@pvium/p2id-core 0.1.0

package/contracts/PviumZKVerifier.sol

@@ -0,0 +1,2465 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright 2022 Aztec
+pragma solidity >=0.8.21;
+
+uint256 constant N = 2097152;
+uint256 constant LOG_N = 21;
+uint256 constant NUMBER_OF_PUBLIC_INPUTS = 19;
+uint256 constant VK_HASH = 0x0e105995ea345a2ad495a8e7707075ad3c49fce617f1391d13c50cd3b0b823aa;
+library HonkVerificationKey {
+    function loadVerificationKey() internal pure returns (Honk.VerificationKey memory) {
+        Honk.VerificationKey memory vk = Honk.VerificationKey({
+            circuitSize: uint256(2097152),
+            logCircuitSize: uint256(21),
+            publicInputsSize: uint256(19),
+            ql: Honk.G1Point({ 
+               x: uint256(0x0c0fbd2b1460a0b0bccb62d4527f91523b2167690e98976144216c342ddfd54b),
+               y: uint256(0x11544daf1bcdccd74d55d7cb84308b67f818b5565fb5885960ec5f082e8dc820)
+            }),
+            qr: Honk.G1Point({ 
+               x: uint256(0x28dbee3939797ad1b0f4947260120a96095bc6f8a2efc27dc4e7c6f5b07b8a97),
+               y: uint256(0x2f76eb999a1f10bc5ca0c5a2ca944e2d9575859a660b068a35a4cd762258ac13)
+            }),
+            qo: Honk.G1Point({ 
+               x: uint256(0x0db8223a888b42240f95090755c9867ec384f53ea27018038a5931ace533f6b3),
+               y: uint256(0x0a477836a16431cd37acf0750e34e1194c09332d4832390c7ade2fb126473311)
+            }),
+            q4: Honk.G1Point({ 
+               x: uint256(0x2065bcb9724f9ee952a41a41e3977d61b192006edddc00c1772e112c1bc03e48),
+               y: uint256(0x190168907b7d58c8cca95bf235ca510b355c951adecb8fbb7327267ccf52df42)
+            }),
+            qm: Honk.G1Point({ 
+               x: uint256(0x2b9cd7175e6a8085b12aa1fefa7f780b4b59aeaa3d77ad7b0c80256db165dc0a),
+               y: uint256(0x19bd7797f55aad7e70dc26950f398332014eb44a43d7b4d3accb73f2fb81e72a)
+            }),
+            qc: Honk.G1Point({ 
+               x: uint256(0x1d6198f289227dc183297f8bfcb42f8c66abf8e786260d29e0a362189c794a8c),
+               y: uint256(0x17c99fef6cc097e7ef6edbb53602d8973c2a037967895bafc6fe771c885767b6)
+            }),
+            qLookup: Honk.G1Point({ 
+               x: uint256(0x1f21d9f47259d55071f21e0a600cce11be50b6261e9b911ffdac6f59f6215e2b),
+               y: uint256(0x2745f5f6e1b51445152840adeb0495c242b71185d64d7fbcde817feeed67993c)
+            }),
+            qArith: Honk.G1Point({ 
+               x: uint256(0x2fa49e34485452c477e319ce0b98195742b4bdb3a0f75a35f1f98711fb6552d5),
+               y: uint256(0x062d5c1410518684f3e85b7e0d75d228378df2c5cc53df106ea7170de36e2ac2)
+            }),
+            qDeltaRange: Honk.G1Point({ 
+               x: uint256(0x0bea0cfb68dc8f81f2c244a6344319cb27efb8f7109a3ba1dc50ba345f6aa15f),
+               y: uint256(0x1fea9d5d1eda129a40e56346e0367bcb46e82fae3a847eeb595c4a7893051a6c)
+            }),
+            qElliptic: Honk.G1Point({ 
+               x: uint256(0x0000000000000000000000000000000000000000000000000000000000000000),
+               y: uint256(0x0000000000000000000000000000000000000000000000000000000000000000)
+            }),
+            qMemory: Honk.G1Point({ 
+               x: uint256(0x1d8486efda2cbf96cee91c579c809f3d1f404b82d8b9c517d35ddcbcfaea20f1),
+               y: uint256(0x177229d5cabcb16f5e7ca9a56fcd912b345c98d9e2b89a283e0556b185703bdf)
+            }),
+            qNnf: Honk.G1Point({ 
+               x: uint256(0x04a36167f7e1fa03b2d3b5f85f52dddfef2ed6ed1e89eab9c6d6a7807a30c1a5),
+               y: uint256(0x11c5181b7f5dcea80d4c44fce4a065c68ed26049e92f54e337cf59c692f84a4c)
+            }),
+            qPoseidon2External: Honk.G1Point({ 
+               x: uint256(0x0000000000000000000000000000000000000000000000000000000000000000),
+               y: uint256(0x0000000000000000000000000000000000000000000000000000000000000000)
+            }),
+            qPoseidon2Internal: Honk.G1Point({ 
+               x: uint256(0x0000000000000000000000000000000000000000000000000000000000000000),
+               y: uint256(0x0000000000000000000000000000000000000000000000000000000000000000)
+            }),
+            s1: Honk.G1Point({ 
+               x: uint256(0x25944439c5706db4d4bf17d0cd03f71be02afd91f9bf1239399b1dc1a4457cc0),
+               y: uint256(0x1a15f305835907343f73c005ed56ac4eb883d0f52c3fbd09fa9a9498eb510e62)
+            }),
+            s2: Honk.G1Point({ 
+               x: uint256(0x2fed2d66b9d97e6a101d28c6d37936a013103bc6439b6dedeea05476c2a78743),
+               y: uint256(0x16ece6ba7dd5f4dba4c042852c1860d65cfd42e2aba596e056ea0b5862ecaadd)
+            }),
+            s3: Honk.G1Point({ 
+               x: uint256(0x029823cc2b4de546d728162738c23fe5f85654f2f262126c38c1f44a94f6159f),
+               y: uint256(0x1f8fa4b575751192e7e4636d3a804c573f0835bc0d599587b4665d7c16eddc13)
+            }),
+            s4: Honk.G1Point({ 
+               x: uint256(0x15bd3c01ea657ed5332fac7f5d249ed2ab95ff7612998fa2ed97a18e1911aeb2),
+               y: uint256(0x1bc51c2d4fc014f4b3347da10bf7728cf9e5bbbe4a03fbf8eb6def18abd39743)
+            }),
+            t1: Honk.G1Point({ 
+               x: uint256(0x0303f2fae9d426bcacf03a3324f365ab85a56ac0de12398d52a7925ad09a7934),
+               y: uint256(0x200446d603ee01a4aa4dd1aeea063ac302c4d6fade4212ca043cffb5adb17093)
+            }),
+            t2: Honk.G1Point({ 
+               x: uint256(0x2faa880e21816f6e84f9a63c69bd2f641ccf46a3c046edf5bfd8da9445aaef02),
+               y: uint256(0x1b79ad9778d05ded45f962653dd8b2143aff5c962a60fb865635e76fd26bed28)
+            }),
+            t3: Honk.G1Point({ 
+               x: uint256(0x1f890e8812cc36372bc6f2e02f72d826c6735b38b62f4f19476afbb148b46124),
+               y: uint256(0x291180bac307a9a350ed326f0f8c23dcc8efbb2f2b37c00cde6d0f2a0c8babb8)
+            }),
+            t4: Honk.G1Point({ 
+               x: uint256(0x1236e0212395b26cd85777136008cc4caf0066037c702b301becb0e0c35d77f4),
+               y: uint256(0x2bd26f5b3b40d4a7a4d8277693f487d458bd5f784307e93b33de6de026db1366)
+            }),
+            id1: Honk.G1Point({ 
+               x: uint256(0x26d67c272def0127082be77c31e099a9f9811a35712b6bb1b0d86a9da8ba8682),
+               y: uint256(0x2393540006d4df8339c1a01889cd436986995291781382e59ce5363c9d91d60c)
+            }),
+            id2: Honk.G1Point({ 
+               x: uint256(0x12e6fe2df8d9f5b64817faa8f113db27de14e800a1dc2ff37b4838fd18023777),
+               y: uint256(0x018b58768dafe2b28cae54316b30ce28a70f7d257d23b3f85d90c36882e9f911)
+            }),
+            id3: Honk.G1Point({ 
+               x: uint256(0x263c75326cd89fdd00fb443a4b17b32ee882d427d1254bdb65ddcee2ab059a49),
+               y: uint256(0x1ff9f006ee2ea95c6c8b003b6f7f1607e06f5234e666d0e61fb780aad36ed85e)
+            }),
+            id4: Honk.G1Point({ 
+               x: uint256(0x0336650f6db1ba47a4eea770d57476a426fc36b7c33582e94dfcfcc6ad8fdd2c),
+               y: uint256(0x10116c0793907dad66484d1c071dc462133812c5562aafa3452a1a56ece2778e)
+            }),
+            lagrangeFirst: Honk.G1Point({ 
+               x: uint256(0x2a56ce41f6b0be13b9c26747621b821eee81b23a887f299049b14c11e98460d6),
+               y: uint256(0x1aa98f2de3ddda547d8f6de4e725ded5827d6338c78656c0d12ca1aea6ef2c7c)
+            }),
+            lagrangeLast: Honk.G1Point({ 
+               x: uint256(0x14232344c2648e564db86dd72f84cb5435120efb4ba0095da25343c75fe30ddc),
+               y: uint256(0x24206e1fd8e64ba30eb1c6f720748fd02ef8d915a969596faebec5b98bd0614c)
+            })
+        });
+        return vk;
+    }
+}
+
+pragma solidity ^0.8.27;
+
+interface IVerifier {
+    function verify(bytes calldata _proof, bytes32[] calldata _publicInputs) external view returns (bool);
+}
+
+/**
+ * @notice  Library of error codes
+ * @dev     You can run `forge inspect Errors errors` to get the selectors for the optimised verifier
+ */
+library Errors {
+    error ValueGeLimbMax();
+    error ValueGeGroupOrder();
+    error ValueGeFieldOrder();
+
+    error InvertOfZero();
+    error NotPowerOfTwo();
+    error ModExpFailed();
+
+    error ProofLengthWrong();
+    error ProofLengthWrongWithLogN(uint256 logN, uint256 actualLength, uint256 expectedLength);
+    error PublicInputsLengthWrong();
+    error SumcheckFailed();
+    error ShpleminiFailed();
+
+    error PointAtInfinity();
+
+    error ConsistencyCheckFailed();
+    error GeminiChallengeInSubgroup();
+}
+
+type Fr is uint256;
+
+using {add as +} for Fr global;
+using {sub as -} for Fr global;
+using {mul as *} for Fr global;
+
+using {notEqual as !=} for Fr global;
+using {equal as ==} for Fr global;
+
+uint256 constant SUBGROUP_SIZE = 256;
+uint256 constant MODULUS = 21888242871839275222246405745257275088548364400416034343698204186575808495617; // Prime field order
+uint256 constant P = MODULUS;
+Fr constant SUBGROUP_GENERATOR = Fr.wrap(0x07b0c561a6148404f086204a9f36ffb0617942546750f230c893619174a57a76);
+Fr constant SUBGROUP_GENERATOR_INVERSE = Fr.wrap(0x204bd3277422fad364751ad938e2b5e6a54cf8c68712848a692c553d0329f5d6);
+Fr constant MINUS_ONE = Fr.wrap(MODULUS - 1);
+Fr constant ONE = Fr.wrap(1);
+Fr constant ZERO = Fr.wrap(0);
+// Instantiation
+
+library FrLib {
+    bytes4 internal constant FRLIB_MODEXP_FAILED_SELECTOR = 0xf8d61709;
+
+    function invert(Fr value) internal view returns (Fr) {
+        uint256 v = Fr.unwrap(value);
+        require(v != 0, Errors.InvertOfZero());
+
+        uint256 result;
+
+        // Call the modexp precompile to invert in the field
+        assembly {
+            let free := mload(0x40)
+            mstore(free, 0x20)
+            mstore(add(free, 0x20), 0x20)
+            mstore(add(free, 0x40), 0x20)
+            mstore(add(free, 0x60), v)
+            mstore(add(free, 0x80), sub(MODULUS, 2)) 
+            mstore(add(free, 0xa0), MODULUS)
+            let success := staticcall(gas(), 0x05, free, 0xc0, 0x00, 0x20)
+            if iszero(success) {
+                mstore(0x00, FRLIB_MODEXP_FAILED_SELECTOR)
+                revert(0, 0x04)
+            }
+            result := mload(0x00)
+            mstore(0x40, add(free, 0xc0))
+        }
+
+        return Fr.wrap(result);
+    }
+
+    function pow(Fr base, uint256 v) internal view returns (Fr) {
+        uint256 b = Fr.unwrap(base);
+        // Only works for power of 2
+        require(v > 0 && (v & (v - 1)) == 0, Errors.NotPowerOfTwo());
+        uint256 result;
+
+        // Call the modexp precompile to invert in the field
+        assembly {
+            let free := mload(0x40)
+            mstore(free, 0x20)
+            mstore(add(free, 0x20), 0x20)
+            mstore(add(free, 0x40), 0x20)
+            mstore(add(free, 0x60), b)
+            mstore(add(free, 0x80), v) 
+            mstore(add(free, 0xa0), MODULUS)
+            let success := staticcall(gas(), 0x05, free, 0xc0, 0x00, 0x20)
+            if iszero(success) {
+                mstore(0x00, FRLIB_MODEXP_FAILED_SELECTOR)
+                revert(0, 0x04)
+            }
+            result := mload(0x00)
+            mstore(0x40, add(free, 0xc0))
+        }
+
+        return Fr.wrap(result);
+    }
+
+    function div(Fr numerator, Fr denominator) internal view returns (Fr) {
+        unchecked {
+            return numerator * invert(denominator);
+        }
+    }
+
+    function sqr(Fr value) internal pure returns (Fr) {
+        unchecked {
+            return value * value;
+        }
+    }
+
+    function unwrap(Fr value) internal pure returns (uint256) {
+        unchecked {
+            return Fr.unwrap(value);
+        }
+    }
+
+    function neg(Fr value) internal pure returns (Fr) {
+        unchecked {
+            return Fr.wrap(MODULUS - Fr.unwrap(value));
+        }
+    }
+
+    function from(uint256 value) internal pure returns (Fr) {
+        unchecked {
+            require(value < MODULUS, Errors.ValueGeFieldOrder());
+            return Fr.wrap(value);
+        }
+    }
+
+    function fromBytes32(bytes32 value) internal pure returns (Fr) {
+        unchecked {
+            uint256 v = uint256(value);
+            require(v < MODULUS, Errors.ValueGeFieldOrder());
+            return Fr.wrap(v);
+        }
+    }
+
+    function toBytes32(Fr value) internal pure returns (bytes32) {
+        unchecked {
+            return bytes32(Fr.unwrap(value));
+        }
+    }
+}
+
+// Free functions
