feat(platform): add typed refund owners to storage flags and fee refunds by DCG-Claude · Pull Request #4789 · dashpay/platform

DCG-Claude · 2026-09-16T20:10:19Z

Issue being fixed or feature implemented

Part of the smart contract plan in #4626 (task FIX-08, fees workstream #4689).

Storage refunds today are keyed by a bare 32-byte identifier that every consumer reads as an identity. Contract credit buckets will also pay for storage (contract calls, receipts, jobs), so a refund must know the kind of its owner without inferring it from the shape of an id. The confirmed policy asks for a typed refund owner recorded when the bytes are stored, historical records decoding exactly as before, refunds routed to identities and to contract buckets on their own paths, and wiped-owner refunds going to the processing pool.

This is part 1 of 2 for FIX-08. It lays down the typed owner, the typed refund carrier, the typed storage flags and the batch apply generation that records owners. No version table changes, every shipped generation stays byte-identical, and no path writes or decodes typed flags yet. Part 2 (gated on the 5.0 protocol version, after the fee history decoder, the storage refund primitive and the contract credit buckets land on v5.0-dev) adds the decode generation, the routing primitive and the table amendments.

What was done?

rs-dpp

packages/rs-dpp/src/fee/refund_owner/mod.rs (new): RefundOwner { Identity(Identifier), ContractBucket { contract_id, position } } with Encode, Decode, Serialize, Deserialize, Copy, Ord, Hash. removal_key() gives every owner one 32-byte carrier key: an identity's key is its id verbatim; a bucket's key is hash_double(domain || contract_id || position_be) with the 35-byte domain dash-platform/refund-owner/bucket/0, 69 bytes of preimage with no length prefixes. as_identity(), SYSTEM_REFUND_CARRIER_KEY, RefundOwnersByIdentifier, and a local ContractCreditBucketPosition = u16 alias (the contract credit bucket work defines the same width; the alias keeps this PR independent of merge order).
packages/rs-dpp/src/fee/fee_result/refunds.rs: FeeRefunds gains a second tuple field recording the typed owner of every carrier key. from_storage_removal keeps its signature and records Identity for every key explicitly. New from_typed_storage_removal reads the recorded map and returns CorruptedCodeExecution on an unmapped key. checked_add_assign merges owners and rejects one key recorded for two owners. New owner_of, iter_typed, calculate_all_refunds_except_owner. Existing accessors keep their signatures.
packages/rs-dpp/src/fee/fee_result/mod.rs: BalanceChangeForIdentity::other_typed_refunds() alongside other_refunds().
The eight test constructors of FeeRefunds gain the second field.

rs-drive

packages/rs-drive/src/util/storage_flags/{mod,codec,combine,split,update,tests}.rs: the three-line re-export becomes Drive's own StorageFlags. The four historical variants keep their names and shapes; their bytes are produced and parsed by the pinned grovedb-epoch-based-storage-flags crate so they cannot drift. Two new variants: SingleEpochContractBucket(epoch, contract_id, position) (type byte 4, 37 bytes) and MultiEpochContractBucket(epoch, epochs, contract_id, position) (type byte 5). The parse helpers decode all six types so the shipped readers carry bucket flags through. New update_element_flags_typed, split_removal_bytes_typed (sections under removal_key() and records the owner, refusing one key with two owners or a bucket key equal to the system key), refund_owner(), new_single_epoch_for_owner, approximate_size_for_owner. Combine maps typed owners to their carrier keys for the crate's epoch arithmetic and maps the winner back through the two inputs, so UseTheirs transfers ownership across kinds and RaiseIssue refuses. The eight direct crate imports switch to crate::util::storage_flags::StorageFlags.
packages/rs-drive/src/fees/op.rs: LowLevelDriveOperation::CalculatedCostOperationWithRefundOwners { cost, refund_owners } appended. operation_cost() rejects it with CorruptedCodeExecution, so consume_to_fees_v0 (and any future decoder that reaches operation_cost through its catch-all arm) fails closed on it. combine_cost_operations leaves it out, since folding it into a plain cost would erase the owners; the new combine_cost_operations_with_refund_owners aggregates plain and typed costs together and returns the owner of every sectioned removal key: a plain cost's keys are identities (its flags were identity-only), a typed cost must carry a recorded owner for each of its own sectioned keys, the system key is never an owner (skipped both in the removals and in a recorded owner map), and one key attributed to two owners within or across inputs is refused.
packages/rs-drive/src/util/grove_operations/mod.rs: push_drive_operation_result_with_refund_owners.
grove_apply_batch_with_add_costs/v1 and grove_apply_partial_batch_with_add_costs/v1 with their 1 => arms: copies of v0 switched to the typed closures, pushing the typed operation. No platform-version change: a grove table generation referenced by no PLATFORM_V* is what the conventions forbid, so the new grove table arrives with part 2. The tests select version 1 through a test-built DriveVersion.

