-language specification written to be implemented by coding agents. In the StrongDM Attractor repository, NLSpecs define desired behavior for a unified LLM client, a coding agent loop, and a DOT pipeline engine. NLSpec An NLSpec is a natural-language specification written to be implemented by coding agents. In the StrongDM Attractor repository, NLSpecs define desired behavior for a unified LLM client, a coding agent loop, and a DOT pipeline engine. ## Role In The Factory - It is the contract between human intent and generated implementation. - It is durable enough to rebuild implementation code from scratch. - It can define interfaces, validation behavior, error cases, tool contracts, and extension points. ## Risk inference: NLSpecs are stronger than casual prompts but weaker than executable tests. A dark factory needs NLSpecs plus validation: conformance tests, holdouts, static checks, and observability.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "related" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "related" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "related", "wikilink" ] } ] }, { "title": "Software Factory", "slug": "software-factory", "type": "concept", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium", "tags": [ "software-factory", "industrialization", "automation" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "dark-factory", "strongdm-attractor", "agentic-pipeline-runner" ], "wikilinks": [ "dark-factory" ], "tensions": [ "A software factory can still include human reading and manual approval; dark factory is a more aggressive variant." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "A software factory is a repeatable production system for turning intent into software artifacts. In this wiki, it is the broader category; dark-factory is the lights-off variant that tries to avoid human code reading.", "headings": [ { "level": "1", "title": "Software Factory" }, { "level": "2", "title": "Components" }, { "level": "2", "title": "Relation To Dark Factory" } ], "markdown_path": "concepts/software-factory.md", "html_id": "page-software-factory", "body_text": "Software Factory A software factory is a repeatable production system for turning intent into software artifacts. In this wiki, it is the broader category; dark-factory is the lights-off variant that tries to avoid human code reading. ## Components - Input artifacts: goals, specs, product requirements, examples, acceptance criteria. - Execution machinery: coding agents, tool registries, isolated worktrees, CI, runners, and queues. - Validation machinery: tests, review agents, holdouts, policy gates, observability, evidence bundles. - Feedback machinery: retries, fix loops, failure diagnosis, escalation, and learning. ## Relation To Dark Factory inference: A software factory can improve throughput without eliminating human review. A dark factory treats human diff review as the bottleneck to remove, so it invests more heavily in validation contracts and automated observability.", "search_text": "Software Factory software-factory concept medium software-factory industrialization automation dan-shapiro-you-dont-write-code-2026-02-13 strongdm-attractor-nlspec jleechanorg-dark-factory-repo dark-factory strongdm-attractor agentic-pipeline-runner dark-factory A software factory is a repeatable production system for turning intent into software artifacts. In this wiki, it is the broader category; dark-factory is the lights-off variant that tries to avoid human code reading. Software Factory A software factory is a repeatable production system for turning intent into software artifacts. In this wiki, it is the broader category; dark-factory is the lights-off variant that tries to avoid human code reading. ## Components - Input artifacts: goals, specs, product requirements, examples, acceptance criteria. - Execution machinery: coding agents, tool registries, isolated worktrees, CI, runners, and queues. - Validation machinery: tests, review agents, holdouts, policy gates, observability, evidence bundles. - Feedback machinery: retries, fix loops, failure diagnosis, escalation, and learning. ## Relation To Dark Factory inference: A software factory can improve throughput without eliminating human review. A dark factory treats human diff review as the bottleneck to remove, so it invests more heavily in validation contracts and automated observability.", "backlinks": [ { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "related" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "related" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "related" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "wikilink" ] } ] }, { "title": "Agent Orchestrator", "slug": "agent-orchestrator", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "orchestration", "agents", "worktrees", "ci", "pr-workflow" ], "sources": [ "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "jleechanorg-agent-orchestrator-repo" ], "source_links": [], "related": [ "agentic-pipeline-runner", "code-review-vs-validation", "tool-nodes-vs-llm-nodes", "software-factory" ], "wikilinks": [ "dark-factory", "dot-pipeline" ], "tensions": [ "It automates worker feedback loops, but its README still reserves merge/review and portability decisions for human judgment." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Agent Orchestrator is a fleet/session orchestration system for AI coding agents. The jleechanorg fork manages workers in isolated git worktrees and routes CI failures, review comments, and status changes back to those workers.", "headings": [ { "level": "1", "title": "Agent Orchestrator" }, { "level": "2", "title": "System Role" }, { "level": "2", "title": "Relation To Dark Factory" } ], "markdown_path": "systems/agent-orchestrator.md", "html_id": "page-agent-orchestrator", "body_text": "Agent Orchestrator Agent Orchestrator is a fleet/session orchestration system for AI coding agents. The jleechanorg fork manages workers in isolated git worktrees and routes CI failures, review comments, and status changes back to those workers. ## System Role - Spawn and manage agent sessions. - Allocate isolated worktrees and branches. - Track session metadata under deterministic runtime paths. - Integrate with GitHub PRs, CI, review comments, notifications, and dashboards. - Add fork-specific evidence gates, skeptic verification, Beads, OpenClaw integration, and additional agent/runtime plugins. ## Relation To Dark Factory inference: Agent Orchestrator can serve as the agent/session substrate under a dark-factory, especially for PR-shaped work. It does not replace dot-pipeline; it manages worker lifecycles that a graph runner might invoke.", "search_text": "Agent Orchestrator agent-orchestrator system high orchestration agents worktrees ci pr-workflow jleechanorg-agent-orchestrator-repo agentic-pipeline-runner code-review-vs-validation tool-nodes-vs-llm-nodes software-factory dark-factory dot-pipeline Agent Orchestrator is a fleet/session orchestration system for AI coding agents. The jleechanorg fork manages workers in isolated git worktrees and routes CI failures, review comments, and status changes back to those workers. Agent Orchestrator Agent Orchestrator is a fleet/session orchestration system for AI coding agents. The jleechanorg fork manages workers in isolated git worktrees and routes CI failures, review comments, and status changes back to those workers. ## System Role - Spawn and manage agent sessions. - Allocate isolated worktrees and branches. - Track session metadata under deterministic runtime paths. - Integrate with GitHub PRs, CI, review comments, notifications, and dashboards. - Add fork-specific evidence gates, skeptic verification, Beads, OpenClaw integration, and additional agent/runtime plugins. ## Relation To Dark Factory inference: Agent Orchestrator can serve as the agent/session substrate under a dark-factory, especially for PR-shaped work. It does not replace dot-pipeline; it manages worker lifecycles that a graph runner might invoke.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "related", "wikilink" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "related", "wikilink" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "related" ] } ] }, { "title": "Agentic Pipeline Runner", "slug": "agentic-pipeline-runner", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "runner", "pipeline", "agents", "dot" ], "sources": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_refs": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "dot-pipeline", "attractor-pattern", "agent-orchestrator", "cxdb", "healer" ], "wikilinks": [ "dot-pipeline", "jleechanorg-dark-factory-repo", "agent-orchestrator" ], "tensions": [ "The runner can be considered disposable relative to the graph, but runner correctness is still load-bearing for safety and reproducibility." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "An agentic pipeline runner executes a dot-pipeline by parsing the graph, validating structure, invoking node handlers, routing outcomes, recording events, and coordinating agents or tools.", "headings": [ { "level": "1", "title": "Agentic Pipeline Runner" }, { "level": "2", "title": "Expected Responsibilities" }, { "level": "2", "title": "Concrete Source" }, { "level": "2", "title": "Boundary" } ], "markdown_path": "systems/agentic-pipeline-runner.md", "html_id": "page-agentic-pipeline-runner", "body_text": "Agentic Pipeline Runner An agentic pipeline runner executes a dot-pipeline by parsing the graph, validating structure, invoking node handlers, routing outcomes, recording events, and coordinating agents or tools. ## Expected Responsibilities - Parse graph source into nodes, edges, attributes, and routing conditions. - Validate unsupported shapes, missing prompts, invalid gates, and unsafe graph constructs. - Initialize execution context: workdir, backend, feature, run ID, CXDB, evidence bundle, and environment. - Execute deterministic nodes, LLM/coding-agent nodes, gates, human gates, joins, retries, and exits. - Finalize run status and emit evidence for debugging and audit. ## Concrete Source The jleechanorg-dark-factory-repo is the concrete implementation source in this wiki. It includes Python runner code, prompt catalogs, DOT pipelines, CXDB logging, Healer diagnosis, benchmark projects, and backend adapters for multiple agent systems. ## Boundary agent-orchestrator is not the same thing. It manages workers, sessions, worktrees, and PR feedback loops. A pipeline runner manages a process graph and may call an orchestrator as one backend.", "search_text": "Agentic Pipeline Runner agentic-pipeline-runner system high runner pipeline agents dot strongdm-attractor-nlspec 2389-dark-factory-dot-file-2026-03-09 jleechanorg-dark-factory-repo dot-pipeline attractor-pattern agent-orchestrator cxdb healer dot-pipeline jleechanorg-dark-factory-repo agent-orchestrator An agentic pipeline runner executes a dot-pipeline by parsing the graph, validating structure, invoking node handlers, routing outcomes, recording events, and coordinating agents or tools. Agentic Pipeline Runner An agentic pipeline runner executes a dot-pipeline by parsing the graph, validating structure, invoking node handlers, routing outcomes, recording events, and coordinating agents or tools. ## Expected Responsibilities - Parse graph source into nodes, edges, attributes, and routing conditions. - Validate unsupported shapes, missing prompts, invalid gates, and unsafe graph constructs. - Initialize execution context: workdir, backend, feature, run ID, CXDB, evidence bundle, and environment. - Execute deterministic nodes, LLM/coding-agent nodes, gates, human gates, joins, retries, and exits. - Finalize run status and emit evidence for debugging and audit. ## Concrete Source The jleechanorg-dark-factory-repo is the concrete implementation source in this wiki. It includes Python runner code, prompt catalogs, DOT pipelines, CXDB logging, Healer diagnosis, benchmark projects, and backend adapters for multiple agent systems. ## Boundary agent-orchestrator is not the same thing. It manages workers, sessions, worktrees, and PR feedback loops. A pipeline runner manages a process graph and may call an orchestrator as one backend.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "related" ] }, { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "related" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "related", "wikilink" ] }, { "slug": "cxdb", "title": "CXDB", "type": "system", "kinds": [ "related" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "related", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "related" ] }, { "slug": "healer", "title": "Healer", "type": "system", "kinds": [ "related" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "related" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "related" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "related" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "related" ] } ] }, { "title": "CXDB", "slug": "cxdb", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "observability", "event-log", "sqlite", "diagnosis" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "healer", "agentic-pipeline-runner", "code-review-vs-validation" ], "wikilinks": [ "healer" ], "tensions": [ "Observability can diagnose failures after the fact, but it does not by itself prove generated code is safe." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "CXDB is the observability/event-log layer described by Shapiro and implemented in the local dark-factory repository as a SQLite event log for pipeline runs.", "headings": [ { "level": "1", "title": "CXDB" }, { "level": "2", "title": "Role" }, { "level": "2", "title": "Design Implication" } ], "markdown_path": "systems/cxdb.md", "html_id": "page-cxdb", "body_text": "CXDB CXDB is the observability/event-log layer described by Shapiro and implemented in the local dark-factory repository as a SQLite event log for pipeline runs. ## Role - Record node visits, outcomes, errors, and run metadata. - Support post-run diagnosis and clustering through healer. - Provide an audit trail for factory operation. ## Design Implication inference: A dark factory needs run-level provenance as much as code provenance. Without event history, failed autonomous runs are hard to debug and successful runs are hard to trust.", "search_text": "CXDB cxdb system medium-high observability event-log sqlite diagnosis dan-shapiro-you-dont-write-code-2026-02-13 jleechanorg-dark-factory-repo healer agentic-pipeline-runner code-review-vs-validation healer CXDB is the observability/event-log layer described by Shapiro and implemented in the local dark-factory repository as a SQLite event log for pipeline runs. CXDB CXDB is the observability/event-log layer described by Shapiro and implemented in the local dark-factory repository as a SQLite event log for pipeline runs. ## Role - Record node visits, outcomes, errors, and run metadata. - Support post-run diagnosis and clustering through healer. - Provide an audit trail for factory operation. ## Design Implication inference: A dark factory needs run-level provenance as much as code provenance. Without event history, failed autonomous runs are hard to debug and successful runs are hard to trust.