A high-citation research cohort, not a census of every university.
The snapshot takes the first 1,000 OpenAlex records returned for type=education and works_count>1000 when sorted by cited_by_count descending.
1,000 institution records · 64 country/territory codes · 2026-11-14T04:36:28.146Z
Final manual eligibility decisions: 0 / 1,000
Reproducible scoring does not mean the source observations are verified.
The 20 November 2026 full-cohort comparison matched 1000/1,000 OpenAlex IDs. Using a current/frozen ratio outside 0.8–1.2, 0 works counts, 999 citation counts, 754 H-indices and 471 two-year mean-citedness values differ materially. The root cause is unresolved.
Current OpenAlex metadata also marks 27 rows inactive and 7 as super-systems; 50 rows have another published cohort entity in current lineage. These are manual-review signals, not automatic exclusion rules. The edition remains frozen and positions are not silently rewritten.
Four research proxies are active in the published score.
Citation-volume impact
45%Within-cohort percentile of log1p(cited_by_count). This is aggregate citation volume, not field-normalized citation impact.
Research-quality proxy
25%75% H-index percentile plus 25% OpenAlex two-year mean-citedness percentile.
Research output
20%Within-cohort percentile of log1p(works_count).
Recent activity continuity
10%Number of non-zero recent counts_by_year entries multiplied by 20. Every institution in this snapshot has five active years, so this component does not change rank order.
score = 0.45 × impact + 0.25 × quality + 0.20 × output + 0.10 × continuity
Each raw input is transformed with log1p and then converted to a within-cohort percentile from 0 to 100. Continuity is active recent years multiplied by 20. Equal raw values receive their average occupied percentile. Ordering uses the unrounded score; rounding is for display.
The current edition shows sequential positions, not shared tied ranks.
The published snapshot sorts by unrounded score, then cited_by_count descending, then stable OpenAlex ID. Displayed ordinal positions are sequential rather than shared tied ranks. Rows without a name, country code or country are excluded during collection. Numeric fields absent from the source response are currently converted to zero before scoring; the published 2026 snapshot contains no zero-valued scored inputs. Future releases must preserve missing values explicitly.
0 rows contain a zero scored input; distinct active-year values: 5
The index cannot support teaching or overall-quality conclusions.
- Selection by aggregate citations favors larger, older and well-covered institutions.
- Aggregate citations and works are not normalized by field, institutional size or contribution.
- H-index reflects both scale and accumulated history; two-year mean citedness is only one quarter of the quality proxy.
- The continuity component has no variance in this snapshot and must be removed, redefined or replaced before the next edition.
- Final manual eligibility decisions are incomplete for all 1,000 rows; OpenAlex type=education alone does not prove standalone degree-awarding status.
- Key source fields show cohort-wide systematic differences; the edition cannot be promoted as current data until original responses and transformations are resolved.
Edition, source date, review status and page updates are separate facts.
Suggested citation: WorldMetra (2026), Global University Research Index frozen historical dataset, method v1.0, OpenAlex source snapshot marked as retrieved 2026-11-14, source-integrity review open; add your access date.
Dataset, full-cohort audit and checksums → · V2.0 comprehensive-ranking research preview (not formally released) →