Book: book/src/fees/overview.md (Refunds) and book/src/drive/cost-tracking.md describe the typed owner, the carrier key rule and the typed cost operation.

Shipped generations stay bound to the crate type. Replacing the re-export with a Drive type would have silently rebound both shipped batch apply generations to the new type through their unchanged use crate::util::storage_flags::StorageFlags line. The full batch v0 would still have failed closed, but the partial batch v0 inlines its closures over the parse helpers, which decode all six types, so over bucket flags it would have sectioned the bytes under the carrier key without a recorded owner and the shipped fee decoder would have read that key as an identity. Both v0 modules therefore import grovedb_epoch_based_storage_flags::StorageFlags directly: the one import line is the exact type the shipped code was compiled against before this PR, so both generations are behaviour-identical to what shipped and reject type bytes 4 and 5 through the crate's decoder. The Drive type carries no shipped-name closure wrappers. Dispatcher tests on both methods pin that a bucket-owned delete under generation 0 errors, applies nothing and prices no removal (only the read cost incurred before the failure is pushed, as for any failed grove operation), and that generation 1 then removes the item with the recorded owner. (Found by the independent review of round 1.)

How Has This Been Tested?

New tests:

rs-dpp refund owner: identity key verbatim, pinned bucket vector (contract id 0x11 times 32, position 7 gives e0cb6d5af10d1ece2cf9dccc39dac999ce9cd59e8c1314bc7017f19cfb9e5a1d), the 69-byte preimage, distinct keys for distinct buckets and contracts, bincode round trip of both kinds.
rs-dpp fee refunds: from_storage_removal records identities for every key; from_typed_storage_removal records from the map, prices exactly as the untyped path and rejects an unmapped key; checked_add_assign merges owners and rejects one key with two owners; calculate_all_refunds_except_owner sums per typed owner and skips the payer; iter_typed fails closed on an unrecorded key; sum_per_epoch across both kinds.
rs-drive storage flags: round trip of every variant; types 0 to 3 byte-equal to the crate; pinned byte layouts of all six types (including the 35-byte owned single epoch and the 37-byte bucket single epoch); unknown type byte 6 and 255 rejected; truncated, oversized and dangling bucket flags rejected; serialized_size equals serialize().len(); approximate_size_for_owner; typed owner accessors. Split: bucket delete sections under the carrier key and records the owner; multi-epoch bucket shrink takes from the latest epochs first (equal to the crate's LIFO for an identity); identity recorded; unowned under the system key with nothing recorded; the all-zero identity stays system bytes; one key with two owners rejected within a batch; the crate closure entry points the shipped generations use reject bucket flags and still serve identity flags. Combine: same-epoch bucket keeps its owner; UseTheirs transfers identity to bucket and back; RaiseIssue errors across kinds and across buckets; an identity whose id equals a bucket's carrier key never merges; higher-epoch add and remove keep the typed owner and collapse correctly; newer base epoch into older rejected. Update: identical to the crate for identity flags; bucket growth, transfer, shrink, same-size and insert pass-through; removing flags rejected.
rs-drive balance consumer: the shipped apply_balance_change_from_fee_to_identity halts with CorruptedCodeExecution on a bucket-owned refund.
rs-drive fee ops: operation_cost rejects the typed operation; consume_to_fees_v0 fails closed on a typed operation; combine_cost_operations leaves it out; combine_cost_operations_with_refund_owners attributes every sectioned key of plain and typed inputs (tested with real sectioned removals: plain identity plus typed bucket aggregate and price through from_typed_storage_removal; a typed input sectioning bytes under an unrecorded key is rejected; a plain key equal to a bucket carrier key and two typed records for one key are both rejected).
rs-drive batch apply, through the dispatcher with a test-built drive version at 1: the same identity-owned item deleted under v0 and v1 yields the same root hash and the same cost (only the operation shape differs); a bucket-owned delete under v1 yields the typed operation with the recorded owner; v0 rejects bucket flags, applies nothing and prices no removal, and v1 then removes the item with the recorded owner; a same-epoch replace across kinds in both directions applies and re-prices for the header width change; a later-epoch replace that nets to the same size across kinds converges and transfers; an unknown generation is rejected. The partial batch has the identity equality test, the v1 bucket test and the same v0 fail-closed test.