+function add(Fr a, Fr b) pure returns (Fr) {
+    unchecked {
+        return Fr.wrap(addmod(Fr.unwrap(a), Fr.unwrap(b), MODULUS));
+    }
+}
+
+function mul(Fr a, Fr b) pure returns (Fr) {
+    unchecked {
+        return Fr.wrap(mulmod(Fr.unwrap(a), Fr.unwrap(b), MODULUS));
+    }
+}
+
+function sub(Fr a, Fr b) pure returns (Fr) {
+    unchecked {
+        return Fr.wrap(addmod(Fr.unwrap(a), MODULUS - Fr.unwrap(b), MODULUS));
+    }
+}
+
+function notEqual(Fr a, Fr b) pure returns (bool) {
+    unchecked {
+        return Fr.unwrap(a) != Fr.unwrap(b);
+    }
+}
+
+function equal(Fr a, Fr b) pure returns (bool) {
+    unchecked {
+        return Fr.unwrap(a) == Fr.unwrap(b);
+    }
+}
+
+uint256 constant CONST_PROOF_SIZE_LOG_N = 25;
+
+uint256 constant NUMBER_OF_SUBRELATIONS = 29;
+uint256 constant BATCHED_RELATION_PARTIAL_LENGTH = 8;
+uint256 constant ZK_BATCHED_RELATION_PARTIAL_LENGTH = 9;
+uint256 constant NUMBER_OF_ENTITIES = 41;
+// The number of entities added for ZK (gemini_masking_poly)
+uint256 constant NUM_MASKING_POLYNOMIALS = 1;
+uint256 constant NUMBER_OF_ENTITIES_ZK = NUMBER_OF_ENTITIES + NUM_MASKING_POLYNOMIALS;
+uint256 constant NUMBER_UNSHIFTED = 36;
+uint256 constant NUMBER_UNSHIFTED_ZK = NUMBER_UNSHIFTED + NUM_MASKING_POLYNOMIALS;
+uint256 constant NUMBER_TO_BE_SHIFTED = 5;
+uint256 constant PAIRING_POINTS_SIZE = 8;
+
+uint256 constant FIELD_ELEMENT_SIZE = 0x20;
+uint256 constant GROUP_ELEMENT_SIZE = 0x40;
+
+// Powers of alpha used to batch subrelations (alpha, alpha^2, ..., alpha^(NUM_SUBRELATIONS-1))
+uint256 constant NUMBER_OF_ALPHAS = NUMBER_OF_SUBRELATIONS - 1;
+
+// ENUM FOR WIRES
+enum WIRE {
+    Q_M,
+    Q_C,
+    Q_L,
+    Q_R,
+    Q_O,
+    Q_4,
+    Q_LOOKUP,
+    Q_ARITH,
+    Q_RANGE,
+    Q_ELLIPTIC,
+    Q_MEMORY,
+    Q_NNF,
+    Q_POSEIDON2_EXTERNAL,
+    Q_POSEIDON2_INTERNAL,
+    SIGMA_1,
+    SIGMA_2,
+    SIGMA_3,
+    SIGMA_4,
+    ID_1,
+    ID_2,
+    ID_3,
+    ID_4,
+    TABLE_1,
+    TABLE_2,
+    TABLE_3,
+    TABLE_4,
+    LAGRANGE_FIRST,
+    LAGRANGE_LAST,
+    W_L,
+    W_R,
+    W_O,
+    W_4,
+    Z_PERM,
+    LOOKUP_INVERSES,
+    LOOKUP_READ_COUNTS,
+    LOOKUP_READ_TAGS,
+    W_L_SHIFT,
+    W_R_SHIFT,
+    W_O_SHIFT,
+    W_4_SHIFT,
+    Z_PERM_SHIFT
+}
+
+library Honk {
+    struct G1Point {
+        uint256 x;
+        uint256 y;
+    }
+
+    struct VerificationKey {
+        // Misc Params
+        uint256 circuitSize;
+        uint256 logCircuitSize;
+        uint256 publicInputsSize;
+        // Selectors
+        G1Point qm;
+        G1Point qc;
+        G1Point ql;
+        G1Point qr;
+        G1Point qo;
+        G1Point q4;
+        G1Point qLookup; // Lookup
+        G1Point qArith; // Arithmetic widget
+        G1Point qDeltaRange; // Delta Range sort
+        G1Point qMemory; // Memory
+        G1Point qNnf; // Non-native Field
+        G1Point qElliptic; // Auxillary
+        G1Point qPoseidon2External;
+        G1Point qPoseidon2Internal;
+        // Copy constraints
+        G1Point s1;
+        G1Point s2;
+        G1Point s3;
+        G1Point s4;
+        // Copy identity
+        G1Point id1;
+        G1Point id2;
+        G1Point id3;
+        G1Point id4;
+        // Precomputed lookup table
+        G1Point t1;
+        G1Point t2;
+        G1Point t3;
+        G1Point t4;
+        // Fixed first and last
+        G1Point lagrangeFirst;
+        G1Point lagrangeLast;
+    }
+
+    struct RelationParameters {
+        // challenges
+        Fr eta;
+        Fr beta;
+        Fr gamma;
+        // derived
+        Fr publicInputsDelta;
+    }
+
+    struct Proof {
+        // Pairing point object
+        Fr[PAIRING_POINTS_SIZE] pairingPointObject;
+        // Free wires
+        G1Point w1;
+        G1Point w2;
+        G1Point w3;
+        G1Point w4;
+        // Lookup helpers - Permutations
+        G1Point zPerm;
+        // Lookup helpers - logup
+        G1Point lookupReadCounts;
+        G1Point lookupReadTags;
+        G1Point lookupInverses;
+        // Sumcheck
+        Fr[BATCHED_RELATION_PARTIAL_LENGTH][CONST_PROOF_SIZE_LOG_N] sumcheckUnivariates;
+        Fr[NUMBER_OF_ENTITIES] sumcheckEvaluations;
+        // Shplemini
+        G1Point[CONST_PROOF_SIZE_LOG_N - 1] geminiFoldComms;
+        Fr[CONST_PROOF_SIZE_LOG_N] geminiAEvaluations;
+        G1Point shplonkQ;
+        G1Point kzgQuotient;
+    }
+
+    /// forge-lint: disable-next-item(pascal-case-struct)
+    struct ZKProof {
+        // Pairing point object
+        Fr[PAIRING_POINTS_SIZE] pairingPointObject;
+        // ZK: Gemini masking polynomial commitment (sent first, right after public inputs)
+        G1Point geminiMaskingPoly;
+        // Commitments to wire polynomials
+        G1Point w1;
+        G1Point w2;
+        G1Point w3;
+        G1Point w4;
+        // Commitments to logup witness polynomials
+        G1Point lookupReadCounts;
+        G1Point lookupReadTags;
+        G1Point lookupInverses;
+        // Commitment to grand permutation polynomial
+        G1Point zPerm;
+        G1Point[3] libraCommitments;
+        // Sumcheck
+        Fr libraSum;
+        Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH][CONST_PROOF_SIZE_LOG_N] sumcheckUnivariates;
+        Fr libraEvaluation;
+        Fr[NUMBER_OF_ENTITIES_ZK] sumcheckEvaluations; // Includes gemini_masking_poly eval at index 0 (first position)
+        // Shplemini
+        G1Point[CONST_PROOF_SIZE_LOG_N - 1] geminiFoldComms;
+        Fr[CONST_PROOF_SIZE_LOG_N] geminiAEvaluations;
+        Fr[4] libraPolyEvals;
+        G1Point shplonkQ;
+        G1Point kzgQuotient;
+    }
+}
+
+// ZKTranscript library to generate fiat shamir challenges, the ZK transcript only differest
+/// forge-lint: disable-next-item(pascal-case-struct)
+struct ZKTranscript {
+    // Oink
+    Honk.RelationParameters relationParameters;
+    Fr[NUMBER_OF_ALPHAS] alphas; // Powers of alpha: [alpha, alpha^2, ..., alpha^(NUM_SUBRELATIONS-1)]
+    Fr[CONST_PROOF_SIZE_LOG_N] gateChallenges;
+    // Sumcheck
+    Fr libraChallenge;
+    Fr[CONST_PROOF_SIZE_LOG_N] sumCheckUChallenges;
+    // Shplemini
+    Fr rho;
+    Fr geminiR;
+    Fr shplonkNu;
+    Fr shplonkZ;
+    // Derived
+    Fr publicInputsDelta;
+}
+
+library ZKTranscriptLib {
+    function generateTranscript(
+        Honk.ZKProof memory proof,
+        bytes32[] calldata publicInputs,
+        uint256 vkHash,
+        uint256 publicInputsSize,
+        uint256 logN
+    ) external pure returns (ZKTranscript memory t) {
+        Fr previousChallenge;
+        (t.relationParameters, previousChallenge) =
+            generateRelationParametersChallenges(proof, publicInputs, vkHash, publicInputsSize, previousChallenge);
+
+        (t.alphas, previousChallenge) = generateAlphaChallenges(previousChallenge, proof);
+
+        (t.gateChallenges, previousChallenge) = generateGateChallenges(previousChallenge, logN);
+        (t.libraChallenge, previousChallenge) = generateLibraChallenge(previousChallenge, proof);
+        (t.sumCheckUChallenges, previousChallenge) = generateSumcheckChallenges(proof, previousChallenge, logN);
+
+        (t.rho, previousChallenge) = generateRhoChallenge(proof, previousChallenge);
+
+        (t.geminiR, previousChallenge) = generateGeminiRChallenge(proof, previousChallenge, logN);
+
+        (t.shplonkNu, previousChallenge) = generateShplonkNuChallenge(proof, previousChallenge, logN);
+
+        (t.shplonkZ, previousChallenge) = generateShplonkZChallenge(proof, previousChallenge);
+        return t;
+    }
+
+    function splitChallenge(Fr challenge) internal pure returns (Fr first, Fr second) {
+        uint256 challengeU256 = uint256(Fr.unwrap(challenge));
+        // Split into two equal 127-bit chunks (254/2)
+        uint256 lo = challengeU256 & 0x7FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF; // 127 bits
+        uint256 hi = challengeU256 >> 127;
+        first = FrLib.from(lo);
+        second = FrLib.from(hi);
+    }
+
+    function generateRelationParametersChallenges(
+        Honk.ZKProof memory proof,
+        bytes32[] calldata publicInputs,
+        uint256 vkHash,
+        uint256 publicInputsSize,
+        Fr previousChallenge
+    ) internal pure returns (Honk.RelationParameters memory rp, Fr nextPreviousChallenge) {
+        (rp.eta, previousChallenge) = generateEtaChallenge(proof, publicInputs, vkHash, publicInputsSize);
+
+        (rp.beta, rp.gamma, nextPreviousChallenge) = generateBetaGammaChallenges(previousChallenge, proof);
+    }
+
+    function generateEtaChallenge(
+        Honk.ZKProof memory proof,
+        bytes32[] calldata publicInputs,
+        uint256 vkHash,
+        uint256 publicInputsSize
+    ) internal pure returns (Fr eta, Fr previousChallenge) {
+        // Size: 1 (vkHash) + publicInputsSize + 8 (geminiMask(2) + 3 wires(6))
+        bytes32[] memory round0 = new bytes32[](1 + publicInputsSize + 8);
+        round0[0] = bytes32(vkHash);
+
+        for (uint256 i = 0; i < publicInputsSize - PAIRING_POINTS_SIZE; i++) {
+            require(uint256(publicInputs[i]) < P, Errors.ValueGeFieldOrder());
+            round0[1 + i] = publicInputs[i];
+        }
+        for (uint256 i = 0; i < PAIRING_POINTS_SIZE; i++) {
+            round0[1 + publicInputsSize - PAIRING_POINTS_SIZE + i] = FrLib.toBytes32(proof.pairingPointObject[i]);
+        }
+
+        // For ZK flavors: hash the gemini masking poly commitment (sent right after public inputs)
+        round0[1 + publicInputsSize] = bytes32(proof.geminiMaskingPoly.x);
+        round0[1 + publicInputsSize + 1] = bytes32(proof.geminiMaskingPoly.y);
+
+        // Create the first challenge
+        // Note: w4 is added to the challenge later on
+        round0[1 + publicInputsSize + 2] = bytes32(proof.w1.x);
+        round0[1 + publicInputsSize + 3] = bytes32(proof.w1.y);
+        round0[1 + publicInputsSize + 4] = bytes32(proof.w2.x);
+        round0[1 + publicInputsSize + 5] = bytes32(proof.w2.y);
+        round0[1 + publicInputsSize + 6] = bytes32(proof.w3.x);
+        round0[1 + publicInputsSize + 7] = bytes32(proof.w3.y);
+
+        previousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(round0))) % P);
+        (eta,) = splitChallenge(previousChallenge);
+    }
+
+    function generateBetaGammaChallenges(Fr previousChallenge, Honk.ZKProof memory proof)
+        internal
+        pure
+        returns (Fr beta, Fr gamma, Fr nextPreviousChallenge)
+    {
+        bytes32[7] memory round1;
+        round1[0] = FrLib.toBytes32(previousChallenge);
+        round1[1] = bytes32(proof.lookupReadCounts.x);
+        round1[2] = bytes32(proof.lookupReadCounts.y);
+        round1[3] = bytes32(proof.lookupReadTags.x);
+        round1[4] = bytes32(proof.lookupReadTags.y);
+        round1[5] = bytes32(proof.w4.x);
+        round1[6] = bytes32(proof.w4.y);
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(round1))) % P);
+        (beta, gamma) = splitChallenge(nextPreviousChallenge);
+    }
+
+    // Alpha challenges non-linearise the gate contributions
+    function generateAlphaChallenges(Fr previousChallenge, Honk.ZKProof memory proof)
+        internal
+        pure
+        returns (Fr[NUMBER_OF_ALPHAS] memory alphas, Fr nextPreviousChallenge)
+    {
+        // Generate the original sumcheck alpha 0 by hashing zPerm and zLookup
+        uint256[5] memory alpha0;
+        alpha0[0] = Fr.unwrap(previousChallenge);
+        alpha0[1] = proof.lookupInverses.x;
+        alpha0[2] = proof.lookupInverses.y;
+        alpha0[3] = proof.zPerm.x;
+        alpha0[4] = proof.zPerm.y;
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(alpha0))) % P);
+        Fr alpha;
+        (alpha,) = splitChallenge(nextPreviousChallenge);
+
+        // Compute powers of alpha for batching subrelations
+        alphas[0] = alpha;
+        for (uint256 i = 1; i < NUMBER_OF_ALPHAS; i++) {
+            alphas[i] = alphas[i - 1] * alpha;
+        }
+    }
+
+    function generateGateChallenges(Fr previousChallenge, uint256 logN)
+        internal
+        pure
+        returns (Fr[CONST_PROOF_SIZE_LOG_N] memory gateChallenges, Fr nextPreviousChallenge)
+    {
+        previousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(Fr.unwrap(previousChallenge)))) % P);
+        (gateChallenges[0],) = splitChallenge(previousChallenge);
+        for (uint256 i = 1; i < logN; i++) {
+            gateChallenges[i] = gateChallenges[i - 1] * gateChallenges[i - 1];
