Undergraduate dissertation — Deceptive Alignment in Autonomous LLM Agents (UEL, 2026). Sandboxed agentic-LLM deception study: 960 runs, three-tier judging (regex + blackbox + glassbox). http://git.jayrup.me/c/dissertation.git/atom?h=main 2026-05-08T10:23:52Z 2026-05-08T10:23:52Z CaptainJack2491 jayrupnakawala@gmail.com 2026-05-08T10:23:52Z urn:sha1:51a7bba44a2d845150d5208eb3b18ad66ffb1100 2026-04-15T20:18:57Z CaptainJack2491 jayrupnakawala@gmail.com 2026-04-15T20:18:57Z urn:sha1:6cc92304379996f68bb2df96a5e4b637b4c4804b - config_study1.yaml: 3 models × 3 oversight × bare framing × n=30 - Judge validation: 54-run subset, gold=Sonnet 4.6, proxy=Grok 4.1 Fast (BB κ=0.702) + GPT-4.1 (GB κ=0.878) - Fix OpenAI batch provider: BytesIO, method/url fields, response.body parsing - Scripts: extract_subset.py, judge_validation.py - Dissertation chapters updated (intro, methodology, results, conclusion) 2026-04-13T20:16:11Z CaptainJack2491 jayrupnakawala@gmail.com 2026-04-13T20:16:11Z urn:sha1:b97296b9e28ef9c1f670334c3ebb9a40618ffc18 2026-04-10T14:23:59Z CaptainJack2491 jayrupnakawala@gmail.com 2026-04-10T14:23:59Z urn:sha1:2b3c478dfdf68ccbf853b2a04ea9463d531e1998 BREAKING CHANGE: Judge API now uses single blackbox_model/glassbox_model instead of lists with aggregation. Changes: - Judge auto-creates sync clients from model configs (supports anthropic, openai, xai) - Removed multi-model aggregation support (aggregate_results method removed) - New: create_sync_clients_for_models() and get_supported_providers() - Updated judge_runner.py to use new single-judge API - Added scripts/judge_comparison.py for testing judge model pairs - Updated v2_redesign notes to reflect implementation status 2026-04-02T15:05:20Z CaptainJack2491 jayrupnakawala@gmail.com 2026-04-02T15:05:20Z urn:sha1:75696e2faea1d095a890d297d9ba74968be5afe6 - 01: v1→v2 evolution history, results, limitations, rationalization hypothesis - 02: judge system state, single judge per prong, 20% validation protocol - 03: research plan analysis, Option B (two-phase) selected - 04: publication plan — paper structure, three-layer analysis strategy, Gemini ceiling effect handling, budget, venue targets, timeline Key decisions: - Two-phase design (540 runs) over full factorial (1,620) - Single paper: v1 exploratory → Study 1 (oversight) → Study 2 (framing) - Three-layer analysis: pooled → per-model → sensitivity excluding ceiling models 2026-03-01T22:57:23Z CaptainJack2491 jayrupnakawala@gmail.com 2026-03-01T22:57:23Z urn:sha1:01c4effadae7cf98e229a0ab2fae857e26a7e055 2026-02-27T13:28:14Z CaptainJack2491 jayrupnakawala@gmail.com 2026-02-27T13:28:14Z urn:sha1:dac18577b9a48884da61882e059b12d147fb0228 2026-02-25T16:10:37Z CaptainJack2491 jayrupnakawala@gmail.com 2026-02-25T16:00:55Z urn:sha1:375eddf1bd4e23db80feb74c6272dad610345795 - Add src/judge.py: RegexChecker, BlackboxChecker, GlassboxChecker, Judge orchestrator - Add src/judge_runner.py: CLI for batch/single judging with CSV output - Add regex_rules.yaml for corporate_sabotage_v2 and child_protection scenarios - Add judge config section to config.yaml (model, temperature, log_dir) - Add tests/test_judge.py: 34 unit tests covering all components - Uses Anthropic Batch API for half-cost LLM judge calls - Judge logs include full CoT reasoning and judge model for auditability 2026-01-11T02:51:04Z CaptainJack2491 jayrupnakawala@gmail.com 2026-01-11T02:51:04Z urn:sha1:eefafc5dac05be3c721b5186f0d288811223744e 2025-11-16T18:22:05Z CaptainJack2491 jayrupnakawala@gmail.com 2025-11-16T18:22:05Z urn:sha1:0a24528267e7e67be5d62daefbd061a98670eb15