Model-1 … Model-5
A planned five-model cohort. The initialization seeds were locked before any results were seen. Training never started, so no model results exist.
Status
Model-1 to Model-5 were never trained
The seed manifest is experimental-design evidence, not a model result. The evaluation tables for Model-1..5 are empty because the models were never produced, not because they performed badly.
| Planned model | Initialization seed | Training status | Evaluation status |
|---|---|---|---|
| Model-1 | 668309859 | NOT TRAINED | No model result exists |
| Model-2 | 1658757876 | NOT TRAINED | No model result exists |
| Model-3 | 315576475 | NOT TRAINED | No model result exists |
| Model-4 | 1003306941 | NOT TRAINED | No model result exists |
| Model-5 | 744367755 | NOT TRAINED | No model result exists |
There is deliberately no empty results table on this page. An empty results grid would invite the reading that these models were measured and scored nothing. They were not measured, because they do not exist.
Why training never started
- status recorded as WIP, not admitted, training not started
- three intentional RED tests had not been closed
- driver and runtime were not installed on the new machine
- a 10 GiB boot disk was insufficient for safe preparation
- long training and the population were never started
None of these reasons is a result about model quality. They are operational and readiness constraints recorded in the handoff.
Why a locked seed manifest still matters
Locking the initialization seeds before any results were seen is the part of this record that carries scientific value. It means that if this cohort is ever trained, the initializations were not chosen after the fact to produce a desired outcome.
That is experimental-design evidence. It is not evidence about how the models would perform, and a seed locked is not a model trained.
An independent initialization cohort — varying only init_seed under the same architecture, data, data order and training contract — is one of the outstanding requirements listed under Limits.
Source: Scientific Evidence Inventory v1.0 (evidence cutoff 30 August 2026), §8.6, §11.