Reports and write-ups on agent reliability, AI control and model evaluation.https://ebinbt.dev/en-gbhttps://github.com/ebt55/incidentgate/blob/main/docs/verification.mdhttps://github.com/ebt55/incidentgate/blob/main/docs/verification.mdA dated log of what each measurement actually establishes, including the first time a monitor was shown a covert action and what its rationale reveals about how it decided.Sat, 22 Aug 2026 00:00:00 GMTwrite-upGitHub — incidentgatehttps://github.com/ebt55/digital-grimace-scale/blob/main/notes/paper.mdhttps://github.com/ebt55/digital-grimace-scale/blob/main/notes/paper.mdThe preregistered primary test failed and is published as a FAIL; a re-preregistered iteration found an answer-margin channel that survives being trained out of the model’s language.Tue, 18 Aug 2026 00:00:00 GMTreportGitHub — digital-grimace-scalehttps://github.com/ebt55/proofpack/blob/main/docs/KNOWN-GAPS.mdhttps://github.com/ebt55/proofpack/blob/main/docs/KNOWN-GAPS.mdAn engineering audit of my own prototype, priority-ranked, written so a team inheriting the repository knows exactly where the edges are.Mon, 17 Aug 2026 00:00:00 GMTwrite-upGitHub — proofpackhttps://github.com/ebt55/whose-voice/blob/main/REPORT.mdhttps://github.com/ebt55/whose-voice/blob/main/REPORT.mdHidden principals can be recovered from poisoned training data at 12–44% top-1 of 47 candidates in a dense regime, and not at the poison densities real attacks use.Mon, 27 Jul 2026 00:00:00 GMTreportApart Research × Formation Research — Secret Loyalties hackathon