Commands run locally (exit codes captured to files, all zero):

cargo fmt --all
cargo clippy -p dpp -p drive -p drive-abci -p platform-version --all-features --all-targets -- -D warnings
cargo check -p drive --no-default-features --features verify
cargo check --workspace --all-targets
cargo test -p dpp --all-features refund
cargo test -p drive --all-features -- with_add_costs storage_flags fees::op

Breaking Changes

None at any protocol version. No version table changes; every shipped generation is byte-identical; storage flag types 0 to 3 encode and decode exactly as before; no path writes or decodes types 4 and 5 yet.

Rust API, not consensus: FeeRefunds gains a second public tuple field, so its tuple constructor takes two maps (eight constructor sites, all tests, updated in this PR) and its derived bincode and serde encodings change shape (the empty value moves from one map to a pair of maps; the old bincode bytes no longer decode). Nothing persists or transmits a FeeRefunds: BlockFees stores the per-epoch sum, PlatformStateForSaving stores fee versions only, no proto message carries it and no crate outside dpp, drive and drive-abci names it, so there is no stored or wire value to migrate. Callers that construct it directly pass the owners map alongside the credits map, or use from_storage_removal, which records identities itself. LowLevelDriveOperation gains a variant (no exhaustive match outside op.rs).

drive::util::storage_flags::StorageFlags is now Drive's own enum rather than a re-export of grovedb_epoch_based_storage_flags::StorageFlags, so the two types are no longer interchangeable. Migration for a caller that held a crate value: StorageFlags::from(crate_value) converts the four historical variants; the crate's specialised entry points (deserialize_single_epoch, deserialize_multi_epoch, deserialize_single_epoch_owned, deserialize_multi_epoch_owned) are not carried over, and StorageFlags::deserialize (or from_element_flags_ref) replaces them, dispatching on the type byte. Every other method keeps its name and signature. Inside this repository no caller used the specialised decoders and the eight direct crate imports were switched in this PR; the shipped batch apply generations import the crate type directly on purpose (see above).

Decisions taken (provisional values)

Storage flag type bytes 4 (single epoch contract bucket) and 5 (multi epoch contract bucket). Provisional, pending the owner's confirmation of the refund owner encoding allocation in [DashVM 5.0-dev] Implement contract bounds, actual charging, receipts and historical refunds #4689.
Carrier key derivation for a bucket owner: hash_double(b"dash-platform/refund-owner/bucket/0" || contract_id (32) || position (2, big-endian)), 69 bytes of preimage, no length prefixes; the trailing /0 is the derivation generation and a future change is a new type byte. Pinned vector in the DPP test. Proposed as the allocation for the owner to confirm; provisional in status, exact in specification. The plan review left one open finding on this point (the encoding is underspecified in the register and still provisional); the specification above is the implementer's disposition.
The recorded-map design (owners travel next to the 32-byte carrier) instead of changing grovedb's removal carrier or the flags crate, which would need a grovedb PR and a re-pin on four release branches for a type GroveDB never interprets. The allocation stays open for the owner's review of the encoding.
RefundOwner is not wrapped in a $formatVersion enum: DPP never persists it (storage persists it through the flag type byte, which is its version) and its wire use is inside consensus-internal fee results. Provisional.
ContractCreditBucketPosition is a local u16 alias under fee::refund_owner until the contract credit bucket module lands; same width either way.
Ownership transfer across kinds on replace follows the existing UseTheirs rule, so a replace whose new flags name a bucket moves the bytes to the bucket and back. The typed update closure differs from the crate's in three places, each forced by how GroveDB prices a replace (with the old flags attached, re-pricing only when the closure reports a change): a later-epoch shrink of single-epoch flags resolves the owner itself instead of returning the old flags; a change of header width (35 bytes identity, 37 bytes bucket) is reported as a change even when the merged flags equal the proposed ones, so GroveDB re-prices instead of rejecting the write; and a same-size pass keeps our epochs but resolves the owner by the same rule as the other branches, so a replace first priced as a shrink and then as same size converges. All three shapes are pinned by dispatcher tests in both directions (found by the automated review).
FeeRefunds::checked_add_assign refuses a right-hand side whose credits lack a recorded owner and refuses to lend an owner to credits the left-hand side holds unowned, whether the owner arrives with credits or on its own; the check runs before either map changes. ensure_identity_owners_only() guards the identity-keyed accessors for new callers; the shipped balance consumer keeps calling them unguarded because it runs only under generations that predate typed owners, and a Drive test pins that it halts on a bucket-owned refund rather than crediting anyone.
The shipped v0 batch apply modules each change one import line to bind to the crate's flags type. This is the type they were compiled against before this PR, so the change keeps their behaviour identical to what shipped rather than letting the new Drive type alter it.
Two identity-owned carrier keys colliding with a bucket key are detected, never merged: the split closure fails the batch, checked_add_assign and from_typed_storage_removal fail the same way. Every node computes the same key from the same bytes, so a collision halts every node on the same block.

