外Waitan LudiCHINA POLICY · IMPLEMENTATION EVIDENCE
DIRECTION X Research edition
Search the full corpus
All research tools
Corpus overviewAsk the evidenceProgrammesFirm profilesDirective tracerWho may applyProvince coveragePolicy monitorResearch reportsOriginal source library

KNOW WHAT YOU ARE RELYING ON

Useful evidence.
Visible limits.

The workbench turns existing documentary research into an inspectable workflow. It does not turn incomplete coverage into certainty.

01 / SAMPLE

Twenty selected cases.

The project-chain pilot covers charging, medical devices, machine tools and automotive supply chains. Six cases have selected structured findings reviewed in the local snapshot dated 24 October 2026. The other cases are intake leads.

This is a purposive sample. It cannot support national totals, sector rankings, provincial completeness rates or causal effects. Counts of documents and evidence stages are not measures of policy effectiveness.

Inspect the broader coverage register ↗
02 / STAGES

Each step needs evidence.

A rule is not an award; an award is not a signed contract; a contract is not acceptance or payment. Reported execution is not independent bank confirmation. An original annex can be a tender attachment rather than a signed contract.

“Not established” means this review does not establish the stage. It does not mean the stage never occurred, a legal duty was breached or a project is late.

03 / RECEIPTS

Wording and truth are different checks.

Claim receipts retain the original wording, locator, source identifier, source hash and interpretation boundary. Literal matching checks whitespace-normalised saved text. Visual and structured-table transcriptions remain separately labelled.

A matching quotation does not establish semantic entailment, independent corroboration or current status. The review snapshot is not represented as independent human validation.

04 / ORIGINALS

Preserve the source version.

The local source library retains originals and extracted text. Original downloads check the file’s SHA-256 against the database. The case screen checks source metadata against the review snapshot separately.

A source may be missing, unreadable or changed. Availability does not establish authenticity, the truth of a source assertion or whether a newer external version exists.

Open source library ↗
05 / FINANCE

Never add incompatible amounts.

Procurement consideration, proposed grants, programme allocations, company cash receipts, income recognition and deferred balances have different meanings. A consortium award is not each member’s revenue. A product brand is not a resolved manufacturer.

The Heyuan and Dengfeng controls are recomputed using decimal arithmetic from the reviewed rows. Financial filing reconciliations are displayed as existing recorded checks; they do not establish a specific government award was paid.

Explore the financial lens ↗
06 / YOUR BRIEF

Keep judgment attributable.

Personal briefs retain analyst-authored notes separately from documentary findings. They are stored in this browser, not shared with colleagues. JSON exports preserve the evidence snapshot; Markdown exports preserve readable receipts and limitations.

“Check evidence changes” compares against the current local data. An unchanged fingerprint means only that the compared local content is unchanged. It does not check the web or establish no new event occurred.

THE NEXT VALIDATION GATE

Independent expert review remains open.

A defensible next step is a blinded task evaluation with Chinese-reading analysts: citation correctness, numerical interpretation, appropriate abstention, time to a useful brief and reproduction from originals. Automated tests and model-authored reviews do not substitute for that evaluation.

You can restart this tour at any time from “Take a tour”.