| license: other | |
| license_name: heuremen-attribution-no-training | |
| license_link: https://github.com/Heuremen-LLC/indahl.ai/tree/main/code/faultsguards/blob/main/LICENSE.txt | |
| language: | |
| - en | |
| pretty_name: Faults and Guards | |
| tags: | |
| - evaluation | |
| - llm-failure-modes | |
| - measurement | |
| - prompting | |
| size_categories: | |
| - n<1K | |
| # Faults and Guards | |
| Faults is the catalogue of failure modes a language model commits doing measurement work. Guards is the set of procedures built against them. Published by Heuremen LLC as PDFs at https://heuremen.org/faults.pdf and https://heuremen.org/guards.pdf. | |
| `entries.jsonl` has one row per fault or guard: `document` (faults or guards), `number`, `title`, `part` (the family or part it sits in), and `text`, exactly as published. `faults.json` and `guards.json` hold the whole documents with their introductions and sections. The numbers are stable identifiers and do not run in family order. | |
| The Python package that loads this text into prompts, function-calling schemas and an MCP server is at https://github.com/Heuremen-LLC/indahl.ai/tree/main/code/faultsguards. It generates this data from the published sources and tests that it does. | |
| ## Terms | |
| Free to download, read, run and cite with attribution (Heuremen LLC, https://heuremen.org). Use inside a prompt, a tool call, a retrieval step or a review at the time a model answers is permitted. Text and data mining, and use of this text to train, fine-tune or build AI models, are not permitted. This is an express reservation of rights. See LICENSE.txt and https://indahl.ai/terms#ai-training. | |