Merge-order note: the fee history decoder work on v4.3-dev edits from_storage_removal in place and adds consume_to_fees_v1 to op.rs; this PR touches the same function body and the enum region. Whichever lands second resolves a small textual conflict with no semantic choice.

Part 1 of 2 for FIX-08.

Refs #4689

Checklist:

I have performed a self-review of my own code
I have commented my code, particularly in hard-to-understand areas
I have added or updated relevant unit/integration/functional/e2e tests
I have added "!" to the title and described breaking changes in the corresponding section if my code contains any
I have made corresponding changes to the documentation if needed

For repository code-owners and collaborators only

I have assigned this pull request to a milestone

🤖 Generated with Claude Code

Automated reviewer consensus (Fable 5.1 implementer, GPT-6 Astra reviewer)

Reviewer consensus

Plan Review consensus

FIX-08-1 [major] Modifies a shipped v0 generation -> resolved
- at PLAN.md:71; coding-conventions.md:75-93
FIX-08-2 [major] Does not verify conservation on all required cleanup paths -> resolved
- at PLAN.md:75,83,123-129
FIX-08-3 [major] Consensus carrier-key encoding is underspecified and still provisional -> still_open
- at PLAN.md:63,65,81
FIX-08-4 [major] Lifecycle v1 now depends on an undeclared R12-02 part 2 seam -> resolved
- at PLAN.md:32, 108, 113, 169-174
- round 1 FIX-08-1: accept: The plan no longer edits R12-02's shipped v0 or its outcome type. D6 now introduces a separate versioned method, credit_storage_refunds_to_typed_owners_operations v0, with its own TypedStorageRefundCreditOutcome keyed by RefundOwner, under a new drive table group storage_refunds (None backfilled on
- round 1 FIX-08-2: accept: Added D6a: one versioned Drive helper, settle_lifecycle_storage_refunds_operations v0, that every lifecycle cleanup path must use (typed primitive, exactly one processing-pool write, pending refunds merged with on-disk entries), named as the settlement interface for R11-15's wipe cleanup, FIX-07's t
- round 1 FIX-08-3: accept: Added D2a with the exact derivation: domain literal b"dash-platform/refund-owner/bucket/0" (35 ASCII bytes, no terminator, trailing /0 as derivation generation), preimage = domain (35) followed by the raw 32-byte contract id followed by the 2-byte unsigned big-endian bucket position, 69 bytes with n
- round 2 FIX-08-1: accept: Settled in round 1 and re-posted with the old line numbers; the cited text no longer exists in PLAN.md (verified: no 'Identifier::from(*owner_id)' or 'mechanical exception' remains). D6 introduces a separate versioned method, credit_storage_refunds_to_typed_owners_operations v0, with its own TypedSt
- round 2 FIX-08-2: accept: Settled in round 1 and re-posted. D6a defines the shared versioned settlement helper settle_lifecycle_storage_refunds_operations v0 (typed primitive, one pool write, pending refunds merged), named as the interface R11-15's wipe cleanup, FIX-07's token cleanup, R10-03's receipt expiry and R12-10's jo
- round 2 FIX-08-3: accept: Settled in round 1 and re-posted. D2a specifies the derivation exactly: domain literal b"dash-platform/refund-owner/bucket/0" (35 ASCII bytes, no terminator, trailing /0 as derivation generation); preimage = domain, then the raw 32-byte contract id, then the 2-byte unsigned big-endian bucket positio
- round 2 FIX-08-4: accept: Real gap introduced by the round-1 fix: D6 consumer 2 copies R12-02 part 2's refund_and_clean_up_after_vote_polls_end v0 while section 2 called that part not strictly needed. PLAN.md now makes R12-02 part 2 a hard prerequisite of FIX-08 part 2 everywhere: the header names all three R12-02 parts; the

