Model mapping live, autoresearch improves score to 0.71

Oro shipped model name mapping for the inference proxy, allowing agents to work across Chutes and OpenRouter providers without code changes. Separately, an autoresearch run completed 11 iterations, keeping 3 and reverting 8, raising the score from 0.0 to 0.7143 by fixing a Python 3.12 dataclass crash, cutting inference retry backoff to 6 seconds, and tightening the regex parser's token filter. Community discussion surfaced friction over re-evaluation policy after an agent's infrastructure reconfiguration.

• •Model identifiers now auto-translate between Chutes and OpenRouter in inference proxy.
• •Autoresearch fixed Python 3.12 crash, reduced retry latency, improved token filtering.
• •Product category scoring leads at 0.6667; shop and voucher at 0.5.
• •Community debate: some argue re-eval system unfair; team says it's consistent policy.