Supply-Chain & Relationship Intelligence

What data it carries

Twenty two endpoints over the commercial relationship graph, plus the derived concentration, dependency and multi-hop datasets built on it.

• Eight relationship dimensions · competitor, customer, supplier, partner, investment, joint venture, licensing and distribution, queryable together or one at a time
• Per-edge metadata · trust tier, confidence, first and last observed dates, counterparty entity kind, and whether evidence is available
• Evidence · the filing, news or curated source behind a relationship, with excerpts and years: the answer to “why do you assert this?”
• Disclosed economic value · contract, deal and commitment values as a point-in-time observation series
• Supplier concentration & customer concentration · top-1/3/5 and HHI, computed over the value-covered subset and labelled as such
• Relationship risk · integrated counterparty exposure for a single company
• Comparison · the counterparties of several companies in one call: which are shared and which are unique
• Graph neighbourhood · the multi-hop network around a company as nodes and edges, not a flat list
• Connection paths · every disclosed route between two companies, so a dependency two or three hops away becomes visible
• Indirect exposure · exposure propagated across the graph, with an industry rollup and the coverage behind it

Competitive position, measured honestly

The same graph records who a company competes with, which turns relationship data into competitive analysis.

• Competitor set · who the filings actually name as competitors, rather than an industry code standing in for a peer group
• Market share · each company measured against the combined revenue of its competitors, with rank and coverage
• Position over time · how a company’s standing has moved across reporting periods
• Comparable dimensions · revenue, growth and margins side by side, so leadership is explained rather than asserted

How it behaves: stated, not implied

• Reported before inferred. Every edge carries a trust tier. Filing-grounded, curated and news-grounded edges are returned by default; model- and legacy-derived edges are opt-in, never mixed in silently.
• Both directions of disclosure. A relationship is found whether the company names its counterparty or the counterparty names the company: a supplier’s own filing naming this company as a customer counts, and the direction is reported per edge.
• Concentration is labelled by basis. Counts, disclosed values and revenue share are three separate bases with different coverage. They are returned as distinct, named blocks rather than blended into one headline number.
• Coverage is disclosed, not hidden. Where a basis is unavailable for a company it is returned as an explicit typed status, never as a zero or a silent omission.
• Stable identifiers. Each relationship carries a rel_uid that stays constant across graph rebuilds, so a saved reference keeps resolving.

Best real-life usage

• Concentration risk: find issuers whose disclosed revenue depends on a handful of named customers.
• Second-order exposure: trace who supplies the suppliers before judging a shock’s reach.
• Diligence with citations: pull the relationship, then pull the filing text that supports it into the memo.
• Competitor mapping: build a peer set from disclosed commercial reality rather than an industry code.

Combine with other datasets

Each analysis below pairs this dataset with another to answer a question neither can alone.

Over MCP with AI agents

Connect an AI agent to the CSIMarket MCP server (mcp.csimarket.com) and it can call these endpoints as tools, no glue code. Ask in plain language; the agent picks the endpoints, fetches the data, and composes the answer.

• “Map the major customers and suppliers connected to NVIDIA, and show me the evidence behind the largest relationships.”
• “Which companies disclose a dependency on a single customer for a large share of revenue?”