Foundational Architecture

Why General AI Fails at Entitlement Diligence

Unstructured RAG models and general LLMs treat municipal codes as loose text blocks. They hallucinate thresholds, miss overlapping environmental overlays, and ignore hidden cross-references. LocalCode replaces probabilistic guessing with a deterministic legal data graph.

Generic AI / Vector Retrieval

Probabilistic parsing of unstructured code text.

json

// Vector Search Result for: "Austin MF-3 Max Density"
{
  "query": "What is the maximum density for MF-3?",
  "llm_response": "Based on § 25-2-64, the maximum density is 36 units per acre.",
  "confidence": "High (98%)",
  "source_anchoring": "Text snippet from PDF page 105",
  "CRITICAL_HALLUCINATION": {
    "status": "FAILED",
    "reason": "The model completely missed Chapter 25-8 (SOS Watershed Cap), which gut-lines net impervious cover to 15% on this specific recharge zone, capping true physical yield at 8 units per acre."
  }
}

The LocalCode Intelligence Engine

Relational execution across our multi-layered legal graph.

json

// LocalCode Authoritative Contract Body
{
  "subject": {
    "citation": "§ 25-2-64",
    "base_zoning": "MF-3"
  },
  "confidence_state": "context_required",
  "confirmed_attributes": [
    { "attribute_key": "applicability_rule", "scope": "zoning_jurisdiction" }
  ],
  "confirmed_standards": [
    { "standard_key": "max_density_ceiling", "value_num": 36, "unit": "du_ac" }
  ],
  "standard_groups": [
    {
      "group_key": "austin_watershed_bua_constraints",
      "binding_limit": "15% Net Site Impervious Cover",
      "source_section": "Ch. 25-8 Article 12 (Save Our Springs)"
    }
  ],
  "unresolved_dependencies": [
    { "source": "§ 25-2-64", "target": "§ 25-4-232", "status": "blocked" }
  ],
  "allowed_claims": [
    "The baseline zoning table permits up to 36 du/ac."
  ],
  "prohibited_claims": [
    "Claiming 36 du/ac is achievable without modeling the Chapter 25-8 environmental cap."
  ]
}

Engine Mechanics

Three Schema Layers That Eliminate Guesswork

Semantic Classification (codes_attr)

We index the legal function of every municipal provision. Before any numeric values are evaluated, our engine isolates whether a clause is a definition, a measurement rule, a procedure, or an overlay restriction. This prevents the system from misinterpreting measurement methods as hard regulatory caps.

Structured Standards (codes_standards)

We break down dense legal prose into queryable scalar rows containing strict comparators, numeric values, and exact conditions. Because these rules are evaluated using a strict relational structure rather than loose prompt summaries, they are calculation-ready and perfectly predictable.

Absolute Provenance & History

Every single standard and attribute assignment is anchored back to an immutable text block via 0-based, end-exclusive character offsets. We track the official ordinance numbers, adoption dates, and full amendment lineages, providing land-use teams with an absolute, auditable paper trail directly back to primary sources.

Built for the Age of Bounded Cognition

If your team or platform is building AI workflows over land use data, do not let an LLM directly read raw text documents. Integrate LocalCode to serve as your deterministic verification layer — ensuring your agents operate within bounded legal realities, backed by verifiable municipal truth.

Explore Austin Feasibility Pilot →