Board decisions issued

p. 25. This is not a ruled table but an indented leader-dot breakdown list; no column headings are printed, so header_rows is empty. Four columns were inferred from the horizontal positions of the figures: col 1 holds the most deeply indented values (380, 26, 14), col 2 holds 435, and col 3 holds the outermost values (1,065 and 691). Text above the list ("...dered during fiscal year 1996." and the lead-in "A breakdown of Board decisions follows:") is running narrative/caption and was omitted. Horizontal rules under 1,065 (double rule) and under 691 (single rule) are decorative underscores, not data. A small stray dash-like mark appears to the right of the "Supplemental / 26" line, in the area of col 3; it is almost certainly a scanning speck or a broken leader dot, not a value, and was transcribed as an empty cell. The crop is cut off at the bottom: the breakdown clearly continues (representation-case decisions etc. are missing), and no total/footnote block for the list is visible. All figures read cleanly.

p. 26. Continuation page: no column headings are printed on this crop, so header_rows is empty. The crop also begins mid-row-group — the first visible label ("Determinations in jurisdictional disputes") is the last detail line of a group whose parent (and any Total / "Contested decisions" lines) is on the preceding page. This is a "stepped" table: it has no printed column rules, and a value's horizontal position encodes the indent level of its row label. I read four distinct numeric positions on this page and have mapped them to four columns, left to right: (1) the deepest detail level, holding 15, 2, 49, 202, 3, 0, 0, 182, 188, 4; (2) "Representation decisions" 253; (3) "Other decisions" 3; (4) "Noncontested decisions" 374. Caveat: positions 2, 3 and 4 are separated by roughly 75-95 px each while position 1 sits about 140 px left of position 2, so the step widths are not uniform, and the label indents do not line up with this reading — "Other decisions" is printed at essentially the same left margin as "Noncontested decisions" (both near flush left) yet its value sits one step to the left of 374, while "Representation decisions" is printed about one em further right than "Other decisions" and its value sits one further step left again. A transcriber who treats "Representation decisions" and "Other decisions" as siblings would merge columns 2 and 3 into a single column, giving three numeric columns instead of four. I kept the four positions I can actually see on the page rather than normalising to the apparent hierarchy, but this is the main uncertainty in the grid. A short horizontal rule is printed across the rightmost numeric position only, between the "Noncontested decisions" row and the "Unfair labor practice" row; it carries no label or value, so it is not transcribed as a row. "374" is set in bold; the other values are roman. The three rows below that rule are unindented relative to the other deepest-level rows and sum to 374 (182 + 188 + 4), consistent with their being the breakdown of "Noncontested decisions" despite the intervening rule. Likewise 2 + 49 + 202 = 253 and 3 + 0 + 0 = 3, so the printed subtotals on this page are internally consistent. All digits are legible; there are no footnote markers, dollar/percent signs or punctuation in any cell. The right edge of the crop falls immediately after "374", so if the table carries a further total column to the right of that position it would be cut off, though no such column shows any value on this page.