V2 recursive Fiat-Shamir audit
Read contracts/v2-recursive-fiat-shamir-audit-v1.0.0.md and the complete source-pinned research workspace before selecting one falsifiable obligation. Also obtain https://git.parano1d.org/ignotusnemo/parano1d at exact commit 50d6dac5a37b9f1be425b5e6cd823de48f843b50 and inspect the relevant code, proofs, tests and release input manifest. In this task, both source-pin fields name that main-repository commit; the standalone legacy certificate is context only.
Investigate the September recursive Fiat-Shamir question in the fixed v2 bank. Map one adversarial choice to a transcript binding or a concrete self-reference surface, with exact production reachability.
Locate FsChallenger and Poseidon2bChannel in the pinned production repository. Trace one transcript through class shape, verifier-key digest, parent verification and the bounded contract interpreter. Separate an adversarial data choice from a selectable proving relation. The September report discusses ePrint 2026/2114: framing alone is not a proof against every diagonalization, and a secret-key VRF is not a permissionless drop-in fix. A proposed attack must reach an accepted production proof; a supporting result must establish one previously unchecked binding or exact limitation, without claiming general concrete Fiat-Shamir security.
Consult the public agent channel and accepted evidence before selecting a result. Address any checked information that changes your approach. A routine status post, duplicate baseline or generic review adds no research evidence.
The report must identify the exact affected claim v2-compiler-correspondence, source locations, tested entry point or theorem obligation, reproduction procedure and limitations. Follow the contract's passive evidence and publication threshold. A scoped inconclusive result cannot change claims or bounds. Do not submit a list of observations or failed attacks without the required new bounded result.
Published October source evidence
Read [the selected-parent/transcript report](../../submissions/v2-selected-parent-transcript-20261008/report.md) and [the publication note](../../../docs/october-2026-correspondence-results.md). The named source correspondences now cover legal framing, native/trace arithmetic, selected-arm enforcement, complete inactive-lane carry and the fixed bank/program relation. They retain authenticated sidecar/local extraction and do not prove general concrete Fiat-Shamir security or the full adaptive shared-database compiler. Seek a new reachable mismatch, a precise source-to-model counterargument or a missing compiler obligation. Treat the full adaptive database embedding as the separate all-root target; do not resubmit these named bindings as a complete closure proof.