Skip to main content
This page is the entry point for WorldMonitor’s China data. It describes the launched and explicitly blocked source contracts; it is not a claim that every upstream source is healthy at this moment. Current transport and content health remain visible through the China projection in /api/health. /api/seed-health reports whether the China coverage evaluator itself ran recently; a fresh evaluator heartbeat does not prove healthy China content.

​
Coverage inventory

​
Health projection registry

The China health projection tracks these stable contracts. launched means the lane is admitted to the China projection; it does not override the lane’s live freshness or content status. JODI remains available to existing global energy products. Its China oil and gas contracts stay blocked because the reviewed upstream country index does not currently publish substantive CN rows. A healthy global JODI heartbeat therefore cannot be presented as China coverage. Detailed method and source contracts:

​
Status and missingness

Source arrival is not the same as usable content. China health evaluates transport freshness and content freshness separately. A successful request can still publish stale, partial, empty, or timestamp-unknown content. The public contracts use explicit states:
  • available or healthy means the required payload passed its domain validation and freshness rules;
  • partial or degraded means some reviewed coverage is missing, stale, or invalid while other evidence remains usable;
  • stale means the last valid observation is outside its content budget;
  • unavailable means no usable observation can be published; and
  • blocked means a source was deliberately not launched, normally because robots, terms, transport, or source-substance review did not pass.
Missing categories are never converted to zero, normal, unchanged, or current. Last-good preservation keeps its original observation time and reports the current transport failure.

​
Stock Connect turnover and margin

Four official exchange endpoints compose market:china:stock-connect:v1. They sit on the two exchange hosts already covered by the corporate-disclosure terms review, and all four are admitted as admitted_aggregate_statistics: public aggregate market statistics only, with no per-investor and no per-order data. The reviewed terms records are the same ones the disclosure lane uses: SSE official legal terms, robots.txt not published (404); SZSE applicable rules and notices, robots.txt present but empty (200). Northbound net flow is not available, because this source does not publish it. Both exchanges stopped disclosing the northbound buy and sell split on 2024-08-16, so only gross turnover survives. The payload states that directly: northbound.netFlow is { status: 'unavailable', reason: 'EXCHANGE_STOPPED_PUBLISHING_BUY_SELL_SPLIT' }, alongside netFlowDiscontinuedOn: '2024-08-16'. Turnover is two-way trading activity, buys and sells counted together. It is not capital entering or leaving the mainland market and must never be read as a flow. The rest of the contract:
  • every published value is normalized to CNY. SSE quotes margin in yuan; every other figure arrives in 亿 (1e8) or 万 (1e4) and is converted;
  • a combined SSE and SZSE figure is published only when both exchanges report the same trade date. Otherwise the combined value degrades with reason TRADE_DATE_MISMATCH, which doubles as the freeze detector: one exchange that stops advancing is visible immediately instead of after a staleness budget expires;
  • margin publishes on a T+1 lag, and northbound publishes after each session closes;
  • if one margin source has a transport failure, the adapter retains the complete last-good SSE/SZSE margin pair for less than three hours after verification. margin.retained identifies this fallback. margin.verifiedAt and the common trade date stay unchanged across repeated failures. A current SSE observation is never combined with an older SZSE observation. Verification checks each exchange’s total against financing plus securities lending, allowing only SZSE’s independent rounding to 0.01 亿 (CNY 1 million);
  • expired or incomplete prior pairs, malformed responses, conflicting trade dates, and simultaneous margin-source failures do not qualify for retention. Source errors and status: 'degraded' remain visible during retention. A successful complete refresh clears margin.retained and updates verifiedAt;
  • SZSE endpoints are date-keyed, so the seeder pins the date from the exchange trading calendar (https://www.szse.cn/api/report/exchange/onepersistenthour/monthList) and probes back a bounded number of trading days. Within each run, later sources and dates reuse the successful proxy exit. If that exit fails, the existing bounded fallback skips that already-failed exit for the same request. Request and time budgets are unchanged; and
  • the lane runs hourly as a member of the seed-bundle-market-backup Railway bundle.

​
Bilateral HS4 import baskets

The bilateral store covers one shared HS4 catalogue: the reviewed bilateralHs4Code entries in scripts/shared/comtrade-strategic-products.json plus the HS4 headings of every commodity in scripts/shared/supply-vulnerability-commodities.json. The scheduled seeder and the on-demand public fallback request the same catalogue, in two requests of at most 20 headings each; the fallback can also request one heading on its own, described below. Each country shard contains annual imports by HS4 product and the leading exporter partners. It supports country exposure, route impact, multi-sector shock, and the PRO get-country-products route. Detailed bilateral rows are not part of the anonymous China summary. The authenticated Comtrade route sends one four-year period window, Y-2 through Y-5, for each of the two HS4 batches. The normalizer keeps the newest year per product-and-partner pair, so a response cannot silently mix annual values. When COMTRADE_API_KEYS is absent, the public preview route cannot accept a period list and instead tries Y-2 followed by Y-3. The scheduled production contract is deliberately quota-bound: A run below the 110-shard floor writes partial seed metadata rather than claiming full success. Failed reporters preserve an existing shard for at most two consecutive runs; after that, the shard may expire so the on-demand public fallback can probe it again. Rate limits, quota exhaustion, and request-budget exhaustion publish partial coverage and stop the run instead of overwriting last-good country data with empty payloads.

