The record

Every place that has been read out of the scans. Each number links back to the page it came from.

Where the indexed record goes quiet

A missing catalogue entry does not establish that reporting stopped. These title-derived gaps are checked against broader collection activity to test for a collection-wide scanning cutoff; each individual cause remains unknown.

Ask Jay

Jay answers only from the extracted record. It cannot answer from memory, every number it reports carries the page it was read from, and it will not describe a title-series gap without the collection-wide control and its limits.

Whose effluent was in your water

Treatment plants tied to the nearest river gauge, then ordered by catchment area — which necessarily grows downstream, so the towns sort themselves without any flow routing. Read each list top to bottom the way the water runs.

How accurate is this?

Measured against ground truth a human read off the same scans by hand. Published including the failures — an archive misread at scale is worse than one never read, because the errors look like findings.

What one more document would answer

Eleven million pages cannot be read alphabetically, and reading by subject only serves whoever picked the subject. This orders them by what reading them would unlock — because a question one document away is a question somebody already wanted.

It is also the honest replacement for a progress bar. "You processed 40 documents" is a fact about you. "You made the Grand River answerable, and it had been waiting on one town since 1961" is a fact about the world.

Who decided, and who voted against

A title-keyword census found roughly 13,600 minutes, agendas, hansard and commission-hearing items. The civic parser is promising, but it does not yet have the measurement path's hand-read benchmark. Such records can still be cheap to read: "It was moved by X and seconded by Y that Z. CARRIED." is a form that has barely changed in a century, so a pattern finds it. No model, no GPU, and a person can check it.

The control is the clerk's own tally. That "-16." at the end of a roll was written by someone in the room; if the names don't add up to it, the roll was misread and it says so rather than publishing a division quietly missing two people.

Where the readings disagree

Every claim here keeps its cited archive page. Prose uses sentence evidence; experimental table records retain cell locators but abstain without localized cell proof. When two source-backed readings still disagree, nobody decides between them — both remain visible with page links and crops when available. The evidence check establishes what the cited page says, not which interpretation is correct.

Flagging one tells us people think it is wrong. It does not change the record: an objection with no evidence cannot outrank a sentence on a page, because then somebody would have to judge the objection. To change what is shown, bring a page and a sentence.