NCT04004988 — SoA extraction uncertainty report

Prompt 3.8.1, single pass. Source: NCT04004988_soa.pdf (PDF pages 1-3 = document pages 9-11 per PAGEMAP.md). No protocol markdown available.

Decisions needed (3)

#	where	call made	alternative	detail
D1	p.11, general note "If multiple procedures take place at the same time point..." (marker g2)	Kept as one table-wide note, anchored to header row 1 for traceability only	Bind the note to the seven activities it names (Safety 12-lead ECG, Supine Vital Signs, PK Sampling, Clinical Laboratory Tests, Immunogenicity, Blood glucose monitoring, Pharmacogenetic Sample)	4
D2	p.10, header row 2	Row named "Study Day" (synthesised); the printed "Procedure" in its label cell is read as the heading of the activity-name column	Use the printed "Procedure" as the property name	3
D3	p.10, header row 1, column 15 ("ED^c")	ED recorded as a value of the Epoch row, with an empty study-day cell below	Treat ED as a visit (early-discontinuation visit) rather than an epoch value	3

Recorded, not open (7)

§2 / taxonomy "one table spanning both pages": the page-11 reprint of the header is one table across a page break, not a continuation. Header rows were de-duplicated.
§2: main_soa. Periods 1 and 2 share one printed day grid for the same participants (a crossover), so they were not split into tracks.
§5: L = 1 label column (Procedure). The first data column is position 2 and data columns run 2-15. The Comments column (16) is a notes column and is not in the grid.
§6 notes column: the 12 non-empty Comments cells became footnote annotations with synthesised markers n1-n12, each bound to its own row (method: synthesized).
§6 abbreviations: the abbreviation block (ADA, AE, CRU, D/d, ECG, ED, PK, TE ADA) produced no annotations. No term is printed as a marker, and binding by word overlap is forbidden.
§6 header-cell footnotes: marker b sits on the study-day cells D-1, D1 and D36 (±1) (cols 3, 4, 14). Marker c sits on the ED cell (row 1, col 15).
§6: n6 "See Appendix 2 for details. Day 1 predose sample is for baseline only." both points and explains, so it is typed footnote rather than source_note. n10 contains Section 9.4.6 and 9.7 references inside explanatory text, so it stays a footnote.

1. Per table

Table 1: "Study Schedule Protocol I8F-MC-GPGS" - Type main_soa, document pages 10-11. - 14 data columns (2-15): D-28 to D-2, D-1, D1, D2, D3, D4, D5, D6, D7, D8, D15, D21, D36 (±1), ED. - 2 header rows and 20 activity rows (rows 3-22). The table is flat: every row is a level-0 activity with marks (indentation_method: assumed_flat). - Activity rows per page: - p.10: 17 rows (rows 3-19). - p.11: 3 rows (rows 20-22). - p.9: 0 rows. It is outside the declared range: it carries only the section heading "2. Schedule of Activities", with no table content. - 280 activity_schedule cells, 90 of them non-empty. - Cell values that are timing text rather than marks were transcribed literally: "0 hour", "Predose", "Predose, 12", "Predose, 8, 12", "24, 36", and the hour values 24/48/72/96/120/144/168/336/480. They were not split or converted to X.

2. Merged-mark decisions

No merged marks in the body. Every body cell has its own vertical rules; checked on a 250 dpi render.
Header row 1 "Periods 1 and 2 Study Days – at least 35 days washout between Day 1 doses" is merged across columns 3:14. Its right boundary is the rule between D36 (±1) and ED, confirmed on the render.

3. Synthesised

Property names: "Epoch" for row 1, where the label cell is empty. "Study Day" for row 2, where the label cell reads "Procedure" (see D2).
ED placement: see D3.
Annotation markers: n1-n12 (Comments column, one per row). g1 and g2 are the two unmarked general notes below the table.

4. General notes

g1 ("Note: All sampling times ...") and g2 (order of procedures) are table-scope.
Each has one schedule_property location on row 1 with method: synthesized, following the §6 table-scope convention. The marker is deliberately not placed on any element's annotation_markers.
g2 names specific procedures, so it could be bound to them instead (D1).

5. Mechanical mark-check

The PDF has a text layer; it is not glyph-spread.
Every mark and timing token was binned to the nearest header column centre using pdftotext -bbox on pages 10-11.
The result matches the visual read cell for cell, with no disagreements.
Notable check: the Immunogenicity X on p.11 sits under D15 (x ≈ 437.5). In the plain -layout text dump it appeared shifted left, and the bbox check confirms D15.

6. Annotation text integrity

Not glyph-spread, so no reconstruction was needed.
Each Comments note was read whole from its rule-bounded cell. Each cell sits beside exactly one activity row; no note spans several rows.
No annotation text is contained in another.

7. Low-confidence calls

D2 and D3 (header naming and ED typing).
Footnote b describes Period 2 omissions. Its scope is recorded only on the cells where it is printed (D-1, D1, D36 header cells). It was not extended to the individual activities it names (pregnancy test, physical examination, clinical laboratory).

8. Orphan risk

None. Every annotation has at least one marker location, and every marker is defined in the source.

9. Method provenance

indentation_method: assumed_flat on all 20 activities.
method: synthesized on the marker locations of n1-n12 (activity_name), and of g1 and g2 (schedule_property row 1).
No unresolved locations.