+        }
+        nextPreviousChallenge = previousChallenge;
+    }
+
+    function generateLibraChallenge(Fr previousChallenge, Honk.ZKProof memory proof)
+        internal
+        pure
+        returns (Fr libraChallenge, Fr nextPreviousChallenge)
+    {
+        // 2 comm, 1 sum, 1 challenge
+        uint256[4] memory challengeData;
+        challengeData[0] = Fr.unwrap(previousChallenge);
+        challengeData[1] = proof.libraCommitments[0].x;
+        challengeData[2] = proof.libraCommitments[0].y;
+        challengeData[3] = Fr.unwrap(proof.libraSum);
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(challengeData))) % P);
+        (libraChallenge,) = splitChallenge(nextPreviousChallenge);
+    }
+
+    function generateSumcheckChallenges(Honk.ZKProof memory proof, Fr prevChallenge, uint256 logN)
+        internal
+        pure
+        returns (Fr[CONST_PROOF_SIZE_LOG_N] memory sumcheckChallenges, Fr nextPreviousChallenge)
+    {
+        for (uint256 i = 0; i < logN; i++) {
+            Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH + 1] memory univariateChal;
+            univariateChal[0] = prevChallenge;
+
+            for (uint256 j = 0; j < ZK_BATCHED_RELATION_PARTIAL_LENGTH; j++) {
+                univariateChal[j + 1] = proof.sumcheckUnivariates[i][j];
+            }
+            prevChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(univariateChal))) % P);
+
+            (sumcheckChallenges[i],) = splitChallenge(prevChallenge);
+        }
+        nextPreviousChallenge = prevChallenge;
+    }
+
+    // We add Libra claimed eval + 2 libra commitments (grand_sum, quotient)
+    function generateRhoChallenge(Honk.ZKProof memory proof, Fr prevChallenge)
+        internal
+        pure
+        returns (Fr rho, Fr nextPreviousChallenge)
+    {
+        uint256[NUMBER_OF_ENTITIES_ZK + 6] memory rhoChallengeElements;
+        rhoChallengeElements[0] = Fr.unwrap(prevChallenge);
+        uint256 i;
+        for (i = 1; i <= NUMBER_OF_ENTITIES_ZK; i++) {
+            rhoChallengeElements[i] = Fr.unwrap(proof.sumcheckEvaluations[i - 1]);
+        }
+        rhoChallengeElements[i] = Fr.unwrap(proof.libraEvaluation);
+        i += 1;
+        rhoChallengeElements[i] = proof.libraCommitments[1].x;
+        rhoChallengeElements[i + 1] = proof.libraCommitments[1].y;
+        i += 2;
+        rhoChallengeElements[i] = proof.libraCommitments[2].x;
+        rhoChallengeElements[i + 1] = proof.libraCommitments[2].y;
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(rhoChallengeElements))) % P);
+        (rho,) = splitChallenge(nextPreviousChallenge);
+    }
+
+    function generateGeminiRChallenge(Honk.ZKProof memory proof, Fr prevChallenge, uint256 logN)
+        internal
+        pure
+        returns (Fr geminiR, Fr nextPreviousChallenge)
+    {
+        uint256[] memory gR = new uint256[]((logN - 1) * 2 + 1);
+        gR[0] = Fr.unwrap(prevChallenge);
+
+        for (uint256 i = 0; i < logN - 1; i++) {
+            gR[1 + i * 2] = proof.geminiFoldComms[i].x;
+            gR[2 + i * 2] = proof.geminiFoldComms[i].y;
+        }
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(gR))) % P);
+
+        (geminiR,) = splitChallenge(nextPreviousChallenge);
+    }
+
+    function generateShplonkNuChallenge(Honk.ZKProof memory proof, Fr prevChallenge, uint256 logN)
+        internal
+        pure
+        returns (Fr shplonkNu, Fr nextPreviousChallenge)
+    {
+        uint256[] memory shplonkNuChallengeElements = new uint256[](logN + 1 + 4);
+        shplonkNuChallengeElements[0] = Fr.unwrap(prevChallenge);
+
+        for (uint256 i = 1; i <= logN; i++) {
+            shplonkNuChallengeElements[i] = Fr.unwrap(proof.geminiAEvaluations[i - 1]);
+        }
+
+        uint256 libraIdx = 0;
+        for (uint256 i = logN + 1; i <= logN + 4; i++) {
+            shplonkNuChallengeElements[i] = Fr.unwrap(proof.libraPolyEvals[libraIdx]);
+            libraIdx++;
+        }
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(shplonkNuChallengeElements))) % P);
+        (shplonkNu,) = splitChallenge(nextPreviousChallenge);
+    }
+
+    function generateShplonkZChallenge(Honk.ZKProof memory proof, Fr prevChallenge)
+        internal
+        pure
+        returns (Fr shplonkZ, Fr nextPreviousChallenge)
+    {
+        uint256[3] memory shplonkZChallengeElements;
+        shplonkZChallengeElements[0] = Fr.unwrap(prevChallenge);
+
+        shplonkZChallengeElements[1] = proof.shplonkQ.x;
+        shplonkZChallengeElements[2] = proof.shplonkQ.y;
+
+        nextPreviousChallenge = FrLib.from(uint256(keccak256(abi.encodePacked(shplonkZChallengeElements))) % P);
+        (shplonkZ,) = splitChallenge(nextPreviousChallenge);
+    }
+
+    function loadProof(bytes calldata proof, uint256 logN) internal pure returns (Honk.ZKProof memory p) {
+        uint256 boundary = 0x0;
+
+        // Pairing point object
+        for (uint256 i = 0; i < PAIRING_POINTS_SIZE; i++) {
+            uint256 limb = uint256(bytes32(proof[boundary:boundary + FIELD_ELEMENT_SIZE]));
+            // lo limbs (even index) < 2^136, hi limbs (odd index) < 2^120
+            require(limb < 2 ** (i % 2 == 0 ? 136 : 120), Errors.ValueGeLimbMax());
+            p.pairingPointObject[i] = FrLib.from(limb);
+            boundary += FIELD_ELEMENT_SIZE;
+        }
+
+        // Gemini masking polynomial commitment (sent first in ZK flavors, right after pairing points)
+        p.geminiMaskingPoly = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+
+        // Commitments
+        p.w1 = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.w2 = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.w3 = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+
+        // Lookup / Permutation Helper Commitments
+        p.lookupReadCounts = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.lookupReadTags = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.w4 = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.lookupInverses = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.zPerm = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.libraCommitments[0] = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+
+        p.libraSum = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+        boundary += FIELD_ELEMENT_SIZE;
+        // Sumcheck univariates
+        for (uint256 i = 0; i < logN; i++) {
+            for (uint256 j = 0; j < ZK_BATCHED_RELATION_PARTIAL_LENGTH; j++) {
+                p.sumcheckUnivariates[i][j] = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+                boundary += FIELD_ELEMENT_SIZE;
+            }
+        }
+
+        // Sumcheck evaluations (includes gemini_masking_poly eval at index 0 for ZK flavors)
+        for (uint256 i = 0; i < NUMBER_OF_ENTITIES_ZK; i++) {
+            p.sumcheckEvaluations[i] = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+            boundary += FIELD_ELEMENT_SIZE;
+        }
+
+        p.libraEvaluation = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+        boundary += FIELD_ELEMENT_SIZE;
+
+        p.libraCommitments[1] = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        p.libraCommitments[2] = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+
+        // Gemini
+        // Read gemini fold univariates
+        for (uint256 i = 0; i < logN - 1; i++) {
+            p.geminiFoldComms[i] = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+            boundary += GROUP_ELEMENT_SIZE;
+        }
+
+        // Read gemini a evaluations
+        for (uint256 i = 0; i < logN; i++) {
+            p.geminiAEvaluations[i] = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+            boundary += FIELD_ELEMENT_SIZE;
+        }
+
+        for (uint256 i = 0; i < 4; i++) {
+            p.libraPolyEvals[i] = bytesToFr(proof[boundary:boundary + FIELD_ELEMENT_SIZE]);
+            boundary += FIELD_ELEMENT_SIZE;
+        }
+
+        // Shplonk
+        p.shplonkQ = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+        boundary += GROUP_ELEMENT_SIZE;
+        // KZG
+        p.kzgQuotient = bytesToG1Point(proof[boundary:boundary + GROUP_ELEMENT_SIZE]);
+    }
+}
+
+library RelationsLib {
+    struct EllipticParams {
+        // Points
+        Fr x_1;
+        Fr y_1;
+        Fr x_2;
+        Fr y_2;
+        Fr y_3;
+        Fr x_3;
+        // push accumulators into memory
+        Fr x_double_identity;
+    }
+
+    // Parameters used within the Memory Relation
+    // A struct is used to work around stack too deep. This relation has alot of variables
+    struct MemParams {
+        Fr memory_record_check;
+        Fr partial_record_check;
+        Fr next_gate_access_type;
+        Fr record_delta;
+        Fr index_delta;
+        Fr adjacent_values_match_if_adjacent_indices_match;
+        Fr adjacent_values_match_if_adjacent_indices_match_and_next_access_is_a_read_operation;
+        Fr access_check;
+        Fr next_gate_access_type_is_boolean;
+        Fr ROM_consistency_check_identity;
+        Fr RAM_consistency_check_identity;
+        Fr timestamp_delta;
+        Fr RAM_timestamp_check_identity;
+        Fr memory_identity;
+        Fr index_is_monotonically_increasing;
+    }
+
+    // Parameters used within the Non-Native Field Relation
+    // A struct is used to work around stack too deep. This relation has alot of variables
+    struct NnfParams {
+        Fr limb_subproduct;
+        Fr non_native_field_gate_1;
+        Fr non_native_field_gate_2;
+        Fr non_native_field_gate_3;
+        Fr limb_accumulator_1;
+        Fr limb_accumulator_2;
+        Fr nnf_identity;
+    }
+
+    struct PoseidonExternalParams {
+        Fr s1;
+        Fr s2;
+        Fr s3;
+        Fr s4;
+        Fr u1;
+        Fr u2;
+        Fr u3;
+        Fr u4;
+        Fr t0;
+        Fr t1;
+        Fr t2;
+        Fr t3;
+        Fr v1;
+        Fr v2;
+        Fr v3;
+        Fr v4;
+        Fr q_pos_by_scaling;
+    }
+
+    struct PoseidonInternalParams {
+        Fr u1;
+        Fr u2;
+        Fr u3;
+        Fr u4;
+        Fr u_sum;
+        Fr v1;
+        Fr v2;
+        Fr v3;
+        Fr v4;
+        Fr s1;
+        Fr q_pos_by_scaling;
+    }
+
+    Fr internal constant GRUMPKIN_CURVE_B_PARAMETER_NEGATED = Fr.wrap(17); // -(-17)
+    uint256 internal constant NEG_HALF_MODULO_P = 0x183227397098d014dc2822db40c0ac2e9419f4243cdcb848a1f0fac9f8000000;
+
+    // Constants for the Non-native Field relation
+    Fr internal constant LIMB_SIZE = Fr.wrap(uint256(1) << 68);
+    Fr internal constant SUBLIMB_SHIFT = Fr.wrap(uint256(1) << 14);
+
+    function accumulateRelationEvaluations(
+        Fr[NUMBER_OF_ENTITIES] memory purportedEvaluations,
+        Honk.RelationParameters memory rp,
+        Fr[NUMBER_OF_ALPHAS] memory subrelationChallenges,
+        Fr powPartialEval
+    ) external pure returns (Fr accumulator) {
+        Fr[NUMBER_OF_SUBRELATIONS] memory evaluations;
+
+        // Accumulate all relations in Ultra Honk - each with varying number of subrelations
+        accumulateArithmeticRelation(purportedEvaluations, evaluations, powPartialEval);
+        accumulatePermutationRelation(purportedEvaluations, rp, evaluations, powPartialEval);
+        accumulateLogDerivativeLookupRelation(purportedEvaluations, rp, evaluations, powPartialEval);
+        accumulateDeltaRangeRelation(purportedEvaluations, evaluations, powPartialEval);
+        accumulateEllipticRelation(purportedEvaluations, evaluations, powPartialEval);
+        accumulateMemoryRelation(purportedEvaluations, rp, evaluations, powPartialEval);
+        accumulateNnfRelation(purportedEvaluations, evaluations, powPartialEval);
+        accumulatePoseidonExternalRelation(purportedEvaluations, evaluations, powPartialEval);
+        accumulatePoseidonInternalRelation(purportedEvaluations, evaluations, powPartialEval);
+
+        // batch the subrelations with the precomputed alpha powers to obtain the full honk relation
+        accumulator = scaleAndBatchSubrelations(evaluations, subrelationChallenges);
+    }
+
+    /**
+     * Aesthetic helper function that is used to index by enum into proof.sumcheckEvaluations, it avoids
+     * the relation checking code being cluttered with uint256 type casting, which is often a different colour in code
+     * editors, and thus is noisy.
