Glossary

Terms & Acronyms

A

AI Context: The text provided to an LLM (like Claude) before a user’s query. Includes location data, plant profiles, and relevant reference information.

API: Application Programming Interface – a way for software to communicate with external services.

C

Cache Hit Rate: Percentage of queries that retrieve pre-computed data from cache instead of computing fresh. Target: 70%+.

Climate Region: One of 8 U.S. gardening regions (e.g., Southeast, Pacific Coast) that group states by similar growing conditions.

CDO: Climate Data Online – NOAA’s service for accessing climate normals and historical weather data.

D

Differential Storage: Storing only what differs from defaults, not full duplicates. Example: Store “heat-tolerant varieties” for Southeast, use default timing for all regions.

DDL: Data Definition Language – SQL commands for creating tables (CREATE TABLE, ALTER TABLE, etc.).

E

EDDMapS: Early Detection & Distribution Mapping System – database of invasive species by state.

Embedding: Vector representation of text used for semantic search. Example: Plant profile → 1536-dimension vector.

F

Frost-Relative: Planting timing expressed relative to last frost date (e.g., “6 weeks before last frost”) rather than absolute dates.

H

Hierarchical Retrieval: Loading data from most-specific to least-specific (Zone → Region → State) to minimize tokens.

L

Lazy Loading: Loading data only when needed, not upfront. Example: Load pollinator details only if user asks “tell me more”.

M

Monolithic Document: Single large file containing all requirements (2,092 lines). Replaced with modular structure.

Modular Structure: Breaking large document into focused files (200-600 lines each) for token-efficient retrieval.

N

NCEI: National Centers for Environmental Information – NOAA’s data center.

NRCS: Natural Resources Conservation Service – USDA agency providing soil survey data.

P

PHZM: Plant Hardiness Zone Map – USDA’s map of climate zones for plant selection.

Pinecone: Vector database service for semantic search. Alternative: self-hosted solutions.

PRISM: Parameter-elevation Relationships on Independent Slopes Model – produces authoritative USDA zone data.

PWS: Personal Weather Station – home weather stations that report to Weather Underground.

R

RAG: Retrieval-Augmented Generation – AI technique that retrieves relevant chunks before generating responses. Reduces tokens by 90%.

Regional Override: State or region-specific data that overrides national defaults. Example: “Choose heat-tolerant varieties” for Southeast tomatoes.

S

Semantic Search: Finding similar content by meaning (vector similarity) not keywords. Example: “When to plant” finds “planting windows” and “frost dates”.

SSURGO: Soil Survey Geographic Database – detailed soil data by county/mapunit.

Sparse Storage: Storing only data that exists (not empty entries). Example: Regional notes for 10% of plants, not all plants.

T

Token: Unit of text for LLMs. Roughly 1 token = 4 characters. Example: “Zone 7b” = ~2 tokens.

Token Budget: Maximum tokens allowed in AI context. Target: 600 tokens input.

TTL: Time To Live – how long cached data stays valid before expiring.

U

USDA: United States Department of Agriculture.

USDA Zone: Hardiness zone (e.g., “7b”) based on average annual minimum temperature.

V

Vector DB: Database storing embeddings for semantic search. Example: Pinecone, Weaviate, self-hosted.

W

WU: Weather Underground – weather data service with 180K+ U.S. personal weather stations.

---

Common Phrases

“93% token reduction”: Optimization from 8,000 tokens (monolithic) to 300-900 tokens (modular) per query.

“Cache hit”: Successfully retrieving pre-computed data from cache instead of computing fresh.

“Climate region clustering”: Grouping 50 states into 8 regions to reduce data duplication.

“Decision Gate”: Milestone where project is validated before proceeding. Gates at weeks 4, 12, and 20.

“On-demand retrieval”: Loading data only when specific query needs it, not loading everything upfront.

“P0/P1/P2”: Priority levels. P0 = critical path, P1 = high value, P2 = enhancement.

“Query-specific context”: Building AI context tailored to specific query, not generic dump.

“Token-efficient”: Minimizing tokens used while maintaining quality. Core goal of this architecture.

---

Acronym Quick Reference

Acronym	Full Name	Category
AI	Artificial Intelligence	Technology
API	Application Programming Interface	Technology
CDO	Climate Data Online	Data Source
DDL	Data Definition Language	Database
EDDMapS	Early Detection & Distribution Mapping System	Data Source
LLM	Large Language Model	Technology
NCEI	National Centers for Environmental Information	Organization
NOAA	National Oceanic and Atmospheric Administration	Organization
NRCS	Natural Resources Conservation Service	Organization
PHZM	Plant Hardiness Zone Map	Data Source
PRISM	Parameter-elevation Relationships on Independent Slopes Model	Data Source
PWS	Personal Weather Station	Technology
RAG	Retrieval-Augmented Generation	Technology
SSURGO	Soil Survey Geographic Database	Data Source
TTL	Time To Live	Caching
UI/UX	User Interface / User Experience	Design
USDA	United States Department of Agriculture	Organization
WU	Weather Underground	Data Source

---

Last Updated: February 11, 2026
Purpose: Reference for developers working on 50-state expansion
Note: Add new terms as project evolves