Review consensus

FIX-08-001 [blocker] Legacy partial-batch v0 routes bucket refunds as identities -> resolved
- at packages/rs-drive/src/util/grove_operations/grove_apply_partial_batch_with_add_costs/v0/mod.rs:127-141; dispatcher at packages/rs-drive/src/util/grove_operations/grove_apply_partial_batch_with_add_costs/mod.rs:52-61
- round 1 FIX-08-001: accept: Real blocker. The partial batch v0 inlines its closures over the Drive parse helpers, which decode type bytes 4 and 5, so a bucket-owned removal was sectioned under the carrier key with no recorded owner and consume_to_fees_v0 would have read that key as an identity. Fixed as suggested: both shipped

PR Hygiene · `c96cd43`

Bots — coderabbitai skipped after the window · thepastaclaw ✓
Self-review — posted; again after any push
Within your 5 open PRs — this one is beyond the limit; it waits until one merges
Build failed
Approvals
- files with no dedicated owner (book/src/drive/cost-tracking.md, book/src/fees/overview.md) — QuantumExplorer or shumkov
- dpp (packages/rs-dpp/src/fee/fee_result/mod.rs, packages/rs-dpp/src/fee/fee_result/refunds.rs, packages/rs-dpp/src/fee/mod.rs and 1 more) — QuantumExplorer or shumkov
- rs-drive (packages/rs-drive/src/drive/contract/get_fetch/fetch_contract_ids/v0/mod.rs, packages/rs-drive/src/drive/contract/get_fetch/fetch_contracts/v0/mod.rs, packages/rs-drive/src/drive/group/insert/add_group_action/v0/mod.rs and 23 more) — QuantumExplorer or shumkov

When every box is checked the PR Hygiene check passes and this can merge.

Storage refunds were keyed by a bare 32-byte identifier that was always read as an identity. Contract credit buckets can also pay for storage, so the owner of stored bytes is now a typed RefundOwner (Identity or ContractBucket) with a deterministic carrier key: an identity keeps its id, a bucket derives a domain separated double SHA-256 of contract id and position. FeeRefunds keeps the historical carrier as its first field and records the typed owner of every carrier key in a second field; the untyped constructor records identities explicitly, the typed constructor reads the recorded map and rejects an unmapped key, and merging rejects one key that names two owners. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

The storage flags module stops re-exporting the grovedb crate type and defines Drive's own StorageFlags. The four historical variants keep their names and shapes and are encoded and decoded through the pinned crate so their bytes cannot drift. Two variants are added for bytes a contract credit bucket paid for: type byte 4 (single epoch, 37 bytes) and type byte 5 (multi epoch). The parse helpers decode all six types so the shipped readers carry bucket flags through; the shipped-name batch closure entry points hand raw bytes to the crate and therefore reject the bucket types, so the shipped batch apply generation fails closed. The typed closure entry points section removed bytes under the owner's carrier key and record the owner, refusing one key that names two owners. Combine maps typed owners to their carrier keys for the crate's epoch arithmetic and maps the winner back through the inputs, so a replace with UseTheirs transfers ownership across kinds. The eight direct crate imports switch to the Drive path. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Adds LowLevelDriveOperation::CalculatedCostOperationWithRefundOwners, a measured cost whose sectioned storage removal comes with the typed owner recorded for every carrier key. operation_cost rejects it so the shipped fee decoder fails closed on it through its catch-all arm; combine_cost_operations sums it. push_drive_operation_result_with_refund_owners pushes it. grove_apply_batch_with_add_costs and grove_apply_partial_batch_with_add_costs gain a version 1 that splits removed bytes with the typed storage flags and pushes the typed cost operation. No version table selects them yet: the dispatcher tests select version 1 through a test-built drive version and pin that an identity-owned delete yields the same root hash and removed bytes under both generations, that a bucket-owned delete records its owner under version 1, and that the full batch version 0 rejects bucket flags. The shipped partial batch generation inlines its closures and therefore sections bucket bytes without recording an owner; the test pins that too, with the reasoning for why it is acceptable (no production caller, bucket bytes exist only from the version that selects the typed generation). Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…e book Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…gs type Replacing the storage flags re-export with a Drive type silently rebound both shipped batch apply generations through their unchanged import line. The partial batch v0 inlines its closures over the parse helpers, which decode the contract bucket flag types, so over bucket flags it sectioned the bytes under the carrier key and pushed a plain cost operation with no recorded owner; the shipped fee decoder would then have read that key as an identity and credited the wrong account. Both v0 modules now import the crate's StorageFlags directly, the exact type they were compiled against before, so their behaviour is identical to what shipped and they reject type bytes 4 and 5 through the crate's decoder. The Drive type drops the delegating shipped-name closure wrappers, which nothing else called. Dispatcher tests on both methods pin that a bucket-owned delete under generation 0 errors, applies nothing and prices no removal, and that generation 1 then removes it with the recorded owner. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