+     */
+    function wire(Fr[NUMBER_OF_ENTITIES] memory p, WIRE _wire) internal pure returns (Fr) {
+        return p[uint256(_wire)];
+    }
+
+    /**
+     * Ultra Arithmetic Relation
+     *
+     */
+    function accumulateArithmeticRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        // Relation 0
+        Fr q_arith = wire(p, WIRE.Q_ARITH);
+        {
+            Fr neg_half = Fr.wrap(NEG_HALF_MODULO_P);
+
+            Fr accum = (q_arith - Fr.wrap(3)) * (wire(p, WIRE.Q_M) * wire(p, WIRE.W_R) * wire(p, WIRE.W_L)) * neg_half;
+            accum = accum + (wire(p, WIRE.Q_L) * wire(p, WIRE.W_L)) + (wire(p, WIRE.Q_R) * wire(p, WIRE.W_R))
+                + (wire(p, WIRE.Q_O) * wire(p, WIRE.W_O)) + (wire(p, WIRE.Q_4) * wire(p, WIRE.W_4)) + wire(p, WIRE.Q_C);
+            accum = accum + (q_arith - ONE) * wire(p, WIRE.W_4_SHIFT);
+            accum = accum * q_arith;
+            accum = accum * domainSep;
+            evals[0] = accum;
+        }
+
+        // Relation 1
+        {
+            Fr accum = wire(p, WIRE.W_L) + wire(p, WIRE.W_4) - wire(p, WIRE.W_L_SHIFT) + wire(p, WIRE.Q_M);
+            accum = accum * (q_arith - Fr.wrap(2));
+            accum = accum * (q_arith - ONE);
+            accum = accum * q_arith;
+            accum = accum * domainSep;
+            evals[1] = accum;
+        }
+    }
+
+    function accumulatePermutationRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Honk.RelationParameters memory rp,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        Fr grand_product_numerator;
+        Fr grand_product_denominator;
+
+        {
+            Fr num = wire(p, WIRE.W_L) + wire(p, WIRE.ID_1) * rp.beta + rp.gamma;
+            num = num * (wire(p, WIRE.W_R) + wire(p, WIRE.ID_2) * rp.beta + rp.gamma);
+            num = num * (wire(p, WIRE.W_O) + wire(p, WIRE.ID_3) * rp.beta + rp.gamma);
+            num = num * (wire(p, WIRE.W_4) + wire(p, WIRE.ID_4) * rp.beta + rp.gamma);
+
+            grand_product_numerator = num;
+        }
+        {
+            Fr den = wire(p, WIRE.W_L) + wire(p, WIRE.SIGMA_1) * rp.beta + rp.gamma;
+            den = den * (wire(p, WIRE.W_R) + wire(p, WIRE.SIGMA_2) * rp.beta + rp.gamma);
+            den = den * (wire(p, WIRE.W_O) + wire(p, WIRE.SIGMA_3) * rp.beta + rp.gamma);
+            den = den * (wire(p, WIRE.W_4) + wire(p, WIRE.SIGMA_4) * rp.beta + rp.gamma);
+
+            grand_product_denominator = den;
+        }
+
+        // Contribution 2
+        {
+            Fr acc = (wire(p, WIRE.Z_PERM) + wire(p, WIRE.LAGRANGE_FIRST)) * grand_product_numerator;
+
+            acc = acc
+                - ((wire(p, WIRE.Z_PERM_SHIFT) + (wire(p, WIRE.LAGRANGE_LAST) * rp.publicInputsDelta))
+                    * grand_product_denominator);
+            acc = acc * domainSep;
+            evals[2] = acc;
+        }
+
+        // Contribution 3
+        {
+            Fr acc = (wire(p, WIRE.LAGRANGE_LAST) * wire(p, WIRE.Z_PERM_SHIFT)) * domainSep;
+            evals[3] = acc;
+        }
+
+        // Contribution 4: z_perm initialization check (lagrange_first * z_perm = 0)
+        {
+            Fr acc = (wire(p, WIRE.LAGRANGE_FIRST) * wire(p, WIRE.Z_PERM)) * domainSep;
+            evals[4] = acc;
+        }
+    }
+
+    function accumulateLogDerivativeLookupRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Honk.RelationParameters memory rp,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        Fr table_term;
+        Fr lookup_term;
+
+        // Calculate the write term (the table accumulation)
+        // table_term = table_1 + γ + table_2 * β + table_3 * β² + table_4 * β³
+        {
+            Fr beta_sqr = rp.beta * rp.beta;
+            table_term = wire(p, WIRE.TABLE_1) + rp.gamma + (wire(p, WIRE.TABLE_2) * rp.beta)
+                + (wire(p, WIRE.TABLE_3) * beta_sqr) + (wire(p, WIRE.TABLE_4) * beta_sqr * rp.beta);
+        }
+
+        // Calculate the read term
+        // lookup_term = derived_entry_1 + γ + derived_entry_2 * β + derived_entry_3 * β² + q_index * β³
+        {
+            Fr beta_sqr = rp.beta * rp.beta;
+            Fr derived_entry_1 = wire(p, WIRE.W_L) + rp.gamma + (wire(p, WIRE.Q_R) * wire(p, WIRE.W_L_SHIFT));
+            Fr derived_entry_2 = wire(p, WIRE.W_R) + wire(p, WIRE.Q_M) * wire(p, WIRE.W_R_SHIFT);
+            Fr derived_entry_3 = wire(p, WIRE.W_O) + wire(p, WIRE.Q_C) * wire(p, WIRE.W_O_SHIFT);
+
+            lookup_term = derived_entry_1 + (derived_entry_2 * rp.beta) + (derived_entry_3 * beta_sqr)
+                + (wire(p, WIRE.Q_O) * beta_sqr * rp.beta);
+        }
+
+        Fr lookup_inverse = wire(p, WIRE.LOOKUP_INVERSES) * table_term;
+        Fr table_inverse = wire(p, WIRE.LOOKUP_INVERSES) * lookup_term;
+
+        Fr inverse_exists_xor =
+        wire(p, WIRE.LOOKUP_READ_TAGS) + wire(p, WIRE.Q_LOOKUP)
+            - (wire(p, WIRE.LOOKUP_READ_TAGS) * wire(p, WIRE.Q_LOOKUP));
+
+        // Inverse calculated correctly relation
+        Fr accumulatorNone = lookup_term * table_term * wire(p, WIRE.LOOKUP_INVERSES) - inverse_exists_xor;
+        accumulatorNone = accumulatorNone * domainSep;
+
+        // Inverse
+        Fr accumulatorOne = wire(p, WIRE.Q_LOOKUP) * lookup_inverse - wire(p, WIRE.LOOKUP_READ_COUNTS) * table_inverse;
+
+        Fr read_tag = wire(p, WIRE.LOOKUP_READ_TAGS);
+
+        Fr read_tag_boolean_relation = read_tag * read_tag - read_tag;
+
+        evals[5] = accumulatorNone;
+        evals[6] = accumulatorOne;
+        evals[7] = read_tag_boolean_relation * domainSep;
+    }
+
+    function accumulateDeltaRangeRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        Fr minus_one = ZERO - ONE;
+        Fr minus_two = ZERO - Fr.wrap(2);
+        Fr minus_three = ZERO - Fr.wrap(3);
+
+        // Compute wire differences
+        Fr delta_1 = wire(p, WIRE.W_R) - wire(p, WIRE.W_L);
+        Fr delta_2 = wire(p, WIRE.W_O) - wire(p, WIRE.W_R);
+        Fr delta_3 = wire(p, WIRE.W_4) - wire(p, WIRE.W_O);
+        Fr delta_4 = wire(p, WIRE.W_L_SHIFT) - wire(p, WIRE.W_4);
+
+        // Contribution 6
+        {
+            Fr acc = delta_1;
+            acc = acc * (delta_1 + minus_one);
+            acc = acc * (delta_1 + minus_two);
+            acc = acc * (delta_1 + minus_three);
+            acc = acc * wire(p, WIRE.Q_RANGE);
+            acc = acc * domainSep;
+            evals[8] = acc;
+        }
+
+        // Contribution 7
+        {
+            Fr acc = delta_2;
+            acc = acc * (delta_2 + minus_one);
+            acc = acc * (delta_2 + minus_two);
+            acc = acc * (delta_2 + minus_three);
+            acc = acc * wire(p, WIRE.Q_RANGE);
+            acc = acc * domainSep;
+            evals[9] = acc;
+        }
+
+        // Contribution 8
+        {
+            Fr acc = delta_3;
+            acc = acc * (delta_3 + minus_one);
+            acc = acc * (delta_3 + minus_two);
+            acc = acc * (delta_3 + minus_three);
+            acc = acc * wire(p, WIRE.Q_RANGE);
+            acc = acc * domainSep;
+            evals[10] = acc;
+        }
+
+        // Contribution 9
+        {
+            Fr acc = delta_4;
+            acc = acc * (delta_4 + minus_one);
+            acc = acc * (delta_4 + minus_two);
+            acc = acc * (delta_4 + minus_three);
+            acc = acc * wire(p, WIRE.Q_RANGE);
+            acc = acc * domainSep;
+            evals[11] = acc;
+        }
+    }
+
+    function accumulateEllipticRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        EllipticParams memory ep;
+        ep.x_1 = wire(p, WIRE.W_R);
+        ep.y_1 = wire(p, WIRE.W_O);
+
+        ep.x_2 = wire(p, WIRE.W_L_SHIFT);
+        ep.y_2 = wire(p, WIRE.W_4_SHIFT);
+        ep.y_3 = wire(p, WIRE.W_O_SHIFT);
+        ep.x_3 = wire(p, WIRE.W_R_SHIFT);
+
+        Fr q_sign = wire(p, WIRE.Q_L);
+        Fr q_is_double = wire(p, WIRE.Q_M);
+
+        // Contribution 10 point addition, x-coordinate check
+        // q_elliptic * (x3 + x2 + x1)(x2 - x1)(x2 - x1) - y2^2 - y1^2 + 2(y2y1)*q_sign = 0
+        Fr x_diff = (ep.x_2 - ep.x_1);
+        Fr y1_sqr = (ep.y_1 * ep.y_1);
+        {
+            // Move to top
+            Fr partialEval = domainSep;
+
+            Fr y2_sqr = (ep.y_2 * ep.y_2);
+            Fr y1y2 = ep.y_1 * ep.y_2 * q_sign;
+            Fr x_add_identity = (ep.x_3 + ep.x_2 + ep.x_1);
+            x_add_identity = x_add_identity * x_diff * x_diff;
+            x_add_identity = x_add_identity - y2_sqr - y1_sqr + y1y2 + y1y2;
+
+            evals[12] = x_add_identity * partialEval * wire(p, WIRE.Q_ELLIPTIC) * (ONE - q_is_double);
+        }
+
+        // Contribution 11 point addition, x-coordinate check
+        // q_elliptic * (q_sign * y1 + y3)(x2 - x1) + (x3 - x1)(y2 - q_sign * y1) = 0
+        {
+            Fr y1_plus_y3 = ep.y_1 + ep.y_3;
+            Fr y_diff = ep.y_2 * q_sign - ep.y_1;
+            Fr y_add_identity = y1_plus_y3 * x_diff + (ep.x_3 - ep.x_1) * y_diff;
+            evals[13] = y_add_identity * domainSep * wire(p, WIRE.Q_ELLIPTIC) * (ONE - q_is_double);
+        }
+
+        // Contribution 10 point doubling, x-coordinate check
+        // (x3 + x1 + x1) (4y1*y1) - 9 * x1 * x1 * x1 * x1 = 0
+        // N.B. we're using the equivalence x1*x1*x1 === y1*y1 - curve_b to reduce degree by 1
+        {
+            Fr x_pow_4 = (y1_sqr + GRUMPKIN_CURVE_B_PARAMETER_NEGATED) * ep.x_1;
+            Fr y1_sqr_mul_4 = y1_sqr + y1_sqr;
+            y1_sqr_mul_4 = y1_sqr_mul_4 + y1_sqr_mul_4;
+            Fr x1_pow_4_mul_9 = x_pow_4 * Fr.wrap(9);
+
+            // NOTE: pushed into memory (stack >:'( )
+            ep.x_double_identity = (ep.x_3 + ep.x_1 + ep.x_1) * y1_sqr_mul_4 - x1_pow_4_mul_9;
+
+            Fr acc = ep.x_double_identity * domainSep * wire(p, WIRE.Q_ELLIPTIC) * q_is_double;
+            evals[12] = evals[12] + acc;
+        }
+
+        // Contribution 11 point doubling, y-coordinate check
+        // (y1 + y1) (2y1) - (3 * x1 * x1)(x1 - x3) = 0
+        {
+            Fr x1_sqr_mul_3 = (ep.x_1 + ep.x_1 + ep.x_1) * ep.x_1;
+            Fr y_double_identity = x1_sqr_mul_3 * (ep.x_1 - ep.x_3) - (ep.y_1 + ep.y_1) * (ep.y_1 + ep.y_3);
+            evals[13] = evals[13] + y_double_identity * domainSep * wire(p, WIRE.Q_ELLIPTIC) * q_is_double;
+        }
+    }
+
+    function accumulateMemoryRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Honk.RelationParameters memory rp,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        MemParams memory ap;
+
+        // Compute eta powers locally
+        Fr eta_two = rp.eta * rp.eta;
+        Fr eta_three = eta_two * rp.eta;
+
+        /**
+         * MEMORY
+         *
+         * A RAM memory record contains a tuple of the following fields:
+         *  * i: `index` of memory cell being accessed
+         *  * t: `timestamp` of memory cell being accessed (used for RAM, set to 0 for ROM)
+         *  * v: `value` of memory cell being accessed
+         *  * a: `access` type of record. read: 0 = read, 1 = write
+         *  * r: `record` of memory cell. record = access + index * eta + timestamp * eta_two + value * eta_three
+         *
+         * A ROM memory record contains a tuple of the following fields:
+         *  * i: `index` of memory cell being accessed
+         *  * v: `value1` of memory cell being accessed (ROM tables can store up to 2 values per index)
+         *  * v2:`value2` of memory cell being accessed (ROM tables can store up to 2 values per index)
+         *  * r: `record` of memory cell. record = index * eta + value2 * eta_two + value1 * eta_three
+         *
+         *  When performing a read/write access, the values of i, t, v, v2, a, r are stored in the following wires +
+         * selectors, depending on whether the gate is a RAM read/write or a ROM read
+         *
+         *  | gate type | i  | v2/t  |  v | a  | r  |
+         *  | --------- | -- | ----- | -- | -- | -- |
+         *  | ROM       | w1 | w2    | w3 | -- | w4 |
+         *  | RAM       | w1 | w2    | w3 | qc | w4 |
+         *
+         * (for accesses where `index` is a circuit constant, it is assumed the circuit will apply a copy constraint on
+         * `w2` to fix its value)
+         *
+         *
+         */
+
+        /**
+         * Memory Record Check
+         * Partial degree: 1
+         * Total degree: 4
+         *
+         * A ROM/ROM access gate can be evaluated with the identity:
+         *
+         * qc + w1 \eta + w2 \eta_two + w3 \eta_three - w4 = 0
+         *
+         * For ROM gates, qc = 0
+         */
+        ap.memory_record_check = wire(p, WIRE.W_O) * eta_three;
+        ap.memory_record_check = ap.memory_record_check + (wire(p, WIRE.W_R) * eta_two);
+        ap.memory_record_check = ap.memory_record_check + (wire(p, WIRE.W_L) * rp.eta);
+        ap.memory_record_check = ap.memory_record_check + wire(p, WIRE.Q_C);
+        ap.partial_record_check = ap.memory_record_check; // used in RAM consistency check; deg 1 or 4
+        ap.memory_record_check = ap.memory_record_check - wire(p, WIRE.W_4);
+
+        /**
+         * Contribution 13 & 14
+         * ROM Consistency Check
+         * Partial degree: 1
+         * Total degree: 4
+         *
+         * For every ROM read, a set equivalence check is applied between the record witnesses, and a second set of
+         * records that are sorted.
+         *
+         * We apply the following checks for the sorted records:
+         *
+         * 1. w1, w2, w3 correctly map to 'index', 'v1, 'v2' for a given record value at w4
+         * 2. index values for adjacent records are monotonically increasing
+         * 3. if, at gate i, index_i == index_{i + 1}, then value1_i == value1_{i + 1} and value2_i == value2_{i + 1}
+         *
+         */
+        ap.index_delta = wire(p, WIRE.W_L_SHIFT) - wire(p, WIRE.W_L);
+        ap.record_delta = wire(p, WIRE.W_4_SHIFT) - wire(p, WIRE.W_4);
+
+        ap.index_is_monotonically_increasing = ap.index_delta * (ap.index_delta - Fr.wrap(1)); // deg 2
+
+        ap.adjacent_values_match_if_adjacent_indices_match = (ap.index_delta * MINUS_ONE + ONE) * ap.record_delta; // deg 2
+
+        evals[15] = ap.adjacent_values_match_if_adjacent_indices_match * (wire(p, WIRE.Q_L) * wire(p, WIRE.Q_R))
+            * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 5
+        evals[16] = ap.index_is_monotonically_increasing * (wire(p, WIRE.Q_L) * wire(p, WIRE.Q_R))
+            * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 5
+
+        ap.ROM_consistency_check_identity = ap.memory_record_check * (wire(p, WIRE.Q_L) * wire(p, WIRE.Q_R)); // deg 3 or 7
+
+        /**
+         * Contributions 15,16,17
+         * RAM Consistency Check
+         *
+         * The 'access' type of the record is extracted with the expression `w_4 - ap.partial_record_check`
+         * (i.e. for an honest Prover `w1 * eta + w2 * eta^2 + w3 * eta^3 - w4 = access`.
