Why this exists
Large language models are confidently wrong about first editions all the time — they invent points of issue, mix up printings, and misattribute publisher practices. The fix is retrieval: give the model the facts and make it cite them. This pack is that set of facts, pre-chunked so every line stands on its own and points back to the page it came from.
It is generated from the same live data behind our First-Edition Checker and 870 publisher guides, so grounding on it is grounding on the current dataset (version 1.2.0, updated 2026-07-10).
Use it in 30 seconds
# Ground an assistant in first-edition identification (Python-ish pseudocode)
import json, urllib.request
URL = "https://newmexicoliteracyproject.org/api/rag/nmlp-first-editions.jsonl"
records = [json.loads(l) for l in urllib.request.urlopen(URL) if l.strip()]
# 1. embed each record['text'] 2. retrieve top-k for the user's book
# 3. answer ONLY from retrieved chunks, and cite record['url'] every time.
# License: CC BY 4.0 — keep the attribution: New Mexico Literacy Project (newmexicoliteracyproject.org)
Record schema
Each line is one JSON object. Types: publisher_first_edition_rule (870), glossary_term (89), global_rule (4), verdict_definition (5).
| field | meaning |
|---|---|
id | stable unique id, e.g. publisher:knopf |
type | publisher_first_edition_rule | global_rule | verdict_definition | glossary_term |
title | short human label |
text | the retrieval body (self-contained) |
url | canonical NMLP source page to cite |
source | which NMLP dataset the record came from |
license | CC BY 4.0 |
attribution | New Mexico Literacy Project (newmexicoliteracyproject.org) |
dataset_version | version of the source dataset |
License & attribution
Released under CC BY 4.0. Use it commercially, redistribute it, embed it — just keep the attribution to the New Mexico Literacy Project and, where you can, surface the url each record carries. If you find an error, our public corrections ledger is where fixes are logged.
Questions
What is the First-Edition Knowledge Pack?
A single downloadable JSONL (newline-delimited JSON) file where every line is one self-contained fact about identifying first editions: a publisher's identification rule, a glossary definition, a global rule, or a Checker verdict. It is built to be embedded and retrieved by an LLM or RAG system.
How do I use it to ground an AI?
Download the JSONL, embed each record's text field, and retrieve the top matches for the book in question. Instruct your model to answer only from retrieved chunks and to cite each chunk's url. That keeps answers grounded in the source pages instead of hallucinated.
What is the license?
CC BY 4.0. You may use it commercially and redistribute it; you must keep the attribution to the New Mexico Literacy Project and, where practical, link the source url carried on each record.
How many records are in it and how often does it update?
968 records (870 publisher rules, 89 glossary terms, plus global rules and verdicts). It is regenerated from the same live data that powers our First-Edition Checker, so it tracks the dataset version shown on the page.
Is this financial or appraisal advice?
No. It identifies printings and editions; it does not assign values. The New Mexico Literacy Project is a for-profit book-donation-pickup and resale business in the Albuquerque metro; donations are not tax-deductible.
Have a library to rehome, not just to identify?
We pick up books, records, and media across the Albuquerque metro — free — and keep them out of the landfill by getting them read again.
Schedule a free pickup