Independent evidence audit · direct-link workbench

I find where the evidence stops and the claim keeps going.

Source-traceable stress-testing for research-backed claims: what the source actually establishes, where implementation or interpretation adds load, and which conclusions remain defensible after inspection.

What I audit

Claims are traced back to the layer that actually bears them.

The service is not limited to one statistical technique or one kind of paper. The common unit is a material claim and its evidence trail.

Claim → source

Claim-to-Source Trace

Trace a proposition through the paper, table, figure, equation, measure, supplementary material, code, and earlier versions where relevant.

Construct

Construct–Measurement Fit

Test whether an observed measure, coefficient, proxy, or level of association is being treated as a stronger construct than the design identifies.

Implementation

Robustness & Implementation

Inspect the actual machinery behind a robustness claim instead of treating the existence of code or supplementary analysis as validation.

Metrics

Metric, Ratio & Aggregation

Decompose summary indicators, denominators, shared ratio components, and individual-versus-group inference before accepting the headline interpretation.

Lineage

Claim Lineage & Version Audit

Track how wording, evidence, caveats, and claim status change from working paper to revision, publication, self-citation, and downstream summary.

Writing

Evidentiary Editing

After the substantive position is fixed, align the written product with evidence strength without manufacturing missing theories, evidence, or expert opinion.

Featured case

When a robustness check changes the ranking it was said to preserve.

Pärnamets et al. (2026) · open-mindedness across 68 countries

Archived reliability-correction implementation → bug-only repair → quantified claim consequence

The archived R Markdown recalculates later multi-item reliabilities into one object while continuing to append values from a stale object. A bug-only repair changes the focal open-mindedness versus collective-narcissism ordering for the conspiracy-beliefs outcome. The repaired difference still has a 95% interval that narrowly includes zero, so the audit does not claim the opposite substantive theory is established.

P(OM > CN) · buggy
0.0167%
P(OM > CN) · bug-only repaired
96.67%
Posterior draws
6,000
Best-fit work

Small, decision-relevant scopes first.

one material claimone disputed studyone robustness layerone proxy or constructone short claim lineage

Potential users include researchers, independent experts, boutique research or litigation-support teams, and organisations that want adversarial pre-release review of a consequential claim.

Scope boundary

Evidence audit, not statistical omniscience.

Quantitative verification is offered where a claim can be independently reconstructed from disclosed data, code, or reported quantities. This work is not a substitute for specialist statistical, legal, or subject-matter expert opinion.