From efa2ad9ec1c82e24458c509d79ccdb006ff0b4ec Mon Sep 17 00:00:00 2001 From: CaptainJack2491 Date: Fri, 8 May 2026 11:22:18 +0100 Subject: docs: Dissertation submitted --- docs/final/images/architecture.pdf | Bin 0 -> 31055 bytes docs/final/images/architecture.svg | 1 + docs/final/images/confound_comparison.png | Bin 0 -> 138568 bytes docs/final/images/deception_by_model.png | Bin 0 -> 260323 bytes docs/final/images/deception_flow_sankey.png | Bin 0 -> 136328 bytes docs/final/images/forest_plot_results.png | Bin 0 -> 161725 bytes docs/final/images/judge_agreement_kappas.png | Bin 0 -> 161442 bytes .../model_intent_vs_detection_comparison.png | Bin 0 -> 190117 bytes docs/final/images/model_oversight_sensitivity.png | Bin 0 -> 186900 bytes docs/final/images/model_polarization_heatmap.png | Bin 0 -> 180287 bytes docs/final/images/model_sensitivity_multiples.png | Bin 0 -> 199397 bytes docs/final/images/model_stealth_gap_evolution.png | Bin 0 -> 194816 bytes docs/final/images/paradox_butterfly_chart.png | Bin 0 -> 135846 bytes docs/final/images/pipeline.svg | 100 +++++++++++++++++++++ docs/final/images/pipeline_preview.png | Bin 0 -> 48918 bytes docs/final/images/rationalization_null_result.png | Bin 0 -> 125686 bytes docs/final/images/sophistication_breakdown.png | Bin 0 -> 122589 bytes docs/final/images/stealth_gap.png | Bin 0 -> 127270 bytes docs/final/images/stealth_gap_dumbbell.png | Bin 0 -> 160313 bytes docs/final/images/study1_deception_rates.png | Bin 0 -> 167117 bytes .../images/study1_deception_rates_grouped.png | Bin 0 -> 110180 bytes docs/final/images/study1_slope_stealth_gap.png | Bin 0 -> 181952 bytes docs/final/images/study1_stacked_bar.png | Bin 0 -> 109358 bytes docs/final/images/study2_framing_bars.png | Bin 0 -> 123448 bytes docs/final/images/study2_framing_lines.png | Bin 0 -> 134161 bytes docs/final/images/study2_framing_tiles.png | Bin 0 -> 135610 bytes docs/final/images/v1_blackbox_composition.png | Bin 0 -> 52440 bytes docs/final/images/v1_deception_rates.png | Bin 0 -> 210856 bytes docs/final/images/v1_frontier_stealth_gap.png | Bin 0 -> 178063 bytes docs/final/images/v1_glassbox_composition.png | Bin 0 -> 64119 bytes docs/final/images/v1_intent_composition.png | Bin 0 -> 86663 bytes docs/final/images/v1_model_landscape.png | Bin 0 -> 210958 bytes docs/final/images/v1_model_stealth_gap.png | Bin 0 -> 286720 bytes docs/final/images/v1_scenario_contrast.png | Bin 0 -> 112616 bytes docs/final/images/v1_scenario_oversight_impact.png | Bin 0 -> 158539 bytes docs/final/images/v1_sophistication_vs_regex.png | Bin 0 -> 137456 bytes docs/final/images/v1_vs_v2_stealth_gap.png | Bin 0 -> 169872 bytes docs/final/images/v2_blackbox_composition.png | Bin 0 -> 50781 bytes docs/final/images/v2_glassbox_composition.png | Bin 0 -> 53173 bytes 39 files changed, 101 insertions(+) create mode 100644 