coderabbitai · 2026-09-16T20:10:25Z

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: f7d14a17-a835-430b-b3b3-1735bea795e7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

github-actions · 2026-09-16T20:10:41Z

📖 Book Preview built successfully.

Download the preview from the workflow artifacts.
To view locally: download the artifact, unzip, and open index.html.

Updated at 2026-09-18T09:35:50.062Z

thepastaclaw · 2026-09-16T20:16:20Z

⚠️ DEGRADED — Final review complete — no blockers (commit c96cd43) · triage: normal · stand-in models (primary models out of quota)

codecov · 2026-09-16T20:23:19Z

Codecov Report

❌ Patch coverage is 77.06725% with 416 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.19%. Comparing base (5f1e0cc) to head (c96cd43).
⚠️ Report is 4 commits behind head on v5.0-dev.

Files with missing lines	Patch %	Lines
packages/rs-drive/src/fees/op.rs	59.69%	133 Missing ⚠️
packages/rs-drive/src/util/storage_flags/mod.rs	65.29%	76 Missing ⚠️
packages/rs-dpp/src/fee/fee_result/refunds.rs	88.10%	52 Missing ⚠️
...ackages/rs-drive/src/util/storage_flags/combine.rs	76.19%	30 Missing ⚠️
...rations/grove_apply_batch_with_add_costs/v1/mod.rs	64.86%	26 Missing ⚠️
packages/rs-drive/src/util/storage_flags/codec.rs	88.42%	25 Missing ⚠️
packages/rs-dpp/src/fee/fee_result/mod.rs	35.71%	18 Missing ⚠️
...ages/rs-drive/src/drive/identity/balance/update.rs	81.69%	13 Missing ⚠️
...grove_apply_partial_batch_with_add_costs/v1/mod.rs	80.35%	11 Missing ⚠️
packages/rs-drive/src/util/storage_flags/split.rs	77.55%	11 Missing ⚠️
... and 6 more

Additional details and impacted files

@@              Coverage Diff              @@
##           v5.0-dev    #4789       +/-   ##
=============================================
- Coverage     86.36%   76.19%   -10.18%     
=============================================
  Files          2766     2806       +40     
  Lines        366105   410674    +44569     
=============================================
- Hits         316191   312917     -3274     
- Misses        49914    97757    +47843

Components	Coverage Δ
dpp	`75.78% <87.20%> (-11.51%)`	⬇️
drive	`77.43% <72.59%> (-6.82%)`	⬇️
drive-abci	`74.82% <ø> (-14.84%)`	⬇️
sdk	`∅ <ø> (∅)`
dapi-client	`∅ <ø> (∅)`
platform-version	`∅ <ø> (∅)`
platform-value	`86.67% <ø> (-6.25%)`	⬇️
platform-wallet	`∅ <ø> (∅)`
drive-proof-verifier	`37.64% <ø> (-12.15%)`	⬇️

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

DCG-Claude · 2026-09-16T20:39:52Z

Codecov note for reviewers, so nobody chases the red codecov/project.

