Legacy AI Content Fact-Check Matrix
Legacy AI Content Fact-Check Matrix
Audit result
All 26 preserved legacy Markdown pages were inventoried and reviewed on 2026-09-02. Three MCP pages were rewritten from current primary sources. The other 23 pages remain available as provenance but are excluded from publication because static review cannot establish that old code, dependency ranges, model identifiers, third-party URLs, or vendor-specific workflows still run. “Syntax pass” means a fenced Python block parses; it does not mean imports, integrations, or behavior were executed.
Status meanings
| Status | Meaning |
|---|---|
| Rewritten current | Claims replaced with current, claim-cited guidance from primary sources. |
| Historical code snapshot | Concepts and static syntax reviewed; runtime and dependency compatibility not established. |
| Historical framework snapshot | High-level concepts reviewed; fast-changing APIs are not treated as current. |
| Historical security snapshot | Threat categories reviewed; rankings and example outcomes are not universal facts. |
| Reviewed template | Structure reviewed; no test execution is implied. |
| Deprecated workflow snapshot | Vendor-specific procedure is not current or end-to-end tested. |
| Link ledger audited | Link availability checked; inclusion does not endorse source quality. |
File-by-file disposition
| Preserved file | Status | Verification and correction |
|---|---|---|
MCP Discovery.md | Rewritten current | Replaced legacy HTTP fingerprinting with authorized inventory and the current server/discover operation. Corrected the Knostic result: 1,862 servers were discovered, while 119 were manually sampled and all 119 exposed unauthenticated tool-list access—not all 1,862. S1 |
MCP v1.md | Rewritten current | Replaced obsolete lifecycle guidance with the 2026-07-28 specification: requests are self-contained, capabilities are per-request, and protocol-level sessions plus initialize were removed. S1 |
MCP v2.md | Rewritten current | Re-scoped CVE-2025-6515 to the oatpp-mcp implementation rather than MCP generally, and aligned controls with the current trust and authorization model. S1 |
0. Retrieval-Augmented Generation.md | Historical code snapshot | RAG flow is directionally sound. Referenced local Venv/ scripts are absent; the note now points to preserved RAG-CAS artifacts. Mermaid semantics reviewed as a retrieval/generation flow. S5 |
1. Text Extraction.md | Historical code snapshot | PDF extraction/OCR concepts are sound; one Python block parses. Paths, system packages, and OCR runtime were not executed. S5 |
2. Text Chunking.md | Historical code snapshot | Correctly distinguishes character-based splitting from tokens; one Python block parses. A fixed character count cannot guarantee a token count. S5 |
3. Embedding Chunks and Storing in Vector DB.md | Historical code snapshot | Embedding and similarity-search description is sound; one Python block parses. Model/database compatibility was not executed. S5 |
4. Querying and Retrieving Relevant Context.md | Historical code snapshot | Query embedding and nearest-neighbor retrieval flow is sound; one Python block parses. “Top 3” is a configuration choice, not a generally optimal value. S5 |
5. RAG interface.md | Historical code snapshot | UI/retrieval design is plausible; one Python block parses. Streamlit, Chroma, and SentenceTransformers integration was not run. S5 |
6. main.md | Historical code snapshot | End-to-end local RAG design is plausible; one Python block parses and the Mermaid flow matches the prose. Imports, OCR tools, models, and services were not run. S5 |
7. Ollama.md | Historical code snapshot | One Python block parses. Stored paths and subprocess assumptions are environment-specific; current Ollama API endpoints and context settings should be used for a new build. S6 |
Docker. - Pending/10. Docker.md | Historical code snapshot | Container design is historical. depends_on controls order but does not establish readiness unless readiness conditions and health checks are configured. Floating image tags are not reproducible. S7 |
Docker. - Pending/11. init.sh.md | Historical code snapshot | Shell generator was reviewed statically only. It creates unpinned dependencies and a multi-process container; generated services were not built or run. S7 |
Docker. - Pending/12. requirement.txt.md | Historical code snapshot | Package names are plausible but unpinned, so this is not a reproducible dependency lock. No installation was attempted. |
Docker. - Pending/13. app.py.md | Historical code snapshot | Python block parses. Imports and integration behavior were not executed; LangChain package paths are version-sensitive. S8 |
Docker. - Pending/14. supervisord.conf.md | Historical code snapshot | Supervisor syntax was inspected only. Process readiness, signals, logs, and container shutdown behavior were not integration-tested. |
0. LangChain and LangGraph.md | Historical framework snapshot | Two Python blocks parse and both Mermaid diagrams match the conceptual contrast. Current LangGraph documentation still centers durable execution, persistence, streaming, and human-in-the-loop control; API examples remain version-sensitive. S8 |
2.4.1. Test Logging Template.md | Reviewed template | Useful test-record skeleton; redaction and model/version capture are appropriate. A filled template is still evidence only when artifacts and results are retained. S9 |
2.4.2. AI Red Teaming.md | Historical security snapshot | Broad attack categories align with current OWASP/NIST guidance; the Mermaid decision flow matches the prose. Examples require authorization and model-specific validation. S9 |
2.4.3. LLM Questioning Playbook.md | Historical security snapshot | Test structure is useful, but “typical vulnerability” rankings and expected failure modes are not universal or stable across models. Mermaid semantics reviewed. S9 |
2.4.4. Top Commenting Patterns.md | Historical security snapshot | Preserved prompt-pattern collection. Counts, tiers, and success expectations lack a reproducible benchmark and are not treated as facts. S9 |
CLAUDE_DESKTOP_N8N_MCP_SETUP.md | Deprecated workflow snapshot | Created for a 2025 environment and legacy MCP lifecycle. Stored commands, URLs, package versions, and Claude/n8n UI steps were not verified as current. S1 |
Prompt Downloader.md | Deprecated workflow snapshot | Planning note contains vendor- and repository-specific assumptions. Token scope, data rights, rate limits, and current node schemas require fresh validation. S11 |
complete_setup_guide.md | Deprecated workflow snapshot | Static setup guide only; GitHub, Anthropic, and n8n behavior changes over time. JSON/JavaScript snippets were not executed and credentials must not be committed. S11 |
researcher.md | Deprecated workflow snapshot | Companion workflow JSON is syntactically valid, but Perplexity/Anthropic models, n8n nodes, endpoints, and output behavior were not run end to end. S11 |
90-existing-resources.md | Link ledger audited | Of 45 unique HTTP(S) targets, 28 returned 200, 12 returned 404, one returned 403, and four were malformed or unreachable on 2026-09-02. The page is excluded from publication; use the authoritative ledger instead. |
Cross-cutting findings
- All ten fenced Python blocks compile syntactically with Python 3, all 21 Bash blocks pass
bash -n, all three YAML blocks parse, and three of four JSON blocks parse. The fourth JSON block intentionally contains an ellipsis and is an illustrative fragment. Three of four JavaScript blocks passnode --check; the fourth is a workflow-expression fragment rather than a standalone program. These checks do not validate imports, integrations, credentials, or behavior. - The three rewritten MCP diagrams and five diagrams in historical legacy pages were semantically reviewed against their surrounding text. Historical diagrams are retained as explanatory snapshots, not current API contracts.
- Version-sensitive pages now carry
version_checked,execution_checked, and explicit fact-check status fields. The 23 non-current pages havedg-publish: false. - Raw pre-rewrite MCP pages are retained at
dumpster/AI/2026-09-02-legacy-ai-fact-check/raw-mcp/for provenance.