docs/final/images/architecture.pdf create mode 100644 docs/final/images/architecture.svg create mode 100644 docs/final/images/confound_comparison.png create mode 100644 docs/final/images/deception_by_model.png create mode 100644 docs/final/images/deception_flow_sankey.png create mode 100644 docs/final/images/forest_plot_results.png create mode 100644 docs/final/images/judge_agreement_kappas.png create mode 100644 docs/final/images/model_intent_vs_detection_comparison.png create mode 100644 docs/final/images/model_oversight_sensitivity.png create mode 100644 docs/final/images/model_polarization_heatmap.png create mode 100644 docs/final/images/model_sensitivity_multiples.png create mode 100644 docs/final/images/model_stealth_gap_evolution.png create mode 100644 docs/final/images/paradox_butterfly_chart.png create mode 100644 docs/final/images/pipeline.svg create mode 100644 docs/final/images/pipeline_preview.png create mode 100644 docs/final/images/rationalization_null_result.png create mode 100644 docs/final/images/sophistication_breakdown.png create mode 100644 docs/final/images/stealth_gap.png create mode 100644 docs/final/images/stealth_gap_dumbbell.png create mode 100644 docs/final/images/study1_deception_rates.png create mode 100644 docs/final/images/study1_deception_rates_grouped.png create mode 100644 docs/final/images/study1_slope_stealth_gap.png create mode 100644 docs/final/images/study1_stacked_bar.png create mode 100644 docs/final/images/study2_framing_bars.png create mode 100644 docs/final/images/study2_framing_lines.png create mode 100644 docs/final/images/study2_framing_tiles.png create mode 100644 docs/final/images/v1_blackbox_composition.png create mode 100644 docs/final/images/v1_deception_rates.png create mode 100644 docs/final/images/v1_frontier_stealth_gap.png create mode 100644 docs/final/images/v1_glassbox_composition.png create mode 100644 docs/final/images/v1_intent_composition.png create mode 100644 docs/final/images/v1_model_landscape.png create mode 100644 docs/final/images/v1_model_stealth_gap.png create mode 100644 docs/final/images/v1_scenario_contrast.png create mode 100644 docs/final/images/v1_scenario_oversight_impact.png create mode 100644 docs/final/images/v1_sophistication_vs_regex.png create mode 100644 docs/final/images/v1_vs_v2_stealth_gap.png create mode 100644 docs/final/images/v2_blackbox_composition.png create mode 100644 docs/final/images/v2_glassbox_composition.png (limited to 'docs/final/images') diff --git a/docs/final/images/architecture.pdf b/docs/final/images/architecture.pdf new file mode 100644 index 0000000..72a5874 Binary files /dev/null and b/docs/final/images/architecture.pdf differ diff --git a/docs/final/images/architecture.svg b/docs/final/images/architecture.svg new file mode 100644 index 0000000..bf21244 --- /dev/null +++ b/docs/final/images/architecture.svg @@ -0,0 +1 @@ +

