Institutions
A new view of research impact.
Institutional impact has been measured by citation counts and download logs for decades. Neither describes what now happens most often: a question is asked of a system, some evidence is retrieved, and a great deal of good work is never reached at all. This measures that layer, at department and institution scale.
Portfolios
Institution and department views built on identifiers, not strings.
- 01Institution and department portfolios
- Group output by faculty, department, centre, or funding programme, using registry identifiers rather than free-text affiliation strings. Records that cannot be attributed confidently are reported as unattributed, not assigned by guess.
- 02Consistent roll-up
- A department figure is the sum of records you can inspect. Every aggregate opens into the list that produced it, with the same measurement applied at both levels.
- 03Comparable time windows
- Portfolios are measured over declared windows so a comparison between two units is not an artefact of one having been measured six months later.
Comparison
Field-normalized visibility, with the raw numbers still visible.
A methods group and a large clinical unit cannot be compared on raw counts. They also cannot be compared responsibly on a normalized number that hides how the adjustment was made.
- 01Field normalization
- Subject areas differ enormously in publication volume, citation behaviour, and how often their questions are asked of retrieval systems. Comparisons are normalized by field and publication age before any unit is compared to another.
- 02Raw counts shown alongside
- Normalization can hide as much as it reveals, so raw values are always available next to the normalized ones. A reader should be able to see what the adjustment did.
- 03Small units handled honestly
- A group with a handful of publications does not get a confident number. Where the sample cannot support an estimate, the interface says so instead of rendering a precise-looking figure.
- 04Not a ranking instrument
- Trace is designed to find remediable problems, not to rank people. Visibility measurements are provisional heuristics and should not be used in hiring, promotion, or tenure decisions.
Not for individual assessment
These measurements are provisional product heuristics under active validation. They are intended to find and fix discoverability problems. They are not suitable for evaluating individual researchers, and Trace does not support that use.
Detection
Find the good work that is not being reached.
- 01Under-discovered high-quality work
- Records whose measured discoverability falls well below comparable work in the same field and age band. This is the most useful thing an institution can learn, because it identifies good research that is simply not being reached.
- 02Attributable causes
- For each under-discovered record, the diagnosis separates causes the author controls (terminology, abstract structure, title specificity) from causes the institution controls (repository deposit, identifier registration, affiliation records).
- 03Remediation queues
- Findings are grouped into work queues a research office can actually execute: deposits to make, identifiers to register, records to correct.
Availability
Open availability and repository gaps.
- 01Open-access and availability gaps
- Where a record has no lawful open version, that is recorded as an availability gap. Availability is one of the few levers with a plausible mechanism for improving retrieval, and it is squarely within institutional control.
- 02Repository coverage
- Which portion of the portfolio is deposited in an institutional or subject repository, which deposits are incomplete, and which are present but not discoverable because their metadata is thin.
- 03Policy compliance visibility
- Funder and institutional availability requirements produce a concrete, checkable list of records. The product reports coverage against that list; it does not assert legal compliance.
- 04Version linkage
- Preprints, accepted manuscripts, and versions of record are frequently unlinked, which splits a single work into several weakly discoverable records. Unlinked versions are reported as a defect with a fix.
Integrity
Author identity and metadata integrity.
- 01Author identity integrity
- Persistent author identifiers, name variants, and merged or split records. Ambiguous identity is the most common reason a portfolio undercounts a researcher's work.
- 02Affiliation normalization
- Affiliation strings are resolved against a public organisation registry. Unresolvable strings are surfaced for correction rather than silently dropped or silently claimed.
- 03Metadata completeness
- Missing abstracts, absent subject terms, unregistered funding acknowledgements, unlinked trial registrations, and corrections that are not attached to the record they correct.
- 04Source disagreement is preserved
- Where bibliographic sources disagree about a record, both values are kept and shown. Flattening disagreement into one authoritative value hides exactly the errors a research office needs to find.
Exports
Output a research office can use.
- Research office exports
- Record-level CSV and JSON exports with the measurement, the diagnosis, and the suggested action, suitable for joining to a CRIS or repository system.
- Reporting packs
- Department-level summaries with the methodology version, the run window, and an explicit statement of what the measurement does and does not support.
- Reproducible figures
- Every exported figure records the query set version and provider set, so a number in a report circulated next year can still be re-derived.
Next step
Start with one department.
A single department with a well-known publication record is the fastest way to judge whether the measurement is describing something real. Enterprise pilots are available now.