Prompt 3.8.1, single pass. Source: NCT05324124_soa.pdf (4 PDF pages = document pages 9-12 per PAGEMAP.md). No protocol markdown available; all text from the PDF.
| # | where | call made | alternative | detail |
|---|---|---|---|---|
| D1 | p.12, between rows 19 (Genetic sample) and 20 (Selpercatinib administration) | The black "CCI" redaction box across the full table width is treated as hiding unknown content; no activity rows were created for it. | Add one or more placeholder rows (e.g. "Redacted (CCI)") with no marks so the hidden rows are represented. | 3.1 |
| D2 | p.12, rows 21-22 (AE/SAE review, Concomitant medication review) | The double-headed arrow is treated as one merged cell covering Day 1 to Day 11 (cols 4-14), and "↔" is recorded in each of those 11 columns. | Limit the arrow to the columns the drawn arrow physically crosses (about Day 1 to Day 10). | 3.2 |
| D3 | p.9 (header on every page), rows 2-3, col 16 | The Follow-Up text "within 7 to 10 days after last dose" (one cell merged down across header rows 2-3) is stored on the study-day row 3; row 2 is empty for that column. | Store it as a separate visit-window property, or on row 2. | 3.3 |
| D4 | p.12, abbreviation line below the table | The abbreviation line (CRU, ECG, ED, h; the rest is redacted as CCI) is not emitted as annotations, because no term is printed as a marker. They only appear as ordinary words in labels and cells. | Emit ab1-ab4 (CRU, ECG, ED, h) bound to the labels/cells where those words appear. | 3.4 |
hierarchical_level: null. It does not tell any two columns apart.method: synthesized). cm7 ("See Appendix 10.2, Clinical Laboratory Tests, for details.") is a bare pointer and is typed source_note.Table 01: main_soa, "Schedule of Activities (SoA)" (section 1.3), document pages 9-12.
- Columns: 17 physical columns. Col 1 is Procedure (label). Data columns are 2-16 (Screening; Day -1; Days 1-11; ED; Follow-Up phone call). Col 17 is Comments (notes).
- Header rows: 3 (epoch, a presentational "Days" band, study day).
- Activities: 19 (rows 4-22). All are level 0 (a flat table with no group headers; indentation_method: assumed_flat).
- Activity rows per page: p.9: 6 (rows 4-9); p.10: 3 (rows 10-12); p.11: 6 (rows 13-18); p.12: 4 (rows 19-22). Every page contributes rows.
- Schedule cells: 77, including 22 distributed arrow cells.
- Annotations: 8 (7 footnote, 1 source_note).
pdftotext -bbox. The result matches the visual read cell for cell.method: visual_read).3.1 Redaction (D1). A black box labelled "CCI" (Company Confidential Information) covers the whole table width on p.12 below "Genetic sample". Its height, about 3-4 normal rows, suggests it hides one or more activity rows and their marks. Nothing was invented for it. This is the main place where the extraction could be incomplete.
3.2 Arrow spans (D2). On rows 21-22 the cells for Day 1 through Day 11 have no internal vertical rules in the raster, so they form one merged cell (cols 4-14). The arrow is drawn inside that cell and was distributed across the whole cell (source_range "4:14"), following §5. ED (col 15) has its own X. The Follow-Up column (col 16) is empty for both rows, as printed; it was not filled in.
3.3 Follow-Up window (D3). Col 16 has "Follow-Up phone call" in header row 1 and a cell merged down across rows 2-3 with the window text. The schema has no vertical merge for headers, so the text is on row 3.
3.4 Abbreviations (D4). The line reads "Abbreviations: CRU = clinical research unit; ECG = electrocardiogram; ED = early discontinuation; h = hour;" and the rest is redacted as CCI. None of these terms is a marker, so nothing was emitted, per §6.
merged_cell_range "3:15")."3:15").source_range "4:14").annotation_text_source.method: deglyph_reconstruction). The cell edges come from the rule lines visible in the raster.marker_location, and each marker also appears in that row's annotation_markers.activity_name_source.indentation_method: assumed_flat: all 19 activities.activity_schedule.method: visual_read: the 22 arrow cells (rows 21-22, cols 4-14).schedule_property.structure_method: inferred_from_layout: row 2 ("Days" band).marker_locations.method: synthesized: cm1-cm8 (Comments-column notes).annotation_text_source.method: deglyph_reconstruction: cm8.unresolved locations.