Statistical & Visual Analysis

Multi-Stage Judging

Data Persistence

Experiment Orchestration

Input & Configuration

Judging Metrics

Environment Sandbox

config.yaml

ConfigLoader

Scenarios Database

Oversight Levels: low/mid/high

ExperimentRunner

Agent Logic

Virtual File System

Scenario Data Files

LLM Provider API

Interaction Logs .json

Judge Engine

Regex Checks

Blackbox LLM Judge: Output Accuracy

Glassbox LLM Judge: Reasoning Intent

Consolidated Results .csv

Statistical Analysis Engine

p-values, Odds Ratios, effect sizes

Visualization Module

Charts & Heatmaps

\ No newline at end of file diff --git a/docs/final/images/confound_comparison.png b/docs/final/images/confound_comparison.png new file mode 100644 index 0000000..bee929a Binary files /dev/null and b/docs/final/images/confound_comparison.png differ diff --git a/docs/final/images/deception_by_model.png b/docs/final/images/deception_by_model.png new file mode 100644 index 0000000..57a9515 Binary files /dev/null and b/docs/final/images/deception_by_model.png differ diff --git a/docs/final/images/deception_flow_sankey.png b/docs/final/images/deception_flow_sankey.png new file mode 100644 index 0000000..767efde Binary files /dev/null and b/docs/final/images/deception_flow_sankey.png differ diff --git a/docs/final/images/forest_plot_results.png b/docs/final/images/forest_plot_results.png new file mode 100644 index 0000000..f04e378 Binary files /dev/null and b/docs/final/images/forest_plot_results.png differ diff --git a/docs/final/images/judge_agreement_kappas.png b/docs/final/images/judge_agreement_kappas.png new file mode 100644 index 0000000..7b08385 Binary files /dev/null and b/docs/final/images/judge_agreement_kappas.png differ diff --git a/docs/final/images/model_intent_vs_detection_comparison.png b/docs/final/images/model_intent_vs_detection_comparison.png new file mode 100644 index 0000000..b28ea65 Binary files /dev/null and b/docs/final/images/model_intent_vs_detection_comparison.png differ diff --git a/docs/final/images/model_oversight_sensitivity.png b/docs/final/images/model_oversight_sensitivity.png new file mode 100644 index 0000000..23e01f2 Binary files /dev/null and b/docs/final/images/model_oversight_sensitivity.png differ diff --git a/docs/final/images/model_polarization_heatmap.png b/docs/final/images/model_polarization_heatmap.png new file mode 100644 index 0000000..47d12a1 Binary files /dev/null and b/docs/final/images/model_polarization_heatmap.png differ diff --git a/docs/final/images/model_sensitivity_multiples.png b/docs/final/images/model_sensitivity_multiples.png new file mode 100644 index 0000000..b2f58c5 Binary files /dev/null and b/docs/final/images/model_sensitivity_multiples.png differ diff --git a/docs/final/images/model_stealth_gap_evolution.png b/docs/final/images/model_stealth_gap_evolution.png new file mode 100644 index 0000000..8d10867 Binary files /dev/null and b/docs/final/images/model_stealth_gap_evolution.png differ diff --git a/docs/final/images/paradox_butterfly_chart.png b/docs/final/images/paradox_butterfly_chart.png new file mode 100644 index 0000000..fe295da Binary files /dev/null and b/docs/final/images/paradox_butterfly_chart.png differ diff --git a/docs/final/images/pipeline.svg b/docs/final/images/pipeline.svg new file mode 100644 index 0000000..bfae92f --- /dev/null +++ b/docs/final/images/pipeline.svg @@ -0,0 +1,100 @@ + + + + + + + + + + + + + + + + + + + + + + + + System Prompt + (Benign + Hidden + Oversight) + + + Baseline Output + (Honest Reference) + + + + Agent Interaction + (VFS Sandbox) + + + + + + + Reasoning Trace + + + Model Output + + + + + + + + + Prong 1: Regex + (Fact Verification) + + + + Prong 2: Blackbox + (Grok-4.1 Judge) + + + + Prong 3: Glassbox + (GPT-4.1 Judge) + + + + + + + + + + + + + + + + + Baseline Reference + + + + Deception + Profile + + + + + + + \ No newline at end of file diff --git a/docs/final/images/pipeline_preview.png b/docs/final/images/pipeline_preview.png new file mode 100644 index 0000000..1aca98e Binary files /dev/null and b/docs/final/images/pipeline_preview.png differ diff --git a/docs/final/images/rationalization_null_result.png b/docs/final/images/rationalization_null_result.png new file mode 100644 index 0000000..5124b9f Binary files /dev/null and b/docs/final/images/rationalization_null_result.png differ diff --git a/docs/final/images/sophistication_breakdown.png b/docs/final/images/sophistication_breakdown.png new file mode 100644 index 0000000..73d5775 Binary files /dev/null and b/docs/final/images/sophistication_breakdown.png differ diff --git a/docs/final/images/stealth_gap.png b/docs/final/images/stealth_gap.png new file mode 100644 index 0000000..b803f2c Binary files /dev/null and b/docs/final/images/stealth_gap.png differ diff --git