NCT01797120 — Uncertainty report (PDF_TO_JSON_PROMPT v3.8.1)

Decisions needed (6)

#	where	call made	alternative	detail
D1	p.69, banner above table, header row 1 (marker tb1)	The bold banner "TREATMENT MUST BEGIN ≤ 7 WORKING DAYS FROM RANDOMIZATION" is kept as a footnote (synthesized marker tb1) attached to the Phase header row.	Treat the banner as part of the table title only, or drop it as a non-table heading.	3.1
D2	p.69, header rows 1-2, columns 7-9	The column 7-9 header cells span both header rows; their text is recorded once in row 1 (Phase) and the row-2 (Visit) cells are left empty.	Repeat the same text in row 2 as well, so every column has a Visit label.	3.1
D3	p.69, header row 1, column 7	Row 1 is typed epoch although column 7's value (End of Induction Phase End of Treatment) is an end-of-phase visit rather than a phase.	Type row 1 as other, or model column 7's header as a visit.	3.1
D4	p.69, row 8 (Vital Signs), column 2	The Pre-Study cell is kept literally as "X (includes Ht)" (the qualifier stays inside the mark).	Record the mark as "X" and move "(includes Ht)" into a separate note.	3.2
D5	p.69, header row 1, column 7 (marker pr2)	"Section 5.6" printed under "End of Treatment" is removed from the header label and recorded as a cross-reference note (pr2) on that header cell.	Keep "Section 5.6" as part of the header label text.	3.3
D6	p.71	The table's page range is 69-71 because footnotes k-o, *, ^ and £ print on page 71, which has no activity rows.	Declare the table as pages 69-70 (grid pages only).	3.1

Recorded, not open (6)

§2/type defs: page 70 reprints the header and continues the same untitled table, so it is one table spanning pages, not a continuation.
§1b: the header and property rows reprinted on page 70 were de-duplicated.
§5: one label column (L = 1), so the first data column is position 2 and the data columns are 2-9.
§6: the inline "(see Appendix C)" on Concomitant Meds was removed from the activity name and emitted as source_note pr1.
§6: the header-cell markers *, ^, £ and o sit on the specific schedule_grid cells where they are printed, not on the property rows.
§4: the table is flat. Every row is a level-0 activity with indentation_method: assumed_flat, and none are grouping headers.

1. Tables

One table: Table 01, main_soa, titled "Study Parameters" (section 7). It covers document pages 69-71. - The schedule has 8 data columns in positions 2-9: Pre-Study; Cycle 1 Day 1; Cycle 1 Day 15; Day 1 Each Subsequent Cycle; Every 12 weeks; End of Induction/End of Treatment; Continuation Phase; Follow-Up Off Therapy. - There are 2 header property rows and 20 activities, with 71 scheduled cells and 21 annotations. - Activity rows per page: p.69 has 15 (rows 3-17), p.70 has 5 (rows 18-22) and p.71 has 0. Page 71 holds only footnotes k, l, m, n, o, *, ^ and £, so it has no activity rows. It was not skipped. - There are no other SoA tables in the excerpt.

2. Merged marks

Header: row 1, columns 2-6, "Induction Phase (fulvestrant + everolimus or placebo for a maximum of 12 cycles)" is a horizontally merged cell (merged_cell_range "2:6").
Header: columns 7-9 are merged vertically across header rows 1-2 (see D2).
Body: no merged or spanning marks and no arrows.

3. Details

3.1 Header structure

Row 1 "Phase": the name is synthesized because the label cell is empty. Type is epoch, level 1.
Row 2 "Visit": the name is synthesized because the label cell reads "Procedure", which is the heading of the activity column. Type is visit, level 2.
The banner is captured as footnote tb1 (D1).
Page range: see D6.

3.2 Qualified mark

Vital Signs, Pre-Study: an "X" with footnote a, with "(includes Ht)" printed below it. It is kept literally as "X (includes Ht)" (D4).

3.3 Synthesized markers

tb1: the banner, placed on schedule_property row 1 with method: synthesized.
pr1: "see Appendix C", on activity row 7.
pr2: "Section 5.6", on the header cell at row 1, column 7.

All three locations carry method: synthesized, and each marker is also listed in that element's annotation_markers.

4. Mechanical mark-check

Layer type: the PDF has a text layer and a vector grid, so it is not image-based.
Method: I ran pdftotext -bbox and set column x-centres from the header labels (about 244, 311, 382, 450, 521, 592, 654 and 717 pt). I then matched X tokens with ^[Xx][*a-zA-Z0-9]?$ and assigned each to the nearest column and row y-band.
Result: the bbox matrix matches the visual read on 200-dpi and 100-dpi renders cell for cell, on both pages. There were no disagreements.
Footnote superscripts: these were read from the render and cross-checked against their positions in the pdftotext -layout output.

5. Annotation text integrity

Glyph spacing: the text layer is not glyph-spread, so no words needed rebuilding.
Footnote bounds: the footnotes a-o, *, ^ and £ are separate paragraphs, each with a printed marker, so their start and end were clear.
Overlap: footnote l repeats the activity label "Study Drug Compliance (Pill Diary)" but no two annotations share text.
Source quirks: "HbsAg" in the row 18 label is transcribed as printed; the footnotes spell it "HBsAg".

6. Low-confidence calls

Row 2 type: row 2 mixes visits (Pre-Study, C1D1, C1D15) with recurring occasions ("Day 1 Each Subsequent Cycle", "Every 12 weeks"). It is typed visit as the best overall fit.
Column 7 content: the column-7 header holds the alternative "End of Induction Phase End of Treatment", which is kept as printed.

7. Orphan risk

None. - Every marker that appears in the table is defined on pages 70-71. - Every annotation has at least one marker location. - Footnote o is on both the Follow-Up header cell (row 1, column 9) and the Imaging Scans cell (row 11, column 9).

8. Method provenance

indentation_method: assumed_flat is set on all 20 activities because the table is flat.
method: synthesized is set on the locations of tb1, pr1 and pr2.
There are no unresolved locations and no raster, visual-read or proximity methods.