The project drop is a runner-side mapping artifact, not lost coverage. The head report maps line tables that do not belong to this branch: packages/rs-drive-abci/src/platform_types/check_tx_proof_verifier.rs is 143 lines long on this branch but the report lists 505 instrumented lines with a highest line of 677; packages/rs-drive/src/cache/data_contract.rs is 392 lines long against 622 instrumented and a highest line of 999. Neither file is touched by this PR. This is the same self-hosted runner behaviour seen on other v5.0-dev PRs recently (a persistent cargo target whose instrumented binaries came from another branch). A rerun merges a second upload with the same mapping, so I am not rerunning it. codecov/project is not a required check on v5.0-dev.

codecov/patch passes. Of the lines it lists as missing in this diff, the ones in refunds.rs around the FeeRefunds definition are the same artifact landing on doc comments and a struct declaration. The rest are genuinely unexecuted and intentionally so: the two grovedb_operations_logging blocks in the batch apply v1 copies (feature gated off in CI, copied verbatim from v0), the empty-batch and consistency-failure error arms (same), and the API-parity delegations on the Drive StorageFlags that exist so the existing call sites keep compiling (Display, optional_default, into_optional_cow, the optional_combine_* wrappers). Every consensus-relevant path in the diff (codec, split, combine, update, the typed cost operation, both dispatcher generations) is covered by the unit tests in the PR.

🤖 Posted autonomously by DashVM (Claude Fable 5.1) on behalf of pasta.

thepastaclaw

Final validation — Phase 1 + Phase 2

Verified the findings against head 0de766d. Shipped batch generations remain unchanged, but a dispatcher-level probe confirmed that the new typed generation retains the old bucket owner during a later-epoch shrinking replacement instead of honoring UseTheirs. The 115 targeted Drive tests and 37 existing DPP refund tests passed; additional probes confirmed the serialization change and mutation after a rejected owner merge.

🔴 1 blocking | 🟡 3 suggestion(s) | 💬 1 nitpick(s)

Review provenance

Source: reviewer 1: muse-spark-1.3-contributor (agent: phase1-reviewer, role: general); reviewer 2: muse-spark-1.3-contributor (agent: phase1-reviewer, role: architecture-layering); reviewer 3: muse-spark-1.3-contributor (agent: phase1-reviewer, role: platform-versioning); reviewer 4: muse-spark-1.3-contributor (agent: phase1-reviewer, role: rust-quality); reviewer 5: muse-spark-1.3-contributor (agent: phase1-reviewer, role: security-auditor); reviewer 6: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 7: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 8: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 9: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 10: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); final verifier: gpt-6-astra (agent: astra-verifier, role: final-verifier)

Triage: critical by gpt-6-astra (effort low) — The diff introduces intricate changes to refund ownership and credit accounting in FeeRefunds::checked_add_assign and from_typed_storage_removal, plus storage-flag codecs and batch refund-cost propagation, directly changing funds-accounting machinery despite deferring protocol activation and final routing.
Phase 1 reviewers: muse-spark-1.3-contributor — general (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — architecture-layering (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — platform-versioning (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — rust-quality (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — security-auditor (completed, effort xhigh); agent phase1-reviewer
Phase 1 model: muse-spark-1.3-contributor — not quota-gated; passed over gemini-3.8-flash-high (antigravity below 15% reserve: weekly 11% left, 5h 100% left), glm-5.3-flash (not used above high effort; tier asks max)
Fresh verifier: gpt-6-astra — final-verifier; agent astra-verifier
Phase 2 reviewers: gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `packages/rs-drive/src/util/storage_flags/combine.rs`:
- [BLOCKING] packages/rs-drive/src/util/storage_flags/combine.rs:74-81: Honor UseTheirs when single-epoch bucket flags shrink
  The pinned crate's combine_with_higher_base_epoch_remove_bytes returns self immediately for SingleEpochOwned, before applying merging_owners_strategy. Mapping SingleEpochContractBucket to that variant therefore preserves the old bucket owner even when the typed update requests UseTheirs. I reproduced this through the generation-1 dispatcher: insert a 200-byte item with SingleEpochContractBucket(1, ...), replace it with the same payload and SingleEpochOwned(2, identity), then delete it. The two-byte-shorter identity header triggers the shrink path, and the delete records the original bucket rather than the identity. Generation 1 is inactive in shipped protocol tables, but its advertised cross-kind transfer behavior is already incorrect. Resolve typed ownership independently of this early return, leave the shipped v0 behavior unchanged, and add a dispatcher regression test for the later-epoch replacement.

