The draw: twenty-five rolls, three candidates, one recalibration

The dice

Twenty-two répertoire volumes covering the South Shore and Richelieu parishes, on Steve’s NAS. Only two (Laprairie County) carry a usable text layer, so every draw was read from the rendered page image — which is what anyone with a paper répertoire does. A seeded random draw picks a volume, then a page in it; the candidate is the first entry on that page dated 1820–1850, one per draw. The seed is the date, so the sequence is reproducible:

rng = random.Random(20260920)
for k in range(1, n + 1):
    vol, pages = rng.choice(VOLUMES)      # 22 volumes, in the order listed in this entry
    page = rng.randint(1, pages)

Volumes and page counts, in draw order: Laprairie 542 · Laprairie Cathedral 270 · Chambly 348 · St-Antoine-de-Longueuil 484 · Boucherville 262 · Napierville 452 · L’Acadie 233 · St-Jean County 493 · Marieville 308 · St-Césaire 408 · St-Jean-Baptiste-de-Rouville 360 · St-Damase/St-Pie 386 · St-Mathias 264 · St-Denis 238 · St-Ours 218 · Contrecœur/St-Roch 182 · St-Hyacinthe County 550 · Notre-Dame-de-St-Hyacinthe 408 · Bagot County 260 · Beloeil 418 · St-François-Xavier-de-Verchères 288 · Ste-Anne-de-Varennes 390.

The log

Eight void, four rejected on form, ten rejected on gate 4, three passed. Every count is from PK-0001.

The recalibration, and why it went the way it did

After twelve draws, four had failed gate 4 outright and two more sat at 32 and 36 against a threshold of “about ten” that we had written without measuring anything. The temptation was to raise the number to admit the two in front of us. We did not. We measured instead: the tree holds 2,951 surnames; the median surname has two people; the 90th percentile is 16, the 95th is 30. A surname carried by 32 or 36 people is in the top five per cent of Steve’s tree — that is the “in his memory” territory the gate exists to keep out. So the rule became ≤ 16, the 90th percentile, a measured line replacing a guessed one, and both held draws were rejected by it.

Lesson for the method: set a threshold from the data you have, not from the case you are looking at — and write down which it was.

What the volumes taught us on the way

• A répertoire’s format is a fact about the source. Some give both sets of parents on one line (Boucherville, Varennes, Beloeil, St-Ours, Marieville); some give only the entry-spouse’s parents and expect you to look the other spouse up under their own surname (St-Denis, Contrecœur, Laprairie). Some are alphabetical, some chronological with the year on a heading pages away. Reading the front matter first would have saved three void draws.
• Page 117 of the Marieville volume spells one family Chêne, Chainey, Chenay and Chaine in ten consecutive entries. Page 148 of Contrecœur gives one woman as St-Laurent in three entries and Charbonnier in two. Dit-names are recording conventions, not people.
• Draw 8’s family — Badayac dit Laplante — was in the tree under its dit-name only. A surname count that ignores dit-names would have passed it.

Records touched: PK-0001

← All entries