Skip to content

Research

Research you can inspect, reproduce and challenge.

OpenDeco publishes papers, technical reports, methods notes, benchmark analyses and data work with the sources and limitations kept close to the result.

Index state

Published entries
0
Output types
7
Filter axes
7
Dataset records
4
Model records
3

Filter research

Research is filed under these seven axes. Every count reads 0 while the index is empty.

Year

Taken from the publication date of each version.

Model family

The registry holds 3 model records and no family record.

Dataset

Every dataset version an entry reads, drawn from the 4 records in the data library.

Topic

Assigned by the author at publication.

Research index

0 results

No research is published yet.

OpenDeco has released nothing. The first entry appears when a publication clears the release gate, and this count stays at 0 until then.

The public site reads a release-scoped view that cannot reach draft, internal-review or frozen publications. That view is empty because no publication has been released.

What a row will carry

Six lines, in the order specification 619 fixes. A line is dropped when the fact behind it is absent, never to save space.

  1. Type and status.
  2. The title, linked to the article.
  3. One sentence of summary. A row grows before it truncates a scientific qualification.
  4. The authors.
  5. The date, with the DOI or persistent identifier once one is minted.
  6. Links to the data, the code and the benchmark release, each shown only when it is real.

What OpenDeco publishes here

Specification 11 fixes seven output types. An entry is filed as exactly one of them, and a paper is one of the seven.

Paper
A formal manuscript submitted to or published in a journal, carrying its full methods and its data statement.
Preprint
A manuscript posted before or during peer review, versioned so that an earlier citation keeps resolving.
Methods note
A short account of one decision: a normalisation rule, an outcome definition, a metric choice.
Research question
A question written down before the work starts, with the analysis plan and what would count as an answer.
Benchmark analysis
A reading of one benchmark release: the configurations evaluated, the dataset versions used, and where the models disagree.
Data note
A description of one dataset version: what arrived, how it was normalised, what is missing, what may be redistributed.
Technical report
An engineering or operational report that carries evidence and never reaches a journal.

The standard an entry meets

Ten properties, fixed by specification 3. An OpenDeco publication carries all ten of them or it stays unpublished.

Reproducible
The inputs, the code identity and the command that regenerates the result ship with it.
Versioned
The article, the datasets it reads and the model configurations it scores all carry versions.
Source-linked
Every number resolves to the dataset version and the run that produced it.
Explicit about assumptions
Assumptions appear in the article and are named as assumptions.
Explicit about uncertainty
Intervals and sample sizes travel with each estimate.
Explicit about data leakage
Where a model was calibrated on the data it is scored against, the article says so.
Explicit about model lineage
Family, version, implementation and configuration stay apart. A number attaches to the configuration.
Explicit about licensing
Each dataset carries its licence and its redistribution status.
Explicit about missing data
Missing profile resolution is reported as missing and never interpolated into existence.
Explicit about confirmatory or exploratory intent
An exploratory reading carries the word exploratory.

Two further rules shape how a result reads here. Performance measured on calibration data is never reported as independent validation. One league-table score never stands in for uncertainty.

Where the work is now

Six surfaces carry what exists today. Each one states its own emptiness where it applies.