For analysts who research China’s innovation landscape

# China’s defense innovation network, mapped before your agents ask.

You have tried buying the procurement data. It took weeks to search, it missed the military and high-tech work, and it was stale by the time it shipped.

We deliver the network itself: resolved, ranked and sourced.

Chinese Academy of Sciences sideNAURA (Beijing NAURA Microelectronics Equipment, a chip-equipment maker) sidetied to both200 of 25,805 entities drawn, 2,278 ties

What you may have already tried

### The flat-file vendor

Top-level procurement notices, little military or high-tech signal, and no way to join them to your own data. It sat unused.

### Building it in-house

Chinese-language analysts and scrapers take months to stand up, the best sources sit behind the firewall, and a crawl breaks when a site changes.

### The big platform

Opaque scores, thin coverage of Chinese procurement and marketplaces, and a seat licence for a screen your agents cannot use.

### Pointing an LLM at the web

It cannot reach behind the firewall, answers fluently with no record behind the claim, and starts from zero next session.

Your analysts now work through agents. Agents need precomputed context they can load, cite and act on. A flat file gives them none of it.

### Analyst weeks burned

Every question starts with weeks of searching and resolving by hand before anyone reads a result.

### Lost contracts

The government customer picks the team that shows the mapped network over the team that shows a spreadsheet.

The offer

## Networks for your agents.

A standing set of China networks, pulled from sources behind the firewall, resolved across Latin and Chinese names, scored once on the whole graph, and delivered in the format your agents read.

- **Reach.** We collect from sources behind the firewall: 13,346,769 procurement pages, read through 24 September 2026.
- **Resolve and precompute.** Name variants become one entity. PageRank, paths to the key institutions and ties by type are computed once, on the whole network.
- **Deliver.** The network ships as files your agents load, with an analyst who builds the target network with you.

Page counts read 24 September 2026.

The first thing your agent reads

## One entity, as delivered.

Institute of Microelectronics sits one hop from both the Chinese Academy of Sciences and NAURA, the Beijing chip-equipment maker. Your agent gets its identity, its rank in the whole network and its strongest tie of each kind, with the source behind each one.

- **1** Both names, so a search in either script lands on the same entity.
- **2** Every rank carries its denominator: 2 of 25,805.
- **3** The full tie count stays visible, 2,208, so a short list never reads as the whole.
- **4** Each tie names its source and its first and last date.

context/ent_b8f3cab26136.json

```
{ "id": "ent_b8f3cab26136", "name": "Institute of Microelectronics",1 "name_zh": "中国科学院微电子研究所", "roles": ["buyer", "paper_institution", "patent_applicant", "winning_bidder"], "sources": ["doclake_procurement", "epo", "openalex"], "hops": {"cas": 1, "naura": 1}, "pagerank": {"value": 0.01478, "rank": 2, "of": 25805},2 "ties": {"total": 2208, "by_relation": {"co_publication": 2075, "award": 99, "agent_for": 5, "candidate": 11, "co_patent": 18}},3 "strongest_of_each": [ {"with": "Chinese Academy of Sciences", "relation": "co_publication", "documents": 3346, "sources": ["openalex"], "first": "1988-11-01", "last": "2026-09-15"},4 {"with": "真芯（北京）半导体有限责任公司", "relation": "co_patent", "documents": 230, "sources": ["epo"], "first": "2022-01-07", "last": "2024-02-20"}, {"with": "东方国际招标有限责任公司", "relation": "agent_for", "documents": 346, "sources": ["doclake_procurement"], "first": "2017-04-21", "last": "2026-09-15"}, {"with": "苏州四方杰芯电子科技有限公司", "relation": "award", "documents": 7, "sources": ["doclake_procurement"], "first": "2024-11-15", "last": "2025-08-25"}, {"with": "深圳市优迪泰自动化科技有限公司", "relation": "candidate", "documents": 2, "sources": ["doclake_procurement"], "first": "2014-11-24", "last": "2014-11-24"} ], "release": "semiconductors_full-2026-09-21"}
```

Release semiconductors_full-2026-09-21. This release names the source and document count behind each tie; a link to each document ships with the evidentiary packages below.

Over time, your agents stop starting from zero. Each question begins at a mapped network, and the analyst time goes to the answer.

What we have built so far

### China’s semiconductor network

25,805 entities and 226,453 ties within two hops of the Chinese Academy of Sciences and NAURA, from research, procurement and patent records. Scored once on the whole union, drawn at 200.

[Open the network](https://blackgrove.ai/china-semiconductor-network)

How it arrives

## Two things, both working today.

Bulk files

JSON Lines (JSONL) or Parquet, keyed by stable entity ids, dropped where your pipeline reads them.

An analyst in the loop

A named analyst builds the target network with you and answers questions on the evidence.

Name one target. We send its network and evidence back in a week. Keep it either way.

Set that against what you pay now: a flat-file licence, plus the analyst weeks it takes to make the file usable.

What to tell your team

“Blackgrove sends the China network already resolved, ranked and sourced, in files our agents load directly. We start from the network instead of a spreadsheet.”

[Open the semiconductor network](https://blackgrove.ai/china-semiconductor-network)

or [name one target](https://blackgrove.ai/demo/new)
