Dashboard review guide¶

Pilot studies and Synthetic data have separate navigation sections. Pilots support review against published submission packages; the study report summarizes their results and opens each review workspace. Synthetic workspaces support source exploration and development of the EDC-to-submission workflow.

The links beside each study name return to the study report and open that study's source folder in this repository.

Compare pilot datasets¶

The workspace starts with the first three subjects. Open Subjects to search identifiers, choose individual subjects or a larger group, or select all. There is no subject selection limit. Click outside a picker, move focus outside it, or press Escape to close it.

Select domain opens multiple input datasets. Names include the domain and description, such as DM - Demographics. The output Select domain control adds output datasets beside the current output. Specification chooses the YAML currently shown; selecting a column in another output focuses that output's specification. Previously selected comparisons remain open.

Each table starts with up to ten matching rows. Its Rows control independently selects 10, 25, 50, or 100 rows per page. Use its arrow controls to explore all remaining records. Table heights grow to show the selected page without an inner vertical scrollbar; scroll the page to review larger row counts. A subject without events can correctly have no event rows. Subject-independent lookup tables remain visible. Search matches values across the displayed datasets and links records with matching subject identifiers.

Follow columns and resize the workspace¶

Select an output column to reveal its connected input datasets automatically. The connection can pass through another derived dataset into its upstream sources. Related columns narrows tables to linked variables while retaining the subject identifier and declared keys. Purple Key badges and cells identify keys from each dataset's Define-XML or yamaa specification. An ID badge marks the configured subject identifier when no key is declared. Blue marks the selected column and amber marks related columns; keys keep their purple background. All columns restores the complete schemas. Manually selected comparison datasets remain available. Input headers and YAML blocks also select declared connections. Clear highlight restores the manual input selection and removes the column focus.

The YAML pane shows the upstream rules needed to derive the selected variable. It hides output.columns, keeps the relevant input declarations and intermediate definitions, and narrows column, row and intermediate derivations to related entries. Matching keys, filters, grouping and lookup selection order stay with their relevant rule. Row-driver declarations remain available even for literal values. The selected definition is highlighted, and the caption shows how many source lines are visible.

Other study settings and verifications remain visible for context. Original line numbers are retained; inline input and output mappings show the relevant fields. All columns or Clear highlight restores the complete original specification.

The specification's View control switches between the original source files and the engine's Resolved model (inspection). The resolved model expands inherited declarations, copied producer columns and named windows. Source files retain their original text and line numbers; the resolved model is an inspection artifact with its own line numbers. Its normalized representation is not an authoring template. Open source opens the selected original file.

Focused views mark omitted line ranges. Copy source file copies the complete selected file, including hidden lines. Copy model copies the complete inspection model. The selected view is retained in the shared review URL. For authoring conventions and verification examples, see Writing specifications.

Connections include declared values, joins, filters, keys, and ordering. They describe column dependencies; they do not identify contributing input rows. Unresolved references are reported above the specification.

Matching keys are included for the input or lookup a derivation actually reads. Unused inputs and sibling lookup fields do not become sources of every column. Row filters, grouping and lookup selection order remain part of the trace; final output sorting does not add value dependencies. For example, TRT01AN traces to DM, while an exposure-derived TRTSDT also opens EX. Manually selected comparison panes remain open independently of automatic lineage.

Drag the specification separator to change its width, or the horizontal separator between input and output areas to change their heights. Dataset panels can be resized using their bottom-right corners or the width buttons in their headers. Drag a column's right edge to change its width. Focus a separator and use arrow keys for keyboard resizing. Reset layout restores the defaults. Additional panels can be reached with the area's scrollbars. On a phone, panels stack vertically.

Verify documentation observations¶

The Consistency findings tab lists actionable issues. Each row names the problem, summarizes the difference, and shows measured impact and agent review. The Consistency metric counts issues; the listing separately counts the supporting automated checks. Successful execution stays in the study metrics.

Search a variable, issue, document, quote or rule ID. Issue type filters conflicts, divergences and gaps. Agent review separately filters agreed, disputed, mixed, stale and pending assessments. Agreement is not resolution. Several checks can support one issue; independent problems in one variable stay separate. A shared check can support more than one issue without duplicating it in the supporting-check count.

Expand Trace evidence to see:

expected and observed behavior;
the relevant source quotations and specification clauses, each shown once;
selected data comparisons, with record counts and representative value pairs;
the proposed action.

Full context reveals the complete quotation or specification with original line numbers. Supporting checks and agent review retains the rule statements, complete check evidence and the adjudicator's reasons. Source quotations stay literal. View source opens the retained document at the cited location.

Impact counts distinct subjects across issue-specific comparisons. Shared and alternative comparisons are labelled as context and excluded from impact. Impact not measured is different from No differences in tested data. Reference differences do not by themselves identify an error in the derived data.

Open evidence in workspace selects the variable and the subjects named by its measured impact. Without a measured subject set, the current subject selection is retained. Old links naming an individual check open its first related issue.

Label differences and missing columns remain review issues. Not reviewed identifies an authored question with a verified source excerpt. Missing evidence means the expected source or excerpt could not be verified. Narrative coverage is partial. See the verification policy.

Review documented rules¶

Reviewed datasets add Documented rules issues. Study-specific grouping is explicit: each issue has a short title, expected and observed behavior, a proposed action, and links to its supporting checks. Changes to check evidence require the grouping to be reviewed before publication. Datasets without a grouping file retain one issue per check.

Issue types are conflict (documents disagree with data or one another), divergence (a documented rule is expressed differently), gap (a rule or clause is undocumented), and evidence (a quote was not found). A divergence can matter even when the current data do not distinguish the rules. Agent review and human resolution are separate from these types.

View source documents¶

Each retained file has its own tab in Source documents. Choose a file and use Compare side by side to open a second independent viewer. Switch files within either pane; Close comparison returns to one. On narrow screens, the two viewers stack vertically. Use arrow keys, Home, or End to move between file tabs.

PDF pages retain their original layout. Excel sheet tables retain coordinates, merged cells and formulas. XML uses the publisher's retained stylesheet. Each viewer has a Source link to that file in the repository at the displayed study revision. Extracted reading copies and repository paths are omitted from this interface. Selected files and the primary cited location are preserved in the review URL. Missing documents remain explicitly unavailable.

Review synthetic workflow readiness¶

Synthetic workspaces inventory staged ODM archives and show the availability of source, SDTM, ADaM, and submission-metadata files. Missing stages are shown as Not staged. Archive members, sizes, and repository links help locate source material. These dashboards describe source availability; they do not claim successful derivations or passed reference comparisons.

Copy the browser URL to preserve the specification, subjects, compared datasets, row-count choices, column focus, searches, and open evidence. Review comments appears below pilot dashboard content with a separate GitHub Discussion for the focused dataset. GitHub sign-in is required to post. A review-issue link remains available if the embedded discussion cannot load.