docs/china-official-macro-policy.mdx

Rama main · réplica de lectura del repositorio original.

---
title: "China Official Macro and Policy"
description: "Official China macro observations, release-calendar verification, policy-event taxonomy, source admission, revisions, and missingness."
---

WorldMonitor maintains two independent official-document lanes: revision-aware
numeric macro observations and typed policy/enforcement events. They share the
[decision-signal provenance contract](/decision-signal-provenance), but they
do not share fetchers, scoring, or freshness rules.

## Official macro observations

The launched macro snapshot requires five series:

| Series ID | Publisher | Pillar | Unit | Maximum observation age |
| --- | --- | --- | --- | --- |
| `nbs_industrial_value_added_yoy` | National Bureau of Statistics of China | Activity | Percent | 75 days |
| `nbs_fixed_asset_investment_yoy` | National Bureau of Statistics of China | Investment/property | Percent | 75 days |
| `nbs_real_estate_investment_yoy` | National Bureau of Statistics of China | Investment/property | Percent | 75 days |
| `safe_fx_reserves` | State Administration of Foreign Exchange | External pressure | USD 100 million | 45 days |
| `safe_bank_fx_settlement` | State Administration of Foreign Exchange | External pressure | CNY 100 million | 60 days |

NBS is collected from `www.stats.gov.cn/english/PressRelease/`; SAFE is
collected from `www.safe.gov.cn/safe/`. Both sources have bounded request
budgets, manual redirect handling, response-size limits, robots preflight,
source-host allowlists, and reviewed attribution requirements. The macro
seeder does not use a proxy to bypass a rejected source decision.

The following reviewed candidates remain explicit unavailable observations
instead of being silently omitted:

- `pboc_aggregate_financing_flow`
- `pboc_new_rmb_loans`
- `pboc_policy_liquidity_operation`
- `gacc_exports`
- `gacc_imports`
- `gacc_trade_balance`

PBoC candidates remain blocked when `www.pbc.gov.cn` robots policy disallows
the reviewed paths or terms review is incomplete. GACC candidates remain
blocked when `english.customs.gov.cn` cannot pass the TLS/source preflight.
Their missing observations remain `unavailable`.

### M2 through an official republication

`pboc_m2_yoy` has one additional reviewed source binding at the
[Shanghai Financial Portal](https://jrj.sh.gov.cn/SCGK194/index.html). The
portal credits the People's Bank of China. The observation and its provenance
identify Shanghai as the actual republisher and retain the PBOC credit.
This binding does not open direct PBOC access or supply other PBOC series.

The [August report](https://jrj.sh.gov.cn/SCGK194/20260915/8b8f76bb432e4c4b8aa290646d19231b.html)
states August 2026 M2 growth of 7.5%, with a Shanghai publication date of
15 September 2026. This is a day-precision republication date, not an original
PBOC release timestamp or an NBS database publication date. The subsequent
quarter report supplies September monthly M2 growth of 7.2% and is dated
15 October 2026. The adapter discovers the newest national report on the
reviewed listing, including half-year and quarter titles, and extracts only
the M2 month-end observation. It does not confuse the balance, M1 or M0 with
the M2 growth rate.

The [HTTPS robots endpoint](https://jrj.sh.gov.cn/robots.txt) returned a normal
404 on 24 October 2026. The existing runtime records this as
`no_rules_published`. Each run checks robots before the listing and article,
uses verified HTTPS with manual redirects, and has a six-request limit and a
15-second total fetch deadline for this optional source.
The [portal's copyright statement](https://jrj.sh.gov.cn/BQSM/index.html),
section 2, requires users of republications to retain the credited source.
Ingestion is limited to numeric facts and publication metadata with source
credit and a link to the article. Article prose and third-party media are not
redistributed.

