From the bench: how we audit, what we look for, and what we keep learning.
August 1, 2026
A two-sided vault only earns when both sides are matched, yet it pays that yield across every share on a side. One unmatched deposit siphons rewards from the stakers who generated them.
Read the note →July 29, 2026
Bug bounties settle duplicates by timestamp, so they pay for speed rather than understanding. Cheap generated submissions made that worse, and submission fees made researchers pay for the privilege. Why Cimora spends its capacity on scoped engagements instead.
Read the note →July 16, 2026
A batched Merkle accumulator seals each subtree by snapshotting a storage root that the insert loop only writes back after it exits. When a subtree fills partway through a call, the snapshot captures the previous call's root and the newest leaf is lost.
Read the note →July 5, 2026
A pool derives a note's spend key differently on its withdraw path and its transfer path, so marking the note spent on one never blocks the other. The same balance can be drained twice, and every single-path test still passes.
Read the note →June 23, 2026
A privileged setter rewrites a vault's fee destination and emits nothing, so the change leaves no log for an indexer to catch. The state is correct and the audit trail never existed.
Read the note →June 12, 2026
A view helper promised a redeemable amount that the redeem path then rejected. On getters that round one way while the action rounds another, and why an honest zero beats a confident revert.
Read the note →