Methodology
How we calculate the conflict-strength score.
Every Arm's Length Grid ships with a single 0–100 number. This page documents exactly how we get there — the data sources, the weights, and the audit trail. No secret sauce.
The score is a sum of three signals
ORCID timeline overlap
ORCID Public API v3 Weight
18 + 2 × years
30 / overlap
Co-employment at the same institution is the strongest structural conflict signal. The base score reflects "they were colleagues"; each additional overlapping year compounds the risk.
PubMed collaborative publication
PubMed E-utilities Weight
12 base · 22 if last-author senior
no cap on count
Joint authorship in the last 7–10 years signals working relationship. Last-author senior status (lab head / senior corresponding) further indicates direct mentorship/supervision.
Mutual conference / talk page
Google web search (15 pages)
Weight
8 high · 3 medium confidence
no cap on count
Co-presence at conferences, panels, or workshops is a weaker but real signal. Confidence tier reflects whether both names AND a conference keyword appear on the same indexed page.
Total score is the sum of all contributing factors, capped at 100. ORCID contributions are individually capped at 30 points each so a single very long co-employment cannot saturate the score alone.
Tier interpretation
0 – 19
Arm's length
No meaningful conflict surfaced. Safe to proceed.
20 – 49
Caution
Light overlap. Review individual findings before assigning.
50 – 79
Strong conflict
Material overlap on one or more axes. Consider an alternate evaluator.
80 – 100
Severe conflict
Disqualifying overlap. Direct supervisor or recent co-author.
Worked example
Candidate Dr. Jane Chen evaluated against Prof. Mark Whitfield:
| Finding | Points |
|---|---|
| ORCID · co-employed at MIT, 3 yrs overlap | +24 |
| PubMed · 2022 paper, Whitfield = last author | +22 |
| PubMed · 2024 paper, both co-authors | +12 |
| Google · NeurIPS '22 panel page (high) | +8 |
| Total · Strong tier | 66 / 100 |
Audit trail
Every finding links back
Each scored item displays the verification URL to the ORCID profile, PubMed record, or Google page that produced it.
Senior status detection
Last-author detection is name-match against the publication's terminal author. We flag both directions: candidate-as-senior and evaluator-as-senior.
Institution match logic
ORCID overlaps require case-insensitive substring match on institution name AND year-range intersection ≥ 1 year.
Honest limitations
- ORCID coverage varies. Researchers in early career, the humanities, or non-Anglophone institutions may lack a populated ORCID record. Always cross-check the CV against the score.
- Name collisions exist. Common surnames (e.g. "John Smith") can return false PubMed positives. Where possible, the platform incorporates ORCID disambiguation; for high-stakes decisions, review the underlying PubMed records via the verification link.
- Google depth is bounded at 15 pages. Beyond that, signal-to-noise collapses. Conference results carry the lightest weight by design.
- The score is advisory, not adjudicative. It surfaces structural overlap that a human reviewer can verify. Final arm's-length determinations always require human judgment per your institution's policy.
Enterprise
Need a custom scoring rubric?
Universities 50+ tier deployments can configure per-institution weights, add policy-specific signals (e.g. dissertation supervision lookup), and export audit logs to compliance systems. Pricing on the institutional plan page.
See institutional plans