+         * This is validated by requiring `access` to be boolean
+         *
+         * For two adjacent entries in the sorted list if _both_
+         *  A) index values match
+         *  B) adjacent access value is 0 (i.e. next gate is a READ)
+         * then
+         *  C) both values must match.
+         * The gate boolean check is
+         * (A && B) => C  === !(A && B) || C ===  !A || !B || C
+         *
+         * N.B. it is the responsibility of the circuit writer to ensure that every RAM cell is initialized
+         * with a WRITE operation.
+         */
+        Fr access_type = (wire(p, WIRE.W_4) - ap.partial_record_check); // will be 0 or 1 for honest Prover; deg 1 or 4
+        ap.access_check = access_type * (access_type - Fr.wrap(1)); // check value is 0 or 1; deg 2 or 8
+
+        // reverse order we could re-use `ap.partial_record_check`  1 -  ((w3' * eta + w2') * eta + w1') * eta
+        // deg 1 or 4
+        ap.next_gate_access_type = wire(p, WIRE.W_O_SHIFT) * eta_three;
+        ap.next_gate_access_type = ap.next_gate_access_type + (wire(p, WIRE.W_R_SHIFT) * eta_two);
+        ap.next_gate_access_type = ap.next_gate_access_type + (wire(p, WIRE.W_L_SHIFT) * rp.eta);
+        ap.next_gate_access_type = wire(p, WIRE.W_4_SHIFT) - ap.next_gate_access_type;
+
+        Fr value_delta = wire(p, WIRE.W_O_SHIFT) - wire(p, WIRE.W_O);
+        ap.adjacent_values_match_if_adjacent_indices_match_and_next_access_is_a_read_operation =
+            (ap.index_delta * MINUS_ONE + ONE) * value_delta * (ap.next_gate_access_type * MINUS_ONE + ONE); // deg 3 or 6
+
+        // We can't apply the RAM consistency check identity on the final entry in the sorted list (the wires in the
+        // next gate would make the identity fail).  We need to validate that its 'access type' bool is correct. Can't
+        // do  with an arithmetic gate because of the  `eta` factors. We need to check that the *next* gate's access
+        // type is  correct, to cover this edge case
+        // deg 2 or 4
+        ap.next_gate_access_type_is_boolean =
+            ap.next_gate_access_type * ap.next_gate_access_type - ap.next_gate_access_type;
+
+        // Putting it all together...
+        evals[17] = ap.adjacent_values_match_if_adjacent_indices_match_and_next_access_is_a_read_operation
+            * (wire(p, WIRE.Q_O)) * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 5 or 8
+        evals[18] = ap.index_is_monotonically_increasing * (wire(p, WIRE.Q_O)) * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 4
+        evals[19] = ap.next_gate_access_type_is_boolean * (wire(p, WIRE.Q_O)) * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 4 or 6
+
+        ap.RAM_consistency_check_identity = ap.access_check * (wire(p, WIRE.Q_O)); // deg 3 or 9
+
+        /**
+         * RAM Timestamp Consistency Check
+         *
+         * | w1 | w2 | w3 | w4 |
+         * | index | timestamp | timestamp_check | -- |
+         *
+         * Let delta_index = index_{i + 1} - index_{i}
+         *
+         * Iff delta_index == 0, timestamp_check = timestamp_{i + 1} - timestamp_i
+         * Else timestamp_check = 0
+         */
+        ap.timestamp_delta = wire(p, WIRE.W_R_SHIFT) - wire(p, WIRE.W_R);
+        ap.RAM_timestamp_check_identity = (ap.index_delta * MINUS_ONE + ONE) * ap.timestamp_delta - wire(p, WIRE.W_O); // deg 3
+
+        /**
+         * Complete Contribution 12
+         * The complete RAM/ROM memory identity
+         * Partial degree:
+         */
+        ap.memory_identity = ap.ROM_consistency_check_identity; // deg 3 or 6
+        ap.memory_identity =
+            ap.memory_identity + ap.RAM_timestamp_check_identity * (wire(p, WIRE.Q_4) * wire(p, WIRE.Q_L)); // deg 4
+        ap.memory_identity = ap.memory_identity + ap.memory_record_check * (wire(p, WIRE.Q_M) * wire(p, WIRE.Q_L)); // deg 3 or 6
+        ap.memory_identity = ap.memory_identity + ap.RAM_consistency_check_identity; // deg 3 or 9
+
+        // (deg 3 or 9) + (deg 4) + (deg 3)
+        ap.memory_identity = ap.memory_identity * (wire(p, WIRE.Q_MEMORY) * domainSep); // deg 4 or 10
+        evals[14] = ap.memory_identity;
+    }
+
+    function accumulateNnfRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        NnfParams memory ap;
+
+        /**
+         * Contribution 12
+         * Non native field arithmetic gate 2
+         * deg 4
+         *
+         *             _                                                                               _
+         *            /   _                   _                               _       14                \
+         * q_2 . q_4 |   (w_1 . w_2) + (w_1 . w_2) + (w_1 . w_4 + w_2 . w_3 - w_3) . 2    - w_3 - w_4   |
+         *            \_                                                                               _/
+         *
+         *
+         */
+        ap.limb_subproduct = wire(p, WIRE.W_L) * wire(p, WIRE.W_R_SHIFT) + wire(p, WIRE.W_L_SHIFT) * wire(p, WIRE.W_R);
+        ap.non_native_field_gate_2 =
+            (wire(p, WIRE.W_L) * wire(p, WIRE.W_4) + wire(p, WIRE.W_R) * wire(p, WIRE.W_O) - wire(p, WIRE.W_O_SHIFT));
+        ap.non_native_field_gate_2 = ap.non_native_field_gate_2 * LIMB_SIZE;
+        ap.non_native_field_gate_2 = ap.non_native_field_gate_2 - wire(p, WIRE.W_4_SHIFT);
+        ap.non_native_field_gate_2 = ap.non_native_field_gate_2 + ap.limb_subproduct;
+        ap.non_native_field_gate_2 = ap.non_native_field_gate_2 * wire(p, WIRE.Q_4);
+
+        ap.limb_subproduct = ap.limb_subproduct * LIMB_SIZE;
+        ap.limb_subproduct = ap.limb_subproduct + (wire(p, WIRE.W_L_SHIFT) * wire(p, WIRE.W_R_SHIFT));
+        ap.non_native_field_gate_1 = ap.limb_subproduct;
+        ap.non_native_field_gate_1 = ap.non_native_field_gate_1 - (wire(p, WIRE.W_O) + wire(p, WIRE.W_4));
+        ap.non_native_field_gate_1 = ap.non_native_field_gate_1 * wire(p, WIRE.Q_O);
+
+        ap.non_native_field_gate_3 = ap.limb_subproduct;
+        ap.non_native_field_gate_3 = ap.non_native_field_gate_3 + wire(p, WIRE.W_4);
+        ap.non_native_field_gate_3 = ap.non_native_field_gate_3 - (wire(p, WIRE.W_O_SHIFT) + wire(p, WIRE.W_4_SHIFT));
+        ap.non_native_field_gate_3 = ap.non_native_field_gate_3 * wire(p, WIRE.Q_M);
+
+        Fr non_native_field_identity =
+        ap.non_native_field_gate_1 + ap.non_native_field_gate_2 + ap.non_native_field_gate_3;
+        non_native_field_identity = non_native_field_identity * wire(p, WIRE.Q_R);
+
+        // ((((w2' * 2^14 + w1') * 2^14 + w3) * 2^14 + w2) * 2^14 + w1 - w4) * qm
+        // deg 2
+        ap.limb_accumulator_1 = wire(p, WIRE.W_R_SHIFT) * SUBLIMB_SHIFT;
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 + wire(p, WIRE.W_L_SHIFT);
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 + wire(p, WIRE.W_O);
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 + wire(p, WIRE.W_R);
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 + wire(p, WIRE.W_L);
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 - wire(p, WIRE.W_4);
+        ap.limb_accumulator_1 = ap.limb_accumulator_1 * wire(p, WIRE.Q_4);
+
+        // ((((w3' * 2^14 + w2') * 2^14 + w1') * 2^14 + w4) * 2^14 + w3 - w4') * qm
+        // deg 2
+        ap.limb_accumulator_2 = wire(p, WIRE.W_O_SHIFT) * SUBLIMB_SHIFT;
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 + wire(p, WIRE.W_R_SHIFT);
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 + wire(p, WIRE.W_L_SHIFT);
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 + wire(p, WIRE.W_4);
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 * SUBLIMB_SHIFT;
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 + wire(p, WIRE.W_O);
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 - wire(p, WIRE.W_4_SHIFT);
+        ap.limb_accumulator_2 = ap.limb_accumulator_2 * wire(p, WIRE.Q_M);
+
+        Fr limb_accumulator_identity = ap.limb_accumulator_1 + ap.limb_accumulator_2;
+        limb_accumulator_identity = limb_accumulator_identity * wire(p, WIRE.Q_O); //  deg 3
+
+        ap.nnf_identity = non_native_field_identity + limb_accumulator_identity;
+        ap.nnf_identity = ap.nnf_identity * (wire(p, WIRE.Q_NNF) * domainSep);
+        evals[20] = ap.nnf_identity;
+    }
+
+    function accumulatePoseidonExternalRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        PoseidonExternalParams memory ep;
+
+        ep.s1 = wire(p, WIRE.W_L) + wire(p, WIRE.Q_L);
+        ep.s2 = wire(p, WIRE.W_R) + wire(p, WIRE.Q_R);
+        ep.s3 = wire(p, WIRE.W_O) + wire(p, WIRE.Q_O);
+        ep.s4 = wire(p, WIRE.W_4) + wire(p, WIRE.Q_4);
+
+        ep.u1 = ep.s1 * ep.s1 * ep.s1 * ep.s1 * ep.s1;
+        ep.u2 = ep.s2 * ep.s2 * ep.s2 * ep.s2 * ep.s2;
+        ep.u3 = ep.s3 * ep.s3 * ep.s3 * ep.s3 * ep.s3;
+        ep.u4 = ep.s4 * ep.s4 * ep.s4 * ep.s4 * ep.s4;
+        // matrix mul v = M_E * u with 14 additions
+        ep.t0 = ep.u1 + ep.u2; // u_1 + u_2
+        ep.t1 = ep.u3 + ep.u4; // u_3 + u_4
+        ep.t2 = ep.u2 + ep.u2 + ep.t1; // 2u_2
+        // ep.t2 += ep.t1; // 2u_2 + u_3 + u_4
+        ep.t3 = ep.u4 + ep.u4 + ep.t0; // 2u_4
+        // ep.t3 += ep.t0; // u_1 + u_2 + 2u_4
+        ep.v4 = ep.t1 + ep.t1;
+        ep.v4 = ep.v4 + ep.v4 + ep.t3;
+        // ep.v4 += ep.t3; // u_1 + u_2 + 4u_3 + 6u_4
+        ep.v2 = ep.t0 + ep.t0;
+        ep.v2 = ep.v2 + ep.v2 + ep.t2;
+        // ep.v2 += ep.t2; // 4u_1 + 6u_2 + u_3 + u_4
+        ep.v1 = ep.t3 + ep.v2; // 5u_1 + 7u_2 + u_3 + 3u_4
+        ep.v3 = ep.t2 + ep.v4; // u_1 + 3u_2 + 5u_3 + 7u_4
+
+        ep.q_pos_by_scaling = wire(p, WIRE.Q_POSEIDON2_EXTERNAL) * domainSep;
+        evals[21] = evals[21] + ep.q_pos_by_scaling * (ep.v1 - wire(p, WIRE.W_L_SHIFT));
+
+        evals[22] = evals[22] + ep.q_pos_by_scaling * (ep.v2 - wire(p, WIRE.W_R_SHIFT));
+
+        evals[23] = evals[23] + ep.q_pos_by_scaling * (ep.v3 - wire(p, WIRE.W_O_SHIFT));
+
+        evals[24] = evals[24] + ep.q_pos_by_scaling * (ep.v4 - wire(p, WIRE.W_4_SHIFT));
+    }
+
+    function accumulatePoseidonInternalRelation(
+        Fr[NUMBER_OF_ENTITIES] memory p,
+        Fr[NUMBER_OF_SUBRELATIONS] memory evals,
+        Fr domainSep
+    ) internal pure {
+        PoseidonInternalParams memory ip;
+
+        Fr[4] memory INTERNAL_MATRIX_DIAGONAL = [
+            FrLib.from(0x10dc6e9c006ea38b04b1e03b4bd9490c0d03f98929ca1d7fb56821fd19d3b6e7),
+            FrLib.from(0x0c28145b6a44df3e0149b3d0a30b3bb599df9756d4dd9b84a86b38cfb45a740b),
+            FrLib.from(0x00544b8338791518b2c7645a50392798b21f75bb60e3596170067d00141cac15),
+            FrLib.from(0x222c01175718386f2e2e82eb122789e352e105a3b8fa852613bc534433ee428b)
+        ];
+
+        // add round constants
+        ip.s1 = wire(p, WIRE.W_L) + wire(p, WIRE.Q_L);
+
+        // apply s-box round
+        ip.u1 = ip.s1 * ip.s1 * ip.s1 * ip.s1 * ip.s1;
+        ip.u2 = wire(p, WIRE.W_R);
+        ip.u3 = wire(p, WIRE.W_O);
+        ip.u4 = wire(p, WIRE.W_4);
+
+        // matrix mul with v = M_I * u 4 muls and 7 additions
+        ip.u_sum = ip.u1 + ip.u2 + ip.u3 + ip.u4;
+
+        ip.q_pos_by_scaling = wire(p, WIRE.Q_POSEIDON2_INTERNAL) * domainSep;
+
+        ip.v1 = ip.u1 * INTERNAL_MATRIX_DIAGONAL[0] + ip.u_sum;
+        evals[25] = evals[25] + ip.q_pos_by_scaling * (ip.v1 - wire(p, WIRE.W_L_SHIFT));
+
+        ip.v2 = ip.u2 * INTERNAL_MATRIX_DIAGONAL[1] + ip.u_sum;
+        evals[26] = evals[26] + ip.q_pos_by_scaling * (ip.v2 - wire(p, WIRE.W_R_SHIFT));
+
+        ip.v3 = ip.u3 * INTERNAL_MATRIX_DIAGONAL[2] + ip.u_sum;
+        evals[27] = evals[27] + ip.q_pos_by_scaling * (ip.v3 - wire(p, WIRE.W_O_SHIFT));
+
