Kaart se keersy, 1319/1896 (verso), 2017

THE DEDUCTION TABLE. Two subdivisions taken off Portion 2, ninety-six years apart: PORTION 3, named BOTER KRAAL OOST and then struck through and renumbered "Ptn. 3", 1732 morgen 24 square roods, diagram 13/1920, transferred 12 July 1920 on transfer 9891; and PORTION 6, 1026.3570 ha, survey record 836/2016, diagram 1570/2016, transfer 136029/2017, marked D.C.

ITS OWN DATE, NOT ITS FRONT'S. 8 says a verso inherits its front's dates because it is physically the same sheet, and that is right for a table drawn with the diagram - but this one accumulated for 121 years and its latest entry is a 2017 transfer, so title_edtf is set to 2017 and the gate reads that.

RELEASED WITH publish_ok=yes UNDER 11: a deduction table is a list of portion numbers, names, extents, diagrams and deed numbers. It is a record of GROUND, not of people, and the 1936 proxy measures the wrong thing on it. The sheet was opened and read before the flag was set.

[14 Aug 2026] publish_ok WITHDRAWN, AND THE LEAK TEST WAS RIGHT. It was set to "yes" under 11 - a deduction table is a record of ground, not of people, so the 1936 proxy measures the wrong thing on it. leak_tests L14 refused anyway: this sheet is a PDF whose only content date is 2017, and L14 applies a bare 1936 cut-off to PDFs WITHOUT consulting publish_ok. L15 does consult it, which is why the forty-odd post-1936 versos already released here pass - every one of them is a .jpg.

THE FLAG WAS NOT FORCED THROUGH. A backstop that a flag can override is not a backstop, and this one exists because a PDF can carry unbounded text that no check in this repo reads. NOTHING IS LOST THAT MATTERS: the deduction table is transcribed in full in the note above - both subdivisions, their diagrams, extents, transfer numbers and dates - and the front of the sheet, dated 1896, publishes normally. LEAD-0195 records the asymmetry for a decision made calmly rather than during a deploy.