chore(bb): refresh stale pinned Chonk IVC inputs (fixes nightly debug build)

[AztecBot]

Problem

The Nightly Debug Build (barretenberg-nightly-debug-build.yml) has been failing on next for several days.

• Kickoff source (aztec-claude mirror): https://github.com/AztecProtocol/aztec-claude/actions/runs/27084944508
• Real aztec-packages run: https://github.com/AztecProtocol/aztec-packages/actions/runs/27054717437

Every step of the debug build passes except one test:

FAILED: barretenberg/cpp/scripts/run_test.sh bbapi_tests ChonkPinnedIvcInputsTest.AllPinnedFlows (code: 1)

Root cause

The pinned Chonk IVC inputs are stale. barretenberg/cpp/scripts/chonk-inputs.hash = 209dde8e69a27c9f, last bumped 2026-05-21 (2ab01bb). Multiple circuit-affecting feat!: changes have merged to next since then with no accompanying ci-refresh-chonk.

In the failing flow the 10 accumulated circuits prove fine until circuit 10 (hiding_kernel), where the pinned precomputed VK no longer matches the circuit built from current next:

Pub inputs offset mismatch: 669 vs 673
Commitment mismatch: Q_M / Q_C / ... / SIGMA_* / ID_* / LAGRANGE_* / Q_ECC_OP_QUEUE
Does the precomputed vk match with the one generated during accumulation? false
...
Assertion failed: (kernel_return_data_match)
Reason: kernel_return_data mismatch: proof contains { 0x1832..., 0x0 }
        but kernel_calldata commitment is { 0x01, 0x02 }

The hiding-kernel public-input layout grew by 4 (669 → 673), so the frozen VK is wrong and the databus / kernel_return_data consistency check asserts. This assert is BB_ASSERT_DEBUG, compiled out of release/default builds — so only the debug nightly exercises it, which is why regular PR CI stayed green while the protocol circuits drifted. This is fixture staleness, not a code regression; the fix is to regenerate and re-pin the inputs.

Why this wasn't already fixed

A prior automated attempt, #23744 (2026-05-30), diagnosed the same issue but never ran CI: it was a draft labeled ci-refresh-chonk but missing ci-draft, and CI only runs on drafts that carry ci-draft (.github/workflows/ci3.yml line 44). The refresh post-action therefore never executed, and the stale-draft bot auto-closed it on 2026-06-04. The pin stayed stale and the nightly kept failing.

Fix

This PR triggers the standard refresh path with both required labels: ci-refresh-chonk (regenerate + re-pin) and ci-draft (so CI actually runs on the draft). CI regenerates and uploads fresh pinned inputs, proves the smallest refreshed flow in C++ and bb.js, and pushes back the updated chonk-inputs.hash (refresh commit carries --ci-skip).

The trailing-newline change to chonk-inputs.hash is just a placeholder diff for CI to overwrite; the final diff should be a single-line hash bump. Do not merge until the ci-refresh-chonk job has pushed the regenerated hash and the pinned-input smoke test is green. A maintainer can also reproduce locally with barretenberg/cpp/scripts/chonk_inputs.sh update (see the chonk-inputs skill).

---

Created by claudebox · group: slackbot

[AztecBot]

Pinned Chonk inputs refreshed to 9ac5fb8d1bb1c372. The update commit includes a head-commit --ci-skip marker so the automatic follow-up run does not repeat CI.

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.