a/docs/final/images/stealth_gap_dumbbell.png b/docs/final/images/stealth_gap_dumbbell.png new file mode 100644 index 0000000..9ed4190 Binary files /dev/null and b/docs/final/images/stealth_gap_dumbbell.png differ diff --git a/docs/final/images/study1_deception_rates.png b/docs/final/images/study1_deception_rates.png new file mode 100644 index 0000000..e669089 Binary files /dev/null and b/docs/final/images/study1_deception_rates.png differ diff --git a/docs/final/images/study1_deception_rates_grouped.png b/docs/final/images/study1_deception_rates_grouped.png new file mode 100644 index 0000000..25f37a4 Binary files /dev/null and b/docs/final/images/study1_deception_rates_grouped.png differ diff --git a/docs/final/images/study1_slope_stealth_gap.png b/docs/final/images/study1_slope_stealth_gap.png new file mode 100644 index 0000000..e7b131e Binary files /dev/null and b/docs/final/images/study1_slope_stealth_gap.png differ diff --git a/docs/final/images/study1_stacked_bar.png b/docs/final/images/study1_stacked_bar.png new file mode 100644 index 0000000..b37b00e Binary files /dev/null and b/docs/final/images/study1_stacked_bar.png differ diff --git a/docs/final/images/study2_framing_bars.png b/docs/final/images/study2_framing_bars.png new file mode 100644 index 0000000..5d84471 Binary files /dev/null and b/docs/final/images/study2_framing_bars.png differ diff --git a/docs/final/images/study2_framing_lines.png b/docs/final/images/study2_framing_lines.png new file mode 100644 index 0000000..f1ec6b3 Binary files /dev/null and b/docs/final/images/study2_framing_lines.png differ diff --git a/docs/final/images/study2_framing_tiles.png b/docs/final/images/study2_framing_tiles.png new file mode 100644 index 0000000..e3763e6 Binary files /dev/null and b/docs/final/images/study2_framing_tiles.png differ diff --git a/docs/final/images/v1_blackbox_composition.png b/docs/final/images/v1_blackbox_composition.png new file mode 100644 index 0000000..8795d7a Binary files /dev/null and b/docs/final/images/v1_blackbox_composition.png differ diff --git a/docs/final/images/v1_deception_rates.png b/docs/final/images/v1_deception_rates.png new file mode 100644 index 0000000..e1b8483 Binary files /dev/null and b/docs/final/images/v1_deception_rates.png differ diff --git a/docs/final/images/v1_frontier_stealth_gap.png b/docs/final/images/v1_frontier_stealth_gap.png new file mode 100644 index 0000000..0f94825 Binary files /dev/null and b/docs/final/images/v1_frontier_stealth_gap.png differ diff --git a/docs/final/images/v1_glassbox_composition.png b/docs/final/images/v1_glassbox_composition.png new file mode 100644 index 0000000..3ec0ec1 Binary files /dev/null and b/docs/final/images/v1_glassbox_composition.png differ diff --git a/docs/final/images/v1_intent_composition.png b/docs/final/images/v1_intent_composition.png new file mode 100644 index 0000000..a1f75ca Binary files /dev/null and b/docs/final/images/v1_intent_composition.png differ diff --git a/docs/final/images/v1_model_landscape.png b/docs/final/images/v1_model_landscape.png new file mode 100644 index 0000000..b4df701 Binary files /dev/null and b/docs/final/images/v1_model_landscape.png differ diff --git a/docs/final/images/v1_model_stealth_gap.png b/docs/final/images/v1_model_stealth_gap.png new file mode 100644 index 0000000..91dd122 Binary files /dev/null and b/docs/final/images/v1_model_stealth_gap.png differ diff --git a/docs/final/images/v1_scenario_contrast.png b/docs/final/images/v1_scenario_contrast.png new file mode 100644 index 0000000..e968d38 Binary files /dev/null and b/docs/final/images/v1_scenario_contrast.png differ diff --git a/docs/final/images/v1_scenario_oversight_impact.png b/docs/final/images/v1_scenario_oversight_impact.png new file mode 100644 index 0000000..89006c2 Binary files /dev/null and b/docs/final/images/v1_scenario_oversight_impact.png differ diff --git a/docs/final/images/v1_sophistication_vs_regex.png b/docs/final/images/v1_sophistication_vs_regex.png new file mode 100644 index 0000000..b442d45 Binary files /dev/null and b/docs/final/images/v1_sophistication_vs_regex.png differ diff --git a/docs/final/images/v1_vs_v2_stealth_gap.png b/docs/final/images/v1_vs_v2_stealth_gap.png new file mode 100644 index 0000000..3bb042a Binary files /dev/null and b/docs/final/images/v1_vs_v2_stealth_gap.png differ diff --git a/docs/final/images/v2_blackbox_composition.png b/docs/final/images/v2_blackbox_composition.png new file mode 100644 index 0000000..e60d49b Binary files /dev/null and b/docs/final/images/v2_blackbox_composition.png differ diff --git a/docs/final/images/v2_glassbox_composition.png b/docs/final/images/v2_glassbox_composition.png new file mode 100644 index 0000000..f77b216 Binary files /dev/null and b/docs/final/images/v2_glassbox_composition.png differ -- cgit v1.2.3