Building the measurement layer for human data.
METER is a reusable measurement engine for questionnaire and assessment data. It turns a new dataset into governed scores and measurement diagnostics without rebuilding a psychometric model for every instrument, population or study.
A pretrained engine: measurement capabilities learned in advance and applied to new data without fitting a new model each time, with uncertainty, diagnostics and explicit capability boundaries.
Read the preprint →Tay & Chen (2026) · Research Square
See what pretrained measurement means
Most psychometric workflows begin again with each new dataset: specify a model, fit it, check it, extract the outputs, and document the analysis. METER takes a different approach. It learns measurement capabilities in advance and applies them to new data when the evidence supports doing so.
Conventional fitting
- New dataset
- Specify model
- Fit model
- Check estimation
- Extract outputs
METER
- New dataset
- Pretrained inference
- Governed output
A worked comparison
On a held-out evaluation of 20,000 people, 50 items and five declared factors:
Specialist confirmatory model
- Runtime
- 662.1 seconds
- Dataset-specific fitting
- Yes
METER
- Runtime
- 0.78 seconds
- Dataset-specific fitting
- No
Median person-score agreement: r = 0.9791
METER reproduced the target person-score information without being fitted, tuned or calibrated on the target dataset. What changes is the estimation step—not the need for sound measurement.
How to read this result
This benchmark demonstrates transfer without target-dataset refitting. It does not imply that METER is universally more accurate than specialist estimators, and the runtime difference applies to this specific evaluation.
Built for measurement that repeats
A study team may administer twelve measures across eight sites and five waves. A survey platform may support hundreds of assessments. A research organisation may need the same instrument scored across several populations. METER is one measurement layer across those analyses, rather than a new fitted workflow every time.
A pipeline per measure
- PHQ-9 → its own scoring pipeline
- CES-D → its own scoring pipeline
- IPIP → its own scoring pipeline
- A new wave → refit and recheck
- A new country → refit and recheck
- A new battery → another model
One measurement layer
- Every supported measure → the same pretrained engine
- Scores, diagnostics and a recorded decision, every time
- Each additional dataset costs seconds, not a fitting run
The saving is in time and in setup, and it compounds: the same one workflow applies to the next instrument, the next population and the next wave. Scores from separate runs are not on a common scale, so to compare waves or groups, score them together in one run.
Clinical outcomes research
Repeated patient-reported measures across studies and sites.
Population research
Common measures across cohorts, countries and waves.
Assessment platforms
Reusable scoring and measurement checks across many assessments.
Using measurement repeatedly across studies, populations or teams? Talk to us about design-partner access.
Write to hello@meterresearch.ai →Watch a real run
Choose an example to see METER check the data, determine what it can support and return the result.
Waiting to start…
- Read
- Check
- Decide
- Measure
How METER works
Bring a CSV, Excel, SPSS or Stata file. Nothing you upload is retained.
Choose the items, the structure, and what METER should estimate.
Each requested claim gets its own capability decision.
Scores, diagnostics and agreement with classical scoring.
The brief, item diagnostics, warnings and what to watch.
Scores, provenance, Methods text and CSVs.
What can METER support?
Score people
✓ supportedSupported for eligible measures. What counts as eligible depends on the measure and on what you ask METER to infer.
Understand individual items
✓ research previewSupported as a research-preview capability for single-factor measures.
Handle multi-factor measures
person scoring onlyPerson scoring is available within the applicable supplied-structure contract; full multidimensional item inference remains under development.
Discover the factor structure
not supportedNot supported. The user supplies the intended structure.
Support depends on what you ask METER to infer. The same dataset may support person scoring while METER withholds a less-established item-level claim.
Built for trustworthy measurement
METER uses the same frozen model across supported datasets, checks whether each requested inference falls within its evidence base, and records how every result was produced.
Evidence before claims
METER has been evaluated on synthetic known-truth problems and real external datasets, including more than 66,000 respondents across 28 countries. Results, limitations and failed evaluations are published in the evidence record.
Explore the evidence →Read the preprint →
METER is a measurement platform currently released for research use. It is not a diagnostic, screening or triage system, and no output may be used to make a decision about an individual person.
Starting with questionnaire and assessment data, METER is being developed as a common measurement layer across instruments, populations and studies. The current research release focuses on pretrained scoring, governed measurement inference, diagnostics, explicit capability boundaries and auditable provenance.
Try METER
Upload a response dataset and see which measurement operations METER can support. Nothing you upload is retained.