AquariumHQ
Search

Methodology

AquariumHQ is a knowledge graph with a website attached, not a website with a database behind it. This page explains the parts of that which affect what you read.

Identity is opaque and immutable

Aquarium subject matter punishes name-derived identity harder than most domains. Scientific names are revised, trade names collide, common names are regional, and domestic strains are marketed under names that are not taxonomic at all.

So every record carries an opaque token allocated once and never changed. Slugs, scientific names, common names and classification are all mutable attributes hanging off that identity — which means a taxonomic revision changes a label rather than destroying every relationship pointing at the record.

Relationships are governed

32 declared edge types, each stating which entity kinds it may connect and whether it makes a factual claim.

The factual flag is the one with teeth. An edge asserting that a species occurs in a river must carry provenance. An edge saying “this page is related to that page” is navigation and must not pretend to be evidence. An audit walks every edge and fails the build on any factual edge without support.

Edge typeConnectsFactual
Belongs toorganism, plant → taxonYes
Native toorganism, plant → jurisdictionYes
Natural habitatorganism, plant → waterConceptYes
Temperatureorganism, plant → waterConceptYes
pHorganism, plant → waterConceptYes
Hardnessorganism, plant → waterConceptYes
Water floworganism, plant → waterConceptYes
Substrateorganism, plant → equipmentConceptYes
Lightplant → equipmentConceptYes
CO₂plant → waterConceptYes
Swimming zoneorganism → husbandryConceptYes
Social requirementorganism → husbandryConceptYes
Dietorganism → husbandryConceptYes
Eatsorganism → organism, plant, husbandryConceptYes
Preys onorganism → organismYes
Compatibilityorganism → organismYes
Damagesorganism → plantYes
Grazes onorganism → plant, waterConceptYes

Compatibility is derived, not stored

A single boolean per pair is what every compatibility chart in the hobby produces, and it is wrong in both directions at once — it says no to pairings that work in a large tank with the right group sizes, and yes to pairings that fail when a fish reaches adult size or starts spawning.

AquariumHQ stores traits and derives verdicts. The engine evaluates 21 dimensions, runs predation and harassment checks in both directions, and returns the reasoning alongside the verdict. Confidence is a separate axis from the conclusion, and “insufficient evidence” is a legitimate output.

  • temperature
  • ph
  • hardness
  • predation
  • size disparity
  • aggression
  • territory
  • fin nipping
  • social requirement
  • zone competition
  • food competition
  • activity mismatch
  • flow
  • substrate
  • diel conflict
  • breeding aggression
  • invertebrate risk
  • plant impact
  • bioload
  • tank size
  • domain

The opportunity registry decides which pages exist

This is the part that is load-bearing rather than descriptive.

Every candidate intent is scored from the underlying data on nine axes and passed through a gate with fixed thresholds. Compatibility route generation then reads the ACCEPTED opportunities and builds exactly those pages — a pair the gate rejected has no URL, cannot be linked to and cannot appear in the sitemap.

An opportunity and a page are deliberately different things. “How big does a neon tetra get” is a real question and a terrible page; it is accepted as an opportunity and delivered as an anchored section of the species page. That separation is what stops a registry of intents turning into an instruction to generate pages.

GateThreshold
min distinct intent3
min unique information3
min usefulness3
max doorway risk2
max duplication risk2
min data availability3
min evidence availability2

The results, including every rejection reason, are on the coverage page.

What AquariumHQ deliberately does not do

  • Reproduce, mirror or reconstruct another aquarium database, including its compatibility matrix.
  • Publish product recommendations, prices or affiliate links.
  • Publish a stocking capacity figure.
  • Fabricate a citation for a value that has no single citable source.
  • Invent search volume figures.
  • Publish health content without veterinary-grade sourcing.
  • Generate a page for every possible combination of two records.