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "related" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "related", "wikilink" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "kinds": [ "related" ] }, { "slug": "healer", "title": "Healer", "type": "system", "kinds": [ "related", "wikilink" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "related", "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "related", "wikilink" ] } ] }, { "title": "Digital Twin Universe", "slug": "digital-twin-universe", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "testing", "simulation", "validation", "enterprise-saas" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13" ], "source_links": [], "related": [ "dark-factory", "code-review-vs-validation", "cxdb" ], "wikilinks": [], "tensions": [ "Simulation can improve validation fidelity, but it can also hide production-only failure modes if the twin is stale or incomplete." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "The digital twin universe is StrongDM's reported local reproduction of enterprise SaaS systems used to validate generated software against realistic external behavior.", "headings": [ { "level": "1", "title": "Digital Twin Universe" }, { "level": "2", "title": "Factory Role" }, { "level": "2", "title": "Key Inference" } ], "markdown_path": "systems/digital-twin-universe.md", "html_id": "page-digital-twin-universe", "body_text": "Digital Twin Universe The digital twin universe is StrongDM's reported local reproduction of enterprise SaaS systems used to validate generated software against realistic external behavior. ## Factory Role It lets agents exercise code in a rich environment without touching production. This matters because dark factories need outcome evidence when humans stop reading every diff. ## Key Inference inference: As code review declines, environment fidelity becomes a primary quality lever. A weak test environment turns \"no human code reading\" into blind trust; a strong twin makes automated validation more meaningful.", "search_text": "Digital Twin Universe digital-twin-universe system medium-high testing simulation validation enterprise-saas dan-shapiro-you-dont-write-code-2026-02-13 dark-factory code-review-vs-validation cxdb The digital twin universe is StrongDM's reported local reproduction of enterprise SaaS systems used to validate generated software against realistic external behavior. Digital Twin Universe The digital twin universe is StrongDM's reported local reproduction of enterprise SaaS systems used to validate generated software against realistic external behavior. ## Factory Role It lets agents exercise code in a rich environment without touching production. This matters because dark factories need outcome evidence when humans stop reading every diff. ## Key Inference inference: As code review declines, environment fidelity becomes a primary quality lever. A weak test environment turns \"no human code reading\" into blind trust; a strong twin makes automated validation more meaningful.", "backlinks": [ { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "related" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "related", "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "related" ] } ] }, { "title": "Healer", "slug": "healer", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "diagnosis", "repair-loop", "observability" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "cxdb", "dark-factory", "agentic-pipeline-runner" ], "wikilinks": [ "cxdb" ], "tensions": [ "Repair loops reduce human debugging burden, but a wrong diagnosis can send agents into repeated ineffective fixes." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Healer is the diagnosis and repair-support system layered on top of cxdb. In the dark-factory repository, df-healer clusters failures from the event log into actionable diagnoses.", "headings": [ { "level": "1", "title": "Healer" }, { "level": "2", "title": "Role" }, { "level": "2", "title": "Design Implication" } ], "markdown_path": "systems/healer.md", "html_id": "page-healer", "body_text": "Healer Healer is the diagnosis and repair-support system layered on top of cxdb. In the dark-factory repository, df-healer clusters failures from the event log into actionable diagnoses. ## Role - Inspect failed runs after gates or execution errors. - Cluster similar failures so repeated breakages are visible. - Produce a diagnosis artifact that can feed the next fix loop. ## Design Implication inference: Automated repair requires structured failure memory. A healer is the factory equivalent of an engineer reading CI logs across many attempts.", "search_text": "Healer healer system medium-high diagnosis repair-loop observability dan-shapiro-you-dont-write-code-2026-02-13 jleechanorg-dark-factory-repo cxdb dark-factory agentic-pipeline-runner cxdb Healer is the diagnosis and repair-support system layered on top of cxdb. In the dark-factory repository, df-healer clusters failures from the event log into actionable diagnoses. Healer Healer is the diagnosis and repair-support system layered on top of cxdb. In the dark-factory repository, df-healer clusters failures from the event log into actionable diagnoses. ## Role - Inspect failed runs after gates or execution errors. - Cluster similar failures so repeated breakages are visible. - Produce a diagnosis artifact that can feed the next fix loop. ## Design Implication inference: Automated repair requires structured failure memory. A healer is the factory equivalent of an engineer reading CI logs across many attempts.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "related" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "related", "wikilink" ] }, { "slug": "cxdb", "title": "CXDB", "type": "system", "kinds": [ "related", "wikilink" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "related", "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "related", "wikilink" ] } ] }, { "title": "StrongDM Attractor", "slug": "strongdm-attractor", "type": "system", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "strongdm", "attractor", "nlspec" ], "sources": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09" ], "source_refs": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09" ], "source_links": [], "related": [ "nlspec", "attractor-pattern", "dot-pipeline", "agentic-pipeline-runner" ], "wikilinks": [ "attractor-pattern" ], "tensions": [ "Attractor is described as a system, but the cited repository is primarily a specification corpus." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "StrongDM Attractor is represented in the cited repository as a set of NLSpecs for building a software factory stack.", "headings": [ { "level": "1", "title": "StrongDM Attractor" }, { "level": "2", "title": "Specified Layers" }, { "level": "2", "title": "Wiki Position" } ], "markdown_path": "systems/strongdm-attractor.md", "html_id": "page-strongdm-attractor", "body_text": "StrongDM Attractor StrongDM Attractor is represented in the cited repository as a set of NLSpecs for building a software factory stack. ## Specified Layers - Unified LLM client: provider-agnostic API, streaming, tool use, retries, adapters, middleware. - Coding agent loop: tool registry, execution environment, filesystem/search/command tools, provider profiles, subagents. - DOT pipeline engine: parse graph, validate graph, run node handlers, route edges, checkpoint/resume, support human gates and model stylesheets. ## Wiki Position Treat Attractor as a reference specification for the attractor-pattern. Do not assume the repository itself is a shipped runner unless a concrete implementation source is added.", "search_text": "StrongDM Attractor strongdm-attractor system high strongdm attractor nlspec strongdm-attractor-nlspec 2389-dark-factory-dot-file-2026-03-09 nlspec attractor-pattern dot-pipeline agentic-pipeline-runner attractor-pattern StrongDM Attractor is represented in the cited repository as a set of NLSpecs for building a software factory stack. StrongDM Attractor StrongDM Attractor is represented in the cited repository as a set of NLSpecs for building a software factory stack. ## Specified Layers - Unified LLM client: provider-agnostic API, streaming, tool use, retries, adapters, middleware. - Coding agent loop: tool registry, execution environment, filesystem/search/command tools, provider profiles, subagents. - DOT pipeline engine: parse graph, validate graph, run node handlers, route edges, checkpoint/resume, support human gates and model stylesheets. ## Wiki Position Treat Attractor as a reference specification for the attractor-pattern. Do not assume the repository itself is a shipped runner unless a concrete implementation source is added.", "backlinks": [ { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "related", "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "kinds": [ "related" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "related" ] }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "kinds": [ "related", "wikilink" ] } ] }, { "title": "Code Review vs Validation", "slug": "code-review-vs-validation", "type": "comparison", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "code-review", "validation", "gates", "security" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_links": [], "related": [ "dark-factory", "digital-twin-universe", "cxdb", "healer" ], "wikilinks": [ "dark-factory", "cxdb", "healer" ], "tensions": [ "Human code review catches design and security issues that tests may miss; dark factory systems try to move that assurance into automated gates and independent review." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Code review asks humans or reviewers to inspect code changes directly. Validation asks systems to prove the change satisfies behavior, safety, policy, and evidence requirements.", "headings": [ { "level": "1", "title": "Code Review vs Validation" }, { "level": "2", "title": "Factory Shift" }, { "level": "2", "title": "Boundary" } ], "markdown_path": "comparisons/code-review-vs-validation.md", "html_id": "page-code-review-vs-validation", "body_text": "Code Review vs Validation Code review asks humans or reviewers to inspect code changes directly. Validation asks systems to prove the change satisfies behavior, safety, policy, and evidence requirements. ## Factory Shift In a dark-factory, human diff review is treated as the bottleneck. The replacement is not blind trust. It is validation infrastructure: - public tests and sealed holdouts; - adversarial independent review; - structural isolation from evaluator internals; - CI and evidence gates; - telemetry through cxdb; - diagnosis through healer; - human escalation for judgment-heavy cases. ## Boundary inference: Validation can replace routine diff reading only where acceptance criteria, threat models, and evidence standards are strong enough. For novel architecture, security posture, or ambiguous product behavior, human review remains a necessary gate.", "search_text": "Code Review vs Validation code-review-vs-validation comparison medium-high code-review validation gates security dan-shapiro-you-dont-write-code-2026-02-13 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo dark-factory digital-twin-universe cxdb healer dark-factory cxdb healer Code review asks humans or reviewers to inspect code changes directly. Validation asks systems to prove the change satisfies behavior, safety, policy, and evidence requirements. Code Review vs Validation Code review asks humans or reviewers to inspect code changes directly. Validation asks systems to prove the change satisfies behavior, safety, policy, and evidence requirements. ## Factory Shift In a dark-factory, human diff review is treated as the bottleneck. The replacement is not blind trust. It is validation infrastructure: - public tests and sealed holdouts; - adversarial independent review; - structural isolation from evaluator internals; - CI and evidence gates; - telemetry through cxdb; - diagnosis through healer; - human escalation for judgment-heavy cases. ## Boundary inference: Validation can replace routine diff reading only where acceptance criteria, threat models, and evidence standards are strong enough. For novel architecture, security posture, or ambiguous product behavior, human review remains a necessary gate.", "backlinks": [ { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "related" ] }, { "slug": "cxdb", "title": "CXDB", "type": "system", "kinds": [ "related" ] }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "related" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "kinds": [ "related" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "related" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "related", "wikilink" ] } ] }, { "title": "LLM Wiki vs RAG", "slug": "llm-wiki-vs-rag", "type": "comparison", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "llm-wiki", "rag", "knowledge-management" ], "sources": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_refs": [ "karpathy-llm-wiki-gist-2026-04-04" ], "source_links": [], "related": [ "llm-wiki-pattern", "SCHEMA" ], "wikilinks": [], "tensions": [ "A wiki preserves synthesis but can go stale; RAG preserves access to raw material but can repeatedly miss the same synthesis." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "RAG retrieves source chunks at question time. An LLM wiki compiles knowledge into maintained Markdown pages before the next question arrives.", "headings": [ { "level": "1", "title": "LLM Wiki vs RAG" }, { "level": "2", "title": "Difference" }, { "level": "2", "title": "Use Together" } ], "markdown_path": "comparisons/llm-wiki-vs-rag.md", "html_id": "page-llm-wiki-vs-rag", "body_text": "LLM Wiki vs RAG RAG retrieves source chunks at question time. An LLM wiki compiles knowledge into maintained Markdown pages before the next question arrives. ## Difference - RAG is strongest for broad retrieval over many raw documents. - An LLM wiki is strongest for stable concepts, repeated questions, curated contradictions, and agent handoff. - RAG can find evidence; a wiki preserves prior reasoning. ## Use Together Use RAG or web search to inspect raw sources when the source set changes. Update the wiki afterward so the next agent starts from compiled knowledge instead of raw rediscovery.", "search_text": "LLM Wiki vs RAG llm-wiki-vs-rag comparison high llm-wiki rag knowledge-management karpathy-llm-wiki-gist-2026-04-04 llm-wiki-pattern SCHEMA RAG retrieves source chunks at question time. An LLM wiki compiles knowledge into maintained Markdown pages before the next question arrives. LLM Wiki vs RAG RAG retrieves source chunks at question time. An LLM wiki compiles knowledge into maintained Markdown pages before the next question arrives. ## Difference - RAG is strongest for broad retrieval over many raw documents. - An LLM wiki is strongest for stable concepts, repeated questions, curated contradictions, and agent handoff. - RAG can find evidence; a wiki preserves prior reasoning. ## Use Together Use RAG or web search to inspect raw sources when the source set changes. Update the wiki afterward so the next agent starts from compiled knowledge instead of raw rediscovery.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "kinds": [ "related" ] } ] }, { "title": "Specs vs Code", "slug": "specs-vs-code", "type": "comparison", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "specs", "code", "durable-artifacts" ], "sources": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_refs": [ "strongdm-attractor-nlspec", "2389-dark-factory-dot-file-2026-03-09", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "nlspec", "dot-pipeline", "what-is-the-durable-artifact-in-a-dark-factory" ], "wikilinks": [], "tensions": [ "Treating code as disposable is useful for generated implementations, but production systems still depend on correct executable code at runtime." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Specs describe intent, behavior, process, and validation. Code is the executable artifact produced from those specs. Dark factory sources push the durable boundary toward specs, graphs, tests, and evidence rather than generated code.", "headings": [ { "level": "1", "title": "Specs vs Code" }, { "level": "2", "title": "Durable Specs" }, { "level": "2", "title": "Disposable Code" }, { "level": "2", "title": "Design Implication" } ], "markdown_path": "comparisons/specs-vs-code.md", "html_id": "page-specs-vs-code", "body_text": "Specs vs Code Specs describe intent, behavior, process, and validation. Code is the executable artifact produced from those specs. Dark factory sources push the durable boundary toward specs, graphs, tests, and evidence rather than generated code. ## Durable Specs - NLSpecs define expected behavior and interfaces. - spec.md captures task intent and acceptance criteria. - DOT graphs encode the development process. - Holdouts and scoring contracts define validation. ## Disposable Code Generated code can be rebuilt, replaced, or regenerated. But \"disposable\" does not mean unimportant. It means the factory should be able to recreate code from durable intent and process artifacts. ## Design Implication inference: The safer framing is not \"code does not matter\"; it is \"code should be reproducible from higher-level durable artifacts and judged by stronger validation.\"", "search_text": "Specs vs Code specs-vs-code comparison medium-high specs code durable-artifacts strongdm-attractor-nlspec 2389-dark-factory-dot-file-2026-03-09 jleechanorg-dark-factory-repo nlspec dot-pipeline what-is-the-durable-artifact-in-a-dark-factory Specs describe intent, behavior, process, and validation. Code is the executable artifact produced from those specs. Dark factory sources push the durable boundary toward specs, graphs, tests, and evidence rather than generated code. Specs vs Code Specs describe intent, behavior, process, and validation. Code is the executable artifact produced from those specs. Dark factory sources push the durable boundary toward specs, graphs, tests, and evidence rather than generated code. ## Durable Specs - NLSpecs define expected behavior and interfaces. - spec.md captures task intent and acceptance criteria. - DOT graphs encode the development process. - Holdouts and scoring contracts define validation. ## Disposable Code Generated code can be rebuilt, replaced, or regenerated. But \"disposable\" does not mean unimportant. It means the factory should be able to recreate code from durable intent and process artifacts. ## Design Implication inference: The safer framing is not \"code does not matter\"; it is \"code should be reproducible from higher-level durable artifacts and judged by stronger validation.\"", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "kinds": [ "related" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "related" ] } ] }, { "title": "Tool Nodes vs LLM Nodes", "slug": "tool-nodes-vs-llm-nodes", "type": "comparison", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "dot", "tools", "llm", "pipeline-design" ], "sources": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_refs": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo" ], "source_links": [], "related": [ "dot-pipeline", "agentic-pipeline-runner", "how-do-dot-pipelines-map-to-agent-workflows" ], "wikilinks": [], "tensions": [ "Overusing tools can make the graph rigid; overusing LLM nodes can make runs expensive, nondeterministic, and hard to debug." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Tool nodes run deterministic commands, checks, transformations, or integrations. LLM nodes delegate reasoning, planning, code generation, review, or synthesis to a model-backed agent.", "headings": [ { "level": "1", "title": "Tool Nodes vs LLM Nodes" }, { "level": "2", "title": "Design Rule" }, { "level": "2", "title": "Why It Matters" } ], "markdown_path": "comparisons/tool-nodes-vs-llm-nodes.md", "html_id": "page-tool-nodes-vs-llm-nodes", "body_text": "Tool Nodes vs LLM Nodes Tool nodes run deterministic commands, checks, transformations, or integrations. LLM nodes delegate reasoning, planning, code generation, review, or synthesis to a model-backed agent. ## Design Rule Use deterministic tool nodes for setup, parsing, validation, deployment, scoring, and evidence collection. Use LLM nodes when the task needs judgment, synthesis, code generation, or review that cannot be encoded cheaply as a tool. ## Why It Matters inference: The dark factory becomes more trustworthy when nondeterminism is isolated. Keep the graph deterministic around gates and evidence, and let LLMs operate inside bounded work nodes.", "search_text": "Tool Nodes vs LLM Nodes tool-nodes-vs-llm-nodes comparison high dot tools llm pipeline-design 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec jleechanorg-dark-factory-repo dot-pipeline agentic-pipeline-runner how-do-dot-pipelines-map-to-agent-workflows Tool nodes run deterministic commands, checks, transformations, or integrations. LLM nodes delegate reasoning, planning, code generation, review, or synthesis to a model-backed agent. Tool Nodes vs LLM Nodes Tool nodes run deterministic commands, checks, transformations, or integrations. LLM nodes delegate reasoning, planning, code generation, review, or synthesis to a model-backed agent. ## Design Rule Use deterministic tool nodes for setup, parsing, validation, deployment, scoring, and evidence collection. Use LLM nodes when the task needs judgment, synthesis, code generation, or review that cannot be encoded cheaply as a tool. ## Why It Matters inference: The dark factory becomes more trustworthy when nondeterminism is isolated. Keep the graph deterministic around gates and evidence, and let LLMs operate inside bounded work nodes.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "related", "wikilink" ] }, { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "related" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "related" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "related" ] }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "kinds": [ "related" ] } ] }, { "title": "2389: The Dark Factory Is a .dot file", "slug": "2389-dark-factory-dot-file-2026-03-09", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "dot", "pipeline", "attractor", "dark-factory" ], "sources": [ "https://2389.ai/posts/the-dark-factory-is-a-dot-file/" ], "source_refs": [], "source_links": [ { "label": "https://2389.ai/posts/the-dark-factory-is-a-dot-file/", "url": "https://2389.ai/posts/the-dark-factory-is-a-dot-file/", "origin": "frontmatter" } ], "related": [ "dot-pipeline", "attractor-pattern", "agentic-pipeline-runner", "tool-nodes-vs-llm-nodes" ], "wikilinks": [ "attractor-pattern", "dot-pipeline", "tool-nodes-vs-llm-nodes", "specs-vs-code", "how-do-dot-pipelines-map-to-agent-workflows", "dan-shapiro-you-dont-write-code-2026-02-13" ], "tensions": [ "The post says the factory code is disposable, but its examples also show substantial runner architecture and validation behavior that must be implemented carefully.", "This source mixes product narrative, implementation notes, and advocacy. Treat claims about what is \"the product\" as a position, not a settled industry standard.", "The post says pipelines are the durable artifact, while dan-shapiro-you-dont-write-code-2026-02-13 emphasizes validation systems and factory operation." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "StrongDM's Attractor NLSpecs describe three layers: unified LLM client, coding agent loop, and DOT-based pipeline engine.", "headings": [ { "level": "1", "title": "2389: The Dark Factory Is a .dot file" }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/2389-dark-factory-dot-file-2026-03-09.md", "html_id": "page-2389-dark-factory-dot-file-2026-03-09", "body_text": "2389: The Dark Factory Is a .dot file Source URL: 2389 Research post https://2389.ai/posts/the-dark-factory-is-a-dot-file/ Published: 2026-03-09. Captured: 2026-06-17. Provenance: primary blog post by Harper Reed describing DOT pipelines, Attractor convergence, and 2389 runner implementations. ## Compact Takeaways - StrongDM's Attractor NLSpecs describe three layers: unified LLM client, coding agent loop, and DOT-based pipeline engine. - Independent implementations such as Kilroy, Mammoth, Smasher, and Tracker converged on the same three-layer architecture, which the post calls the attractor-pattern. - The reusable factory artifact is the dot-pipeline: a Graphviz DOT graph describing workflow stages, LLM calls, human gates, forks, and verification commands. - The post favors deterministic tool nodes for setup, validation, and deployment, reserving LLM nodes for work that genuinely needs model reasoning. - Complex pipelines can encode the entire software development lifecycle, including human gates, TDD loops, retries, multi-model review, and shipping decisions. ## Wiki Implications - tool-nodes-vs-llm-nodes should be treated as a design axis for cost, reproducibility, and operational confidence. - specs-vs-code should distinguish disposable generated runner code from durable pipeline blueprints. - how-do-dot-pipelines-map-to-agent-workflows should use DOT as a process definition, not merely a visualization format. ## Caveats - This source mixes product narrative, implementation notes, and advocacy. Treat claims about what is \"the product\" as a position, not a settled industry standard. - The post says pipelines are the durable artifact, while dan-shapiro-you-dont-write-code-2026-02-13 emphasizes validation systems and factory operation.", "search_text": "2389: The Dark Factory Is a .dot file 2389-dark-factory-dot-file-2026-03-09 source high source dot pipeline attractor dark-factory https://2389.ai/posts/the-dark-factory-is-a-dot-file/ dot-pipeline attractor-pattern agentic-pipeline-runner tool-nodes-vs-llm-nodes attractor-pattern dot-pipeline tool-nodes-vs-llm-nodes specs-vs-code how-do-dot-pipelines-map-to-agent-workflows dan-shapiro-you-dont-write-code-2026-02-13 StrongDM's Attractor NLSpecs describe three layers: unified LLM client, coding agent loop, and DOT-based pipeline engine. 2389: The Dark Factory Is a .dot file Source URL: 2389 Research post https://2389.ai/posts/the-dark-factory-is-a-dot-file/ Published: 2026-03-09. Captured: 2026-06-17. Provenance: primary blog post by Harper Reed describing DOT pipelines, Attractor convergence, and 2389 runner implementations. ## Compact Takeaways - StrongDM's Attractor NLSpecs describe three layers: unified LLM client, coding agent loop, and DOT-based pipeline engine. - Independent implementations such as Kilroy, Mammoth, Smasher, and Tracker converged on the same three-layer architecture, which the post calls the attractor-pattern. - The reusable factory artifact is the dot-pipeline: a Graphviz DOT graph describing workflow stages, LLM calls, human gates, forks, and verification commands. - The post favors deterministic tool nodes for setup, validation, and deployment, reserving LLM nodes for work that genuinely needs model reasoning. - Complex pipelines can encode the entire software development lifecycle, including human gates, TDD loops, retries, multi-model review, and shipping decisions. ## Wiki Implications - tool-nodes-vs-llm-nodes should be treated as a design axis for cost, reproducibility, and operational confidence. - specs-vs-code should distinguish disposable generated runner code from durable pipeline blueprints. - how-do-dot-pipelines-map-to-agent-workflows should use DOT as a process definition, not merely a visualization format. ## Caveats - This source mixes product narrative, implementation notes, and advocacy. Treat claims about what is \"the product\" as a position, not a settled industry standard. - The post says pipelines are the durable artifact, while dan-shapiro-you-dont-write-code-2026-02-13 emphasizes validation systems and factory operation.