Monitor
Research visibility changes. Measure the change.
Engines are updated, indexes are rebuilt, new papers enter the field, and the questions people ask move. A single measurement expires. Trace Monitor re-runs a fixed question set on a schedule and keeps the full history, so a difference is a difference in the world rather than in the instrument.
Association, never causation
Score history
Every run is kept, with its version attached.
Components are stored alongside the composite, because a flat composite can hide a rise in retrieval cancelled by a fall in fidelity. Scores computed under an older methodology version stay labelled with it and are never silently restated.
Demo workspace
Run history for the fictional TRACE-101 trial
| Run date | Retrieved | Cited | Annotation |
|---|---|---|---|
| 2026-02-04 | 54% | 35% | First baseline recorded. |
| 2026-04-12 | 58% | 38% | Accepted manuscript deposited in repository. |
| 2026-06-09 | 61% | 41% | Structured abstract published on the record. |
| 2026-08-01 | 62% | 43% | Engine model version changed. |
Illustrative figures. The annotation column records what happened near a run. It does not claim that the annotated event produced the change, and no ordering of the two is presented as evidence.
New competitor detection
Notice when a new paper takes the question.
Each run records which sources were cited on each question. When a source appears that was absent from previous runs, it is surfaced with the questions it now answers and the date it was first seen.
Demo workspace
First seen this quarter
10.5555/novera.metaanalysis.2026
Now cited on 12 safety questions
10.5555/novera.consensus.2026
Now cited on 7 comparative questions
Fictional records created for demonstration.
Query drift
The questions themselves do not hold still.
A question set fixed in February may no longer describe what people ask in August. Drift is tracked separately from performance so the two are never confused.
The frozen set
The original questions are re-run unchanged. This is the only comparison that supports a trend line.
The current set
A refreshed set reflects how the topic is now discussed, and starts its own series.
The delta
Questions that entered or left the topic are reported as a change in the field, not a change in your score.
Alerting
Few alerts, each one worth reading.
Thresholds are set against measured run-to-run variance rather than a round number, so normal noise does not page anyone. Every alert links to the run and the source evidence behind it.
- Component change
- A component moves beyond its measured run-to-run variance across two consecutive runs.
- New competing paper
- A source not previously seen begins appearing on questions your paper should answer.
- Fidelity regression
- A summary drops a limitation or generalises the population beyond what was studied.
- Availability failure
- An identifier stops resolving, or full text stops being reachable.
- Integrity event
- A correction, expression of concern, or retraction is detected on a tracked record.
Portfolio
One paper is a case. A portfolio is a pattern.
Departments, publishers, and research organisations track many papers at once. Portfolio views group by programme, journal, or therapeutic area and report the distribution rather than a single headline figure — an average across unlike papers is not a finding.
Portfolio views
Reported per group
- Distribution of each component, not only its mean
- Papers whose availability checks are failing
- Papers where fidelity has regressed since the last run
- Topics where competing sources are gaining ground
- Coverage gaps: questions no paper in the portfolio answers
TVI TVI_v0.1. The Trace Visibility Index is a product heuristic under active validation, not a validated scientific metric.