The offline, honesty-bound, script-agnostic resume toolkit. Any format in, clean LaTeX/PDF/DOCX out — every writing system first-class (CJK, Cyrillic, Arabic, Hebrew, Devanagari, Greek, Latin…), no subscription, no scraping, no inflation.
Honesty, sycophancy, calibration, and factuality evals for LLMs.
Heuristic signals on whether an OpenAI-/Anthropic-compatible endpoint serves the model it claims — catch model-swapping, quantization & silent context truncation. Zero-dependency Python CLI. Signals, not proof.
Investigating honesty, deception and steering in large language models. Replicating and extending the MASK honesty benchmark on frontier models, working toward internal representation analysis and activation steering for honesty.
ARCHIVED hologram/duplicate. Canonical: https://github.com/szl-holdings/szl-receipt
ARCHIVED hologram/duplicate. Canonical: https://github.com/szl-holdings/szl-provctl
ARCHIVED duplicate/hologram. Canonical: https://github.com/szl-holdings/szl-kernels
To associate your repository with the honesty topic, visit your repo's landing page and select "manage topics."