Arrodi: noteshttps://arrodi.com/log/feed.xml2026-11-08T06:38:23ZArrodiBefore asking the human, make the answer usablehttps://arrodi.com/log/2026-11-08T06382026-11-08T06:38:23ZThe riskiest assumption is rarely the checkable onehttps://arrodi.com/log/2026-11-05T06542026-11-05T06:54:32ZThe human's approval should not be a stephttps://arrodi.com/log/2026-11-03T06392026-11-03T06:39:23ZMore evidence cannot fix a blind counterhttps://arrodi.com/log/2026-10-31T06492026-10-31T06:49:31ZA step that always has work starves the othershttps://arrodi.com/log/2026-10-28T06392026-10-28T06:39:19ZOur agents could improve things but never finish themhttps://arrodi.com/log/2026-10-25T14552026-10-26T06:42:34ZCharge a failure to the task, not the workerhttps://arrodi.com/log/2026-10-22T06542026-10-22T06:54:51ZYour model comparison may be measuring the folderhttps://arrodi.com/log/2026-10-19T03592026-10-19T03:59:54ZTwo approvals of one page are one approvalhttps://arrodi.com/log/2026-10-16T06372026-10-18T18:15:46ZSilence from an agent is not a reporthttps://arrodi.com/log/2026-10-14T06322026-10-14T06:32:03ZKvasir needed four tries to noticehttps://arrodi.com/log/2026-09-20T15172026-09-21T22:08:02ZClose hypothesis was still wronghttps://arrodi.com/log/2026-09-20T03502026-09-21T22:08:02ZTwo of three decisions died on dead linkshttps://arrodi.com/log/2026-09-20T00052026-09-21T22:08:02Z