What this collection is.
Research Fund Flows is an independent public-record research collection about selected research awards and relevant organizational relationships. The first two editions cover three deliberately chosen cases: a documented historical subaward path, an affiliation control, and a technology-project status/identity test.
It is not a representative sample, a live grants database, a legal finding or a certification of every relationship. It cannot estimate worldwide transfers to any country. No black-box suspicion or exposure scores are assigned.
Keep the claim types separate.
Funding
An award record, subaward disclosure, audit table or equivalent source must support the financial relationship. Preserve the record’s own measure.
Ownership & control
Use a filing or equivalent organizational source. Record the direction, percentage/threshold and date. An ownership edge does not move a grant upstream.
Institutional affiliation
A partnership, lab affiliation or board seat is a nonfinancial relationship unless an award-specific financial bridge is independently established.
State, military & party
PRC-state institutional affiliation, PLA relationships and CCP governance are not interchangeable. Each needs its own dated evidence and award-relevance statement.
Names, nationalities, addresses, ethnicity, coauthorship and shared directors do not by themselves establish financial transfer, covert direction or misconduct.
Name the financial measure.
- Proposed/selected: an announced candidate can remain conditional. It is not automatically an executed award.
- Offered/committed: preserve that source’s contractual or budgetary meaning; neither label alone proves disbursement.
- Obligated: an agency commitment, distinct from cash paid.
- Reported subaward: onward-award disclosure; an action date is not automatically a bank-transfer date.
- Spend: retain the source’s spending field and coverage. It is not a comprehensive transaction audit.
- Withdrawn/cancelled: distinguish a database status from an executed cancellation instrument, reason and date.
Do not add alternative measures of the same award, pass the whole prime amount through every related organization, or mix GBP and USD. A blank, a literal zero and an unsuccessful search are three different observations.
Identity and time are part of the evidence.
Prefer stable award identifiers and recipient/company numbers. Retain the original name alongside any canonical match. Explain a human-reviewed institutional match when a legal identifier is unavailable.
Edition 01 preserves the long WIV source label and its institutional—not Chinese-registry-ID—match. Edition 02 keeps the original “Photoncis” spelling and joins company 04905488 to its official registry identity. Similar QRVCSELs titles do not merge projects 133181 and 103444.
A source’s document date, a relationship’s valid date, a notification date, and our retrieval date mean different things. Current ownership is not silently backdated to a historical award. URL migration folders do not override original document dates.
Let the evidence stop.
Public disclosures can omit lower tiers, vendors, indirect costs, nonpublic agreements and ultimate uses. Corporate reporting can stop before ultimate beneficial ownership. We do not bypass access controls or infer concealed activity from a failed request.
A nonfinding describes the specific source set and search. It is not “nothing exists,” “no money ever moved,” or a zero-dollar estimate. The Duke/Kunshan control and QRVCSELs status case show different reasons a proposed funding arrow must not be drawn.
Source and edition review.
Each edition includes precise source locators, retrieval timestamps, a statement of what the source supports, and a limit. Structured registers retain original-source hashes; API records include the request where needed. These support reproducibility, not a claim that a changing source will always hash identically.
Edition 01’s core source receipts were captured on 12 September 2026 at 15:17–15:19 UTC. Edition 02’s receipts were captured that day at 16:04–16:05 UTC; its workbook was published earlier, on 3 September. A later link check is not a later research cutoff.
The research and presentation receive source/claim review and local browser checks. Neither edition claims external peer review or complete military/ownership coverage. The website’s verification covers mobile and desktop layouts, keyboard/source disclosures, no-JavaScript reading, local/download links and print output. Automated accessibility checks are not full WCAG certification.
Sources, downloads and reuse.
Our downloads contain curated factual extracts, original analysis and attribution. Original institutions retain any applicable rights in their publications. We link to original HTML/PDF/workbook sources instead of serving indiscriminate third-party mirrors. Necessary source labels and short excerpts are attributed; no blanket licence over institutional material is claimed.
When reusing extracted facts, keep source IDs, dates, amount definitions, original labels and uncertainty notes. The raw research cache, full third-party files and private working notes are not part of the production website.
English text uses system fonts. Chinese source locators use a small self-hosted Noto Sans SC subset under the SIL Open Font License; its provenance is included. There are no remote fonts, trackers, analytics or browser-side data requests.
Deliberate updates, visible corrections.
These are semi-static, dated editions. A later source can change a conclusion only through a reviewed revision. Preserve superseded values, identify the changed claim and explain why it changed. Refreshing a source is not automatic publication.