+        ip.v4 = ip.u4 * INTERNAL_MATRIX_DIAGONAL[3] + ip.u_sum;
+        evals[28] = evals[28] + ip.q_pos_by_scaling * (ip.v4 - wire(p, WIRE.W_4_SHIFT));
+    }
+
+    // Batch subrelation evaluations using precomputed powers of alpha
+    // First subrelation is implicitly scaled by 1, subsequent ones use powers from the subrelationChallenges array
+    function scaleAndBatchSubrelations(
+        Fr[NUMBER_OF_SUBRELATIONS] memory evaluations,
+        Fr[NUMBER_OF_ALPHAS] memory subrelationChallenges
+    ) internal pure returns (Fr accumulator) {
+        accumulator = evaluations[0];
+
+        for (uint256 i = 1; i < NUMBER_OF_SUBRELATIONS; ++i) {
+            accumulator = accumulator + evaluations[i] * subrelationChallenges[i - 1];
+        }
+    }
+}
+
+library CommitmentSchemeLib {
+    using FrLib for Fr;
+
+    // Avoid stack too deep
+    struct ShpleminiIntermediates {
+        Fr unshiftedScalar;
+        Fr shiftedScalar;
+        Fr unshiftedScalarNeg;
+        Fr shiftedScalarNeg;
+        // Scalar to be multiplied by [1]₁
+        Fr constantTermAccumulator;
+        // Accumulator for powers of rho
+        Fr batchingChallenge;
+        // Linear combination of multilinear (sumcheck) evaluations and powers of rho
+        Fr batchedEvaluation;
+        Fr[4] denominators;
+        Fr[4] batchingScalars;
+        // 1/(z - r^{2^i}) for i = 0, ..., logSize, dynamically updated
+        Fr posInvertedDenominator;
+        // 1/(z + r^{2^i}) for i = 0, ..., logSize, dynamically updated
+        Fr negInvertedDenominator;
+        // ν^{2i} * 1/(z - r^{2^i})
+        Fr scalingFactorPos;
+        // ν^{2i+1} * 1/(z + r^{2^i})
+        Fr scalingFactorNeg;
+        // Fold_i(r^{2^i}) reconstructed by Verifier
+        Fr[] foldPosEvaluations;
+    }
+
+    // Compute the evaluations Aₗ(r^{2ˡ}) for l = 0, ..., m-1
+    function computeFoldPosEvaluations(
+        Fr[CONST_PROOF_SIZE_LOG_N] memory sumcheckUChallenges,
+        Fr batchedEvalAccumulator,
+        Fr[CONST_PROOF_SIZE_LOG_N] memory geminiEvaluations,
+        Fr[] memory geminiEvalChallengePowers,
+        uint256 logSize
+    ) internal view returns (Fr[] memory) {
+        Fr[] memory foldPosEvaluations = new Fr[](logSize);
+        for (uint256 i = logSize; i > 0; --i) {
+            Fr challengePower = geminiEvalChallengePowers[i - 1];
+            Fr u = sumcheckUChallenges[i - 1];
+
+            Fr batchedEvalRoundAcc = ((challengePower * batchedEvalAccumulator * Fr.wrap(2)) - geminiEvaluations[i - 1]
+                    * (challengePower * (ONE - u) - u));
+            // Divide by the denominator
+            batchedEvalRoundAcc = batchedEvalRoundAcc * (challengePower * (ONE - u) + u).invert();
+
+            batchedEvalAccumulator = batchedEvalRoundAcc;
+            foldPosEvaluations[i - 1] = batchedEvalRoundAcc;
+        }
+        return foldPosEvaluations;
+    }
+
+    function computeSquares(Fr r, uint256 logN) internal pure returns (Fr[] memory) {
+        Fr[] memory squares = new Fr[](logN);
+        squares[0] = r;
+        for (uint256 i = 1; i < logN; ++i) {
+            squares[i] = squares[i - 1].sqr();
+        }
+        return squares;
+    }
+}
+
+uint256 constant Q = 21888242871839275222246405745257275088696311157297823662689037894645226208583; // EC group order. F_q
+
+// Fr utility
+
+function bytesToFr(bytes calldata proofSection) pure returns (Fr scalar) {
+    scalar = FrLib.fromBytes32(bytes32(proofSection));
+}
+
+// EC Point utilities
+function bytesToG1Point(bytes calldata proofSection) pure returns (Honk.G1Point memory point) {
+    uint256 x = uint256(bytes32(proofSection[0x00:0x20]));
+    uint256 y = uint256(bytes32(proofSection[0x20:0x40]));
+    require(x < Q && y < Q, Errors.ValueGeGroupOrder());
+
+    // (0,0) is the canonical EIP-196 encoding of the identity. It is accepted here
+    // because polynomial commitments to identically-zero polynomials (e.g. unused
+    // selector or table polys) are legitimately the identity. On-curve validation
+    // (y² = x³ + 3) is handled by the ecAdd/ecMul precompiles per EIP-196.
+    point = Honk.G1Point({x: x, y: y});
+}
+
+function negateInplace(Honk.G1Point memory point) pure returns (Honk.G1Point memory) {
+    // When y == 0 (order-2 point), negation is the same point. Q - 0 = Q which is >= Q.
+    if (point.y != 0) {
+        point.y = Q - point.y;
+    }
+    return point;
+}
+
+/**
+ * Convert the pairing points to G1 points.
+ *
+ * The pairing points are serialised as an array of 2 limbs representing two points
+ * (P0 and P1, used for lhs and rhs of pairing operation).
+ *
+ * There are 2 limbs (lo, hi) for each coordinate, so 4 limbs per point, 8 total.
+ * Layout: [P0.x_lo, P0.x_hi, P0.y_lo, P0.y_hi, P1.x_lo, P1.x_hi, P1.y_lo, P1.y_hi]
+ *
+ * @param pairingPoints The pairing points to convert.
+ * @return lhs P0 point
+ * @return rhs P1 point
+ */
+function convertPairingPointsToG1(Fr[PAIRING_POINTS_SIZE] memory pairingPoints)
+    pure
+    returns (Honk.G1Point memory lhs, Honk.G1Point memory rhs)
+{
+    // P0 (lhs): x = lo | (hi << 136)
+    uint256 lhsX = Fr.unwrap(pairingPoints[0]);
+    lhsX |= Fr.unwrap(pairingPoints[1]) << 136;
+
+    uint256 lhsY = Fr.unwrap(pairingPoints[2]);
+    lhsY |= Fr.unwrap(pairingPoints[3]) << 136;
+
+    // P1 (rhs): x = lo | (hi << 136)
+    uint256 rhsX = Fr.unwrap(pairingPoints[4]);
+    rhsX |= Fr.unwrap(pairingPoints[5]) << 136;
+
+    uint256 rhsY = Fr.unwrap(pairingPoints[6]);
+    rhsY |= Fr.unwrap(pairingPoints[7]) << 136;
+
+    // Reconstructed coordinates must be < Q to prevent malleability.
+    // Without this, two different limb encodings could map to the same curve point
+    // (via mulmod reduction in on-curve checks) but produce different transcript hashes.
+    require(lhsX < Q && lhsY < Q && rhsX < Q && rhsY < Q, Errors.ValueGeGroupOrder());
+
+    lhs.x = lhsX;
+    lhs.y = lhsY;
+    rhs.x = rhsX;
+    rhs.y = rhsY;
+}
+
+/**
+ * Hash the pairing inputs from the present verification context with those extracted from the public inputs.
+ *
+ * @param proofPairingPoints Pairing points from the proof - (public inputs).
+ * @param accLhs Accumulator point for the left side - result of shplemini.
+ * @param accRhs Accumulator point for the right side - result of shplemini.
+ * @return recursionSeparator The recursion separator - generated from hashing the above.
+ */
+function generateRecursionSeparator(
+    Fr[PAIRING_POINTS_SIZE] memory proofPairingPoints,
+    Honk.G1Point memory accLhs,
+    Honk.G1Point memory accRhs
+) pure returns (Fr recursionSeparator) {
+    // hash the proof aggregated X
+    // hash the proof aggregated Y
+    // hash the accum X
+    // hash the accum Y
+
+    (Honk.G1Point memory proofLhs, Honk.G1Point memory proofRhs) = convertPairingPointsToG1(proofPairingPoints);
+
+    uint256[8] memory recursionSeparatorElements;
+
+    // Proof points
+    recursionSeparatorElements[0] = proofLhs.x;
+    recursionSeparatorElements[1] = proofLhs.y;
+    recursionSeparatorElements[2] = proofRhs.x;
+    recursionSeparatorElements[3] = proofRhs.y;
+
+    // Accumulator points
+    recursionSeparatorElements[4] = accLhs.x;
+    recursionSeparatorElements[5] = accLhs.y;
+    recursionSeparatorElements[6] = accRhs.x;
+    recursionSeparatorElements[7] = accRhs.y;
+
+    recursionSeparator = FrLib.from(uint256(keccak256(abi.encodePacked(recursionSeparatorElements))) % P);
+}
+
+/**
+ * G1 Mul with Separator
+ * Using the ecAdd and ecMul precompiles
+ *
+ * @param basePoint The point to multiply.
+ * @param other The other point to add.
+ * @param recursionSeperator The separator to use for the multiplication.
+ * @return `(recursionSeperator * basePoint) + other`.
+ */
+function mulWithSeperator(Honk.G1Point memory basePoint, Honk.G1Point memory other, Fr recursionSeperator)
+    view
+    returns (Honk.G1Point memory)
+{
+    Honk.G1Point memory result;
+
+    result = ecMul(recursionSeperator, basePoint);
+    result = ecAdd(result, other);
+
+    return result;
+}
+
+/**
+ * G1 Mul
+ * Takes a Fr value and a G1 point and uses the ecMul precompile to return the result.
+ *
+ * @param value The value to multiply the point by.
+ * @param point The point to multiply.
+ * @return result The result of the multiplication.
+ */
+function ecMul(Fr value, Honk.G1Point memory point) view returns (Honk.G1Point memory) {
+    Honk.G1Point memory result;
+
+    assembly {
+        let free := mload(0x40)
+        // Write the point into memory (two 32 byte words)
+        // Memory layout:
+        // Address    |  value
+        // free       |  point.x
+        // free + 0x20|  point.y
+        mstore(free, mload(point))
+        mstore(add(free, 0x20), mload(add(point, 0x20)))
+        // Write the scalar into memory (one 32 byte word)
+        // Memory layout:
+        // Address    |  value
+        // free + 0x40|  value
+        mstore(add(free, 0x40), value)
+
+        // Call the ecMul precompile, it takes in the following
+        // [point.x, point.y, scalar], and returns the result back into the free memory location.
+        let success := staticcall(gas(), 0x07, free, 0x60, free, 0x40)
+        if iszero(success) {
+            revert(0, 0)
+        }
+        // Copy the result of the multiplication back into the result memory location.
+        // Memory layout:
+        // Address    |  value
+        // result     |  result.x
+        // result + 0x20|  result.y
+        mstore(result, mload(free))
+        mstore(add(result, 0x20), mload(add(free, 0x20)))
+
+        mstore(0x40, add(free, 0x60))
+    }
+
+    return result;
+}
+
+/**
+ * G1 Add
+ * Takes two G1 points and uses the ecAdd precompile to return the result.
+ *
+ * @param lhs The left hand side of the addition.
+ * @param rhs The right hand side of the addition.
+ * @return result The result of the addition.
+ */
+function ecAdd(Honk.G1Point memory lhs, Honk.G1Point memory rhs) view returns (Honk.G1Point memory) {
+    Honk.G1Point memory result;
+
+    assembly {
+        let free := mload(0x40)
+        // Write lhs into memory (two 32 byte words)
+        // Memory layout:
+        // Address    |  value
+        // free       |  lhs.x
+        // free + 0x20|  lhs.y
+        mstore(free, mload(lhs))
+        mstore(add(free, 0x20), mload(add(lhs, 0x20)))
+
+        // Write rhs into memory (two 32 byte words)
+        // Memory layout:
+        // Address    |  value
+        // free + 0x40|  rhs.x
+        // free + 0x60|  rhs.y
+        mstore(add(free, 0x40), mload(rhs))
+        mstore(add(free, 0x60), mload(add(rhs, 0x20)))
+
+        // Call the ecAdd precompile, it takes in the following
+        // [lhs.x, lhs.y, rhs.x, rhs.y], and returns their addition back into the free memory location.
+        let success := staticcall(gas(), 0x06, free, 0x80, free, 0x40)
+        if iszero(success) { revert(0, 0) }
+
+        // Copy the result of the addition back into the result memory location.
+        // Memory layout:
+        // Address    |  value
+        // result     |  result.x
+        // result + 0x20|  result.y
+        mstore(result, mload(free))
+        mstore(add(result, 0x20), mload(add(free, 0x20)))
+
+        mstore(0x40, add(free, 0x80))
+    }
+
+    return result;
+}
+
+function rejectPointAtInfinity(Honk.G1Point memory point) pure {
+    require((point.x | point.y) != 0, Errors.PointAtInfinity());
+}
+
+/**
+ * Check if pairing point limbs are all zero (default/infinity).
+ * Default pairing points indicate no recursive verification occurred.
