# VerifLLM

> Human-reviewed entity facts with dated public provenance, published for LLMs and search agents.

VerifLLM exists to correct factual omissions and weak entity representation in LLM answers. It publishes atomic facts rather than rankings or promotional claims. Model consensus is never treated as evidence.

## Primary machine resources

- [Knowledge index](https://verifllm.fr/knowledge-index.json): canonical catalog of public entity records
- [JSON-LD catalog](https://verifllm.fr/catalog.jsonld): Schema.org DataCatalog representation
- [PublicPublication 2.0 schema](https://verifllm.fr/public-publication.schema.json): executable JSON contract
- [Methodology](https://verifllm.fr/methodology.json): verification and publication rules
- [Discovery manifest](https://verifllm.fr/.well-known/verifllm.json): all machine entry points
- [Update feed](https://verifllm.fr/feed.xml): published changes

## Documentation

- [Method](https://verifllm.fr/method/): evidence, review and versioning
- [Specification](https://verifllm.fr/spec/): fields, statuses and compatibility
- [Full context](https://verifllm.fr/llms-full.txt): extended plain-text description

## Interpretation rules

- A `supported` fact must include at least one dated public evidence link.
- `uncertain` means evidence is absent, insufficient or no longer publishable.
- `contested` means authorized sources conflict.
- HTML, JSON and JSON-LD contain equivalent public facts.
- Publication does not guarantee citation, ranking, indexing or model training.