Skip to content

Public REST · release-bound MCP 1.9.0 · no key

A China economy API that can still say “not enough evidence.”

Palimpsest exposes source-bound observations and evidence state, not a magic macro score. The public files and six public MCP tools retain provenance, knowledge clocks, revisions, source coverage and explicit abstentions.

Choose the smallest sufficient surface

REST for exact artifacts; MCP for bounded questions

Static REST

Archive and reproduce

Fetch immutable-format JSON or JSONL, validate it against published schemas, and retain exact bytes for your research record.

MCP 1.9.0

Discover and filter

Let an agent list readings, fetch one bounded signal, inspect evidence desks or query the fixed economic ledger without accepting caller-supplied source URLs.

REST quickstart

Start from the machine index

curl -sS https://palimpsest.info/readings/china-index-latest.json \
  | python3 -m json.tool

The index joins current desks, source states, release monitors, evidence distributions, forecast status and readiness gates without converting them into a true-GDP estimate or causal story.

ArtifactUse it forBoundary
China indexCoverage, source states, releases, current desk metrics and readinessNo broad composite while gates fail
Observation manifestExact JSONL digest, record count, series and clock coverageCurrent rows are the named CFETS family, not the entire registry
Observation ledgerPoint-in-time aggregate rows with revisions and provenancePublisher rights remain attached
Stock ConnectOfficial aggregate cross-border turnover and southbound flowNo northbound net estimate; no general capital-market coverage
Forecast statusPseudo-real-time backtest gates and abstentionsRead the artifact; the 24 August snapshot had zero promoted targets

MCP quickstart

Query one series as known at one time

curl -sS https://api.seiche.info/palimpsest/mcp \
  -H 'Content-Type: application/json' \
  -d '{
    "jsonrpc":"2.0",
    "id":1,
    "method":"tools/call",
    "params":{
      "name":"query_economic_observations",
      "arguments":{
        "series_id":"cn.cfets.shibor_on",
        "as_of":"2026-08-21T23:59:59Z",
        "revision_view":"latest-as-of",
        "limit":20
      }
    }
  }'

The tool verifies the fixed manifest’s exact JSONL bytes, digest and record count before parsing. The as_of cutoff applies to both released_at and collected_at, preventing a later-known vintage from leaking into an earlier question.

Checksum verification establishes consistency with the manifest, not publisher identity. Preserve the row’s evidence URL and source ID.

Interpretation contract

Keep scope, clocks and abstention in the answer

Economic period

What day or interval the observation describes. This is not the collection time.

Release clock

When a vintage became available. If an exact timestamp is absent, metadata states the conservative substitution.

Collection clock

When Palimpsest knew the vintage. Point-in-time analysis must respect this clock too.

Revision view

latest-as-of selects the newest vintage knowable by the cutoff, while all retains every visible revision.

Coverage state

Live, observed, adapter-ready, planned, blocked and out-of-scope are not interchangeable.

Forecast state

warming_up is an abstention. It is not a weak directional signal.

Rights and safe reuse

Public access does not erase source terms

Palimpsest software, schemas and original metadata are MIT-licensed. Source observations retain publisher rights and attribution requirements. Some registered sources are manual, licensed, blocked or out of scope; the API does not imply they have been collected or can be redistributed. Never recruit respondents in China, bypass access controls, or infer restricted microdata from public aggregates.

Read the dataset-by-dataset rights ledger · Inspect source implementation states

Fleet router

One agent entry point, four evidence roles

PalimpsestChina source evidence, revisions, economic-period/release/collection clocks, censorship measurement and explicit evidence gaps.
SeicheSystem funding stress and money/FX/macro-capital transmission. Do not import Palimpsest observations into a Seiche composite without its validation gate.
LiquiLensInstitution-level bank and NBFC risk, supervisory evidence and construction-PIT eligibility boundaries.
UndertowExit liquidity: market depth, slippage, venue concentration and unwind conditions.

Install or inspect the canonical Financial Evidence Agent Skill.

FAQ

Operational questions

Is the Palimpsest China economy API free and keyless?

Yes. The published REST artifacts and the six release-bound MCP 1.9.0 tools are public and read-only, with no API token required.

What is the difference between the REST API and MCP?

REST exposes stable published files and schemas. MCP adds bounded tool discovery and point-in-time filtering over the fixed public ledger. Both return evidence; neither supplies a validated economy-wide forecast.

Can an agent treat missing or stale China data as zero?

No. Missing, stale, blocked, planned and not-collected states are distinct. Source silence is not economic contraction or calm. At the 24 August 2026 snapshot the broad composite was warming_up; re-read the current index before citing status.

Which product should handle a broader money or capital-market question?

Use Palimpsest for source-bound China evidence; Seiche for system funding and money or macro-capital transmission; LiquiLens for institution risk; and Undertow for market-exit liquidity.