Use when ingesting raw artifacts, probes, requirements, or context at the start of any analysis. L0 layer — preserves full variance structure without averaging. Triggers when collecting raw data, exploring project state, or beginning any investigation.
The foundational layer. Every analysis begins here. No exceptions.
Ingest raw artifacts into an unaveraged point cloud. Each datum is a discrete point in a metric space. The variance structure across these points IS the topological signal.
This is non-negotiable. Raw probes, requirements, codebase signals, and constraint statements must never be aggregated, summarized, or flattened before filtration.
Premature averaging is a category error — it destroys the persistence signal that reveals which features are robust versus topological noise. The noise floor is not noise. It is the low-level deductive artifact from which high-level abstraction must be reverse-engineered.
| Input Type | Point Cloud Encoding | What Variance Reveals |
|---|---|---|
| Codebase files | Each file = point, distance = functional dissimilarity |
| Module structure |
| User requirements | Each answer = point, distance = semantic dissimilarity | Requirement clustering |
| Probe traces | Each probe = point, distance = numerical difference | Signal vs noise boundary |
| Zeta zeros | Each gap = point, distance = spectral separation | Arithmetic structure |
Before proceeding to L1, check: does the point cloud have sufficient variance?
/dw-filterIf at any point someone attempts to "summarize what we have so far" or "average the results" before filtration has occurred:
[DW-AXIOM-VIOLATION: NO_AVERAGING]
Raw probes must reach L1 at full resolution.
Averaging before filtration destroys the topological signal.