SDM treats methodology as specs: written once, verified by machine, with all derived output generated automatically. Designed for the AI era — humans set the direction, AI agents execute the process.
Before SDM, methodological artifacts are scattered across Wiki pages, Excel tables, LMS silos — with no single source of truth and no honest answer to "what is covered?".
Competency matrices in Wiki, Excel, LMS — different versions everywhere. Nobody knows which one is current.
There is no reproducible answer to "what covers langchain4j for a Java Middle?". Coverage lives in the expert's head.
Separate circuits: tests for hiring, courses for onboarding — no shared skill graph. Change a skill, update everything by hand.
AI accelerates content production, but without a spec backbone you get more chaos faster. SDM gives AI agents a machine-readable skeleton to work within: humans define ontology and thresholds, agents compute coverage and generate exports. Quality = specs × agent × model.
A single spec chain that works the same way whether a human or an AI agent executes it.
Export (tests, courses, kits, graphs) is delivery mechanics — Presentation layer, not a stage
Skills, topics, concepts in a single graph — no more "which Excel has the latest version".
Fixed gap definition: same ontology + profile → same gaps, every implementation. Reproducible, not guesswork.
One ontology feeds multiple profiles. One question bank, different slices — no duplicated content.
CLI, MCP, Obsidian, web — implementations compete, methodology stays the same. Switch without restarting.
Edit the spec in one place; coverage and exports rebuild controllably. No manual hunt for ripple effects.
Humans state intent in natural language — AI agent clarifies, plans, and executes via MCP/CLI. You review.
A single, machine-verifiable chain of methodological artifacts — independent of who performs the steps: a human or an agent.
Nodes of any kind — skills, topics, concepts, questions, terms, talks — and their relationships in a single graph. Content is part of the ontology.
What is selected from the ontology and to what threshold. The profile defines the target.
What is missing, computed by a fixed measure. Identical across all implementations.
Artifacts, not commands — the same chain, whoever runs it.
A graph of nodes of any kind — skills, questions, terms, topics — and their relationships.
Selection from the ontology with a threshold and depth.
Gap nodes by a fixed definition.
From specification to presentation — each layer has its own pace of change.
| Layer | What | How it changes |
|---|---|---|
| Specification | Ontology (nodes+kinds) + profile + gap definition | Rarely |
| Computation | Coverage calculation by a fixed measure | Iteratively |
| Presentation | Export (MCP / Obsidian / web / CLI / documents) | Fully open |
SDM is built for the AI era from the ground up. Humans describe what they need — AI agents do the heavy lifting within spec boundaries.
"I need a Java Middle profile with AI/LLM skills." The agent clarifies, shows a plan, waits for confirmation. No CLI flags. No YAML by hand.
Agent calls SDM via CLI/MCP: creates skills, adds questions, computes coverage, exports artifacts. Stays within frame — can't violate invariants.
Coverage report, export preview — you say "ok" or "fix this". The agent iterates. You stay in control, the agent does the work.
Agent suggests next steps: close gaps, harden quality, export for a new surface. Same frame, guided by methodology state.
Quality = specs × agent × model. The methodology constrains the agent — it does not choose the gap definition, does not bypass validation.
SDM is not limited to developer profiles. The same spec chain works for curriculum design, conference programs, and any methodological artifact. Each example shows the full Ontology → Profile → Coverage → Export lifecycle.
A complete competency profile with 8 skills, 32 questions, machine-checked coverage, and generated exports. Includes Spring AI and LangChain4j for the AI era.
An ontology with 4 node kinds — topics, skills, concepts, and tools. A course profile sets requirements per topic; coverage shows which modules lack materials.
Talks as first-class nodes linked to skills and speakers. The program profile checks which topics have enough talks — no more spreadsheets.
Clear boundaries help keep the methodology focused.
Methodologists, educators, teachers, bloggers, conference organizers — anyone who works with methodological artifacts and wants to verify their completeness.