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "source" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "source" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "source" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "source" ] }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "kinds": [ "source" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "source" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "source" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "source" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "source" ] } ] }, { "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "dark-factory", "strongdm", "validation" ], "sources": [ "https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/" ], "source_refs": [], "source_links": [ { "label": "https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/", "url": "https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/", "origin": "frontmatter" } ], "related": [ "dark-factory", "software-factory", "digital-twin-universe", "cxdb", "healer" ], "wikilinks": [ "cxdb", "healer", "dark-factory", "code-review-vs-validation", "digital-twin-universe" ], "tensions": [ "The article argues humans must stop reading generated code, while also foregrounding security and quality problems that require stronger validation systems.", "This source is narrative and observational. It reports specific StrongDM practices but does not provide formal specifications or implementation details.", "It presents an aspirational lights-off model, but the examples still imply human judgment at the level of factory design, bootcamps, validation policy, and tool building." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "The core operating shift is: the AI writes code, and reviewing every pull request becomes the bottleneck.", "headings": [ { "level": "1", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either." }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/dan-shapiro-you-dont-write-code-2026-02-13.md", "html_id": "page-dan-shapiro-you-dont-write-code-2026-02-13", "body_text": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either. Source URL: Dan Shapiro blog https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/ Published: 2026-02-13. Captured: 2026-06-17. Provenance: primary article by Dan Shapiro about StrongDM's dark factory work. ## Compact Takeaways - The core operating shift is: the AI writes code, and reviewing every pull request becomes the bottleneck. - The new engineering job is to solve the problems created by not writing or reading code: quality, understanding, observability, and validation. - StrongDM used a local digital twin of enterprise SaaS systems so generated products could be tested against realistic external behavior. - The article names two internal systems: cxdb, an observability layer, and healer, a diagnostic and repair loop built on top of that observability. - The most reusable practice is asking why a human is doing a manual validation step, then turning the answer into a validation rule. ## Wiki Implications - dark-factory should be treated as an operating model, not just an AI coding style. - code-review-vs-validation is the central tradeoff: if humans stop reading code, validation systems must become first-class engineering work. - digital-twin-universe is evidence that high-fidelity test environments become more important as code review becomes less central. ## Caveats - This source is narrative and observational. It reports specific StrongDM practices but does not provide formal specifications or implementation details. - It presents an aspirational lights-off model, but the examples still imply human judgment at the level of factory design, bootcamps, validation policy, and tool building.", "search_text": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either. dan-shapiro-you-dont-write-code-2026-02-13 source high source dark-factory strongdm validation https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/ dark-factory software-factory digital-twin-universe cxdb healer cxdb healer dark-factory code-review-vs-validation digital-twin-universe The core operating shift is: the AI writes code, and reviewing every pull request becomes the bottleneck. Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either. Source URL: Dan Shapiro blog https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/ Published: 2026-02-13. Captured: 2026-06-17. Provenance: primary article by Dan Shapiro about StrongDM's dark factory work. ## Compact Takeaways - The core operating shift is: the AI writes code, and reviewing every pull request becomes the bottleneck. - The new engineering job is to solve the problems created by not writing or reading code: quality, understanding, observability, and validation. - StrongDM used a local digital twin of enterprise SaaS systems so generated products could be tested against realistic external behavior. - The article names two internal systems: cxdb, an observability layer, and healer, a diagnostic and repair loop built on top of that observability. - The most reusable practice is asking why a human is doing a manual validation step, then turning the answer into a validation rule. ## Wiki Implications - dark-factory should be treated as an operating model, not just an AI coding style. - code-review-vs-validation is the central tradeoff: if humans stop reading code, validation systems must become first-class engineering work. - digital-twin-universe is evidence that high-fidelity test environments become more important as code review becomes less central. ## Caveats - This source is narrative and observational. It reports specific StrongDM practices but does not provide formal specifications or implementation details. - It presents an aspirational lights-off model, but the examples still imply human judgment at the level of factory design, bootcamps, validation policy, and tool building.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "source" ] }, { "slug": "cxdb", "title": "CXDB", "type": "system", "kinds": [ "source" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "kinds": [ "source" ] }, { "slug": "healer", "title": "Healer", "type": "system", "kinds": [ "source" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "source" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "source" ] } ] }, { "title": "jleechanorg/agent-orchestrator Repository", "slug": "jleechanorg-agent-orchestrator-repo", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "repository", "orchestration", "worktrees", "agents" ], "sources": [ "https://github.com/jleechanorg/agent-orchestrator" ], "source_refs": [], "source_links": [ { "label": "https://github.com/jleechanorg/agent-orchestrator", "url": "https://github.com/jleechanorg/agent-orchestrator", "origin": "frontmatter" } ], "related": [ "agent-orchestrator", "agentic-pipeline-runner", "code-review-vs-validation", "tool-nodes-vs-llm-nodes" ], "wikilinks": [ "agent-orchestrator", "agentic-pipeline-runner", "code-review-vs-validation", "dark-factory", "software-factory" ], "tensions": [ "The fork is framed as zero-touch automation, but the README still reserves review/merge and portability decisions for human judgment.", "This source is a fork README plus architecture doc, not a stable API contract.", "The fork has environment-specific assumptions; future agents should not copy its defaults without checking portability." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Captured: 2026-06-17. Observed main HEAD: 3923e3cce815b541af59c2756945d184fde4ac25. Provenance: public GitHub repository README, ARCHITECTURE.md, and local shallow clone inspection.", "headings": [ { "level": "1", "title": "jleechanorg/agent-orchestrator Repository" }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/jleechanorg-agent-orchestrator-repo.md", "html_id": "page-jleechanorg-agent-orchestrator-repo", "body_text": "jleechanorg/agent-orchestrator Repository Source URL: jleechanorg/agent-orchestrator https://github.com/jleechanorg/agent-orchestrator Captured: 2026-06-17. Observed main HEAD: 3923e3cce815b541af59c2756945d184fde4ac25. Provenance: public GitHub repository README, ARCHITECTURE.md, and local shallow clone inspection. ## Compact Takeaways - Agent Orchestrator manages fleets of AI coding agents, giving each issue an isolated worktree, branch, session, and PR workflow. - The jleechanorg fork is explicitly not a portable baseline; it targets the maintainer's own workflow and adds fork-specific plugins, CI gates, OpenClaw integration, Beads, Antigravity, MiniMax/Wafer adapters, evidence gates, and skeptic verification. - The operating loop is: start dashboard/orchestrator, spawn workers, route CI/review feedback back to workers, and escalate when human judgment is needed. - ARCHITECTURE.md emphasizes a repo-local config file as source of truth and deterministic runtime state under ~/.agent-orchestrator/, with hash-based namespacing for sessions and worktrees. - The README positions this fork as autonomous coding pipeline infrastructure, while upstream AgentWrapper/agent-orchestrator is a broader orchestration tool. ## Wiki Implications - agent-orchestrator is a lower-level worker/session/PR orchestration layer that can serve as a backend or adjacent execution substrate for agentic-pipeline-runner. - code-review-vs-validation should include external review routing, CI remediation, evidence gates, and skeptic agents as validation infrastructure. - dark-factory and software-factory should distinguish process graph execution from fleet/session management. ## Caveats - This source is a fork README plus architecture doc, not a stable API contract. - The fork has environment-specific assumptions; future agents should not copy its defaults without checking portability.", "search_text": "jleechanorg/agent-orchestrator Repository jleechanorg-agent-orchestrator-repo source high source repository orchestration worktrees agents https://github.com/jleechanorg/agent-orchestrator agent-orchestrator agentic-pipeline-runner code-review-vs-validation tool-nodes-vs-llm-nodes agent-orchestrator agentic-pipeline-runner code-review-vs-validation dark-factory software-factory Captured: 2026-06-17. Observed main HEAD: 3923e3cce815b541af59c2756945d184fde4ac25. Provenance: public GitHub repository README, ARCHITECTURE.md, and local shallow clone inspection. jleechanorg/agent-orchestrator Repository Source URL: jleechanorg/agent-orchestrator https://github.com/jleechanorg/agent-orchestrator Captured: 2026-06-17. Observed main HEAD: 3923e3cce815b541af59c2756945d184fde4ac25. Provenance: public GitHub repository README, ARCHITECTURE.md, and local shallow clone inspection. ## Compact Takeaways - Agent Orchestrator manages fleets of AI coding agents, giving each issue an isolated worktree, branch, session, and PR workflow. - The jleechanorg fork is explicitly not a portable baseline; it targets the maintainer's own workflow and adds fork-specific plugins, CI gates, OpenClaw integration, Beads, Antigravity, MiniMax/Wafer adapters, evidence gates, and skeptic verification. - The operating loop is: start dashboard/orchestrator, spawn workers, route CI/review feedback back to workers, and escalate when human judgment is needed. - ARCHITECTURE.md emphasizes a repo-local config file as source of truth and deterministic runtime state under ~/.agent-orchestrator/, with hash-based namespacing for sessions and worktrees. - The README positions this fork as autonomous coding pipeline infrastructure, while upstream AgentWrapper/agent-orchestrator is a broader orchestration tool. ## Wiki Implications - agent-orchestrator is a lower-level worker/session/PR orchestration layer that can serve as a backend or adjacent execution substrate for agentic-pipeline-runner. - code-review-vs-validation should include external review routing, CI remediation, evidence gates, and skeptic agents as validation infrastructure. - dark-factory and software-factory should distinguish process graph execution from fleet/session management. ## Caveats - This source is a fork README plus architecture doc, not a stable API contract. - The fork has environment-specific assumptions; future agents should not copy its defaults without checking portability.", "backlinks": [ { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "kinds": [ "source" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "source" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "source" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "source" ] } ] }, { "title": "jleechanorg/dark-factory Repository", "slug": "jleechanorg-dark-factory-repo", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "repository", "dark-factory", "dot", "cxdb" ], "sources": [ "https://github.com/jleechanorg/dark-factory" ], "source_refs": [], "source_links": [ { "label": "https://github.com/jleechanorg/dark-factory", "url": "https://github.com/jleechanorg/dark-factory", "origin": "frontmatter" } ], "related": [ "agentic-pipeline-runner", "dot-pipeline", "cxdb", "healer", "agent-orchestrator" ], "wikilinks": [ "attractor-pattern", "cxdb", "healer", "agentic-pipeline-runner", "how-do-dark-factories-validate-without-reading-code", "specs-vs-code" ], "tensions": [ "The README describes runner code as disposable, but the repository contains substantial engine, validation, observability, and benchmark code that future implementations should study.", "This is an implementation repository and contains claims about its own state of the art; use it as implementation evidence, not as independent proof that the pattern is universally validated.", "The observed commit is a point-in-time snapshot. Re-run repository inspection before making current operational claims." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Captured: 2026-06-17. Observed main HEAD: 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae. Provenance: public GitHub repository README and local shallow clone inspection.", "headings": [ { "level": "1", "title": "jleechanorg/dark-factory Repository" }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/jleechanorg-dark-factory-repo.md", "html_id": "page-jleechanorg-dark-factory-repo", "body_text": "jleechanorg/dark-factory Repository Source URL: jleechanorg/dark-factory https://github.com/jleechanorg/dark-factory Captured: 2026-06-17. Observed main HEAD: 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae. Provenance: public GitHub repository README and local shallow clone inspection. ## Compact Takeaways - The repository is a Python implementation of the attractor-pattern with a DOT-based pipeline engine for multi-agent software engineering workflows. - Its README states the durable artifacts are .dot process graphs and spec.md; generated implementation code and runner internals are treated as rebuildable. - Factory execution is split into spec generation and pipeline execution, with plan, implement, test, adversarial review, gates, and sealed holdout stages. - The implementation includes cxdb event logging, healer failure clustering, prompt catalogs, benchmark projects, and multiple backends including Claude, Codex, Antigravity, AO, echo, and mock LLMs. - It explicitly encodes an operator-agent isolation rule: implementing agents should not see holdouts, evaluator internals, or holdout-related environment variables. ## Wiki Implications - agentic-pipeline-runner should include concrete runner phases, backends, sandboxing, event logging, and evidence bundles. - how-do-dark-factories-validate-without-reading-code should include sealed holdouts, adversarial review, and structural blindness, not just ordinary tests. - specs-vs-code should treat specs and DOT graphs as long-lived artifacts and runner implementations as lower-trust/rebuildable machinery. ## Caveats - This is an implementation repository and contains claims about its own state of the art; use it as implementation evidence, not as independent proof that the pattern is universally validated. - The observed commit is a point-in-time snapshot. Re-run repository inspection before making current operational claims.", "search_text": "jleechanorg/dark-factory Repository jleechanorg-dark-factory-repo source high source repository dark-factory dot cxdb https://github.com/jleechanorg/dark-factory agentic-pipeline-runner dot-pipeline cxdb healer agent-orchestrator attractor-pattern cxdb healer agentic-pipeline-runner how-do-dark-factories-validate-without-reading-code specs-vs-code Captured: 2026-06-17. Observed main HEAD: 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae. Provenance: public GitHub repository README and local shallow clone inspection. jleechanorg/dark-factory Repository Source URL: jleechanorg/dark-factory https://github.com/jleechanorg/dark-factory Captured: 2026-06-17. Observed main HEAD: 5449ecc1e5115afda6dca2f8cd49da2dd61a24ae. Provenance: public GitHub repository README and local shallow clone inspection. ## Compact Takeaways - The repository is a Python implementation of the attractor-pattern with a DOT-based pipeline engine for multi-agent software engineering workflows. - Its README states the durable artifacts are .dot process graphs and spec.md; generated implementation code and runner internals are treated as rebuildable. - Factory execution is split into spec generation and pipeline execution, with plan, implement, test, adversarial review, gates, and sealed holdout stages. - The implementation includes cxdb event logging, healer failure clustering, prompt catalogs, benchmark projects, and multiple backends including Claude, Codex, Antigravity, AO, echo, and mock LLMs. - It explicitly encodes an operator-agent isolation rule: implementing agents should not see holdouts, evaluator internals, or holdout-related environment variables. ## Wiki Implications - agentic-pipeline-runner should include concrete runner phases, backends, sandboxing, event logging, and evidence bundles. - how-do-dark-factories-validate-without-reading-code should include sealed holdouts, adversarial review, and structural blindness, not just ordinary tests. - specs-vs-code should treat specs and DOT graphs as long-lived artifacts and runner implementations as lower-trust/rebuildable machinery. ## Caveats - This is an implementation repository and contains claims about its own state of the art; use it as implementation evidence, not as independent proof that the pattern is universally validated. - The observed commit is a point-in-time snapshot. Re-run repository inspection before making current operational claims.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "source", "wikilink" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "source" ] }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "kinds": [ "source" ] }, { "slug": "cxdb", "title": "CXDB", "type": "system", "kinds": [ "source" ] }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "source", "wikilink" ] }, { "slug": "healer", "title": "Healer", "type": "system", "kinds": [ "source" ] }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "kinds": [ "source" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "source" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "source" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "source" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "source" ] } ] }, { "title": "Karpathy LLM Wiki Gist", "slug": "karpathy-llm-wiki-gist-2026-04-04", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "llm-wiki", "knowledge-base", "rag" ], "sources": [ "https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f", "https://gist.githubusercontent.com/karpathy/442a6bf555914893e9891c11519de94f/raw/ac46de1ad27f92b28ac95459c782c07f6b8c964a/llm-wiki.md" ], "source_refs": [], "source_links": [ { "label": "https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f", "url": "https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f", "origin": "frontmatter" }, { "label": "https://gist.githubusercontent.com/karpathy/442a6bf555914893e9891c11519de94f/raw/ac46de1ad27f92b28ac95459c782c07f6b8c964a/llm-wiki.md", "url": "https://gist.githubusercontent.com/karpathy/442a6bf555914893e9891c11519de94f/raw/ac46de1ad27f92b28ac95459c782c07f6b8c964a/llm-wiki.md", "origin": "frontmatter" } ], "related": [ "llm-wiki-pattern", "llm-wiki-vs-rag", "SCHEMA" ], "wikilinks": [ "llm-wiki-pattern", "log", "index", "llm-wiki-vs-rag" ], "tensions": [ "The gist encourages LLM-owned wiki maintenance, but durable usefulness still depends on human source selection and review.", "The gist is an idea file, not a formal standard.", "The pattern does not eliminate the need for source verification, contradiction handling, or human correction." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Created: 2026-04-04. Captured: 2026-06-17. Provenance: primary gist by Andrej Karpathy describing the LLM wiki pattern.", "headings": [ { "level": "1", "title": "Karpathy LLM Wiki Gist" }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/karpathy-llm-wiki-gist-2026-04-04.md", "html_id": "page-karpathy-llm-wiki-gist-2026-04-04", "body_text": "Karpathy LLM Wiki Gist Source URL: LLM Wiki gist https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f Created: 2026-04-04. Captured: 2026-06-17. Provenance: primary gist by Andrej Karpathy describing the LLM wiki pattern. ## Compact Takeaways - The llm-wiki-pattern differs from ordinary RAG because knowledge is compiled into a persistent Markdown wiki rather than re-derived from raw chunks at query time. - The architecture has three layers: immutable raw sources, LLM-maintained wiki pages, and a schema that disciplines maintenance. - Ingest should update source cards, topic pages, backlinks, contradictions, index, and log. - The wiki is not only a storage format. It is a compounding artifact that future sessions can read before returning to raw sources. - The source explicitly treats the human as curator and question-asker, while the LLM handles summarizing, filing, cross-referencing, and bookkeeping. ## Wiki Implications - This repository's reports/wiki/ directory should be maintained as a durable artifact, not as a one-off report. - log and index are operational necessities because future agents need to know what changed and where to start. - llm-wiki-vs-rag should preserve the complementarity: RAG can find raw evidence, while a wiki preserves prior synthesis. ## Caveats - The gist is an idea file, not a formal standard. - The pattern does not eliminate the need for source verification, contradiction handling, or human correction.", "search_text": "Karpathy LLM Wiki Gist karpathy-llm-wiki-gist-2026-04-04 source high source llm-wiki knowledge-base rag https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f https://gist.githubusercontent.com/karpathy/442a6bf555914893e9891c11519de94f/raw/ac46de1ad27f92b28ac95459c782c07f6b8c964a/llm-wiki.md llm-wiki-pattern llm-wiki-vs-rag SCHEMA llm-wiki-pattern log index llm-wiki-vs-rag Created: 2026-04-04. Captured: 2026-06-17. Provenance: primary gist by Andrej Karpathy describing the LLM wiki pattern. Karpathy LLM Wiki Gist Source URL: LLM Wiki gist https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f Created: 2026-04-04. Captured: 2026-06-17. Provenance: primary gist by Andrej Karpathy describing the LLM wiki pattern. ## Compact Takeaways - The llm-wiki-pattern differs from ordinary RAG because knowledge is compiled into a persistent Markdown wiki rather than re-derived from raw chunks at query time. - The architecture has three layers: immutable raw sources, LLM-maintained wiki pages, and a schema that disciplines maintenance. - Ingest should update source cards, topic pages, backlinks, contradictions, index, and log. - The wiki is not only a storage format. It is a compounding artifact that future sessions can read before returning to raw sources. - The source explicitly treats the human as curator and question-asker, while the LLM handles summarizing, filing, cross-referencing, and bookkeeping. ## Wiki Implications - This repository's reports/wiki/ directory should be maintained as a durable artifact, not as a one-off report. - log and index are operational necessities because future agents need to know what changed and where to start. - llm-wiki-vs-rag should preserve the complementarity: RAG can find raw evidence, while a wiki preserves prior synthesis. ## Caveats - The gist is an idea file, not a formal standard. - The pattern does not eliminate the need for source verification, contradiction handling, or human correction.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "kinds": [ "source" ] }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "kinds": [ "source" ] }, { "slug": "llm-wiki-vs-rag", "title": "LLM Wiki vs RAG", "type": "comparison", "kinds": [ "source" ] } ] }, { "title": "StrongDM Attractor NLSpec Repository", "slug": "strongdm-attractor-nlspec", "type": "source", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "source", "strongdm", "attractor", "nlspec", "dot" ], "sources": [ "https://github.com/strongdm/attractor", "https://raw.githubusercontent.com/strongdm/attractor/main/README.md", "https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md", "https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md", "https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md" ], "source_refs": [], "source_links": [ { "label": "https://github.com/strongdm/attractor", "url": "https://github.com/strongdm/attractor", "origin": "frontmatter" }, { "label": "https://raw.githubusercontent.com/strongdm/attractor/main/README.md", "url": "https://raw.githubusercontent.com/strongdm/attractor/main/README.md", "origin": "frontmatter" }, { "label": "https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md", "url": "https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md", "origin": "frontmatter" }, { "label": "https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md", "url": "https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md", "origin": "frontmatter" }, { "label": "https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md", "url": "https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md", "origin": "frontmatter" } ], "related": [ "strongdm-attractor", "nlspec", "dot-pipeline", "agentic-pipeline-runner" ], "wikilinks": [ "nlspec", "software-factory", "dot-pipeline", "strongdm-attractor", "agentic-pipeline-runner", "attractor-pattern" ], "tensions": [ "The repository is framed as Attractor, but the visible artifact is a set of natural-language specifications rather than a runnable implementation.", "The source is a repository with specifications, not proof that any particular generated implementation is correct.", "Claims about production use need separate operational evidence beyond this repo." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Source URLs: GitHub repository, README, Attractor spec, Coding agent loop spec, Unified LLM spec", "headings": [ { "level": "1", "title": "StrongDM Attractor NLSpec Repository" }, { "level": "2", "title": "Compact Takeaways" }, { "level": "2", "title": "Wiki Implications" }, { "level": "2", "title": "Caveats" } ], "markdown_path": "sources/strongdm-attractor-nlspec.md", "html_id": "page-strongdm-attractor-nlspec", "body_text": "StrongDM Attractor NLSpec Repository Source URLs: GitHub repository https://github.com/strongdm/attractor, README https://raw.githubusercontent.com/strongdm/attractor/main/README.md, Attractor spec https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md, Coding agent loop spec https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md, Unified LLM spec https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md Captured: 2026-06-17. Provenance: primary StrongDM repository containing NLSpecs for building Attractor. ## Compact Takeaways - The repository defines nlspec as a human-readable specification intended for coding agents to implement and validate behavior. - The README positions Attractor as a route to building a software-factory, and recommends controlling the agent loop and unified LLM SDK foundation. - The Attractor spec defines a dot-pipeline runner: directed Graphviz graphs, typed attributes, node handlers, edge routing, checkpoint/resume, human gates, model stylesheets, validation, and extensibility. - The coding-agent-loop spec defines provider profiles, tool registries, execution environments, filesystem/search/command tools, and subagent extension points. - The unified LLM spec defines a provider-agnostic LLM client with request translation, streaming, tool use, retries, rate-limit handling, middleware, and adapter contracts. ## Wiki Implications - strongdm-attractor is best represented as a specs-only system definition unless a concrete implementation is separately cited. - agentic-pipeline-runner should include parse, transform, validate, initialize, execute, and finalize phases. - attractor-pattern is plausible because the specs constrain independent implementations toward the same layers. ## Caveats - The source is a repository with specifications, not proof that any particular generated implementation is correct. - Claims about production use need separate operational evidence beyond this repo.", "search_text": "StrongDM Attractor NLSpec Repository strongdm-attractor-nlspec source high source strongdm attractor nlspec dot https://github.com/strongdm/attractor https://raw.githubusercontent.com/strongdm/attractor/main/README.md https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md strongdm-attractor nlspec dot-pipeline agentic-pipeline-runner nlspec software-factory dot-pipeline strongdm-attractor agentic-pipeline-runner attractor-pattern Source URLs: GitHub repository, README, Attractor spec, Coding agent loop spec, Unified LLM spec StrongDM Attractor NLSpec Repository Source URLs: GitHub repository https://github.com/strongdm/attractor, README https://raw.githubusercontent.com/strongdm/attractor/main/README.md, Attractor spec