The OpenRouter campaign · V1
Campaign observation
Sigma5C observed model responses through OpenRouter using a recurring prompt set and rotating task themes. The retained campaign record describes availability and response outcomes under those conditions.
What the campaign was designed to observe
The retained F022 design separated repeatability, breadth, difficulty and domain-focused questions. These are protocol descriptions, not reconstructed campaign outcomes.
| Theme | Declared task design | Observation focus |
|---|---|---|
| Consistency and reliability | Repeat the same prompt set across observation nights. | Compare score variation, response outcomes and timing across repeated conditions. |
| Breadth | Vary prompts across task domains. | Record how responses vary with the question and domain rather than treating a single task as general capability. |
| Difficulty tiers | Use progressively harder prompts, including ambiguous or adversarial cases. | Keep results associated with the declared difficulty tier. |
| Domain-focused work | Concentrate a prompt set on a specialized domain. | Examine responses within that domain and its stated conditions. |
What the records distinguish
| Dimension | Meaning in the retained design |
|---|---|
| Response and verification | Record response outcomes separately from any task-specific scoring or deterministic verification. |
| Repeated observations | Compare variation and drift across repeated prompt conditions. |
| Availability and rate limits | Record unavailable responses and rate-limit events separately from response content. |
| Response timing | Describe timing variation alongside the conditions of the attempted request. |
The public page does not reconstruct a model-by-model campaign outcome table from account activity exports. A listed model, a request entry and a successfully scored response are different records.
Export coverage
The supplied activity exports cover different accounts in different years. Only one account export overlaps this campaign, while the campaign used two routed accounts. Its request totals do not identify campaign membership or successful responses, so they cannot reconstruct the campaign’s recorded successful-call total.
Provider context
Related delivery-route references are maintained separately from this dated observation. A catalog reference does not transfer these results to another model or route.