Kaart se keersy, B.326/1833 (verso), September 1833

Verso subdivision register: E771/1937, diagram 4239/1937, Kleinbotterkraal Ptn 1, 8393.9332 morgen, deed 1937-239-13528. FILE NOT ON DISK. [31 Jul 2026] ID COLLISION REPAIRED. This diagram was assigned DOC-0092 in the 30 Jul inbox, but that id was afterwards issued again to a TANAP corpus inventory. Ids are permanent and never reused, so the diagram takes a NEW id here and the old one keeps whatever it now holds. The five SG diagrams of the Sleutelfontein group had no document rows at all until now, which is why their sources sat unreferenced and their files sat on disk with nothing pointing at them.

[1 Aug 2026, found by a source-file audit] THE SCAN WAS ON DISK ALL ALONG, under the wrong id. It was filed as DOC-0091/DOC-0092, which are the TANAP estate inventories MOOC8/36.35 and MOOC8/32.46, so two ids were each naming two different documents and this row's source_file sat empty. The content had been read and is in the parcel and ownership rows; what was missing was the image on the page. Renamed to its own id and wired up. The audit that caught it compares every file under sources/ against what documents.csv claims, and is worth re-running whenever a batch of scans is filed.

[5 Aug 2026, verso-date sweep] §8: this verso carries endorsements LATER than its front, so it takes its own content dates instead of inheriting. The subdivision register on this back is E771/1937, diagram 4239/1937, deed 1937-239-13528. survey_edtf stays 1833-09 (the sheet's own survey); the 1937 title is the newest ink and the gate reads the max, so it now reads 1937 and is withheld.

[5 Aug 2026] doc_type corrected from SG_diagram to SG_diagram_verso. This row's sg_diagram_no already carried the '(verso)' suffix, so the scan gate read it correctly, but render.DOC_TYPE_LABEL maps the two types to different public labels - the site called this back of a sheet a 'Surveyor-General diagram' - and every per-type count double-counted a front. The 31 Jul verso batch predates the convention 8 records.

[5 Aug 2026] publish_ok=yes. THE SHEET WAS OPENED AND READ, AND IT NAMES NOBODY. One line: survey record E771/1937, diagram 4239/1937, Kleinbotterkraal struck through and renumbered Portion 1, 8393.9332 morgen, deed 1937-239-13528. That is the whole of it: portion numbers, portion NAMES, extents in morgen and square roods, diagram numbers, survey-record numbers, deed numbers and dates. Not one person appears on it. The 1936 scan cut-off exists because a JPEG is opaque to every text check in leak_tests.py, so a document's DATE is the only proxy available for whether the sheet names a living person. On a deduction table that proxy measures the wrong thing: the table is a record of GROUND, and withholding it withholds the farm's own history to protect nobody. Same decision, same reason and same mechanism as DOC-0081 on 3 Aug, when [FAMILY] I was talking about the old farm diagrams only. The date fields stay as they are, because the row has to state what is on the paper and the earlier inheritance from the front was simply wrong. What changes is that the release is now a decision somebody took after looking at the sheet, which is the only thing publish_ok has ever been for.