Thanks for the upgrade. Two real fixes wanted before merge, plus a handful of nits.

Must-fix

1. DoS misclassification — DOS_ERR_NOT_FREE_CONSOLIDATION → ErrTxInvalid (should be ErrTxPolicy)

BDK's core/doserror.hpp lists DOS_ERR_NOT_FREE_CONSOLIDATION with DoS=0, i.e. policy. The mapping in services/validator/ScriptVerifierGoBDK.go classifies it as ErrTxInvalid, and the new test at services/validator/ScriptVerifierGoBDK_test.go:207 encodes that. If a peer relays a non-free-consolidation tx, the node will treat it as invalid and ban — that's a peer-banning consequence for a policy reject. Re-map to ErrTxPolicy (and update the test to match).

2. TestCheckP2SHOutput is a no-op stub (services/validator/TxValidator_test.go:661-676)

The body is now _ = txValidator; _ = txP2SH — no assertions. P2SH-output coverage on the Teranode side drops to zero with this PR. Either restore the assertions against the BDK path, or delete the test if the check now lives entirely in BDK and is covered upstream.

Nits

• Test_ScriptVerificationGoBDK uses zero-defaulted settings.NewPolicySettings(), but TestMaxTxSizePolicy carries a comment saying "BDK rejects values below 99999". One of the two is wrong — either the test will be rejected at construction, or the comment is stale.
• bdkBlockHeight pre-subtracts 1 in policy mode (services/validator/ScriptVerifierGoBDK.go:166-176). Semantically correct vs BDK 1.2.4's consensus=false uses blockHeight+1 era convention, but the rationale isn't obvious at the call site — please add a one-line comment.
• Deleted constants MaxBlockSize, MaxTxSizeConsensus*, MaxTxSigopsCountPolicyAfterGenesis — grep is clean, but please confirm block-validation still has an effective bound from the BDK side (just a sanity-check).
• NewPolicySettings() still returns zero defaults with // TODO set defaults (settings/policy_settings.go:35-39). This PR widens the consumer surface; would be a good moment to either tackle the TODO or document why zero is the right default.
• Stale pushDataCheck reference in docs/references/services/validator_reference.md:230 — pre-existing, but the PR touches the file, easy to fix in-flight.

Happy to re-approve once the two must-fix items are in.

Thanks for the quick turnaround. Both must-fix items resolved cleanly in a3b46755:

• MF1 (DOS_ERR_NOT_FREE_CONSOLIDATION → ErrTxPolicy) — mapping moved into the policy-codes case at ScriptVerifierGoBDK.go:214; test at ScriptVerifierGoBDK_test.go:205 flipped to wantPolicy: true. Locked in across both consensus modes.
• MF2 (TestCheckP2SHOutput) — restored with stronger assertions than the original. NotErrorIs(err, ErrTxPolicy) pins this as a consensus rejection (which it is post-Genesis), and SkipPolicyChecks: true ensures the test doesn't accidentally pass via the policy bypass. Nice fix.

Also liked the 9b3f80f8 cleanup txvalidator interface commit — dropping the GoBT/GoSDK alternates and the two-phase API simplifies the validator surface considerably.

Approving. A few housekeeping items still open from the original review, none gating — fold into a follow-up commit whenever or carry as separate issues:

• bdkBlockHeight pre-subtract at ScriptVerifierGoBDK.go:162-172 could use a one-line comment explaining BDK 1.2.4's consensus=false → blockHeight+1 era convention so the next reader doesn't suspect an off-by-one.
• NewPolicySettings() still returns zero-defaults with // TODO set defaults. Test usage proves BDK treats zero as unlimited, so not broken — but consider renaming to EmptyPolicySettings() or populating from struct tags.
• docs/references/services/validator_reference.md:209 still lists pushDataCheck(tx *bt.Tx) which doesn't exist anywhere in the package.

One operational note worth flagging before prod rollout: bumping to 1.2.4 doesn't fix the testnet 245k-OP_2DUP+OP_CHECKSIGVERIFY hang — that's BDK PR #41, expected in v1.2.5. The go.mod warning comment about #41 was removed in the fix commit; worth a one-line note in the PR description (or an updated go.mod comment) so the next reader understands why the pin moved without #41 having landed. Existing alert regexes matching _Cfunc_ScriptEngine_VerifyScript will also need to be widened to match _Cfunc_TxValidator_ValidateTransaction post-upgrade, since the symbol moved.

Approving — the architectural shift (Teranode → BDK for tx structure/sigops/standardness/output checks) is a clear improvement, and the bdkValidator interface gives clean test injection. Strong coverage in the new boundary tests (TestBDKBlockHeight, TestScriptVerifierGoBDKMapDoSErrors, TestScriptVerifierGoBDKMapScriptErrors).

A few minor cleanups worth addressing in this PR or a follow-up:

Worth a second look before merge:

1. Behavioural change in TestTx5f37c7a38b5e0bc177a4c353481f30c6de1bc46db534019846d7bc829f58254a (services/validator/TxValidator_test.go:743) — historical tx now rejected with ErrTxInvalid under consensus mode. Test expectation flipped without a comment explaining which BDK 1.2.4 check newly rejects it (MvP-4615 input value range?). One-line comment would help future debugging.

2. bdkBlockHeight rejects blockHeight=0 in policy mode (ScriptVerifierGoBDK.go:141). Correct in principle, but a new failure mode — worth confirming no mempool/test path hits this.

Non-blocking polish:

3. Opaque "T14 parity decision" comment in TxValidator.checkInputs — future maintainers won't know what T14 references. Link the tracking issue or restate as "SVNode parity, tracked in MvP-XXXX".

4. Dead params in checkInputs — after the dedup logic was removed, blockHeight and validationOptions are unused (the _ = blockHeight comment claiming "required by the interface" is no longer true). Either drop the params or inline the ~15 remaining lines into ValidateTransaction.

5. BenchmarkConsolidationValidationImpact (TxValidator_benchmark_test.go:135) leaves tv.bdk nil after removing interpreter:. Works today since only checkFees/isConsolidationTx are called, but it's a footgun — set bdk: noopBDKValidator{} for safety.

6. TestMaxTxSigopsCountsPolicy skipped "until a focused BDK ValidateTransaction sigops-policy fixture is added" — given MvP-4613 is a headline behaviour in this PR, please track that fixture in an issue rather than leaving the skip dangling.

7. Direct txValidator.bdk = noopBDKValidator{} mutation in Test_MinFeePolicy and Validator_test.go is awkward — consider a WithBDKValidator(...) option on NewTxValidator if this injection becomes recurring.

8. PR description states DOS_ERR_NOT_FREE_CONSOLIDATION mapping is a "follow-up" item, but the code already maps it to ErrTxPolicy (ScriptVerifierGoBDK.go:165). Either the description is stale or there's a different disagreement still to resolve — please clarify.

9. Stale naming — ScriptVerifierGoBDK.go / scriptVerifierGoBDK no longer "verify scripts"; they adapt to BDK's full transaction validator. Reasonable to defer for diff minimality, but worth a follow-up rename.

10. Please tick the test-plan checkboxes (-tags=bdk and nobdk paths, test/consensus) before merge so reviewers can confirm verification ran.