Disclosed versus inferred: two kinds of China supply-chain and ownership data

Short answer: disclosed data is what a filer published about itself, and it arrives already consolidated at the filer level. Inferred data is constructed by attributing activity across entities, and it is built to be aggregated. The two obey opposite rules. Roll up a disclosed dataset and you double-count. Fail to roll up an inferred one and you see a fraction of the real footprint.

The rule reverses between the two

Disclosed, so do not roll up. ChinaScope C2C Disclosed carries the consolidated disclosures of the listed filer. Aggregating a parent's rows together with its listed subsidiaries' rows double-counts the same underlying relationships, because the parent's consolidated figures already contain them. Treat each filer's disclosure as complete for that filer. (Data cut 2026-05-16.)

Inferred, so do roll up. ChinaScope RFP-Bids is the opposite case. Unlike disclosed filings, it is built to be aggregated: a listed company's true tender footprint is the sum of its own rows and those of its owned entities, much of that activity done by unlisted subsidiaries. Each link is tagged with the ownership relationship (relation_type) and the combined stake (hold_ratio), so you can weight by stake or keep only holdings at or above 50%. (Data cut 2026-05-29.)

Same analyst, same question, two datasets, two opposite correct answers.

What a disclosed row can be, besides a counterparty

Disclosed data carries the filer's own formatting choices, and three of them are routinely mistaken for dirty data.

Anonymised counterparties are real rows. Placeholders such as 客户一 ("Customer One") are how filers withhold a name. The amount and share are still disclosed. Keep them or drop them deliberately, but know that dropping them removes real disclosed value. (ChinaScope C2C Disclosed, data cut 2026-05-16.)

Subtotals sit in the counterparty-name field. Values like 合计 ("total") or 前五名合计 ("top-five total") are the filer's own aggregate lines, not individual counterparties. Filter them out for counterparty-level work, or your top-five analysis counts the top five twice. (Data cut 2026-05-16.)

An aggregate disclosure is a legitimate disclosure. "Top five customers" is not dirty data. It simply is not one counterparty. Decide explicitly whether your analysis includes those rows. (ChinaScope C2C PIT, data cut 2026-08-22.)

Two different answers to "who owns this company"

In ChinaScope Affiliate the parent and controller tables come from the ownership graph, which is control measured by equity structure. The top-10 tables come from the shareholder register, which is the disclosed holder list. A controller need not be a top-10 registered holder, and a top-10 registered holder need not control anything. Picking the wrong table does not produce an error, it produces a different and plausible answer. (Data cut 2026-05-16.)

The same shape appears in Relationships (PIT), where internal_guaranty_detail and internal_guaranty_total, along with their external and entrusted-investment counterparts, are disclosed separately by the filer. They are not a hierarchy to sum across. Adding detail lines to a total double-counts. (Data cut 2026-08-23.)

Inferred data carries its own construction artifacts

ChinaScope SAM's supply-chain graph delivers 66,053 distinct edges across 70,900 rows. 3,122 of those edges arrive on two or three rows, so 4,847 rows, 6.8% of the total, are repeats. Every row has its own id, which is the declared grain and is genuinely unique. But if you join or count on the product pair, dedupe on the triple (primary_code, related_code, relation_category) first or you will inflate by that margin. The repeats differ only in the English relationship_description, and the Chinese field cannot arbitrate because it is populated on about 0.13% of rows. This is a known property of the current V4 graph build. (Data cut 2026-02-26.)

Disclosure depth is a ceiling you cannot aggregate past

Companies disclose each line of business at a single chosen depth. A carmaker reports "passenger vehicles", not "sedans" and "SUVs" separately. In the ChinaScope Product Taxonomy, upper levels 1 to 3 are densely populated in the segment data and levels 5 and below are sparse. Roll up to a parent level for coverage. Aggregation reconstructs upper totals, but it cannot invent detail a company never disclosed. (As described in the July 2026 profile.)

The same ceiling shows up as missing money: about 27% of RFP-Bids awards carry a zero bid_price, and bid_amount on the graph sums only disclosed award values. Screen on positive amounts when you are sizing deals. (Data cut 2026-05-29.)