fix(bb): refresh pinned Chonk inputs to unbreak nightly debug build --ci-refresh-chonk

[AztecBot]

Problem

The Nightly Debug Build has been failing on next. The single failing test is bbapi_tests ChonkPinnedIvcInputsTest.AllPinnedFlows, which aborts on a debug-only assertion during the in-prover sanity-check verify:

HypernovaFoldingVerifier: Failed to recursively verify Sumcheck ...
Assertion failed: (kernel_return_data_match)
Reason: kernel_return_data mismatch: proof contains
  { 0x183227397098d014dc2822db40c0ac2ecbc0b548b438e5469e10460b6c3e7ea4, 0x0 }
but kernel_calldata commitment is { 0x1, 0x2 }

(Chonk::process_public_inputs_and_consistency_checks, barretenberg/cpp/src/barretenberg/chonk/chonk.cpp:169, guarded by BB_ASSERT_DEBUG.)

Why only the nightly debug build catches it

• The databus-continuity check is BB_ASSERT_DEBUG (and the accompanying debug_incoming_circuit VK comparison is verbose-only), so it is compiled out of release builds. Regular release CI — which runs the same ChonkPinnedIvcInputsTest.AllPinnedFlows test (barretenberg/cpp/bootstrap.sh:276) — stays green. Only the debug nightly trips the assert.
• The proof itself is valid; this is not a proving-correctness regression on next.

Root cause: stale pinned inputs

• Last green nightly debug build was run Pw/fix end to end test #92 (2026-05-20, 8b2a320), when the pin was 5d17cc1.
• The pinned Chonk inputs were re-pinned to 209dde8e69a27c9f on 2026-05-21 (2ab01bb, barretenberg/cpp/scripts/chonk-inputs.hash). Every nightly debug build since has failed.
• Breaking protocol-layout changes have landed on next since that pin (e.g. feat!: shift remaining protocol-contract addresses to 1-3 feat!: shift remaining protocol-contract addresses to 1-3 #23218, feat!: demote multi_call_entrypoint to non-protocol contract feat!: demote multi_call_entrypoint to non-protocol contract #23197, feat!: add message signing and fallback keys feat!: add message signing and fallback keys #23510, AZIP-16 v5 defaults chore: AZIP-16 v5 default parameter updates #23764). The failing flow uses MultiCallEntrypoint + EcdsaRAccount, directly hit by these. The prior claudebox investigation (fix(bb): refresh pinned Chonk inputs to unbreak nightly debug build --ci-refresh-chonk #23792) captured the matching Pub inputs offset mismatch: 669 vs 673 + Commitment mismatch: Q_M / Q_C / ... VK signature — the unmistakable fingerprint of pinned VKs/databus commitments no longer lining up with current circuits.

Fix

Refresh the pinned Chonk inputs. This PR carries the ci-refresh-chonk label and the --ci-refresh-chonk head-commit marker. CI regenerates the inputs from current next, uploads the new tarball, and pushes the updated barretenberg/cpp/scripts/chonk-inputs.hash. The 0000000000000000 value here is the sentinel that the refresh post-action overwrites.

Important

The ci-refresh-chonk refresh only runs on a non-draft pull_request. claudebox opens PRs as drafts, so a maintainer must mark this PR Ready for review for the refresh CI to run. This is exactly why the earlier attempt #23792 never completed — it sat as a draft for 4 days (no ci workflow ever ran on it) while its base went stale.

Supersedes #23792 (opened against the 2026-06-02 next, now 4 days behind).

---

Created by claudebox · group: slackbot

[AztecBot]

Pinned Chonk inputs refreshed to bf3e35241d1945ae. The update commit includes a head-commit --ci-skip marker so the automatic follow-up run does not repeat CI.