Skip to content
SemantaSemantaDocs

Reading research results

Distinguish source records, calculation outputs and interpretations.

Last updated

Establish what actually happened

An answer may combine retrieved records, calculation outputs and model explanations. Start by separating their sources.

ResultWhere to check it
Material recordDatabase result card, material ID and detail page
Paper or knowledge-base evidenceDOI, document details and source passages
Calculation resultInputs, node states, logs and output files for that run
Model interpretation or suggestionWhether the cited records, methods and conditions support it

A claim that something was calculated needs an execution record. A successful run, in turn, does not by itself establish a scientific conclusion.

Exercise: check one SiC result

Complete the quick start. Choose a returned record and note its material ID. Open the details, or find it again in the materials database.

Compare the answer with the formula, crystal system, band gap, units and source in the record. A displayed table may round decimal places; check the original value before deciding whether a discrepancy is meaningful.

If the answer combines different structures into one conclusion, ask:

List this query's results separately by material ID. Do not merge different structures.
Distinguish original database fields from your interpretation.
Identify conclusions that still require calculation methods or test conditions.

Check a completed calculation

In Material Expert, open the relevant run from Run history. Confirm the input structure and parameters, then inspect node states, logs and original files. If a node has no output, check whether it executed.

Different tasks require different checks:

  • Geometry optimization: inspect the final structure and whether the configured convergence criteria were met.
  • Bands or density of states: check energy reference, sampling, units and calculation parameters.
  • Molecular dynamics analysis: check the sampling interval, system size and statistical conditions before interpreting curves.
  • Plots or fits: check input columns, units, selection ranges and generated data.

A run status, one fit metric or a smooth curve is not sufficient evidence on its own. Record any checks still needed.

Keep results traceable

Retain material IDs or DOIs, inputs, methods, units and run records with saved charts and files. State differences in conditions before comparing data.

If a download fails, retry from the original result card or run details. A filename mentioned in an answer does not establish that a file was created; use an actual file entry provided by the platform.