Tutorials

The run began with a plain-language question about the prosecution of William Chaloner: how Newton and the Mint built the case against him, and what the depositions, petitions, letters and printed accounts reveal about how contested it was. The timeline shows what the Royal Mint papers hold, year by year; the shaded window sets this page’s evidence period, 1692 to 1700.

Generating a plan costs cents: two probe retrievals of 50 passages each and one planner call. For this run, planning took about a minute and the planner came to $0.06.

The planner proposed five overlapping sections: Newton’s evidence-gathering campaign, the network of informants and witnesses, Chaloner’s own counter-strategy, the trial and execution, and the contradictions and limits of the surviving record. Each card holds a planned overview, a writing goal and a retrieval prompt, with its own time window. Every field is editable, and sections can be reordered, removed or added.

Nothing has been retrieved at scale yet, so this is the cheap moment to reshape the page. Pressing Run starts the pipeline.

Each section sweeps the corpus with a saturation-driven binary search over time windows: 723 searches across the five sections, all finished within two minutes. A relevance gate then reads every candidate passage and decides whether it bears on the section. The gate is the slow step. It runs as a batch job that can take many minutes, and the run card says so rather than looking stuck.

The sweep surfaced 5,051 candidate passages and the gate kept 1,092 of them. The 3,959 it dropped are not deleted: each is kept in an audit file with the gate’s reason.

A section parks after its gate because that is the first moment the cost of writing it is knowable: its evidence is fixed. Each card shows how many passages were kept, how they have been bundled by date into subsections, and an estimate of what writing them will cost. You release sections one at a time or all at once.

The estimate is an expectation, not a quote. Here Archos’s formula put writing all five sections at about $1.71; the cost audit later recorded $7.16 for synthesis, the largest line on the bill.

The finished Evidence Report is scholarly prose with sentence-level footnotes: 1,688 of them across 27 subsections, drawn from 261 distinct primary sources. Hover over or tap a footnote below to see the quotation it rests on and to highlight that passage in the source, here Chaloner’s printed proposals to Parliament of February 1695.

Switch to the Epistemic view to colour each sentence by its part in the argument: the point, the evidence, the explanation, and the link that carries it on. Where the record is thin, or two documents disagree, the report says so rather than resolving it quietly; one of the five sections is given over entirely to contradictions and copying problems in the surviving papers. By the time the report was compiled, the section titles had been rewritten to fit the evidence found.

This Evidence Report cost $12.70 in total, and the cost audit itemises every stage: synthesis $7.16, the relevance gate $3.45, the temporal sweep $1.81 across 723 searches, and the planner $0.06. Every run compiles to HTML with footnote popovers, an A4 PDF in both the scholarly and the epistemic view, and a Word document with real Word footnotes.