npbuilds Bias and Funding Tracer — The Provenance Auditor
Source-triangulator confirms whether sources agree. This skill asks the harder question: are they truly independent of the entity that benefits from the agreement? A high-tier source funded by the company whose product it evaluates is not the same as one with no stake in the answer.
Guiding Principles
- Funding is not the same as truth. Industry-funded research can be correct. But it has a higher prior on motivated conclusions and deserves an independence discount.
- Independence is a spectrum, not a binary. Score it; don't just label it.
- Absence of disclosure is itself a signal. Sources that don't disclose funding get the lowest independence score regardless of what they say.
- Trace upstream. A think-tank report citing a study citing another study — find the original funder.
How to Run
Input
A list of sources from source-triangulator's evidence bundle, plus the claim being verified. For each source:
- Title, URL/DOI
- Authors (if known)
- Publishing organization
- Source type (peer-reviewed / institutional / news / blog / etc.)
Steps
Step 1 — Resolve Funding for Each Source
For each source, search for and record:
- Direct funding disclosure. Most peer-reviewed papers have a "Funding" or "Acknowledgments" section. News articles may have a sponsorship or "in partnership with" line. Reports often list funders on the cover page.
- Institutional funding. If the publishing org is a think-tank, university lab, or NGO, identify its top 3 funders (annual report, "Supporters" page, IRS Form 990 for US nonprofits).
- Author affiliations. Each author's primary employer, board seats, consulting relationships, and stock holdings if disclosed.
If funding cannot be found after a reasonable search, mark funding: undisclosed — do not assume it's none.
Step 2 — Map Stake Relationships
For each funder/affiliation found, ask: does this entity have a financial, regulatory, or reputational stake in the claim being true?
- Direct stake — funder makes/sells the thing being evaluated, or directly profits from the outcome
- Indirect stake — funder is in the same industry, or competes with an alternative the claim disfavors
- Regulatory stake — funder is regulated by an agency the claim affects, or vice versa
- No identifiable stake — funding exists but doesn't bear on this specific claim (e.g., a general operating grant)
Step 3 — Compute Independence Score (0–10)
Per source:
| Signal | Score adjustment |
|---|---|
| Funding fully disclosed, no stake | +4 baseline |
| Funding fully disclosed, indirect stake | +2 baseline |
| Funding fully disclosed, direct stake | -2 baseline |
| Funding undisclosed | 0 baseline (and tag as such) |
| Authors have personal financial interest in outcome (stock, paid consulting) | -2 |
| Authors are independent academics with no industry ties | +2 |
| Source is a registered preregistration / preregistered analysis | +2 |
| Source is post-publication peer-reviewed (e.g., systematic review of the claim) | +1 |
| Source is a press release, marketing, or sponsored content | -3 |
| Source is from an org with documented prior advocacy on this claim | -1 |
Clip to [0, 10]. Document the calculation per source.
Step 4 — Cross-Source Independence Check
After scoring individual sources, examine the SET of sources for the claim:
- Do multiple "independent" sources share a common funder? (e.g., 3 different think-tanks all funded by the same foundation)
- Do they share authors or co-authors on prior work?
- Did they emerge from the same conference, working group, or industry consortium?
If yes, treat the affected sources as effectively ONE source for confidence-tier purposes (per the Confidence Framework's independence definition), and feed this back to evidence-synthesizer for re-tagging.
Output
An augmented evidence bundle with one entry per source:
SOURCE: [title]
funding: [organization] | undisclosed
primary_stake: direct | indirect | regulatory | none | unknown
author_affiliations: [list]
coi_disclosed: yes | no | partial
independence_score: [0-10]
reasoning: [1-2 sentences explaining the score]
CROSS-SOURCE FLAGS:
shared_funders: [list, if any] — affects: [source IDs]
shared_authors: [list, if any] — affects: [source IDs]
effectively_independent_count: [N] (may be lower than total source count)
If cross-source flags reduce the effectively-independent count below the threshold for the preliminary confidence tier (3 for Confirmed, 2 for Likely), recommend a downgrade with explicit reasoning.
Error Handling
No funding info available for any source: Tag the entire claim's evidence as independence: undetermined and recommend evidence-synthesizer downgrade by one tier.
Funder structure is opaque (shell orgs, dark money): Document what you found and what you couldn't trace. Treat as undisclosed for scoring.
Source is too informal to have funding (random blog, social media): Independence score = 0 by default; rely on source-triangulator's tier rating instead.
Scope Boundaries
Does NOT: Allege wrongdoing or fraud. The skill reports stakes; it does not adjudicate intent. Does NOT: Replace methodological review. A poorly-designed independent study is still poorly-designed. Does NOT: Apply uniform skepticism. Industry funding ≠ wrong; this is a calibration tool, not a dismissal tool.