+ */
+function arePairingPointsDefault(Fr[PAIRING_POINTS_SIZE] memory pairingPoints) pure returns (bool) {
+    uint256 acc = 0;
+    for (uint256 i = 0; i < PAIRING_POINTS_SIZE; i++) {
+        acc |= Fr.unwrap(pairingPoints[i]);
+    }
+    return acc == 0;
+}
+
+function pairing(Honk.G1Point memory rhs, Honk.G1Point memory lhs) view returns (bool decodedResult) {
+    bytes memory input = abi.encodePacked(
+        rhs.x,
+        rhs.y,
+        // Fixed G2 point
+        uint256(0x198e9393920d483a7260bfb731fb5d25f1aa493335a9e71297e485b7aef312c2),
+        uint256(0x1800deef121f1e76426a00665e5c4479674322d4f75edadd46debd5cd992f6ed),
+        uint256(0x090689d0585ff075ec9e99ad690c3395bc4b313370b38ef355acdadcd122975b),
+        uint256(0x12c85ea5db8c6deb4aab71808dcb408fe3d1e7690c43d37b4ce6cc0166fa7daa),
+        lhs.x,
+        lhs.y,
+        // G2 point from VK
+        uint256(0x260e01b251f6f1c7e7ff4e580791dee8ea51d87a358e038b4efe30fac09383c1),
+        uint256(0x0118c4d5b837bcc2bc89b5b398b5974e9f5944073b32078b7e231fec938883b0),
+        uint256(0x04fc6369f7110fe3d25156c1bb9a72859cf2a04641f99ba4ee413c80da6a5fe4),
+        uint256(0x22febda3c0c0632a56475b4214e5615e11e6dd3f96e6cea2854a87d4dacc5e55)
+    );
+
+    (bool success, bytes memory result) = address(0x08).staticcall(input);
+    decodedResult = success && abi.decode(result, (bool));
+}
+
+abstract contract BaseZKHonkVerifier is IVerifier {
+    using FrLib for Fr;
+
+    struct PairingInputs {
+        Honk.G1Point P_0;
+        Honk.G1Point P_1;
+    }
+
+    struct SmallSubgroupIpaIntermediates {
+        Fr[SUBGROUP_SIZE] challengePolyLagrange;
+        Fr challengePolyEval;
+        Fr lagrangeFirst;
+        Fr lagrangeLast;
+        Fr rootPower;
+        Fr[SUBGROUP_SIZE] denominators; // this has to disappear
+        Fr diff;
+    }
+
+    // Constants for proof length calculation (matching UltraKeccakZKFlavor)
+    uint256 internal constant NUM_WITNESS_ENTITIES = 8 + NUM_MASKING_POLYNOMIALS;
+    uint256 internal constant NUM_ELEMENTS_COMM = 2; // uint256 elements for curve points
+    uint256 internal constant NUM_ELEMENTS_FR = 1; // uint256 elements for field elements
+    uint256 internal constant NUM_LIBRA_EVALUATIONS = 4; // libra evaluations
+
+    uint256 internal constant LIBRA_COMMITMENTS = 3;
+    uint256 internal constant LIBRA_EVALUATIONS = 4;
+    uint256 internal constant LIBRA_UNIVARIATES_LENGTH = 9;
+
+    uint256 internal constant SHIFTED_COMMITMENTS_START = 30;
+    uint256 internal constant PERMUTATION_ARGUMENT_VALUE_SEPARATOR = 1 << 28;
+
+    uint256 internal immutable $N;
+    uint256 internal immutable $LOG_N;
+    uint256 internal immutable $VK_HASH;
+    uint256 internal immutable $NUM_PUBLIC_INPUTS;
+    uint256 internal immutable $MSMSize;
+
+    constructor(uint256 _N, uint256 _logN, uint256 _vkHash, uint256 _numPublicInputs) {
+        $N = _N;
+        $LOG_N = _logN;
+        $VK_HASH = _vkHash;
+        $NUM_PUBLIC_INPUTS = _numPublicInputs;
+        $MSMSize = NUMBER_UNSHIFTED_ZK + _logN + LIBRA_COMMITMENTS + 2;
+    }
+
+    function verify(bytes calldata proof, bytes32[] calldata publicInputs)
+        public
+        view
+        override
+        returns (bool verified)
+    {
+        // Calculate expected proof size based on $LOG_N
+        uint256 expectedProofSize = calculateProofSize($LOG_N);
+
+        // Check the received proof is the expected size where each field element is 32 bytes
+        require(
+            proof.length == expectedProofSize, Errors.ProofLengthWrongWithLogN($LOG_N, proof.length, expectedProofSize)
+        );
+
+        Honk.VerificationKey memory vk = loadVerificationKey();
+        Honk.ZKProof memory p = ZKTranscriptLib.loadProof(proof, $LOG_N);
+
+        require(publicInputs.length == vk.publicInputsSize - PAIRING_POINTS_SIZE, Errors.PublicInputsLengthWrong());
+
+        // Generate the fiat shamir challenges for the whole protocol
+        ZKTranscript memory t =
+            ZKTranscriptLib.generateTranscript(p, publicInputs, $VK_HASH, $NUM_PUBLIC_INPUTS, $LOG_N);
+
+        // Derive public input delta
+        t.relationParameters.publicInputsDelta = computePublicInputDelta(
+            publicInputs,
+            p.pairingPointObject,
+            t.relationParameters.beta,
+            t.relationParameters.gamma,
+            5 // pubInputsOffset = NUM_DISABLED_ROWS_IN_SUMCHECK + NUM_ZERO_ROWS = 4 + 1
+        );
+
+        // Sumcheck
+        require(verifySumcheck(p, t), Errors.SumcheckFailed());
+        require(verifyShplemini(p, vk, t), Errors.ShpleminiFailed());
+
+        verified = true;
+    }
+
+    function computePublicInputDelta(
+        bytes32[] memory publicInputs,
+        Fr[PAIRING_POINTS_SIZE] memory pairingPointObject,
+        Fr beta,
+        Fr gamma,
+        uint256 offset
+    ) internal view returns (Fr publicInputDelta) {
+        Fr numerator = Fr.wrap(1);
+        Fr denominator = Fr.wrap(1);
+
+        Fr numeratorAcc = gamma + (beta * FrLib.from(PERMUTATION_ARGUMENT_VALUE_SEPARATOR + offset));
+        Fr denominatorAcc = gamma - (beta * FrLib.from(offset + 1));
+
+        {
+            for (uint256 i = 0; i < $NUM_PUBLIC_INPUTS - PAIRING_POINTS_SIZE; i++) {
+                Fr pubInput = FrLib.fromBytes32(publicInputs[i]);
+
+                numerator = numerator * (numeratorAcc + pubInput);
+                denominator = denominator * (denominatorAcc + pubInput);
+
+                numeratorAcc = numeratorAcc + beta;
+                denominatorAcc = denominatorAcc - beta;
+            }
+
+            for (uint256 i = 0; i < PAIRING_POINTS_SIZE; i++) {
+                Fr pubInput = pairingPointObject[i];
+
+                numerator = numerator * (numeratorAcc + pubInput);
+                denominator = denominator * (denominatorAcc + pubInput);
+
+                numeratorAcc = numeratorAcc + beta;
+                denominatorAcc = denominatorAcc - beta;
+            }
+        }
+
+        // Fr delta = numerator / denominator; // TOOO: batch invert later?
+        publicInputDelta = FrLib.div(numerator, denominator);
+    }
+
+    function verifySumcheck(Honk.ZKProof memory proof, ZKTranscript memory tp) internal view returns (bool verified) {
+        Fr roundTargetSum = tp.libraChallenge * proof.libraSum; // default 0
+        Fr powPartialEvaluation = Fr.wrap(1);
+
+        // We perform sumcheck reductions over log n rounds ( the multivariate degree )
+        for (uint256 round; round < $LOG_N; ++round) {
+            Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH] memory roundUnivariate = proof.sumcheckUnivariates[round];
+            Fr totalSum = roundUnivariate[0] + roundUnivariate[1];
+            require(totalSum == roundTargetSum, Errors.SumcheckFailed());
+
+            Fr roundChallenge = tp.sumCheckUChallenges[round];
+
+            // Update the round target for the next rounf
+            roundTargetSum = computeNextTargetSum(roundUnivariate, roundChallenge);
+            powPartialEvaluation =
+                powPartialEvaluation * (Fr.wrap(1) + roundChallenge * (tp.gateChallenges[round] - Fr.wrap(1)));
+        }
+
+        // Last round
+        // For ZK flavors: sumcheckEvaluations has 42 elements
+        // Index 0 is gemini_masking_poly, indices 1-41 are the regular entities used in relations
+        Fr[NUMBER_OF_ENTITIES] memory relationsEvaluations;
+        for (uint256 i = 0; i < NUMBER_OF_ENTITIES; i++) {
+            relationsEvaluations[i] = proof.sumcheckEvaluations[i + NUM_MASKING_POLYNOMIALS]; // Skip gemini_masking_poly at index 0
+        }
+        Fr grandHonkRelationSum = RelationsLib.accumulateRelationEvaluations(
+            relationsEvaluations, tp.relationParameters, tp.alphas, powPartialEvaluation
+        );
+
+        // Row-disabling polynomial: 1 - ∏_{i≥2}(1 - u_i)
+        Fr evaluation = Fr.wrap(1);
+        for (uint256 i = 2; i < $LOG_N; i++) {
+            evaluation = evaluation * (Fr.wrap(1) - tp.sumCheckUChallenges[i]);
+        }
+
+        grandHonkRelationSum =
+            grandHonkRelationSum * (Fr.wrap(1) - evaluation) + proof.libraEvaluation * tp.libraChallenge;
+        verified = (grandHonkRelationSum == roundTargetSum);
+    }
+
+    // Return the new target sum for the next sumcheck round
+    function computeNextTargetSum(Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH] memory roundUnivariates, Fr roundChallenge)
+        internal
+        view
+        returns (Fr targetSum)
+    {
+        Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH] memory BARYCENTRIC_LAGRANGE_DENOMINATORS = [
+            Fr.wrap(0x0000000000000000000000000000000000000000000000000000000000009d80),
+            Fr.wrap(0x30644e72e131a029b85045b68181585d2833e84879b9709143e1f593efffec51),
+            Fr.wrap(0x00000000000000000000000000000000000000000000000000000000000005a0),
+            Fr.wrap(0x30644e72e131a029b85045b68181585d2833e84879b9709143e1f593effffd31),
+            Fr.wrap(0x0000000000000000000000000000000000000000000000000000000000000240),
+            Fr.wrap(0x30644e72e131a029b85045b68181585d2833e84879b9709143e1f593effffd31),
+            Fr.wrap(0x00000000000000000000000000000000000000000000000000000000000005a0),
+            Fr.wrap(0x30644e72e131a029b85045b68181585d2833e84879b9709143e1f593efffec51),
+            Fr.wrap(0x0000000000000000000000000000000000000000000000000000000000009d80)
+        ];
+
+        // To compute the next target sum, we evaluate the given univariate at a point u (challenge).
+
+        // Performing Barycentric evaluations
+        // Compute B(x)
+        Fr numeratorValue = Fr.wrap(1);
+        for (uint256 i = 0; i < ZK_BATCHED_RELATION_PARTIAL_LENGTH; ++i) {
+            numeratorValue = numeratorValue * (roundChallenge - Fr.wrap(i));
+        }
+
+        Fr[ZK_BATCHED_RELATION_PARTIAL_LENGTH] memory denominatorInverses;
+        for (uint256 i = 0; i < ZK_BATCHED_RELATION_PARTIAL_LENGTH; ++i) {
+            denominatorInverses[i] = FrLib.invert(BARYCENTRIC_LAGRANGE_DENOMINATORS[i] * (roundChallenge - Fr.wrap(i)));
+        }
+
+        for (uint256 i = 0; i < ZK_BATCHED_RELATION_PARTIAL_LENGTH; ++i) {
+            targetSum = targetSum + roundUnivariates[i] * denominatorInverses[i];
+        }
+
+        // Scale the sum by the value of B(x)
+        targetSum = targetSum * numeratorValue;
+    }
+
+    function verifyShplemini(Honk.ZKProof memory proof, Honk.VerificationKey memory vk, ZKTranscript memory tp)
+        internal
+        view
+        returns (bool verified)
+    {
+        CommitmentSchemeLib.ShpleminiIntermediates memory mem; // stack
+
+        // - Compute vector (r, r², ... , r²⁽ⁿ⁻¹⁾), where n = log_circuit_size
+        Fr[] memory powers_of_evaluation_challenge = CommitmentSchemeLib.computeSquares(tp.geminiR, $LOG_N);
+        // Arrays hold values that will be linearly combined for the gemini and shplonk batch openings
+        Fr[] memory scalars = new Fr[]($MSMSize);
+        Honk.G1Point[] memory commitments = new Honk.G1Point[]($MSMSize);
+
+        mem.posInvertedDenominator = (tp.shplonkZ - powers_of_evaluation_challenge[0]).invert();
+        mem.negInvertedDenominator = (tp.shplonkZ + powers_of_evaluation_challenge[0]).invert();
+
+        mem.unshiftedScalar = mem.posInvertedDenominator + (tp.shplonkNu * mem.negInvertedDenominator);
+        mem.shiftedScalar =
+            tp.geminiR.invert() * (mem.posInvertedDenominator - (tp.shplonkNu * mem.negInvertedDenominator));
+
+        scalars[0] = Fr.wrap(1);
+        commitments[0] = proof.shplonkQ;
+
+        /* Batch multivariate opening claims, shifted and unshifted
+        * The vector of scalars is populated as follows:
+        * \f[
+        * \left(
+        * - \left(\frac{1}{z-r} + \nu \times \frac{1}{z+r}\right),
+        * \ldots,
+        * - \rho^{i+k-1} \times \left(\frac{1}{z-r} + \nu \times \frac{1}{z+r}\right),
+        * - \rho^{i+k} \times \frac{1}{r} \times \left(\frac{1}{z-r} - \nu \times \frac{1}{z+r}\right),
+        * \ldots,
+        * - \rho^{k+m-1} \times \frac{1}{r} \times \left(\frac{1}{z-r} - \nu \times \frac{1}{z+r}\right)
+        * \right)
+        * \f]
+        *
+        * The following vector is concatenated to the vector of commitments:
+        * \f[
+        * f_0, \ldots, f_{m-1}, f_{\text{shift}, 0}, \ldots, f_{\text{shift}, k-1}
+        * \f]
+        *
+        * Simultaneously, the evaluation of the multilinear polynomial
+        * \f[
+        * \sum \rho^i \cdot f_i + \sum \rho^{i+k} \cdot f_{\text{shift}, i}
+        * \f]
+        * at the challenge point \f$ (u_0,\ldots, u_{n-1}) \f$ is computed.
+        *
+        * This approach minimizes the number of iterations over the commitments to multilinear polynomials
+        * and eliminates the need to store the powers of \f$ \rho \f$.
+        */
+        // For ZK flavors: evaluations array is [gemini_masking_poly, qm, qc, ql, qr, ...]
+        // Start batching challenge at 1, not rho, to match non-ZK pattern
+        mem.batchingChallenge = Fr.wrap(1);
+        mem.batchedEvaluation = Fr.wrap(0);
+
+        mem.unshiftedScalarNeg = mem.unshiftedScalar.neg();
+        mem.shiftedScalarNeg = mem.shiftedScalar.neg();
+
+        // Process all NUMBER_UNSHIFTED_ZK evaluations (includes gemini_masking_poly at index 0)
+        for (uint256 i = 1; i <= NUMBER_UNSHIFTED_ZK; ++i) {
+            scalars[i] = mem.unshiftedScalarNeg * mem.batchingChallenge;
+            mem.batchedEvaluation = mem.batchedEvaluation
+                + (proof.sumcheckEvaluations[i - NUM_MASKING_POLYNOMIALS] * mem.batchingChallenge);
+            mem.batchingChallenge = mem.batchingChallenge * tp.rho;
+        }
+        // g commitments are accumulated at r
+        // For each of the to be shifted commitments perform the shift in place by
+        // adding to the unshifted value.
+        // We do so, as the values are to be used in batchMul later, and as
+        // `a * c + b * c = (a + b) * c` this will allow us to reduce memory and compute.
