Select the statement that best describes the organization's documents
today, one per chapter. The result shows a level for each chapter and the
chapters to read first. Answers are stored in this browser only.
1. Inventory How well does the organization know which documents hold data?Not yet Documents are on the website or in an archive; nobody has listed which hold tables and figures. Foundational An inventory lists the documents, their type, pages, and whether they have a text layer. AI-ready Every document has a catalog record in the document schema, with the series and surveys it draws on. AI-native The inventory is kept current automatically as documents are published, with table and figure counts from layout detection. 2. Scanned How are scanned documents read?Not yet They are not; scans wait, or someone types the tables. Foundational Scans are marked in the inventory and OCR runs with one engine on the priority documents. AI-ready Engines are tested on the organization's own pages per language, every OCR page has a quality score, and pages over the threshold are re-processed or reviewed. AI-native OCR runs as a pipeline stage with the score as a gate, and corrected pages feed the test set. 3. Locate How are tables and figures located in documents?Not yet By a person paging through the document. Foundational Tables are located from the text layer; figures by hand. AI-ready A layout detection model locates tables and figures with bounding boxes, and detections are reviewed on a sample. AI-native Detection runs on every new document, is measured against annotated pages, and the model is retrained on the organization's own documents. 4. Extract How are the data extracted?Not yet Copied by hand into spreadsheets when needed. Foundational Tables are extracted from the text layer into CSV; charts are read by hand. AI-ready Tables and charts are extracted with tools into tidy files that carry page, bounding box, units, and method. AI-native Extraction runs as a pipeline with a vision-language model for charts, and every value carries its provenance. 5. Verify How are extracted values verified?Not yet They are not; the extraction is trusted. Foundational A person checks a sample against the page. AI-ready Values are checked against printed labels, totals, and derived rates, and a sample is reviewed; each value carries a status. AI-native Verification runs in the pipeline, flagged values block publication until corrected, and error rates are reported. 6. Series How are tables from successive documents combined?Not yet They are not; each extracted table stands alone. Foundational Tables of one statistic are mapped to a series and combined, with revisions reviewed. AI-ready Every series has a mapping, units and areas are normalized, breaks are recorded, and the series is published as a record with a values file. AI-native Combination runs as a pipeline stage, with model-proposed mappings and break explanations confirmed by a curator. 7. Publish How are extracted data published?Not yet They are not published; they stay in working files. Foundational Tidy files are published next to the document with the licence of the document. AI-ready Each extracted table has a record in the table schema that links to the document record and to the series it draws on. AI-native Extracted data are served through the catalog API with provenance, and the document record links to every table extracted from it. 8. Cite How can extracted data be found and cited?Not yet Only by finding the document and the page. Foundational Extracted tables appear in the catalog and in search by their printed titles. AI-ready Each table has a citation to the document and page, schema.org markup, and a stable URL. AI-native Each value cites its page and bounding box, corrections are logged as errata, and use is monitored. 9. Operate How does the pipeline run over the backlog?Not yet Document by document, by hand, when time allows. Foundational A status log and a manifest order the work and make runs resumable; time per page is measured. AI-ready Stages run as batch jobs with retries; cost per page is measured per stage; batches are monitored; the configuration is versioned. AI-native Runs are scheduled from the inventory, models are chosen per document type from measurements, and corrections propagate to series and records. 0 of 9 answered Show reading path Reset
Interpretation of the result
A Not yet result in chapter 1 comes first: an inventory is a day of
work and decides where to spend the rest, including how many pages need
the OCR of chapter 2.
Chapters 3 and 4 depend on each other; a team can start with tables from
the text layer (chapter 4, recipe 4.1) before any layout detection.
Chapter 5 is the one not to skip: an extracted value that was never
checked is a transcription, and the chapter's checks use what the
document already offers.
Chapter 9 matters once the first fifty documents are done; before that,
the measurements it asks for do not exist.