Source: NCT02107703_soa.pdf (7 PDF pages = document pages 72–78 per PAGEMAP.md; protocol I3Y-MC-JPBL, Attachment 1). No protocol markdown was provided. Prompt v3.8.1, single pass.
| # | where | call made | alternative | detail |
|---|---|---|---|---|
| D1 | T1 p73–75, rows 7, 10, 14, 19, 27, 37, 40 | Each Procedure Category cell (Study Entry /Enrollment, Medical History, Physical Examination, Tumor Assessment, Lab/Diagnostic Tests, Study Drug, Health Outcomes) is emitted as its own level-0 grouping row with no marks, placed just above its procedures. The procedures are level 1. These grouping rows are not separately printed rows. | Treat the category column as a label column only: no grouping rows, and every procedure at level 0 (a flat table). | 3.2 |
| D2 | T1 p75, rows 38–39 (Fulvestrant / LY2835219 Therapy) | The dosing text cells are spread across columns 6–11: all on-treatment columns and both follow-up columns. The ruled merged cell runs to the right edge of the table. | Limit the text to the on-treatment columns 6–9, reading the extra width as layout only, since no drug is given in postdiscontinuation follow-up. | 3.3 |
| D3 | T1 p73, row 4 (Approximate Duration (days)) | This header row gets no hierarchy level (null), because it does not tell any two columns apart. "Relative day within a cycle" is level 4. | Give Duration level 4 and Relative day level 5, following the printed order. | 3.4 |
| D4 | T1 p73, row 2 (Cycle) | The row is typed as cycle, although its last two cells read "Short-Term Follow-Up" and "Long-Term Follow-Up", which are periods, not cycles. | Type the row other, or treat the follow-up cells as period labels. |
3.4 |
| D5 | T1 p73, subtitle | The subtitle "Perform procedure as indicated." is kept only in table_metadata.notes. It is not an annotation. |
Emit it as a table-scope footnote with a synthesised marker on header row 1. | 3.5 |
| D6 | T2 p78 | The extension-period schedule is typed track with track_label "Extension period". It covers patients who continue treatment after study completion and has its own visits (501–5XX, 901). |
Type it main_soa, as an independent schedule, with no track label. |
4.1 |
| D7 | T2 p78, rows 9–10 | The dosing text for Fulvestrant and LY2835219 is spread across both Day 1 and Day 15 (columns 4–5), matching the ruled merged cell. | Anchor the text to Day 1 only (column 4). | 4.2 |
source_note, de-duplicated by text. Rows with two references (e.g. "Section 7.1 / Attachment 4") were split into separate notes. The synthesised markers pr1–pr20 are numbered once across the study, so "Section 10.3" is pr9 in both tables. Each location has method: synthesized and the marker is also in the row's annotation_markers.source_note. Footnote p ends with "Refer to Section 9.4.1.1.2." but also explains, so it stays a footnote.annotation_text_source.method: visual_transcription on notes a, b and j (see 5.2).pdftotext -bbox and matched X tokens with ^[Xx][*a-zA-Z0-9]?$. I binned each token to a column using the raster column boundaries (269.2 / 297.2 / 323.2 / 346.6 / 377.2 / 413.2 / 471.7 / 534.8 / 606.8 pt on T1), then grouped tokens by row y. On T1 this gave 30 bbox mark rows (p73: 13, p74: 13, p75: 4). They match the 30 marked activity rows cell for cell. The only difference is expected: tokens sitting on a column border (x ≈ 295 at the 4|5 border, x ≈ 350 at the 6|7 border) belong to merged cells whose internal rule is missing in the raster. They were spread across the span rather than binned. On T2, the X tokens at x = 377 (column 4) and x = 524 (column 6) match the visual read. Mechanical and visual reads disagree in no cell.source_note (m), plus 19 protocol-reference source_notes (pr1–pr19).The Procedure Category column is a vertically merged label column. To keep the grouping, each category became a level-0 organisational row with no marks, and its procedures are level 1. The row positions of these grouping rows are placed before their first child, so they are not separate printed rows. Three rows print their name across the merged Category+Procedure cells and carry marks: Survival Information, Adverse Event Collection/CTCAE Grading, and Concomitant Medications (with analgesics). These are standalone level-0 activities. Indentation came from the column position, with category cells bold except "Health Outcomes", which is printed in regular weight. This is recorded as indentation_method: visual_estimate on every row. The category "Medical History" and the procedure "Medical History" (row 11) are distinct printed cells, and both were kept.
| rows | span | evidence |
|---|---|---|
| 8 Informed Consent Form signed (Xᵠ) | 4:5 | no 297 pt rule in the row band |
| 20–23 Tumor measurement, Radiologic imaging, Bone Scintigraphy, X-ray/CT/MRI (Xᶜ/Xᵇ/Xⁱ/Xʲ) | 4:5 | no 297 pt rule |
| 25 Adverse Event Collection (Xᶠ), 26 Concomitant Medications (X) | 6:7 | no 346.6 pt rule (rows 28–29 and 32–36 do have it, so their Cycle 1 marks are split) |
| 38 Fulvestrant Therapy, 39 LY2835219 Therapy (dosing text with ᵍ) | 6:11 | only rules at 269.2 / 323.2 / 606.6 pt in the band. The cell runs to the table edge (D2) |
Shaded empty merged cells (for example Cycle 1 in the tumor rows) carry no value and were not emitted.
The rows are: Study period (epoch, level 1, name synthesised), Cycle (cycle, 2), Visit (visit, 3), Approximate Duration (days) (other, null), and Relative day within a cycle (study_day, 4). Header cells are merged as follows: Baseline 4:5, Patients on Study Treatment 6:9, Postdiscontinuation Follow-Up 10:11. Cycle "BL", Visit "0" and Duration "28" span 4:5, and Cycle 1, Visit 1 and Duration 28 span 6:7. The Relative-day cells in columns 10–11 are shaded and blank, and were emitted as empty grid cells.
The table is typed track with track_label "Extension period". The title limits it to the extension period, which footnote a says "begins after study completion and ends at the end of trial". It has its own visit numbers (501–5XX, 901) and its own columns. This fits the type definition's example of a post-study schedule for participants continuing treatment.
annotation_text_source.method: visual_transcription with a note explaining the repair:None. Every annotation has at least one marker_location, and each such marker also appears in the target row's or cell's annotation_markers. I checked this by script. Every marker printed in either table has a printed definition.
annotation_text_source.method = visual_transcription: T1 notes a, b and j (see 5.2).marker_locations[].method = synthesized: every pr1–pr20 location, anchored to the row whose Protocol Reference cell prints it.activity_name_source.indentation_method = visual_estimate: every activity in both tables (level from the category-column layout, see 3.2).proximity, text_match or unresolved locations. No structure_method values: the header rows all have printed labels, except the epoch row name, which is flagged as synthesised.method. Merged spans come from the raster rules (§1d).