Source: NCT03693430_soa.pdf (4 PDF pages = document pages 9-12 per PAGEMAP.md), protocol NN9536-4378 v3.0, 29 June 2018, section "2 Flowchart". Output: NCT03693430_Table_01_extraction.json (one table). Prompt 3.8.1.
| # | where | call made | alternative | detail |
|---|---|---|---|---|
| D1 | p.11 (and p.12), Visit Window row, V33 column | V33 (End of treatment) window recorded as ±3 days, as printed on pages 9 and 10 | Record ±5 days, as printed on the reprinted header of pages 11 and 12 | 3.1 |
| D2 | p.11 row 44 "Breast neoplasms follow-up"; p.12 row 63 "Training in trial product, pen-handling" | Rows printed above the reprinted Visit header are kept as ordinary activity rows in reading order (Breast neoplasms follow-up under SAFETY; Training under REMINDERS) | Place them in a different section (e.g. Training under TRIAL MATERIAL) or order them differently | 3.2 |
| D3 | p.10 row 37 Control of Eating Questionnaire (CoEQ) | Marks transcribed as lower-case "x", exactly as printed (smaller glyph than every other mark) | Normalise to "X" as a typesetting inconsistency | 3.3 |
| D4 | p.9 row 1 (unlabelled period band) | All six band values typed epoch, including single-column Randomisation (V2), End of treatment (V33), End of trial (V34) | Treat the single-column cells as visit-level labels and keep only Screening / Dose escalation / Maintenance as epochs | 3.4 |
source_note annotations (pr1-pr26, deduplicated by text, split where one label cites several). Sections printed as bare numbers "(6.1)" are rendered "See Section 6.1."; appendix links "See Appendix n.".Table 1 — "Flowchart", main_soa, document pages 9-12. One printed table running across four pages under a reprinted header; no own caption on the later pages, so extracted as one table (not continuations).
3.1 V33 visit window. Pages 9 and 10 print "±3" under V33; the reprinted header on pages 11 and 12 prints "±5". All other header values are identical across the four prints. Recorded ±3 (first print); not resolvable from this excerpt.
3.2 Rows above the reprinted header. On pages 11 and 12 the first body row of the page ("Breast neoplasms follow-up b (9.4)"; "Training in trial product, pen-handling (7.1.1)") is printed directly under the period band, before the Visit / Timing / Window rows reprint. Treated as the next activity in reading order (row 44 follows Technical complaint, row 63 follows Hand out directions for use). Breast neoplasms follow-up (V33, V34) sits naturally with "Colon neoplasms follow-up" in SAFETY. Training sits in REMINDERS by position.
3.3 CoEQ marks are a visibly smaller lower-case "x" (visibly smaller on the render; pixel counts 28-32, at the bottom of the 32-56 range of the other marks). Kept literal.
3.4 Period band typing: see D4.
3.5 Indentation: dark-grey caps rows = level 0; light-grey rows = level 1; white rows printed indented under a grey group row (Height, Body weight, Waist circumference, Systolic/Diastolic Blood Pressure, Pulse, Dispensing visit, Drug accountability) = level 2. indentation_method visual_estimate. Leading spaces in cell_text for level-2 rows represent the visual indent (no text layer exists).
3.6 Visit Window row given hierarchical_level null (does not distinguish any column); Visit = level 2, Weeks = level 3.
3.7 "Attend visit fasting (6.4.1)" is a body row with marks in the REMINDERS section, so it is an activity, not a schedule_property.
Image method (§1a/§1d): vertical rules = pixel columns with ink fraction > 0.5 over the table height (36 rules, identical x on all 4 pages); horizontal rules = pixel rows with ink fraction > 0.85 across the table width. Each cell inset 4 px; mark = count of pixels with intensity < 90. Distribution is cleanly bimodal: empty cells 0, marked cells 28-56; threshold 15. The detector matrix was compared with direct visual reads of every body row on all four pages (dense rows: Concomitant medication, Adverse event, Diet and physical activity counselling; sparse rows: Height, Evaluation of glycaemic status, First date on trial product). No disagreement. Recommend a spot-check of the resolved grid, as for any image-based source.
None: every annotation has at least one marker_location, and every location's marker is present in that row's annotation_markers.