https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md, Coding agent loop spec https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md, Unified LLM spec https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md Captured: 2026-06-17. Provenance: primary StrongDM repository containing NLSpecs for building Attractor. ## Compact Takeaways - The repository defines nlspec as a human-readable specification intended for coding agents to implement and validate behavior. - The README positions Attractor as a route to building a software-factory, and recommends controlling the agent loop and unified LLM SDK foundation. - The Attractor spec defines a dot-pipeline runner: directed Graphviz graphs, typed attributes, node handlers, edge routing, checkpoint/resume, human gates, model stylesheets, validation, and extensibility. - The coding-agent-loop spec defines provider profiles, tool registries, execution environments, filesystem/search/command tools, and subagent extension points. - The unified LLM spec defines a provider-agnostic LLM client with request translation, streaming, tool use, retries, rate-limit handling, middleware, and adapter contracts. ## Wiki Implications - strongdm-attractor is best represented as a specs-only system definition unless a concrete implementation is separately cited. - agentic-pipeline-runner should include parse, transform, validate, initialize, execute, and finalize phases. - attractor-pattern is plausible because the specs constrain independent implementations toward the same layers. ## Caveats - The source is a repository with specifications, not proof that any particular generated implementation is correct. - Claims about production use need separate operational evidence beyond this repo.", "backlinks": [ { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "kinds": [ "source" ] }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "kinds": [ "source" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "source", "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "source", "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "source", "wikilink" ] }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "kinds": [ "source" ] }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "kinds": [ "source" ] }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "kinds": [ "source" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "source" ] }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "kinds": [ "source" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "source" ] }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "kinds": [ "source" ] } ] }, { "title": "How Do Dark Factories Validate Without Reading Code?", "slug": "how-do-dark-factories-validate-without-reading-code", "type": "query", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "query", "validation", "code-review", "gates" ], "sources": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "dan-shapiro-you-dont-write-code-2026-02-13", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_links": [], "related": [ "code-review-vs-validation", "digital-twin-universe", "cxdb", "healer" ], "wikilinks": [ "cxdb", "healer" ], "tensions": [ "Automated validation can reduce code reading, but security-sensitive changes still need human-owned policy and periodic audit." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "They replace routine diff reading with layered outcome evidence.", "headings": [ { "level": "1", "title": "How Do Dark Factories Validate Without Reading Code?" }, { "level": "2", "title": "Validation Stack" }, { "level": "2", "title": "Key Point" } ], "markdown_path": "queries/how-do-dark-factories-validate-without-reading-code.md", "html_id": "page-how-do-dark-factories-validate-without-reading-code", "body_text": "How Do Dark Factories Validate Without Reading Code? They replace routine diff reading with layered outcome evidence. ## Validation Stack - Complete specs with no hidden requirements. - Public tests that agents can use while building. - Sealed holdouts that implementing agents cannot read. - Independent adversarial review by a different backend or model family. - CI, evidence gates, and reviewer verdicts. - Digital twins for realistic external-system behavior. - cxdb event logs and healer diagnosis for failure memory. - Human gates for high-risk, ambiguous, or policy-sensitive decisions. ## Key Point inference: The validation stack must be designed before code generation. If validation is added after a generated patch exists, humans end up reading code again because the factory did not define enough evidence up front.", "search_text": "How Do Dark Factories Validate Without Reading Code? how-do-dark-factories-validate-without-reading-code query medium-high query validation code-review gates dan-shapiro-you-dont-write-code-2026-02-13 jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo code-review-vs-validation digital-twin-universe cxdb healer cxdb healer They replace routine diff reading with layered outcome evidence. How Do Dark Factories Validate Without Reading Code? They replace routine diff reading with layered outcome evidence. ## Validation Stack - Complete specs with no hidden requirements. - Public tests that agents can use while building. - Sealed holdouts that implementing agents cannot read. - Independent adversarial review by a different backend or model family. - CI, evidence gates, and reviewer verdicts. - Digital twins for realistic external-system behavior. - cxdb event logs and healer diagnosis for failure memory. - Human gates for high-risk, ambiguous, or policy-sensitive decisions. ## Key Point inference: The validation stack must be designed before code generation. If validation is added after a generated patch exists, humans end up reading code again because the factory did not define enough evidence up front.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "kinds": [ "wikilink" ] } ] }, { "title": "How Do DOT Pipelines Map To Agent Workflows?", "slug": "how-do-dot-pipelines-map-to-agent-workflows", "type": "query", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "high", "tags": [ "query", "dot", "workflows", "agents" ], "sources": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_refs": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo", "jleechanorg-agent-orchestrator-repo" ], "source_links": [], "related": [ "dot-pipeline", "agentic-pipeline-runner", "agent-orchestrator", "tool-nodes-vs-llm-nodes" ], "wikilinks": [ "agent-orchestrator" ], "tensions": [ "DOT can describe both agent work and deterministic tooling, so the mapping is not one node equals one agent." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "DOT pipelines map workflow structure to executable nodes. Some nodes invoke agents; others run deterministic tools, gates, joins, or human approvals.", "headings": [ { "level": "1", "title": "How Do DOT Pipelines Map To Agent Workflows?" }, { "level": "2", "title": "Mapping" }, { "level": "2", "title": "Agent Orchestrator Boundary" } ], "markdown_path": "queries/how-do-dot-pipelines-map-to-agent-workflows.md", "html_id": "page-how-do-dot-pipelines-map-to-agent-workflows", "body_text": "How Do DOT Pipelines Map To Agent Workflows? DOT pipelines map workflow structure to executable nodes. Some nodes invoke agents; others run deterministic tools, gates, joins, or human approvals. ## Mapping - plan, implement, fix, review: usually LLM/coding-agent nodes. - test, smoke, score, lint, deploy: usually deterministic tool nodes. - gate: verdict nodes that route success, failure, or error. - humangate: explicit human approval or escalation. - waitall or joins: synchronization after fanout. - start and exit: lifecycle markers. ## Agent Orchestrator Boundary inference: A DOT runner can call agent-orchestrator to spawn or manage workers, but the DOT graph remains the process definition. The orchestrator owns session lifecycle; the graph owns workflow logic.", "search_text": "How Do DOT Pipelines Map To Agent Workflows? how-do-dot-pipelines-map-to-agent-workflows query high query dot workflows agents 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec jleechanorg-dark-factory-repo jleechanorg-agent-orchestrator-repo dot-pipeline agentic-pipeline-runner agent-orchestrator tool-nodes-vs-llm-nodes agent-orchestrator DOT pipelines map workflow structure to executable nodes. Some nodes invoke agents; others run deterministic tools, gates, joins, or human approvals. How Do DOT Pipelines Map To Agent Workflows? DOT pipelines map workflow structure to executable nodes. Some nodes invoke agents; others run deterministic tools, gates, joins, or human approvals. ## Mapping - plan, implement, fix, review: usually LLM/coding-agent nodes. - test, smoke, score, lint, deploy: usually deterministic tool nodes. - gate: verdict nodes that route success, failure, or error. - humangate: explicit human approval or escalation. - waitall or joins: synchronization after fanout. - start and exit: lifecycle markers. ## Agent Orchestrator Boundary inference: A DOT runner can call agent-orchestrator to spawn or manage workers, but the DOT graph remains the process definition. The orchestrator owns session lifecycle; the graph owns workflow logic.", "backlinks": [ { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "kinds": [ "wikilink" ] }, { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "kinds": [ "related" ] }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "kinds": [ "related" ] } ] }, { "title": "What Is The Durable Artifact In A Dark Factory?", "slug": "what-is-the-durable-artifact-in-a-dark-factory", "type": "query", "status": "active", "created": "2026-06-17", "updated": "2026-06-17", "confidence": "medium-high", "tags": [ "query", "durable-artifacts", "dark-factory" ], "sources": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo", "dan-shapiro-you-dont-write-code-2026-02-13" ], "source_refs": [ "2389-dark-factory-dot-file-2026-03-09", "strongdm-attractor-nlspec", "jleechanorg-dark-factory-repo", "dan-shapiro-you-dont-write-code-2026-02-13" ], "source_links": [], "related": [ "dot-pipeline", "specs-vs-code", "dark-factory" ], "wikilinks": [], "tensions": [ "Different sources emphasize different durable layers: DOT graphs, specs, validation systems, observability, or the whole factory." ], "bead_refs": [ "dark-factory-rr4" ], "summary": "Short answer: the durable artifact is the factory definition, not just the generated code.", "headings": [ { "level": "1", "title": "What Is The Durable Artifact In A Dark Factory?" }, { "level": "2", "title": "Source-Specific Answers" }, { "level": "2", "title": "Synthesis" } ], "markdown_path": "queries/what-is-the-durable-artifact-in-a-dark-factory.md", "html_id": "page-what-is-the-durable-artifact-in-a-dark-factory", "body_text": "What Is The Durable Artifact In A Dark Factory? Short answer: the durable artifact is the factory definition, not just the generated code. ## Source-Specific Answers - 2389 emphasizes the .dot file: the process graph is the product. - StrongDM Attractor emphasizes NLSpecs for the LLM client, coding loop, and DOT runner. - jleechanorg/dark-factory emphasizes .dot graphs plus spec.md, holdouts, scoring contracts, CXDB logs, and Healer diagnosis. - Shapiro emphasizes validation infrastructure and operational knowledge that let humans stop reading every code change. ## Synthesis inference: The durable artifact is a bundle: - intent specs; - executable process graphs; - prompt and tool contracts; - validation gates and holdouts; - observability logs; - repair and escalation policies. Generated code matters, but it is downstream of the factory definition.", "search_text": "What Is The Durable Artifact In A Dark Factory? what-is-the-durable-artifact-in-a-dark-factory query medium-high query durable-artifacts dark-factory 2389-dark-factory-dot-file-2026-03-09 strongdm-attractor-nlspec jleechanorg-dark-factory-repo dan-shapiro-you-dont-write-code-2026-02-13 dot-pipeline specs-vs-code dark-factory Short answer: the durable artifact is the factory definition, not just the generated code. What Is The Durable Artifact In A Dark Factory? Short answer: the durable artifact is the factory definition, not just the generated code. ## Source-Specific Answers - 2389 emphasizes the .dot file: the process graph is the product. - StrongDM Attractor emphasizes NLSpecs for the LLM client, coding loop, and DOT runner. - jleechanorg/dark-factory emphasizes .dot graphs plus spec.md, holdouts, scoring contracts, CXDB logs, and Healer diagnosis. - Shapiro emphasizes validation infrastructure and operational knowledge that let humans stop reading every code change. ## Synthesis inference: The durable artifact is a bundle: - intent specs; - executable process graphs; - prompt and tool contracts; - validation gates and holdouts; - observability logs; - repair and escalation policies. Generated code matters, but it is downstream of the factory definition.", "backlinks": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "kinds": [ "wikilink" ] }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "kinds": [ "wikilink" ] }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "kinds": [ "related" ] } ] } ], "graph": { "nodes": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "confidence": "high" }, { "slug": "SCHEMA", "title": "Dark Factory Wiki Schema", "type": "schema", "confidence": "high" }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "confidence": "high" }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "confidence": "medium-high" }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "confidence": "medium-high" }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "confidence": "high" }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "confidence": "high" }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "confidence": "high" }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "confidence": "medium" }, { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "confidence": "high" }, { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "confidence": "high" }, { "slug": "cxdb", "title": "CXDB", "type": "system", "confidence": "medium-high" }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "confidence": "medium-high" }, { "slug": "healer", "title": "Healer", "type": "system", "confidence": "medium-high" }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "confidence": "high" }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "confidence": "medium-high" }, { "slug": "llm-wiki-vs-rag", "title": "LLM Wiki vs