← AItopiaOrAImageddon?

The impacts

The impact model being built under PLAN.md §14: one proposal per measured effect of a learned system on people — who, how many, how big, dated, with the source and a quote from it — and one line per ruling on it, kept forever in order. The level on the front page reads from these since the night of 2026-09-06: each entry with a severity ruling and a people count bands to a rung through the band table, and the rule in SCALE.md reads the level from them. The rung column shows what the table makes of each entry, or a dash with the reason it cannot band. Any entry can be re-examined when new data turns up. Candidates a researcher looked at and set aside are here too, with the reason, so a lookup answers whether something was considered.

399 proposals: 124 not-found, 123 validated, 72 reject-definition, 60 exclude-attribution, 16 unverified, 4 quote not found. Statuses: validated — the source was refetched by code and the quote is on it; quote not found — refetched, quote absent, to be re-examined; unverified — the source refused the fetch (a bot wall, usually); reject-definition, exclude-attribution, not-found — the researcher's own ruling on a candidate it set aside; struck — the owner. Every status is a line in impacts/rulings.jsonl; so is every severity ruling and every remedy.