M2 uses a 50-day observation-age limit. The
[July report](https://jrj.sh.gov.cn/SCGK194/20260817/2918b201c7c94119b337ab099c4363ae.html)
was republished on 17 August, 48 days after the preceding June month-end.
The limit includes the macro worker's next 36-hour interval. A newer retrieval
does not make an old observation current. Source failures preserve the last successful publication
and retrieval clocks and expose the failed attempt separately. M2 remains
optional for the five-series launch gate. Existing stored v2 payloads with a
null PBOC M2 row remain valid; the Shanghai binding requires its own matching
publisher preflight. These code contracts do not establish deployment or
production acceptance.

## Revisions and direction

Each official observation retains its observation period, publication time,
retrieval time, source URL, vintage ID, revision sequence, and supersession
state. A correction creates or updates a traceable vintage; it does not rewrite
the historical point-in-time record.

Direction is calculated only when the published series contract supplies a
valid comparison. Missing comparison values remain unavailable. The snapshot
is launch-ready only when all five required observations are valid and within
their individual age budgets. Transport freshness and snapshot completeness
are written as separate seed metadata. Only a published snapshot records a
completed run, so a rejected snapshot is retried on the next macro bundle tick
(08:00 and 09:00 UTC daily) instead of waiting out the 36-hour interval. The
transport metadata carries no content age until the canonical snapshot is
published. The post-publication hook records the published content age before
the completion marker. Validation or publication failures therefore leave
`chinaMacro` health at `STALE_CONTENT` while preserving the last-good snapshot.

`GET /api/economic/v1/get-china-macro-snapshot` reads the canonical
`economic:china:macro:v2` seed and never fans out to upstream publishers on a
cache miss. The macro Railway member runs every 36 hours.

## Release calendar

The release calendar reads the current NBS calendar from `www.stats.gov.cn`.
Loan Prime Rate dates are generated from the reviewed Chinese business-day
calendar and marked `provisional`, then promoted to `verified` only when the
corresponding ChinaMoney/CFETS notice is found on `www.chinamoney.com.cn`.

Failure of the required NBS calendar fails the calendar build. Failure of
ChinaMoney verification leaves LPR dates provisional and records the source
decision; it does not relabel them as verified.

## Policy and enforcement events

The policy lane polls six official agencies:

| Agency | Publisher | Listing host |
| --- | --- | --- |
| `CAC` | Cyberspace Administration of China | `www.cac.gov.cn` |
| `SAMR` | State Administration for Market Regulation | `www.samr.gov.cn` |
| `MIIT` | Ministry of Industry and Information Technology | `www.miit.gov.cn` |
| `MOFCOM` | Ministry of Commerce | `www.mofcom.gov.cn` |
| `NDRC` | National Development and Reform Commission | `zfxxgk.ndrc.gov.cn` |
| `PBOC` | People's Bank of China | `www.pbc.gov.cn` |

Each agency is limited to one listing request and at most three reviewed
document fetches per run, with a 400 ms same-host cadence. Listing responses
are capped at 750 KB and documents at 2 MB. URLs must match the reviewed host
and path contract.

The deterministic taxonomy emits:

- `consultation_draft`
- `announcement_guidance`
- `final_rule`
- `administrative_approval_restriction`
- `investigation`
- `enforcement_penalty`
- `effective_date_change`
- `amendment_correction_repeal_supersession`
- `unknown`

The published event keeps the original Chinese title and text, translation
state, document number, publication and effective dates, reviewed sector/entity
links, content hash, classification evidence, lineage, revisions, and full
provenance. Unknown classification is retained as unknown; it is not promoted
to enforcement.

The Railway macro bundle refreshes this lane every 6 hours. It retains at most
120 events across complete lineages for 180 days. A failed agency preserves
still-valid last-good events with transport freshness changed to `error`;
healthy agencies continue independently.

## Alert boundary

Consultation drafts, guidance, and unknown policy classifications do not alert.
Reviewed policy/enforcement alerts require a new stable lineage. Corrections
and repeat ingestion retain the same lineage for deduplication. The bounded
China decision-signal composition publishes at most four valid items from this
lane.