Fix: MAESTRO layer descriptions still defined the superseded model

[emmanuelgjr]

Plan ID: — (follow-up to #32; reported on issue #31) Ticket: — Wave: —
Constraints honored:

• C4 (transcription only, no taxonomy change): descriptions come from this repository's own canonical architecture table; no control ids change; sub-controls untouched.
• C2: docs/frameworks-registry.js is regenerated with data only; needs approval.
• C1: untouched.

Stacked on #37, like #87: every branch from main currently fails two CI jobs on the date-header bug (Drift 11), and this PR regenerates a docs/*.js bundle. Merge #37, then retarget to main.

Why

#32 corrected the seven MAESTRO layer titles to the CSA model, but left the descriptions of the superseded model behind. Every layer from L4 up had the right name and another layer's definition:

id	title	description before	description now (llm-top10/LLM_MAESTRO.md)
L1	Foundation Models	Base model layer — covers model selection, provenance, fine-tuning security, and model integrity threats.	Base LLMs providing core reasoning and generation
L2	Data Operations	Data pipelines — covers RAG, vector stores, training data, embedding security, and data governance.	Ingestion pipelines, storage, RAG, embeddings, vector stores
L3	Agent Frameworks	Agent runtime — covers agent orchestration, goal management, memory, and planning security.	Orchestration platforms, tool registries, MCP, plugin ecosystems
L4	Deployment & Infrastructure	Tool access — covers MCP, API integrations, plugin security, and tool authorization.	Servers, containers, networks, CI/CD, runtime environments
L5	Evaluation & Observability	Infrastructure — covers containerization, networking, secrets management, and runtime isolation.	Monitoring, logging, telemetry, behavioural baselines
L6	Security & Compliance	Multi-agent coordination — covers inter-agent communication, delegation, consensus, and cascading failure prevention.	Identity, access control, audit, governance, credential management
L7	Agent Ecosystem	Human-agent interface — covers user authentication, output validation, human oversight, and trust management.	Multi-agent interaction, A2A communication, cascade dynamics

• Where the old text shipped: the npm package's registries, the webapp registry bundle, and the OSCAL catalog export. Anyone choosing a layer by its description picked the wrong one.
• Correction to my MAESTRO registry sub-controls follow the superseded layer model #31 comment: it said L1–L3 "agree with the table". They don't quite. L3's old scope omitted MCP and tool registries because the superseded model put those under L4. So all seven are transcribed.
• Tradeoff: the table text is terser than the old L1–L3 prose, e.g. it drops "provenance" and "memory". If that detail matters, enrich the table in LLM_MAESTRO.md and the registry follows. The validator keeps them in lockstep.

Changes

• data/frameworks/maestro.json: 7 layer descriptions; changelog entry.
• docs/frameworks-registry.js: regenerated.
• scripts/validate.js: checkMaestroLayers() now compares layer descriptions as well as titles. Why it missed this: it checked titles only.

Deliberately NOT changed

• The 21 sub-controls, which still follow the superseded model: that's a control-id / taxonomy decision (MAESTRO registry sub-controls follow the superseded layer model #31).
• Mapping rows: all reference top-level ids, so none move.
• The small wording differences between the four MAESTRO mapping files' tables (e.g. Agentic_MAESTRO.md adds "drift detection" to L5). LLM_MAESTRO.md stays canonical.

Human decisions / approvals needed

• C2: data-only regeneration of docs/frameworks-registry.js.

Verify

$ node scripts/validate.js
before: Summary: 0 error(s), 84 warning(s), 312 passed
after:  Summary: 0 error(s), 84 warning(s), 312 passed
negative test — old L5 description restored:
  ✗ [MAESTRO layers] maestro.json L5 description "Infrastructure — covers containerization, …" does not match LLM_MAESTRO.md: "Monitoring, logging, telemetry, behavioural baselines"
  Summary: 1 error(s)
$ npm run stats:check     ✓ current
$ npm test                51/51
$ grep -r "Tool access — covers" (excluding node_modules, reports)  → no matches

Local only: Content Validation runs for PRs into main, so it will run on retarget.

🤖 Generated with Claude Code

https://claude.ai/code/session_014SfR2YzLxRH54DAVzDk8gR