ConservaNode Sourced public facts about wildlife and the laws that protect it.
Early prototype: Sourced public regulatory facts from U.S. Federal Register notices. Intended for release under CC0. See About & Methodology.

How to Read ConservaNode Data

ConservaNode provides structured factual relationships extracted from public Federal Register notices. This guide explains our evidence levels, notice coverage banners, proof pointers, and network traversal rules.

Last reviewed: 2026-10-16

1. Evidence Honesty: STATED vs. INFERRED

Rather than assigning speculative machine-learning confidence scores, ConservaNode categorizes relational facts into two strictly defined evidence levels established in ADR-0004:

● STATED (the notice says this directly)

Definition: The entities co-occur within the exact same sentence or within 150 characters in the primary source document text.

Meaning: The regulatory notice directly connects the two concepts in the text.

Visual: Rendered with a solid green line in relationship graphs.

◌ INFERRED (appears in the same notice; may not state a direct link)

Definition: The entities appear within the same notice context, but are separated by more than 150 characters.

Meaning: The concepts are contextually associated within the regulatory action, but the text may not state a direct grammatical link.

Visual: Rendered with a dashed brown line in relationship graphs.

In our seed dataset, approximately 70.3% of relationships are INFERRED document-level associations and 29.7% are STATED direct sentence links.

Official Source: ConservaNode Architecture & Methodology Specification ↗

2. Notice Coverage Disclosures

Not all documents in our seed release were processed to the same depth:

  • Full Text (First ~3,500 characters): 16 notices (24.2%) were downloaded as complete Federal Register notices and analyzed up to the first 3,500 characters of normalized body text.
  • Title & Abstract Only: 50 notices (75.8%) were acquired from Federal Register search listings, meaning only the official title and publication summary were analyzed.

Every notice page features a prominent coverage banner indicating exactly what text basis was read. Crucial rule: Absence of an entity or edge in ConservaNode is not evidence that the concept is absent from the full underlying notice.

Official Source: ConservaNode Federal Register Notices Catalog ↗

3. Proof Pointers Instead of Quotes

To respect copyright boundaries while enabling complete scientific verification, ConservaNode does not republish full texts or long quotes. Instead, every fact is anchored by a proof pointer:

  • Document SHA-256: The cryptographic digest of the normalized source text.
  • Character Offsets: 0-indexed start and end character positions (e.g. [799..829]) pointing to the exact span in the primary document.
  • Direct Source URL: A clickable link to the official Federal Register record on federalregister.gov ↗.

Official Source: National Archives Federal Register Portal ↗

4. Anti-Hairball Traversal Rules

In large knowledge networks, highly connected hub nodes (such as the country "United States" or the "Endangered Species Act") create dense, uninformative hairballs where everything connects to everything. ConservaNode enforces strict traversal criteria (ADR-0004):

  • Species are connected to other species only through shared geographic range places (RANGE_IN).
  • High-degree places (connected to >15 species) and country-level entities are excluded from species bridging.
  • Statutes and federal agencies are terminal descriptive attributes, never intermediate traversal stepping stones.

Official Source: ConservaNode Graph Quality & Hub Policy (ADR-0004) ↗