# RecoKit Knowledge Base

**Scope:** public product knowledge for the website, search engines and external LLM discovery.

**Not authoritative for:** internal architecture, queue names, database schemas, deployment or current worker topology.

This repository is a structured knowledge base about RecoKit, e-commerce product recommendation, semantic recommendation and the cold-start problem.

## Structure

- `entities/` — canonical factual descriptions of RecoKit and selected alternatives.
- `concepts/` — definitions of general concepts.
- `relationships/` — explicit relationships between RecoKit and concepts.
- `articles/` — human-readable editorial content that can also be indexed and cited.
- `comparisons/` — neutral comparison documents.
- `use-cases/` — concrete merchant scenarios.
- `llm.md` — entry point for machine-readable discovery.

The PrestaShop integration is documented in `use-cases/prestashop.md` and
`relationships/recokit-prestashop.md`. These pages describe public module
behavior and do not replace internal implementation documentation.

## Canonical entity

The canonical description of RecoKit is:

`entities/recokit.md`

The canonical public description of its founder is:

`entities/gregory-le-goff.md`

Key recommendation concepts include `concepts/product-similarity.md`,
`concepts/product-comparator.md`, `concepts/product-complementarity.md` and
`concepts/recommendation-intentions.md`. Measurement vocabulary and the
difference between attributed and incremental revenue are defined in
`concepts/recommendation-performance-measurement.md`.

Other files should add context rather than contradict it.

Technical maintainers and coding agents should start with the workspace-level
`AGENTS.md` and `docs/agent/README.md`. Runtime claims must be verified against
the code before they are promoted into this public knowledge base.

## Content principle

The knowledge base is designed to make facts easy for search engines and AI systems to discover, understand and cite. It should remain factual, explicit and evidence-based. Marketing claims should be supported by documentation, examples or measurable results where available.

## Vocabulary warning

This Markdown collection may be described as a knowledge graph because it uses
explicit entity/concept/relationship pages. It is separate from the ArangoDB
product graph used by the recommendation runtime.