Twee-en-dertig jaar

[6 Aug 2026] Discovery item DEMO-11, written because it is the strongest single humanising fact in the model and it needed no new evidence at all - it is her own fourteen birth dates, already in the model, put in a row beside the land dates already in the model. The site is a record of ground and had no way to say this, because a deed has no column for it.

Written up together with the impossible pair the sibling-spacing check found the same hour (13 May and 20 Nov 1816, 191 days apart). The page states it rather than choosing a date, and reasons about WHICH KIND of error it must be: the count of fourteen is corroborated by the vernoemingspatroon test and by SRC-077's thirteen survivors, so dropping a child would break both, and a wrong date is far likelier than a wrong link. LEAD-0108.

The Meiringspoort day carries [TO VERIFY] in both languages because it rests on a tertiary account (LEAD-0107); the sequence around it is on the model's own dates and is sound.

[6 Aug 2026] kind set to 'anecdote' after the enum check refused 'portrait', which I had invented. It is the least wrong of the five allowed values for a page about one person, and it is not right: this is not an anecdote. The enum has no value for a portrait and probably should, but adding one is a schema decision with renderer consequences and not something to slip in beside a story. Noted rather than done.

[12 Aug 2026] BOLD PASS, both languages. 0 and said it still had lots of bold words and phrases, on a file the voice check had scored 0/3/0 and closed. Rule 4 counts bold carrying a whole SENTENCE and is blind to bolded PHRASES, so a speckled page reads as clean. checks.py now measures DENSITY - bolded words as a share of the page - against the S035 benchmark's 1.1%, plus an en/af asymmetry check, and this story was on that list. Out: dates, extents, survey records, deed and folio numbers, place names and, in S005, whole bolded sentences. STORY-VOICE.md names those exactly, and the benchmark bolds NOTHING in its narrative. NO WORD, FACT, HEDGE, [TO VERIFY], CITATION OR LINK CHANGED - the edit removed emphasis markers. verify_af passing.

[13 Aug 2026] VOICE PASS, both languages. af 4/0/0 -> 0/0/0; en was already reported at 0/0/0 AND WAS NOT CLEAN.

THE CHECK UNDER-REPORTS ENGLISH NARRATION, and this page is the proof. All four faults were present in BOTH languages. The Afrikaans wrote 'hierdie projek' and 'hierdie model', which checks.py matches on the bare phrase; the English wrote 'this project CHECKS its own tree', 'this project TAKES 1859 as its hinge', 'the vernoemingspatroon test this project RAN', and 'the witness rules in this MODEL BOTH TURN on that date' - and the English patterns are verb-bound, r'this project (?:holds|cannot|has never)' and r'this model (?:holds|carries)', so none of them matched. A zero on the English side of this check is not evidence of anything. Fixed in both: 'a tree like this one can be tested against itself', '1859 is where the history turns', 'the vernoemingspatroon test run across the whole sibship'.

THE SCOPE RULE LEAKED INTO THE PROSE AGAIN, the same fault removed from S028 yesterday: '[S016] and the witness rules in this model both turn on that date' explained 6's witness boundary to a reader who came for a woman's fourteen births. The S016 cross-reference is kept, the rule is gone. 'This site is largely a record of ground' -> 'This history is'. 'The page says so rather than choosing one without saying so' -> 'neither is quietly preferred here', which says the same thing about the evidence instead of about the page. The closing note stops saying the births are 'recorded in the model' and 'the same rows'.

FACTS RE-CHECKED AGAINST THE RECORD AND ALL HOLD: fourteen children; first Johanna Jacomina 14 Dec 1808 and last Hendrik Johannes c.1840 are the earliest and latest birth dates on PER-000002's children; sixteen at the first and about forty-eight at the last; and the 191 days between 13 May 1816 and 20 Nov 1816 is exact.

FLAGGED, NOT FIXED, THE SAME ONE AS S028: 'She was fifteen' at the October 1807 marriage. PER-000002 carries birth_edtf 1792 with no month and a baptism of 25 Nov 1792, which makes her fourteen at the wedding and turning fifteen the following month. S023 and S028 agree with each other, so nothing on the site contradicts, and the record holds a year rather than a date.

A REPAIR WORTH RECORDING: a partial string match left the Afrikaans closing sentence ungrammatical, 'staan almal op haar eie bladsy afgelees word'. verify_af passed it, because it compares facts and not sense. Same failure as the S003 Afrikaans in the first batch. Read the sentence, do not trust the gate.

[13 Aug 2026] TWO AGES CORRECTED, from the register itself. The baptism of 25 Nov 1792 that SRC-073 had only ASSERTED is now SEEN on the image, and the parents are confirmed as PER-000120 and PER-000121. Frans settled the reading

She married in October 1807 aged FOURTEEN, not fifteen. A baptism on 25 November 1792 puts her fourteenth birthday five weeks after the wedding at the earliest. The opening now gives the baptism rather than a bare year, and names Swellendam as the marriage place, which UNI-000001 already carried.

The other ages on the page are unaffected and were re-checked: sixteen at the first child (14 Dec 1808), about forty-eight at the last (c.1840), and thirty-two years between them. S028 carried the matching error, 'sixty-six' at her death, and is corrected to sixty-five in the same pass.

THE REGISTER IS NOT YET AN SRC ROW. Citation, transcription and the wider findings are in ../the private research folder/2026-08-13 - Graaff-Reinet baptism register 1792, Bladz 1-3.md. The ark is 3:1:3Q9M-CSV8-D9KN-V, an image ark. verify_af passing.