# AI 前沿情报站

> Source-backed research on AI models, agents, compute, capital and governance.
> Distinguish report assertions, site judgments, market probabilities and review states.
> This graph is a research preview. Read quality flags before relying on a relationship.

## Graph and provenance
- [Graph explorer](https://ai.mimiceco.com/graph)
- [Method and current quality](https://ai.mimiceco.com/graph/method)
- [JSON graph](https://ai.mimiceco.com/graph/kb.json)
- [JSON-LD graph](https://ai.mimiceco.com/graph/kb.jsonld)
- [RDF Turtle](https://ai.mimiceco.com/graph/kb.ttl)
- [Query API](https://ai.mimiceco.com/api/graph?limit=20)

## Query contract
GET /api/graph accepts q (name/alias), type (ontology class), topic
(model, agent, compute, capital, governance, markets), entity (e.g. org:anthropic),
review (pending, auto_accepted or ai_reviewed), from (YYYY-MM-DD), limit (1–200), offset.
Responses include totals, quality, report coverage, source-backed relations and entities.
Identifiers /id/{kind}/{slug} return entity JSON-LD. Unknown entities return 404.

## Interpretation
- pending is unresolved; auto_accepted is Jev source-fidelity review, not independent truth.
- ai_reviewed is an independent Codex source review with literal source witnesses and hash checks; retain the original Jev score and disagreements. It is not human verification.
- Evidence URLs come from the original report. Report snapshots may differ from later revisions.
- proposed, negotiating, announced, signed and completed are distinct transaction states.
- Observation actual, estimate, plan, target and guidance are distinct; preserve scope and units.
- Site claims describe the site's historical judgment; third-party odds are not site predictions.
- AI source audits and human sample checks are counted separately. Do not infer completion.
- Sanlian archive fixtures are excluded from this production graph.