`ram_blowup_regression` is generating invalid proofs which then fail to verify

[TomAFrench]

In the latest noir sync #12445, we switched over ram_blowup_regression to use the blake3 hash rather than sha256. This executes correctly but then we end up generating an invalid proof. The CI log for this is below.

CI Log

11:01:01 FAILED (http://ci.aztec-labs.com/f90c524e26b48f3e): SYS=ultra_honk FLOW=prove_then_verify barretenberg/acir_tests/run_test.sh ram_blowup_regression (10s) (code: 1)
11:01:01 Command: 29975aaf2857168b SYS=ultra_honk FLOW=prove_then_verify barretenberg/acir_tests/run_test.sh ram_blowup_regression
11:01:01 Commit: https://github.com/AztecProtocol/aztec-packages/commit/aae0b015672c6376218b8f88048a98749b64216d
11:01:01 Env: REF_NAME=sync-noir CURRENT_VERSION=0.77.1 CI_FULL=0
11:01:01 Date: Thu Mar  6 10:57:33 UTC 2025
11:01:01 Sys: ARCH=amd64 CPUS=128 MEM=493Gi HOSTNAME=sync-noir_amd64
11:01:01 
11:01:01 10:57:33 bb command: verify
11:01:01 10:57:33   --proof_path: /dev/fd/62
11:01:01 10:57:33   --vk_path: /dev/fd/63
11:01:01 10:57:33   --verbose: true
11:01:01 10:57:33   --scheme: ultra_honk
11:01:01 10:57:33   --crs_path: /home/aztec-dev/.bb-crs
11:01:01 10:57:33   --oracle_hash: poseidon2
11:01:01 10:57:33 Initialized BN254 prover CRS from memory with num points = 0
11:01:01 10:57:33 bb command: write_vk
11:01:01 10:57:33   --scheme: ultra_honk
11:01:01 10:57:33   --bytecode_path: ./target/program.json
11:01:01 10:57:33   --output_path: -
11:01:01 10:57:33   --verbose: true
11:01:01 10:57:33   --crs_path: /home/aztec-dev/.bb-crs
11:01:01 10:57:33   --oracle_hash: poseidon2
11:01:01 10:57:33 using cached grumpkin crs of size 262145 at: "/home/aztec-dev/.bb-crs/grumpkin_g1.dat"
11:01:01 10:57:33 bb command: prove
11:01:01 10:57:33   --scheme: ultra_honk
11:01:01 10:57:33   --bytecode_path: ./target/program.json
11:01:01 10:57:33   --output_path: -
11:01:01 10:57:33   --verbose: true
11:01:01 10:57:33   --crs_path: /home/aztec-dev/.bb-crs
11:01:01 10:57:33   --oracle_hash: poseidon2
11:01:01 10:57:33 using cached grumpkin crs of size 262145 at: "/home/aztec-dev/.bb-crs/grumpkin_g1.dat"
11:01:01 10:57:33 Initialized Grumpkin prover CRS from memory with num points = 65537
11:01:01 10:57:33 Initialized Grumpkin prover CRS from memory with num points = 65537
11:01:01 10:57:36 created circuit
11:01:01 10:57:36 Constructing DeciderProvingKey
11:01:01 10:57:36 Finalized circuit size: 378185
11:01:01 10:57:36 allocating polynomials object in proving key...
11:01:01 10:57:36 populating trace...
11:01:01 10:57:36 Builder failure when we have real witnesses!
11:01:01 10:57:36 created circuit
11:01:01 10:57:36 Constructing DeciderProvingKey
11:01:01 10:57:36 Finalized circuit size: 378185
11:01:01 10:57:36 allocating polynomials object in proving key...
11:01:01 10:57:37 time to construct proving key: 575 ms.
11:01:01 10:57:37 using cached bn254 crs of size 33554433 at "/home/aztec-dev/.bb-crs/bn254_g1.dat"
11:01:01 10:57:37 populating trace...
11:01:01 10:57:37 Initialized BN254 prover CRS from memory with num points = 524289
11:01:01 10:57:37 time to construct proving key: 534 ms.
11:01:01 10:57:37 using cached bn254 crs of size 33554433 at "/home/aztec-dev/.bb-crs/bn254_g1.dat"
11:01:01 10:57:37 Initialized BN254 prover CRS from memory with num points = 524289
11:01:01 10:57:39 created oink proof
11:01:01 10:57:39 starting sumcheck rounds...
11:01:01 10:57:40 completed 19 rounds of sumcheck
11:01:01 10:57:42 executed multivariate-to-univariate reduction
11:01:01 10:57:42 computed opening proof
11:01:01 10:57:43 Sumcheck verification failed.
11:01:01 10:57:43 Proof verification failed
11:01:01 10:57:43 FAILED (10s)
11:01:01 FAILED (http://ci.aztec-labs.com/f90c524e26b48f3e): SYS=ultra_honk FLOW=prove_then_verify barretenberg/acir_tests/run_test.sh ram_blowup_regression (10s) (code: 1)

I've set this test to be skipped by CI here for the time being:

aztec-packages/.test_patterns.yml

Lines 25 to 29 in f13b729

	# Sumcheck is failing for some reason
	- regex: "barretenberg/acir_tests/run_test.sh ram_blowup_regression"
	skip: true
	owners:
	- "U04LLT331NK" # Tom

cc @ledwards2225 as barretenberg is generating a proof which it then ends up unable to verify.