| R1. Finish inside a stated time budget, with an answer or an explanation. | slow and silent models |
|---|---|
| R2. Keep working when a named model is withdrawn. | stale identifiers |
| R3. Do not pay again, in time, for a model that is known to be down. | repeated failure |
| R4. Tell the user what went wrong, in a form that supports a decision. | ambiguous 429s |
| R5. Never spend more than the owner allows, even under concurrency; give each caller a share. | open wallet |
| R6. Treat the model's answer as untrusted and validate structured answers. | unusable output |
| R7. Send a private prompt only where the provider says it is not kept or trained on. | privacy |
AI & evaluation · Updated 2026-11
Lodestone’s seven requirements and the failure each answers
Table 1 in “Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit”: 7 rows, 2 columns.
Cite this chart
Extrafemi Research (2026). Lodestone’s seven requirements and the failure each answers [Chart]. Data from Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit. https://research.extrafemi.com/data/lodestone--data-1