Today was... hectic. TLDR of the issue: If the logic the sequencer runs (in a very efficient way) isnt 100% equivalent to the Cairo Starknet claims to prove , the prover is there to prevent any deviation from the guaraneed logic (and force reorg). An edge case in a transaction today caused such a discrepency. This edge case is now fixed ofc. TLDR of lessons learned: too soon to give a certain answer. Many people took part today in debugging and restoring operations asap, but we will all ask ourselevs what we can learn here, and conduct asap an internal reflection process to brainstorm on prevention and mitigation action items going forward Feel free to ask whatever though! Including now, when the situation is fresh. The builders and I are here to answer any question. Transperancy matters to us.

The last gain from 2025: On Tue, the team detected and applied a config change that reduced the L1 finality time by 20 minutes. This change was not in S-Two itself, but rather in SHARP - the infrastructure that wraps S-Two for aggregation and submission proofs. Two days ago, I said that "now, when we have an active ecosystem of apps, we can focus our work on the ecosystem demands". This is exactly what happened here - we got feedback that L1 latency times are a pain point. We initiated a review process to determine how we can cut it and discovered several low-hanging fruits, as S-Two is far from being the bottleneck. And changes that are much more meaningful than 20 minutes (though a bit more complicated) are coming soon.

From an infra standpoint, the Second half of 2025 was massive: Stwo, decentralization step, pre confirmations. But even more impressively, for the first time, starknet sustained a considerable demand from externals. (We can take much more, though.) - and is now consistently in the top-5 chains on growthepie in the daily fees paid metric. Some of it reflects a long, dedicated effort that already yielded a significant result (Thanks to S-Two, for the first time, Starknet is sustainable while charging less for transfer than many networks without proving). Other parts of it were stepping stones for next year (StarkWare still runs all Apollo nodes. This will change soon, and preconfirmations ought to do better) Now that we have a real demand, we can finally tune Starknet to it. Therefore, 2026 will be significantly more efficient and impactful in terms of execution. We've done the 0-->1. Now it's time for the 1-->100.