How it works
Set up the publication with your reporting team, then reuse the specification for each new release.
Phase A · With the reporting team
Configure the publication
Setup is collaborative and specific to each publication.
- Define the report tree. Set the hierarchy from report to sections, subsections, paragraphs and sentences. Place charts, tables and approved standard text.
- Map sentences to facts. Specify what each sentence must communicate and link its facts to the team’s reporting data.
- Connect original production tables. Build reproducible R transformations for reporting data and derived indicators.
- Save the reusable specification. YAML fact specifications describe the evidence needed. YAML structure files define the hierarchy, order and references used in assembly.
Phase B · For every new release
Produce each release
- Import new production tables. Preserve the source version and run the R pipeline.
- Generate facts and graphics. R performs the calculations and deterministically generates updated YAML fact files, tables and charts.
- Write and verify. AI drafts sentences from assigned facts and supports semantic checking. Numerical and semantic checks flag unsupported claims, missing facts and incorrect values.
- Translate and verify. AI helps create language versions using approved terminology and the same facts. Checks preserve numerical values while allowing locale-specific formatting.
- Assemble progressively. Deterministic code handles ordered assembly where possible, from paragraphs through subsections and sections to the report. Insert tables and charts, save intermediate outputs and check consistency at each level.
- Style and review. Apply institutional presentation rules and produce Word and PDF outputs for final human approval.
Report structure
How sentences use facts
A sentence is the smallest writing unit and can draw on several facts. Teams can revise the specification when their publication or methodology changes.
Report → Section → Subsection
- Paragraph
- Sentence
- Next sentence
Illustrative example · Fictional data
Trace a sentence to its source
Source values
Production index 2023 2024 100.0 104.0 Unit: index points
Reference: fictional table T1R calculation
(104.0 ÷ 100.0 − 1) × 100
Percentage change relative to 2023.
Structured fact
Annual change: +4.0%
Period: 2024 vs. 2023
Unit: percent
Source: T1, both annual valuesGenerated sentence
“The production index increased by 4.0% in 2024 compared with 2023.”
Built for transparent reporting
Reproducible and auditable. R calculations can be rerun from versioned source tables. Text, tables and charts share the same facts, and saved intermediate outputs let reviewers trace claims, check the evidence and recheck affected outputs after revisions.
Consistent across languages and formats. Translation uses approved statistical terminology and keeps the same underlying facts. The organization’s styling carries through to Word and PDF, with final approval resting with the reporting team.
Free publication setup. After testing, AtlasStat intends to offer setup and collaboration free of charge. Any third-party running costs would be discussed beforehand.