l0l1.skelfresearch

Glossary

The vocabulary l0l1 uses for LLM-assisted SQL, validation, and privacy at the warehouse boundary.

Schema-aware validation
Checking a query against the real tables, columns, and relationships of the connected database, rather than against generic SQL grammar alone.
“Mostly correct” SQL
LLM-generated SQL that is locally plausible and syntactically valid but silently wrong against your actual schema or semantics — l0l1’s core reason to exist.
PII detection
Scanning prompts and query text for personally identifiable information (SSNs, emails, phone numbers, card numbers) using Microsoft Presidio plus regex rules.
Anonymization
Replacing detected PII literals with placeholder tokens before a query is stored or forwarded to an external model provider.
Presidio
Microsoft’s open-source data protection and PII detection SDK, which l0l1 uses as the backbone of its scanner.
Pattern learning
Recording successful queries — sanitized of literals — as reusable shape patterns scoped to a workspace, so approved SQL can be suggested later.
Shape pattern
A query stripped of its specific literals, capturing structure (tables, columns, joins, filters) without the sensitive values.
Workspace
The scope within which l0l1 stores and surfaces learned query patterns, so one team’s approved SQL informs completions for that team.
The four boundaries
User-to-tool, tool-to-model-provider, tool-to-warehouse, and warehouse-to-output — the crossings l0l1 reasons about and protects.
LLM reviewer
The step where l0l1 uses OpenAI or Anthropic to reason about a query’s intent and correctness — reviewing SQL rather than authoring it.
Schema introspection
Reading and caching a database’s structure so validation has an accurate, up-to-date picture of what exists.
LSP server
The Language Server Protocol backend that powers l0l1’s VS Code extension, delivering validation and completions inside the editor.