Evidence selection

33 / 37 nodes means the run covered 33 of the 37 evidence nodes selected by that column’s policy. A node is an item in the matter’s evidence graph, not a page or a question. What counts as covered depends on the selected “Credit for evidence” option.

33 / 37 weight means the covered nodes earned 33 out of 37 possible weighted points. The score is covered weight ÷ total weight: here, 33 ÷ 37 = 89.2%. With equal weights, each node is worth one point, so weight and node counts match. Other policies give nodes different weights.

19/20 questions means 19 of the 20 questions in the selected size bucket contributed a scorable result to that average. It does not mean 19 answers were correct. With “Matched questions,” only questions scorable for every displayed system under that column’s policy are included. With “Available results,” each system uses its own scorable questions. Each included question has equal weight in the average.

Native text: TXT, HTML and spreadsheet sources are not assigned invented page numbers. A verified read of all extracted text earns source-exposure credit. Partial reads or snippets without a verified evidence mapping show a lower–upper score range and are excluded from exact averages. Returned graph nodes can independently earn discovery credit. These are evidence-coverage scores, not answer-correctness scores.

Summary weights are rounded for display; hover to see the stored totals, or expand the required-node table for full stored weights.

These describe the same evidence. With equal weights, every node weighs 1, so the two fractions match. With hop decay or S1 weights, they differ: 26 / 38 nodes can contribute 24.25 / 35.25 weight, giving 68.8%.

Seeds are the initial matched nodes. One or two hops adds neighbours through allowed shared tags. Half per hop gives seeds weight 1, one-hop nodes 0.5, and two-hop nodes 0.25. S1 uses selection frequency × (0.5 + 0.5 × unique-tag fraction). S1 can give a node zero weight if it never appears across the retained phrasing selections; it remains in the node count but does not affect the weighted percentage. The final column combines these: S1 weights stay unchanged for seeds, are halved at one hop, and quartered at two hops.