+        // Applied to w1, w2, w3, w4 and zPerm
+        for (uint256 i = 0; i < NUMBER_TO_BE_SHIFTED; ++i) {
+            uint256 scalarOff = i + SHIFTED_COMMITMENTS_START;
+            uint256 evaluationOff = i + NUMBER_UNSHIFTED_ZK;
+
+            scalars[scalarOff] = scalars[scalarOff] + (mem.shiftedScalarNeg * mem.batchingChallenge);
+            mem.batchedEvaluation =
+                mem.batchedEvaluation + (proof.sumcheckEvaluations[evaluationOff] * mem.batchingChallenge);
+            mem.batchingChallenge = mem.batchingChallenge * tp.rho;
+        }
+
+        commitments[1] = proof.geminiMaskingPoly;
+
+        commitments[2] = vk.qm;
+        commitments[3] = vk.qc;
+        commitments[4] = vk.ql;
+        commitments[5] = vk.qr;
+        commitments[6] = vk.qo;
+        commitments[7] = vk.q4;
+        commitments[8] = vk.qLookup;
+        commitments[9] = vk.qArith;
+        commitments[10] = vk.qDeltaRange;
+        commitments[11] = vk.qElliptic;
+        commitments[12] = vk.qMemory;
+        commitments[13] = vk.qNnf;
+        commitments[14] = vk.qPoseidon2External;
+        commitments[15] = vk.qPoseidon2Internal;
+        commitments[16] = vk.s1;
+        commitments[17] = vk.s2;
+        commitments[18] = vk.s3;
+        commitments[19] = vk.s4;
+        commitments[20] = vk.id1;
+        commitments[21] = vk.id2;
+        commitments[22] = vk.id3;
+        commitments[23] = vk.id4;
+        commitments[24] = vk.t1;
+        commitments[25] = vk.t2;
+        commitments[26] = vk.t3;
+        commitments[27] = vk.t4;
+        commitments[28] = vk.lagrangeFirst;
+        commitments[29] = vk.lagrangeLast;
+
+        // Accumulate proof points
+        commitments[30] = proof.w1;
+        commitments[31] = proof.w2;
+        commitments[32] = proof.w3;
+        commitments[33] = proof.w4;
+        commitments[34] = proof.zPerm;
+        commitments[35] = proof.lookupInverses;
+        commitments[36] = proof.lookupReadCounts;
+        commitments[37] = proof.lookupReadTags;
+
+        /* Batch gemini claims from the prover
+         * place the commitments to gemini aᵢ to the vector of commitments, compute the contributions from
+         * aᵢ(−r²ⁱ) for i=1, … , n−1 to the constant term accumulator, add corresponding scalars
+         *
+         * 1. Moves the vector
+         * \f[
+         * \left( \text{com}(A_1), \text{com}(A_2), \ldots, \text{com}(A_{n-1}) \right)
+         * \f]
+        * to the 'commitments' vector.
+        *
+        * 2. Computes the scalars:
+        * \f[
+        * \frac{\nu^{2}}{z + r^2}, \frac{\nu^3}{z + r^4}, \ldots, \frac{\nu^{n-1}}{z + r^{2^{n-1}}}
+        * \f]
+        * and places them into the 'scalars' vector.
+        *
+        * 3. Accumulates the summands of the constant term:
+         * \f[
+         * \sum_{i=2}^{n-1} \frac{\nu^{i} \cdot A_i(-r^{2^i})}{z + r^{2^i}}
+         * \f]
+         * and adds them to the 'constant_term_accumulator'.
+         */
+
+        // Add contributions from A₀(r) and A₀(-r) to constant_term_accumulator:
+        // Compute the evaluations Aₗ(r^{2ˡ}) for l = 0, ..., $LOG_N - 1
+        Fr[] memory foldPosEvaluations = CommitmentSchemeLib.computeFoldPosEvaluations(
+            tp.sumCheckUChallenges,
+            mem.batchedEvaluation,
+            proof.geminiAEvaluations,
+            powers_of_evaluation_challenge,
+            $LOG_N
+        );
+
+        mem.constantTermAccumulator = foldPosEvaluations[0] * mem.posInvertedDenominator;
+        mem.constantTermAccumulator =
+            mem.constantTermAccumulator + (proof.geminiAEvaluations[0] * tp.shplonkNu * mem.negInvertedDenominator);
+
+        mem.batchingChallenge = tp.shplonkNu.sqr();
+        uint256 boundary = NUMBER_UNSHIFTED_ZK + 1;
+
+        // Compute Shplonk constant term contributions from Aₗ(± r^{2ˡ}) for l = 1, ..., m-1;
+        // Compute scalar multipliers for each fold commitment
+        for (uint256 i = 0; i < $LOG_N - 1; ++i) {
+            bool dummy_round = i >= ($LOG_N - 1);
+
+            if (!dummy_round) {
+                // Update inverted denominators
+                mem.posInvertedDenominator = (tp.shplonkZ - powers_of_evaluation_challenge[i + 1]).invert();
+                mem.negInvertedDenominator = (tp.shplonkZ + powers_of_evaluation_challenge[i + 1]).invert();
+
+                // Compute the scalar multipliers for Aₗ(± r^{2ˡ}) and [Aₗ]
+                mem.scalingFactorPos = mem.batchingChallenge * mem.posInvertedDenominator;
+                mem.scalingFactorNeg = mem.batchingChallenge * tp.shplonkNu * mem.negInvertedDenominator;
+                scalars[boundary + i] = mem.scalingFactorNeg.neg() + mem.scalingFactorPos.neg();
+
+                // Accumulate the const term contribution given by
+                // v^{2l} * Aₗ(r^{2ˡ}) /(z-r^{2^l}) + v^{2l+1} * Aₗ(-r^{2ˡ}) /(z+ r^{2^l})
+                Fr accumContribution = mem.scalingFactorNeg * proof.geminiAEvaluations[i + 1];
+                accumContribution = accumContribution + mem.scalingFactorPos * foldPosEvaluations[i + 1];
+                mem.constantTermAccumulator = mem.constantTermAccumulator + accumContribution;
+            }
+            // Update the running power of v
+            mem.batchingChallenge = mem.batchingChallenge * tp.shplonkNu * tp.shplonkNu;
+
+            commitments[boundary + i] = proof.geminiFoldComms[i];
+        }
+
+        boundary += $LOG_N - 1;
+
+        // Finalize the batch opening claim
+        mem.denominators[0] = Fr.wrap(1).div(tp.shplonkZ - tp.geminiR);
+        mem.denominators[1] = Fr.wrap(1).div(tp.shplonkZ - SUBGROUP_GENERATOR * tp.geminiR);
+        mem.denominators[2] = mem.denominators[0];
+        mem.denominators[3] = mem.denominators[0];
+
+        for (uint256 i = 0; i < LIBRA_EVALUATIONS; i++) {
+            Fr scalingFactor = mem.denominators[i] * mem.batchingChallenge;
+            mem.batchingScalars[i] = scalingFactor.neg();
+            mem.batchingChallenge = mem.batchingChallenge * tp.shplonkNu;
+            mem.constantTermAccumulator = mem.constantTermAccumulator + scalingFactor * proof.libraPolyEvals[i];
+        }
+        scalars[boundary] = mem.batchingScalars[0];
+        scalars[boundary + 1] = mem.batchingScalars[1] + mem.batchingScalars[2];
+        scalars[boundary + 2] = mem.batchingScalars[3];
+
+        for (uint256 i = 0; i < LIBRA_COMMITMENTS; i++) {
+            commitments[boundary++] = proof.libraCommitments[i];
+        }
+
+        commitments[boundary] = Honk.G1Point({x: 1, y: 2});
+        scalars[boundary++] = mem.constantTermAccumulator;
+
+        require(
+            checkEvalsConsistency(proof.libraPolyEvals, tp.geminiR, tp.sumCheckUChallenges, proof.libraEvaluation),
+            Errors.ConsistencyCheckFailed()
+        );
+
+        Honk.G1Point memory quotient_commitment = proof.kzgQuotient;
+
+        commitments[boundary] = quotient_commitment;
+        scalars[boundary] = tp.shplonkZ; // evaluation challenge
+
+        PairingInputs memory pair;
+        pair.P_0 = batchMul(commitments, scalars);
+        pair.P_1 = negateInplace(quotient_commitment);
+
+        // Aggregate pairing points (skip if default/infinity — no recursive verification occurred)
+        if (!arePairingPointsDefault(proof.pairingPointObject)) {
+            Fr recursionSeparator = generateRecursionSeparator(proof.pairingPointObject, pair.P_0, pair.P_1);
+            (Honk.G1Point memory P_0_other, Honk.G1Point memory P_1_other) =
+                convertPairingPointsToG1(proof.pairingPointObject);
+
+            // Validate the points from the proof are on the curve
+            rejectPointAtInfinity(P_0_other);
+            rejectPointAtInfinity(P_1_other);
+
+            // accumulate with aggregate points in proof
+            pair.P_0 = mulWithSeperator(pair.P_0, P_0_other, recursionSeparator);
+            pair.P_1 = mulWithSeperator(pair.P_1, P_1_other, recursionSeparator);
+        }
+
+        return pairing(pair.P_0, pair.P_1);
+    }
+
+    function checkEvalsConsistency(
+        Fr[LIBRA_EVALUATIONS] memory libraPolyEvals,
+        Fr geminiR,
+        Fr[CONST_PROOF_SIZE_LOG_N] memory uChallenges,
+        Fr libraEval
+    ) internal view returns (bool check) {
+        Fr one = Fr.wrap(1);
+        Fr vanishingPolyEval = geminiR.pow(SUBGROUP_SIZE) - one;
+        require(vanishingPolyEval != Fr.wrap(0), Errors.GeminiChallengeInSubgroup());
+
+        SmallSubgroupIpaIntermediates memory mem;
+        mem.challengePolyLagrange[0] = one;
+        for (uint256 round = 0; round < $LOG_N; round++) {
+            uint256 currIdx = 1 + LIBRA_UNIVARIATES_LENGTH * round;
+            mem.challengePolyLagrange[currIdx] = one;
+            for (uint256 idx = currIdx + 1; idx < currIdx + LIBRA_UNIVARIATES_LENGTH; idx++) {
+                mem.challengePolyLagrange[idx] = mem.challengePolyLagrange[idx - 1] * uChallenges[round];
+            }
+        }
+
+        mem.rootPower = one;
+        mem.challengePolyEval = Fr.wrap(0);
+        for (uint256 idx = 0; idx < SUBGROUP_SIZE; idx++) {
+            mem.denominators[idx] = mem.rootPower * geminiR - one;
+            mem.denominators[idx] = mem.denominators[idx].invert();
+            mem.challengePolyEval = mem.challengePolyEval + mem.challengePolyLagrange[idx] * mem.denominators[idx];
+            mem.rootPower = mem.rootPower * SUBGROUP_GENERATOR_INVERSE;
+        }
+
+        Fr numerator = vanishingPolyEval * Fr.wrap(SUBGROUP_SIZE).invert();
+        mem.challengePolyEval = mem.challengePolyEval * numerator;
+        mem.lagrangeFirst = mem.denominators[0] * numerator;
+        mem.lagrangeLast = mem.denominators[SUBGROUP_SIZE - 1] * numerator;
+
+        mem.diff = mem.lagrangeFirst * libraPolyEvals[2];
+
+        mem.diff = mem.diff + (geminiR - SUBGROUP_GENERATOR_INVERSE)
+            * (libraPolyEvals[1] - libraPolyEvals[2] - libraPolyEvals[0] * mem.challengePolyEval);
+        mem.diff = mem.diff + mem.lagrangeLast * (libraPolyEvals[2] - libraEval) - vanishingPolyEval * libraPolyEvals[3];
+
+        check = mem.diff == Fr.wrap(0);
+    }
+
+    // This implementation is the same as above with different constants
+    function batchMul(Honk.G1Point[] memory base, Fr[] memory scalars)
+        internal
+        view
+        returns (Honk.G1Point memory result)
+    {
+        uint256 limit = $MSMSize;
+
+        // Identity bases are accepted: VK selector/table polys may be identically zero,
+        // and the ecAdd/ecMul precompiles treat (0,0) as the additive identity per EIP-196.
+        // Soundness against an attacker substituting (0,0) for a non-zero commitment is
+        // upheld by sumcheck/Shplemini, which would fail on inconsistent evaluations.
+
+        bool success = true;
+        assembly {
+            let free := mload(0x40)
+
+            let count := 0x01
+            for {} lt(count, add(limit, 1)) { count := add(count, 1) } {
+                // Get loop offsets
+                let base_base := add(base, mul(count, 0x20))
+                let scalar_base := add(scalars, mul(count, 0x20))
+
+                mstore(add(free, 0x40), mload(mload(base_base)))
+                mstore(add(free, 0x60), mload(add(0x20, mload(base_base))))
+                // Add scalar
+                mstore(add(free, 0x80), mload(scalar_base))
+
+                success := and(success, staticcall(gas(), 7, add(free, 0x40), 0x60, add(free, 0x40), 0x40))
+                // accumulator = accumulator + accumulator_2
+                success := and(success, staticcall(gas(), 6, free, 0x80, free, 0x40))
+            }
+
+            // Return the result
+            mstore(result, mload(free))
+            mstore(add(result, 0x20), mload(add(free, 0x20)))
+        }
+
+        require(success, Errors.ShpleminiFailed());
+    }
+
+    // Calculate proof size based on log_n (matching UltraKeccakZKFlavor formula)
+    function calculateProofSize(uint256 logN) internal pure returns (uint256) {
+        // Witness and Libra commitments
+        uint256 proofLength = NUM_WITNESS_ENTITIES * NUM_ELEMENTS_COMM; // witness commitments
+        proofLength += NUM_ELEMENTS_COMM * 3; // Libra concat, grand sum, quotient comms + Gemini masking
+
+        // Sumcheck
+        proofLength += logN * ZK_BATCHED_RELATION_PARTIAL_LENGTH * NUM_ELEMENTS_FR; // sumcheck univariates
+        proofLength += NUMBER_OF_ENTITIES_ZK * NUM_ELEMENTS_FR; // sumcheck evaluations
+
+        // Libra and Gemini
+        proofLength += NUM_ELEMENTS_FR * 2; // Libra sum, claimed eval
+        proofLength += logN * NUM_ELEMENTS_FR; // Gemini a evaluations
+        proofLength += NUM_LIBRA_EVALUATIONS * NUM_ELEMENTS_FR; // libra evaluations
+
+        // PCS commitments
+        proofLength += (logN - 1) * NUM_ELEMENTS_COMM; // Gemini Fold commitments
+        proofLength += NUM_ELEMENTS_COMM * 2; // Shplonk Q and KZG W commitments
+
+        // Pairing points
+        proofLength += PAIRING_POINTS_SIZE; // pairing inputs carried on public inputs
+
+        return proofLength * 32;
+    }
+
+    function loadVerificationKey() internal pure virtual returns (Honk.VerificationKey memory);
+}
+
+contract PviumZKVerifier is BaseZKHonkVerifier(N, LOG_N, VK_HASH, NUMBER_OF_PUBLIC_INPUTS) {
+     function loadVerificationKey() internal pure override returns (Honk.VerificationKey memory) {
+       return HonkVerificationKey.loadVerificationKey();
+    }
+}