ONTO · Chronology · 2026-10-07
| when | what |
|---|---|
| 41 years | Twenty years of observing the world, then twenty years of collecting scientific data. Out of it came an atlas of the boundaries of the sciences and its formulas. Both were made for people, before AI. |
| about 300 sessions | Working sessions with GPT turned the atlas into a formal scientific base. This base became the canon, GOLD. |
| 2024 | The work was presented to AI, and ONTO was founded to keep it as a standard. |
| 600–700 sessions | ontostandard.org was built as a working stand: scope filters, scoring, before-and-after comparisons. The difference in answer quality with and without the canon was stable on every model tried. |
| January 2026 | First package: onto-standard 1.0 on PyPI. |
| February–April 2026 | The prompt layer: the canon placed in the system prompt of production models, answers scored before and after. Reports CS, DS, ER, F, FO of this period are in the archive. |
| 2026-04-09 | A formal proposal submitted to the Ministry of Digital Development of Uzbekistan. |
| April 2026 | First runs on small open models in the lab. |
| Turning point | The prompt layer worked: on every model tried, the canon made answers better, and the effect reproduced again and again. This finding started the ONTO lab. The whitepaper of March 2026 still said that the method "does not require new models, new training data, or new architectures" and needs discipline "externally enforced". We then chose discipline in the weights, so that it is the model's own: Why we left the prompt layer. |
| The lab, 1,000+ sessions | Research moved into the lab. From session 372 every working session is logged as one line: what was measured, the verdict, the lesson. Thresholds are written before the data, instruments are tested on known answers first, and failed directions stay in the log. |
| October 2026 | Discipline in the weights: stage Beta accepted on two seeds, transferred to OLMo, reproduced from scratch on another GPU. One pre-registered stop condition fired on independent questions; it is published as such. |