RAG", "type": "comparison", "confidence": "high" }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "confidence": "medium-high" }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "confidence": "high" }, { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "confidence": "high" }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "confidence": "high" }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "confidence": "high" }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "confidence": "high" }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "confidence": "high" }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "confidence": "high" }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "confidence": "medium-high" }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "confidence": "high" }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "confidence": "medium-high" } ], "edges": [ { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "attractor-pattern", "kinds": [ "related", "wikilink" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "wikilink" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "dot-pipeline", "kinds": [ "related", "wikilink" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "how-do-dot-pipelines-map-to-agent-workflows", "kinds": [ "wikilink" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "specs-vs-code", "kinds": [ "wikilink" ] }, { "source": "2389-dark-factory-dot-file-2026-03-09", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "related", "wikilink" ] }, { "source": "SCHEMA", "target": "dark-factory", "kinds": [ "wikilink" ] }, { "source": "SCHEMA", "target": "index", "kinds": [ "related" ] }, { "source": "SCHEMA", "target": "karpathy-llm-wiki-gist-2026-04-04", "kinds": [ "source" ] }, { "source": "SCHEMA", "target": "llm-wiki-pattern", "kinds": [ "related", "wikilink" ] }, { "source": "SCHEMA", "target": "log", "kinds": [ "related" ] }, { "source": "agent-orchestrator", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "agent-orchestrator", "target": "code-review-vs-validation", "kinds": [ "related" ] }, { "source": "agent-orchestrator", "target": "dark-factory", "kinds": [ "wikilink" ] }, { "source": "agent-orchestrator", "target": "dot-pipeline", "kinds": [ "wikilink" ] }, { "source": "agent-orchestrator", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source" ] }, { "source": "agent-orchestrator", "target": "software-factory", "kinds": [ "related" ] }, { "source": "agent-orchestrator", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "related" ] }, { "source": "agentic-pipeline-runner", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "agentic-pipeline-runner", "target": "agent-orchestrator", "kinds": [ "related", "wikilink" ] }, { "source": "agentic-pipeline-runner", "target": "attractor-pattern", "kinds": [ "related" ] }, { "source": "agentic-pipeline-runner", "target": "cxdb", "kinds": [ "related" ] }, { "source": "agentic-pipeline-runner", "target": "dot-pipeline", "kinds": [ "related", "wikilink" ] }, { "source": "agentic-pipeline-runner", "target": "healer", "kinds": [ "related" ] }, { "source": "agentic-pipeline-runner", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source", "wikilink" ] }, { "source": "agentic-pipeline-runner", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "attractor-pattern", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "attractor-pattern", "target": "agent-orchestrator", "kinds": [ "wikilink" ] }, { "source": "attractor-pattern", "target": "agentic-pipeline-runner", "kinds": [ "related", "wikilink" ] }, { "source": "attractor-pattern", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "attractor-pattern", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "attractor-pattern", "target": "strongdm-attractor", "kinds": [ "related", "wikilink" ] }, { "source": "attractor-pattern", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "code-review-vs-validation", "target": "cxdb", "kinds": [ "related", "wikilink" ] }, { "source": "code-review-vs-validation", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "code-review-vs-validation", "target": "dark-factory", "kinds": [ "related", "wikilink" ] }, { "source": "code-review-vs-validation", "target": "digital-twin-universe", "kinds": [ "related" ] }, { "source": "code-review-vs-validation", "target": "healer", "kinds": [ "related", "wikilink" ] }, { "source": "code-review-vs-validation", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source" ] }, { "source": "code-review-vs-validation", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "cxdb", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "cxdb", "target": "code-review-vs-validation", "kinds": [ "related" ] }, { "source": "cxdb", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "cxdb", "target": "healer", "kinds": [ "related", "wikilink" ] }, { "source": "cxdb", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "code-review-vs-validation", "kinds": [ "wikilink" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "cxdb", "kinds": [ "related", "wikilink" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "dark-factory", "kinds": [ "related", "wikilink" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "digital-twin-universe", "kinds": [ "related", "wikilink" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "healer", "kinds": [ "related", "wikilink" ] }, { "source": "dan-shapiro-you-dont-write-code-2026-02-13", "target": "software-factory", "kinds": [ "related" ] }, { "source": "dark-factory", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "dark-factory", "target": "code-review-vs-validation", "kinds": [ "related" ] }, { "source": "dark-factory", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "dark-factory", "target": "digital-twin-universe", "kinds": [ "related", "wikilink" ] }, { "source": "dark-factory", "target": "dot-pipeline", "kinds": [ "related", "wikilink" ] }, { "source": "dark-factory", "target": "healer", "kinds": [ "wikilink" ] }, { "source": "dark-factory", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "dark-factory", "target": "software-factory", "kinds": [ "related" ] }, { "source": "digital-twin-universe", "target": "code-review-vs-validation", "kinds": [ "related" ] }, { "source": "digital-twin-universe", "target": "cxdb", "kinds": [ "related" ] }, { "source": "digital-twin-universe", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "digital-twin-universe", "target": "dark-factory", "kinds": [ "related" ] }, { "source": "dot-pipeline", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "dot-pipeline", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "dot-pipeline", "target": "dark-factory", "kinds": [ "wikilink" ] }, { "source": "dot-pipeline", "target": "how-do-dot-pipelines-map-to-agent-workflows", "kinds": [ "related" ] }, { "source": "dot-pipeline", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source", "wikilink" ] }, { "source": "dot-pipeline", "target": "strongdm-attractor-nlspec", "kinds": [ "source", "wikilink" ] }, { "source": "dot-pipeline", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "related" ] }, { "source": "healer", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "healer", "target": "cxdb", "kinds": [ "related", "wikilink" ] }, { "source": "healer", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "healer", "target": "dark-factory", "kinds": [ "related" ] }, { "source": "healer", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "code-review-vs-validation", "kinds": [ "related" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "cxdb", "kinds": [ "related", "wikilink" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "digital-twin-universe", "kinds": [ "related" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "healer", "kinds": [ "related", "wikilink" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source" ] }, { "source": "how-do-dark-factories-validate-without-reading-code", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "agent-orchestrator", "kinds": [ "related", "wikilink" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "how-do-dot-pipelines-map-to-agent-workflows", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "related" ] }, { "source": "index", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "SCHEMA", "kinds": [ "related", "wikilink" ] }, { "source": "index", "target": "agent-orchestrator", "kinds": [ "wikilink" ] }, { "source": "index", "target": "agentic-pipeline-runner", "kinds": [ "related", "wikilink" ] }, { "source": "index", "target": "attractor-pattern", "kinds": [ "wikilink" ] }, { "source": "index", "target": "code-review-vs-validation", "kinds": [ "wikilink" ] }, { "source": "index", "target": "cxdb", "kinds": [ "wikilink" ] }, { "source": "index", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "dark-factory", "kinds": [ "related", "wikilink" ] }, { "source": "index", "target": "digital-twin-universe", "kinds": [ "wikilink" ] }, { "source": "index", "target": "dot-pipeline", "kinds": [ "wikilink" ] }, { "source": "index", "target": "healer", "kinds": [ "wikilink" ] }, { "source": "index", "target": "how-do-dark-factories-validate-without-reading-code", "kinds": [ "wikilink" ] }, { "source": "index", "target": "how-do-dot-pipelines-map-to-agent-workflows", "kinds": [ "wikilink" ] }, { "source": "index", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "karpathy-llm-wiki-gist-2026-04-04", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "llm-wiki-pattern", "kinds": [ "wikilink" ] }, { "source": "index", "target": "llm-wiki-vs-rag", "kinds": [ "wikilink" ] }, { "source": "index", "target": "log", "kinds": [ "related", "wikilink" ] }, { "source": "index", "target": "nlspec", "kinds": [ "wikilink" ] }, { "source": "index", "target": "software-factory", "kinds": [ "wikilink" ] }, { "source": "index", "target": "specs-vs-code", "kinds": [ "wikilink" ] }, { "source": "index", "target": "strongdm-attractor", "kinds": [ "wikilink" ] }, { "source": "index", "target": "strongdm-attractor-nlspec", "kinds": [ "source", "wikilink" ] }, { "source": "index", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "wikilink" ] }, { "source": "index", "target": "what-is-the-durable-artifact-in-a-dark-factory", "kinds": [ "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "agent-orchestrator", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "agentic-pipeline-runner", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "code-review-vs-validation", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "dark-factory", "kinds": [ "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "software-factory", "kinds": [ "wikilink" ] }, { "source": "jleechanorg-agent-orchestrator-repo", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "related" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "agent-orchestrator", "kinds": [ "related" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "agentic-pipeline-runner", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "attractor-pattern", "kinds": [ "wikilink" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "cxdb", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "healer", "kinds": [ "related", "wikilink" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "how-do-dark-factories-validate-without-reading-code", "kinds": [ "wikilink" ] }, { "source": "jleechanorg-dark-factory-repo", "target": "specs-vs-code", "kinds": [ "wikilink" ] }, { "source": "karpathy-llm-wiki-gist-2026-04-04", "target": "SCHEMA", "kinds": [ "related" ] }, { "source": "karpathy-llm-wiki-gist-2026-04-04", "target": "index", "kinds": [ "wikilink" ] }, { "source": "karpathy-llm-wiki-gist-2026-04-04", "target": "llm-wiki-pattern", "kinds": [ "related", "wikilink" ] }, { "source": "karpathy-llm-wiki-gist-2026-04-04", "target": "llm-wiki-vs-rag", "kinds": [ "related", "wikilink" ] }, { "source": "karpathy-llm-wiki-gist-2026-04-04", "target": "log", "kinds": [ "wikilink" ] }, { "source": "llm-wiki-pattern", "target": "SCHEMA", "kinds": [ "related" ] }, { "source": "llm-wiki-pattern", "target": "index", "kinds": [ "wikilink" ] }, { "source": "llm-wiki-pattern", "target": "karpathy-llm-wiki-gist-2026-04-04", "kinds": [ "source" ] }, { "source": "llm-wiki-pattern", "target": "llm-wiki-vs-rag", "kinds": [ "related" ] }, { "source": "llm-wiki-pattern", "target": "log", "kinds": [ "related", "wikilink" ] }, { "source": "llm-wiki-vs-rag", "target": "SCHEMA", "kinds": [ "related" ] }, { "source": "llm-wiki-vs-rag", "target": "karpathy-llm-wiki-gist-2026-04-04", "kinds": [ "source" ] }, { "source": "llm-wiki-vs-rag", "target": "llm-wiki-pattern", "kinds": [ "related" ] }, { "source": "log", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "SCHEMA", "kinds": [ "related", "wikilink" ] }, { "source": "log", "target": "agent-orchestrator", "kinds": [ "wikilink" ] }, { "source": "log", "target": "agentic-pipeline-runner", "kinds": [ "wikilink" ] }, { "source": "log", "target": "attractor-pattern", "kinds": [ "wikilink" ] }, { "source": "log", "target": "code-review-vs-validation", "kinds": [ "wikilink" ] }, { "source": "log", "target": "cxdb", "kinds": [ "wikilink" ] }, { "source": "log", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "dark-factory", "kinds": [ "wikilink" ] }, { "source": "log", "target": "digital-twin-universe", "kinds": [ "wikilink" ] }, { "source": "log", "target": "dot-pipeline", "kinds": [ "wikilink" ] }, { "source": "log", "target": "healer", "kinds": [ "wikilink" ] }, { "source": "log", "target": "how-do-dark-factories-validate-without-reading-code", "kinds": [ "wikilink" ] }, { "source": "log", "target": "how-do-dot-pipelines-map-to-agent-workflows", "kinds": [ "wikilink" ] }, { "source": "log", "target": "index", "kinds": [ "related", "wikilink" ] }, { "source": "log", "target": "jleechanorg-agent-orchestrator-repo", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "karpathy-llm-wiki-gist-2026-04-04", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "llm-wiki-pattern", "kinds": [ "wikilink" ] }, { "source": "log", "target": "llm-wiki-vs-rag", "kinds": [ "wikilink" ] }, { "source": "log", "target": "nlspec", "kinds": [ "wikilink" ] }, { "source": "log", "target": "software-factory", "kinds": [ "wikilink" ] }, { "source": "log", "target": "specs-vs-code", "kinds": [ "wikilink" ] }, { "source": "log", "target": "strongdm-attractor", "kinds": [ "wikilink" ] }, { "source": "log", "target": "strongdm-attractor-nlspec", "kinds": [ "source", "wikilink" ] }, { "source": "log", "target": "tool-nodes-vs-llm-nodes", "kinds": [ "wikilink" ] }, { "source": "log", "target": "what-is-the-durable-artifact-in-a-dark-factory", "kinds": [ "wikilink" ] }, { "source": "nlspec", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "nlspec", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "nlspec", "target": "specs-vs-code", "kinds": [ "related" ] }, { "source": "nlspec", "target": "strongdm-attractor", "kinds": [ "related" ] }, { "source": "nlspec", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "software-factory", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "software-factory", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "software-factory", "target": "dark-factory", "kinds": [ "related", "wikilink" ] }, { "source": "software-factory", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "software-factory", "target": "strongdm-attractor", "kinds": [ "related" ] }, { "source": "software-factory", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "specs-vs-code", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "specs-vs-code", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "specs-vs-code", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "specs-vs-code", "target": "nlspec", "kinds": [ "related" ] }, { "source": "specs-vs-code", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "specs-vs-code", "target": "what-is-the-durable-artifact-in-a-dark-factory", "kinds": [ "related" ] }, { "source": "strongdm-attractor", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "strongdm-attractor", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "strongdm-attractor", "target": "attractor-pattern", "kinds": [ "related", "wikilink" ] }, { "source": "strongdm-attractor", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "strongdm-attractor", "target": "nlspec", "kinds": [ "related" ] }, { "source": "strongdm-attractor", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "strongdm-attractor-nlspec", "target": "agentic-pipeline-runner", "kinds": [ "related", "wikilink" ] }, { "source": "strongdm-attractor-nlspec", "target": "attractor-pattern", "kinds": [ "wikilink" ] }, { "source": "strongdm-attractor-nlspec", "target": "dot-pipeline", "kinds": [ "related", "wikilink" ] }, { "source": "strongdm-attractor-nlspec", "target": "nlspec", "kinds": [ "related", "wikilink" ] }, { "source": "strongdm-attractor-nlspec", "target": "software-factory", "kinds": [ "wikilink" ] }, { "source": "strongdm-attractor-nlspec", "target": "strongdm-attractor", "kinds": [ "related", "wikilink" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "agentic-pipeline-runner", "kinds": [ "related" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "how-do-dot-pipelines-map-to-agent-workflows", "kinds": [ "related" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "tool-nodes-vs-llm-nodes", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "2389-dark-factory-dot-file-2026-03-09", "kinds": [ "source" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "dan-shapiro-you-dont-write-code-2026-02-13", "kinds": [ "source" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "dark-factory", "kinds": [ "related" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "dot-pipeline", "kinds": [ "related" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "jleechanorg-dark-factory-repo", "kinds": [ "source" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "specs-vs-code", "kinds": [ "related" ] }, { "source": "what-is-the-durable-artifact-in-a-dark-factory", "target": "strongdm-attractor-nlspec", "kinds": [ "source" ] } ], "curated_edges": [ { "source": "software-factory", "target": "dark-factory", "label": "lights-off model" }, { "source": "dark-factory", "target": "dot-pipeline", "label": "process artifact" }, { "source": "dark-factory", "target": "code-review-vs-validation", "label": "validation shift" }, { "source": "dark-factory", "target": "digital-twin-universe", "label": "realistic checks" }, { "source": "dot-pipeline", "target": "agentic-pipeline-runner", "label": "execution" }, { "source": "agentic-pipeline-runner", "target": "cxdb", "label": "event log" }, { "source": "cxdb", "target": "healer", "label": "diagnosis" }, { "source": "agentic-pipeline-runner", "target": "agent-orchestrator", "label": "worker substrate" }, { "source": "strongdm-attractor", "target": "attractor-pattern", "label": "spec corpus" }, { "source": "attractor-pattern", "target": "nlspec", "label": "spec layer" }, { "source": "attractor-pattern", "target": "dot-pipeline", "label": "graph layer" }, { "source": "llm-wiki-pattern", "target": "index", "label": "knowledge layer" } ] }, "sections": { "start_here": [ "dark-factory", "what-is-the-durable-artifact-in-a-dark-factory", "dot-pipeline", "agentic-pipeline-runner", "code-review-vs-validation", "agent-orchestrator" ], "source_urls": [ "https://2389.ai/posts/the-dark-factory-is-a-dot-file/", "https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f", "https://gist.githubusercontent.com/karpathy/442a6bf555914893e9891c11519de94f/raw/ac46de1ad27f92b28ac95459c782c07f6b8c964a/llm-wiki.md", "https://github.com/jleechanorg/agent-orchestrator", "https://github.com/jleechanorg/dark-factory", "https://github.com/strongdm/attractor", "https://raw.githubusercontent.com/strongdm/attractor/main/README.md", "https://raw.githubusercontent.com/strongdm/attractor/main/attractor-spec.md", "https://raw.githubusercontent.com/strongdm/attractor/main/coding-agent-loop-spec.md", "https://raw.githubusercontent.com/strongdm/attractor/main/unified-llm-spec.md", "https://www.danshapiro.com/blog/2026/02/you-dont-write-the-code/" ], "tensions": [ { "slug": "index", "title": "Dark Factory LLM Wiki", "type": "index", "text": "This wiki is a curated synthesis; source cards remain the provenance layer for checking exact claims." }, { "slug": "log", "title": "Dark Factory Wiki Log", "type": "log", "text": "The initial ingest compiles point-in-time repository snapshots; repository claims should be refreshed before current operational decisions." }, { "slug": "attractor-pattern", "title": "Attractor Pattern", "type": "concept", "text": "Independent convergence is plausible from the sources, but the evidence is a small cluster of related implementations and specs." }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "text": "Dark factory rhetoric minimizes human code reading, but practical systems still need humans to define intent, gates, threat models, and acceptable risk." }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "text": "Human gates may be necessary, but too many gates recreate the manual bottleneck." }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "text": "Disposable generated code sounds freeing, but the runner, tests, and observability stack become critical infrastructure." }, { "slug": "dark-factory", "title": "Dark Factory", "type": "concept", "text": "\"Do not read code\" creates a security problem unless validation can catch unsafe behavior and supply-chain drift." }, { "slug": "dot-pipeline", "title": "DOT Pipeline", "type": "concept", "text": "DOT is simple and inspectable, but complex software lifecycle graphs can still become hard to reason about without schema validation and visual conventions." }, { "slug": "llm-wiki-pattern", "title": "LLM Wiki Pattern", "type": "concept", "text": "LLM-maintained synthesis compounds learning, but can also compound stale or mistaken synthesis unless source cards and logs stay current." }, { "slug": "nlspec", "title": "NLSpec", "type": "concept", "text": "Natural language specs are accessible to agents and humans, but ambiguity must be controlled with validation and examples." }, { "slug": "software-factory", "title": "Software Factory", "type": "concept", "text": "A software factory can still include human reading and manual approval; dark factory is a more aggressive variant." }, { "slug": "agent-orchestrator", "title": "Agent Orchestrator", "type": "system", "text": "It automates worker feedback loops, but its README still reserves merge/review and portability decisions for human judgment." }, { "slug": "agentic-pipeline-runner", "title": "Agentic Pipeline Runner", "type": "system", "text": "The runner can be considered disposable relative to the graph, but runner correctness is still load-bearing for safety and reproducibility." }, { "slug": "cxdb", "title": "CXDB", "type": "system", "text": "Observability can diagnose failures after the fact, but it does not by itself prove generated code is safe." }, { "slug": "digital-twin-universe", "title": "Digital Twin Universe", "type": "system", "text": "Simulation can improve validation fidelity, but it can also hide production-only failure modes if the twin is stale or incomplete." }, { "slug": "healer", "title": "Healer", "type": "system", "text": "Repair loops reduce human debugging burden, but a wrong diagnosis can send agents into repeated ineffective fixes." }, { "slug": "strongdm-attractor", "title": "StrongDM Attractor", "type": "system", "text": "Attractor is described as a system, but the cited repository is primarily a specification corpus." }, { "slug": "code-review-vs-validation", "title": "Code Review vs Validation", "type": "comparison", "text": "Human code review catches design and security issues that tests may miss; dark factory systems try to move that assurance into automated gates and independent review." }, { "slug": "llm-wiki-vs-rag", "title": "LLM Wiki vs RAG", "type": "comparison", "text": "A wiki preserves synthesis but can go stale; RAG preserves access to raw material but can repeatedly miss the same synthesis." }, { "slug": "specs-vs-code", "title": "Specs vs Code", "type": "comparison", "text": "Treating code as disposable is useful for generated implementations, but production systems still depend on correct executable code at runtime." }, { "slug": "tool-nodes-vs-llm-nodes", "title": "Tool Nodes vs LLM Nodes", "type": "comparison", "text": "Overusing tools can make the graph rigid; overusing LLM nodes can make runs expensive, nondeterministic, and hard to debug." }, { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "text": "The post says the factory code is disposable, but its examples also show substantial runner architecture and validation behavior that must be implemented carefully." }, { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "text": "This source mixes product narrative, implementation notes, and advocacy. Treat claims about what is \"the product\" as a position, not a settled industry standard." }, { "slug": "2389-dark-factory-dot-file-2026-03-09", "title": "2389: The Dark Factory Is a .dot file", "type": "source", "text": "The post says pipelines are the durable artifact, while dan-shapiro-you-dont-write-code-2026-02-13 emphasizes validation systems and factory operation." }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "text": "The article argues humans must stop reading generated code, while also foregrounding security and quality problems that require stronger validation systems." }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "text": "This source is narrative and observational. It reports specific StrongDM practices but does not provide formal specifications or implementation details." }, { "slug": "dan-shapiro-you-dont-write-code-2026-02-13", "title": "Dan Shapiro: You Don't Write the Code. You Don't Read the Code Either.", "type": "source", "text": "It presents an aspirational lights-off model, but the examples still imply human judgment at the level of factory design, bootcamps, validation policy, and tool building." }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "text": "The fork is framed as zero-touch automation, but the README still reserves review/merge and portability decisions for human judgment." }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "text": "This source is a fork README plus architecture doc, not a stable API contract." }, { "slug": "jleechanorg-agent-orchestrator-repo", "title": "jleechanorg/agent-orchestrator Repository", "type": "source", "text": "The fork has environment-specific assumptions; future agents should not copy its defaults without checking portability." }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "text": "The README describes runner code as disposable, but the repository contains substantial engine, validation, observability, and benchmark code that future implementations should study." }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "text": "This is an implementation repository and contains claims about its own state of the art; use it as implementation evidence, not as independent proof that the pattern is universally validated." }, { "slug": "jleechanorg-dark-factory-repo", "title": "jleechanorg/dark-factory Repository", "type": "source", "text": "The observed commit is a point-in-time snapshot. Re-run repository inspection before making current operational claims." }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "text": "The gist encourages LLM-owned wiki maintenance, but durable usefulness still depends on human source selection and review." }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "text": "The gist is an idea file, not a formal standard." }, { "slug": "karpathy-llm-wiki-gist-2026-04-04", "title": "Karpathy LLM Wiki Gist", "type": "source", "text": "The pattern does not eliminate the need for source verification, contradiction handling, or human correction." }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "text": "The repository is framed as Attractor, but the visible artifact is a set of natural-language specifications rather than a runnable implementation." }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "text": "The source is a repository with specifications, not proof that any particular generated implementation is correct." }, { "slug": "strongdm-attractor-nlspec", "title": "StrongDM Attractor NLSpec Repository", "type": "source", "text": "Claims about production use need separate operational evidence beyond this repo." }, { "slug": "how-do-dark-factories-validate-without-reading-code", "title": "How Do Dark Factories Validate Without Reading Code?", "type": "query", "text": "Automated validation can reduce code reading, but security-sensitive changes still need human-owned policy and periodic audit." }, { "slug": "how-do-dot-pipelines-map-to-agent-workflows", "title": "How Do DOT Pipelines Map To Agent Workflows?", "type": "query", "text": "DOT can describe both agent work and deterministic tooling, so the mapping is not one node equals one agent." }, { "slug": "what-is-the-durable-artifact-in-a-dark-factory", "title": "What Is The Durable Artifact In A Dark Factory?", "type": "query", "text": "Different sources emphasize different durable layers: DOT graphs, specs, validation systems, observability, or the whole factory." } ] } }