Glossary#

analysis run#: Complete invocation of compass process that ingests a configuration file, processes jurisdictions, and writes results to the run directory.
clean directory#: Intermediate folder storing cleaned ordinance text used for LLM prompting during feature extraction.
clean text file#: Plain-text excerpt derived from ordinance documents that isolates relevant sections for prompts and validation.
compass process#: CLI command that executes the end-to-end pipeline using the inputs defined in the configuration file.
configuration file#: JSON or JSON5 document that declares inputs, model assignments, concurrency, and output directories for a run.
decision tree#: Hierarchical rubric of questions and outcomes that organizes how ordinance features are extracted and validated.
decision tree prompt#: Structured prompt template that guides the LLM through branching questions to extract quantitative and qualitative ordinance data.
extraction pipeline#: Crawlers, parsers, and feature detectors that transform raw ordinance text into structured records.
INFRA-COMPASS#: End-to-end pipeline that discovers, parses, and validates energy infrastructure ordinances with LLM tooling.
jurisdiction#: County or municipality defined in the jurisdiction CSV that frames the geographic scope of an analysis run.
jurisdiction CSV#: Input spreadsheet whose County and State columns list the locations processed in a run.
LLM#: Large Language Model that interprets ordinance text, classifies features, and answers structured extraction questions.
llm cost tracker#: Runtime utility that multiplies token usage by configured pricing to report estimated spend per model.
llm service#: Abstraction over providers such as OpenAI or Azure OpenAI that enforces authentication, rate limits, and retry policies.
llm service rate limit#: Configuration value that caps tokens per minute for a model to avoid provider throttling.
llm task#: Logical label assigned to prompt templates that maps to a specific model entry within the configuration.
location#: Combination of county and state identifiers that maps to one row in the jurisdiction CSV and produces a single output bundle.
location file log#: Per-location structured log that aggregates runtime diagnostics and JSON exception summaries.
location manifest#: JSON metadata file emitted per location summarizing source documents, extraction status, and validation outcomes.
log directory#: Folder defined by log_dir that stores run-level logs, prompt archives, and timing summaries.
OCR#: Optical Character Recognition stage powered by pytesseract that converts scanned ordinance PDFs into searchable text.
ordinance#: Legal text that governs energy infrastructure within a jurisdiction and feeds the extraction workflows.
ordinance document#: Source PDF or HTML retrieved during crawling that contains the legal language for the targeted technology.
ordinance file directory#: Folder defined by ordinance_file_dir that caches downloaded ordinance PDFs and HTML files.
out directory#: Root folder defined by out_dir where structured results, cleaned text, and logs for each run are written.
Pixi#: Environment manager used to install dependencies, run tasks, and maintain reproducible shells for COMPASS.
Playwright#: Browser automation framework used to crawl web portals and download ordinance documents reliably.
pytesseract#: Python wrapper for the Tesseract OCR engine used to enable text extraction from scanned ordinance documents.
rate limiter#: Token-based throttle that keeps LLM requests within provider quotas while maximizing throughput.
structured record#: Tabular representation of ordinance features, thresholds, and metadata exported for downstream analysis.
technology#: tech configuration key that defines the target infrastructure domain, such as solar or wind.
text splitter#: Utility that chunks ordinance text into overlapping segments sized for LLM context windows.
validation pipeline#: Post-processing stage that verifies extracted features, resolves conflicts, and confirms location metadata.
web search#: Search-and-crawl phase that discovers ordinance links using providers such as Tavily, DuckDuckGo Search, or custom engines.