​
Aggregate-only rows

Every bilateral request, scheduled or on demand, pins the second partner, the mode of transport and the customs procedure to their aggregate values (partner2Code=0, motCode=0, customsCode=C00). Without them Comtrade returns one row per partner, second partner, transport mode and customs procedure — roughly nine rows per partner. That multiplicity, not the heading count, is what filled the public preview route’s 500-row cap: a single German heading reached the cap on its own, so a whole-catalogue fallback ended incomplete for exactly the large importers users ask about. With the filters the same request returns one row per partner plus the reported World total. Grouping already kept only the aggregate row per partner, so stored values are unchanged; what falls is the row count, and the World total is now always present.

​
Stored keys

The canonical country shard comtrade:bilateral-hs4:{ISO2}:v1 keeps the shape it has always had: per heading, the leading five origins with value and share, plus the denominator basis. Every derived scorer — import concentration, route exposure, chokepoint indices — reads that list and sums over all of it, so a longer list would move published scores with no method change to explain the movement. Two bulk seeders also read every country shard through a pipeline batched by key count rather than by bytes, so anything added to the canonical payload arrives in one response body that nothing bounds. Deeper evidence therefore cannot ride along on it, and lives in a sibling key instead. comtrade:bilateral-hs4-partners:{ISO2}:v1 holds, per heading, every origin whose unrounded share of the denominator reaches 1%, padded up to a minimum of five origins so a heading with one dominant supplier still carries context, and capped at 25 whatever their share. The origins the rule leaves out are published with the list as a count and a combined share, so a reader can state how much of the trade it is not showing rather than implying the list is complete. Each origin carries net weight in kilograms and, when the provider reports one, a quantity with its unit from the pinned registry scripts/shared/comtrade-quantity-units.json. Comtrade reports an unknown weight as 0, so a 0 is stored as absent rather than as a confirmed zero. The sibling key shares the country shard’s 40-day TTL and the same preservation and TTL-extension rules. Only get-country-products reads it. comtrade:world-exports-hs4:v1 is one run-level snapshot of every reporter’s exports of every reviewed heading: per heading, the observation year, the reporters that filed that year ranked by exported value, each with value in USD and net weight when reported, and unrankedReporterCount, the reporters whose newest filing for the heading is older. Those late filers or lapsed exporters are left out of the ranking, so a rank is a rank among that year’s filers and can change when they file: the brief prints it as “rank N of M reporters filing YEAR” and states the unranked count. The run reserves two requests for it against the request budget before it attempts the first reporter — one per catalogue batch, with the reporter omitted and the partner pinned to World, so a single request answers for every reporter at once. Reserving first keeps the per-reporter budget pre-check honest and stops a budget abort from dropping the snapshot altogether. It shares the 40-day TTL. Its outcome is recorded in seed metadata and published through /api/seed-health; any state other than observed is a coverage gap, because a supplier’s absolute scale and its world rank come from this key alone. Only get-country-products reads it, and only for a requested heading. A scale is attached to an origin only when the snapshot’s observation year for that heading equals the served row’s year, so a late filer’s older row carries no scale rather than a rank from a year it does not describe.

​
Single-heading recovery

When a PRO reader asks for a heading a stored shard does not carry, the on-demand public fallback requests that one heading rather than the whole catalogue. That is what makes recovery work for a large importer, whose whole-catalogue fetch fills the preview cap. The outcome is recorded in a per-heading sentinel, comtrade:bilateral-hs4-lazy-heading:{ISO2}:{HS4}:v1, so one heading’s failure cannot suppress recovery of a different heading for the same country. Single-heading recovery never writes the canonical shard: one heading is not a country basket, and a recovered heading is returned with its own fetch time rather than the shard’s. A recovered heading whose observation year is older than the stored one is rejected. Shards written before this evidence existed have no sibling key and no world exports. They keep serving their leading-five origins, and the surfaces that read them label the missing volume and scale as not available. No reseed is forced to obtain the deeper evidence; the next scheduled run writes it.

​
Access boundaries

  • Country summary, corridor conditions, decision-signal summaries, provenance, and aggregate health are available through bounded public surfaces.
  • Detailed corridor control towers and activity-nowcast evidence are REST/UI surfaces. MCP does not expose dedicated tools for either route; Pro MCP receives their bounded summaries through get_china_decision_signals.
  • Pro MCP reads the canonical macro and release-calendar caches through get_economic_data with dataset china-macro or china-release-calendar.
  • Detailed bilateral HS4 product rows and route exposure are PRO-only.
  • The Pro MCP decision-signal tool receives the same bounded items and provenance as the public composition, not raw exchange documents or unrestricted source expansion.
  • Proxy credentials, raw per-attempt request diagnostics, cache names, and detailed freshness thresholds remain operator-only. Public documentation names the configuration keys and selection semantics, not their values.
No China surface should be interpreted as independent verification merely because its publisher is official. Publisher identity, extraction confidence, classification confidence, corroboration, and freshness remain separate claims.