The team of Lucnica Ensemble sang Vande Mataram during the welcome in Bratislava. This comes at a time when we are marking 150 years of Vande Mataram and recalling its glorious contribution to India’s history and freedom struggle.

We entered AIRA₃ with an ensemble of models in the live competition, and also assessed it with several others post-hoc. The 8th ranked gold medal entry ensemble was a combination of GPT 5.5 (w/ OpenCode) + Claude 4.8 (w/ ClaudeCode). Post-hoc we assessed with Muse Spark 1.2 (w/ MuseCode), which also performed at a Gold Medal level, as well as Muse Spark 1.1 (w/ OpenCode) and GLM 5.2 (w/ OpenCode), both of which achieved Silver Medal level performance. The post-hoc submissions were also graded externally on the same private test set as those made during the live competition.

The paper argues that for any ensemble whose final output must be one of the member models' answers, including routing, voting, and MoA, the accuracy is fundamentally capped by the co-failure rate β: acc ≤ 1 − β. It also shows that the commonly reported mean pairwise error correlation (ρ) is insufficient to characterize β, so low correlation alone does not imply large ensemble gains. When Does Combining Language Models Help? A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents

Certainty is a symptom of running a single model; calibration comes from an ensemble. Tetlock's forecasting tournaments found that hedgehogs (experts with one big theory who filter all evidence through it) predict worse than "foxes" who hold many small, partial models and aggregate them, despite (in fact because of) the hedgehog's greater confidence. That is precisely why ensemble methods beat single models in statistics. A random forest outperforms one deep decision tree, model-averaging beats any single estimator because averaging many independent, individually-mediocre views cancels their uncorrelated errors. The fox is an ensemble; the hedgehog is a single over-committed model; the felt experience of certainty is the sensation of having stopped averaging. You know your forecast is degrading at the exact moment it starts to feel clean. A borrowing rate is a forecast too. Each lending venue is a single model - one utilization curve, one governance regime, pricing its own local noise as if it were the cost of capital. Accept a fixed rate from any single venue and you've trusted a hedgehog: it feels clean, one number from one source, which is the tell. IRIS prices across venues at once. A solver's fixed rate isn't Aave's estimate or Morpho's but it's what remains once each venue's idiosyncratic noise cancels against the others: the systematic cost of capital, no longer padded for any single venue's local shocks. The ensemble, not the hedgehog. The rate that survived aggregation.