Public docs
How Martenweave Works
Martenweave turns scattered model knowledge into checked, searchable model context.
The workflow is deliberately explicit so humans can inspect each step.
1. Bring Inputs
Inputs can include:
- Excel mappings
- CSV or XLSX datasets
- ticket notes
- validation reports
- decision logs
- SAP table and field context
- source-system extracts
- steward comments
Raw inputs are not treated as truth. They provide evidence.
2. Normalize Into Model Objects
Martenweave represents model knowledge as canonical objects such as:
- domains
- business entities
- entity contexts
- attributes
- attribute usages
- field endpoints
- mappings
- value lists
- validation rules
- issues
- decisions
- evidence
- patch proposals
- change requests
The canonical files are Markdown and YAML. They are readable in Git and can be reviewed without a hosted UI.
3. Validate Consistency
Validation checks object structure and references before trusted indexing.
Checks include:
- ID format
- required fields
- registered object types
- duplicate IDs
- broken references
- reference type mismatches
- ownership and readiness rules
- SAP context rules for SAP starter scenarios
4. Build The Index
Generated outputs are rebuildable.
They can include:
- SQLite index
- search JSONL
- lineage JSONL
- audit logs
- reports
- exports
If generated files disagree with canonical files, canonical files win.
5. Analyze Gaps And Impact
Once the model is indexed, teams can ask practical questions:
- Which dataset columns are missing expected model endpoints?
- Which attributes lack source evidence?
- Which field endpoints depend on a changed rule?
- Which mappings, owners, issues, and decisions are affected?
- What should be reviewed before a migration load or support handover?
6. Propose Changes
AI and import workflows create PatchProposal objects instead of silently changing canonical files.
A good proposal records:
- what evidence triggered the proposal
- which objects are affected
- what operations are proposed
- what assumptions were made
- what validation and impact checks should run
7. Human Approval
Humans approve, reject, or revise proposals.
Approved changes become ChangeRequests and then canonical updates. This keeps the change trail visible.
8. Export Reports And Context
Martenweave can export model context for review, analysis, automation, and AI-assisted work.
Examples include:
- search documents
- lineage exports
- model CSV or XLSX files
- JSON Schema
- gap reports
- impact reports
- GitHub-ready issue or change bundles
Summary
Inputs bring evidence.
Model objects create structure.
Validators create trust.
Indexes create search and analysis.
Patch proposals keep AI and imports reviewable.