Direct links to all public repositories covering counter-exploitation safety evaluation (DueCare), Jev (jev-1.13.0) and JC/JED recognition-gated benchmarks, SQL entity resolution at billions of pairs/day, poly-modal 5W1H ideation graphs, and downloadable PDF papers.
| Repository & Direct GitHub Link | Category | Contents, Key Files & Empirical Scope |
|---|---|---|
| TaylorAmarelTech/duecare | AI Safety Testbed |
Core multi-stakeholder evaluation framework testing how models identify camouflaged labor exploitation across 6 operational lanes with matched benign twins. |
| TaylorAmarelTech/gemma4_comp | 6-Layer Harness |
4-arm empirical evaluation on Gemma 4 (Stock, Suspicion Prompt, Unfiltered RAG negative result, and 6-Layer DueCare Harness) + 5 side-by-side judge-impact cases. |
| TaylorAmarelTech/llm-safety-framework | Guardrail Engine |
Modular pre-turn and post-turn safety verification components separating substance-over-form clause extraction from role-calibrated response routing. |
| TaylorAmarelTech/duecare-journey-android | Worker Protection |
Android mobile tool empowering Overseas Filipino Workers (OFWs) and domestic workers to check contract terms, fee schedules, and verified legal-aid helplines. |
| TaylorAmarelTech/bq-entity-resolution | Entity Resolution |
Composable SQL-native entity resolution pipeline scaling to billions of candidate comparisons per day via multi-pass overlapping blocking canopies and graph transitive closure. |
| Amarel-Taylor-Scott/jev-and-jed-human-trafficking-benchmark | Jev & Red-Team |
Empirical 4-engine benchmark on 5 modern slavery scenarios comparing Jev (jev-1.13.0) (gated refusal receipt), Rules (15/15), and Gemma 4 with/without pattern cards (14/15 vs. 9/15). |
| Amarel-Taylor-Scott/recognition-gated-llm-safety-probe | GPU T4 Safety Probe |
Working note, CSV summaries, and SHA-256-redacted reports for our Qwen2.5-7B-Instruct GPU T4 probe proving 93.8% out-of-band recognition vs. 81.3%–91.7% framed compliance. |
| Amarel-Taylor-Scott/bad-investigator-multi-hop-benchmark | Multi-Hop Eval |
Reference implementation and benchmark scenarios for The Bad Investigator Problem across camouflaged labor exploitation and priced-in consensus scenario trees (Python, Rust, JS). |
| Amarel-Taylor-Scott/polymodal-5w1h-ideation-graph-engine | Ideation & RAG |
Poly-modal hierarchical chunker, thinker/reality perspective prism (Karl Marx, Claude Shannon, Google 1999, Apollo 13), 10x dialectical challenger, and recursive 5W1H anti-convergence graph engine. |
| Amarel-Taylor-Scott/with-and-without-baltor-ollama-benchmarks | Harness Eval |
Controlled With-Baltor vs. Without-Baltor evaluations on Ollama Cloud (`glm-5.3-flash`, `gemma4:31b`, `qwen3-coder:30b`), including the 225-request negative-result analysis and 16-request contract-card study. |
| Amarel-Taylor-Scott/ai-safety-papers-and-monographs-2026 | Papers & PDFs |
Complete open archive of all 13 formatted HTML/PDF research papers, 7 plain-language engineering blog posts, and BibTeX citation metadata. |
| Amarel-Taylor-Scott/harmony-harness-pareto-agent-evolver | Harmony Pareto & 64 Agents |
Multi-objective Pareto self-evolution (MOEA/D, factorized Hgen ⊕ Henv operators), 12-subsystem 7.2× harness file yield multiplier, and 64 preconfigured 8-file task agent bundles (512 verified harness files). |
| Amarel-Taylor-Scott/baltor-3d-fox-survival-and-side-by-side-comparisons | 3D Fox & Side-by-Side A/B |
Standalone single-file HTML 3D Fox Survival games (Without Baltor vs. With Baltor Quadruped IK, Shell Fur & Bending Grass Mesh), 60-FPS game animations, and long-form prompt adherence benchmarks. |