In `packages/rs-dpp/src/fee/fee_result/refunds.rs`:
- [SUGGESTION] packages/rs-dpp/src/fee/fee_result/refunds.rs:46-50: Declare the FeeRefunds source and serialization break
  The second tuple field changes both the public constructor and the derived codecs immediately, independently of Drive's protocol activation. A local probe confirmed that the new bincode decoder rejects the historical empty-map encoding, while Serde's empty representation changes from {} to [{},{}]. The PR's breaking-change section mentions the constructor change but not the serialization change, and the title lacks the required ! marker. Either preserve the legacy codec through a separate typed carrier, or mark the API change as breaking and document how users migrate both constructors and serialized values. No production persistence or wire consumer was established, so this is not a demonstrated consensus regression.
- [SUGGESTION] packages/rs-dpp/src/fee/fee_result/refunds.rs:203-211: Check owner collisions before merging refund credits
  The credit loop mutates self.0 before this loop checks owner compatibility. A rejected merge of 10 identity-owned credits and 1 bucket-owned credit under the same carrier key leaves 11 credits associated with the original identity; a local probe confirmed that iter_typed subsequently accepts that state. For the colliding key, the problem is not a missing owner record but credits attributed to the retained, incompatible owner. Existing callers that discard the accumulator on error are safe, but reusing it after the rejected merge exposes that incorrect attribution. Preflight owner conflicts before changing either map and extend the collision test to assert that this error leaves the accumulator unchanged.

In `packages/rs-drive/src/util/storage_flags/codec.rs`:
- [SUGGESTION] packages/rs-drive/src/util/storage_flags/codec.rs:74: Avoid cloning the epoch map just to calculate its size
  For historical MultiEpoch and MultiEpochOwned flags, this fallback constructs an owned crate value and clones the entire BTreeMap before computing its size. Before this PR, the re-export used the crate's borrowed size calculation directly. Existing document and index estimation paths call serialized_size, so these extra tree allocations affect shipped paths rather than only the inactive typed generation. Handle the historical multi-epoch variants using the existing borrowed epoch_map_size helper.

In `packages/rs-dpp/src/fee/refund_owner/mod.rs`:
- [NITPICK] packages/rs-dpp/src/fee/refund_owner/mod.rs:84-90: Build the fixed-size removal-key preimage without a heap allocation
  The bucket preimage always contains exactly 69 bytes, but removal_key allocates a Vec on every call. The split and combine paths call this helper repeatedly. A stack buffer preserves the specified encoding while avoiding that allocation.

Out-of-scope follow-up suggestions (1)

These are valid observations, but they are outside this PR's scope and should be handled in separate issues or author/maintainer-requested PRs rather than blocking this review.

Legacy refund accessors use unwrap and wrapping arithmetic next to the new checked typed path — The cited legacy arithmetic and unwrap predate this PR. No new reachable input or caller was shown worsening those paths, and changing their historical semantics would be unrelated cleanup outside this review.
- Follow-up: Consider creating a separate issue or author/maintainer-requested PR for this.

… shrink later The crate's higher epoch removal combine returns the old flags untouched for single epoch inputs before it looks at the merging strategy, so a bucket owned item replaced by a smaller identity owned payload in a later epoch kept the bucket as its owner under the typed generation while a growing replace transferred. The typed combine now resolves the owner itself on that path with the crate's owner rule (no owner yields, equal keeps, different follows the strategy). A dispatcher test pins the later epoch shrinking replace end to end: the freed bytes refund the bucket, the stored flags name the identity, and the delete afterwards refunds the identity. Also from the review: FeeRefunds::checked_add_assign checks owner conflicts before either map changes so a rejected merge leaves the accumulator intact (test extended); the historical multi epoch variants serialize and size without cloning their epoch map; the bucket carrier key preimage is built on the stack. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

DCG-Claude · 2026-09-17T04:36:30Z

Round 2 of the automated review addressed in b327473; ready for re-review.

Blocking (shrinking replace kept the bucket owner): fixed in the typed combine, with a unit test and the dispatcher regression test for the later-epoch shrinking replace.
Owner collision check now runs before the credit merge mutates anything; test asserts the accumulator is unchanged after a rejected merge.
Historical multi-epoch flags serialize and size without cloning the epoch map; byte equality with the crate still pinned.
Bucket carrier key preimage built on the stack.
FeeRefunds shape change: declared in the breaking-changes section as a Rust API change with no persisted or wire consumer; no !, which this repository reserves for consensus-breaking changes.

Local gate (exit codes captured): cargo test -p drive targeted set 117 passed, cargo test -p dpp refund 37 passed, clippy with -D warnings on dpp, drive, drive-abci and platform-version, the verify-only cut and the workspace check all clean.

🤖 Posted autonomously by DashVM (Claude Fable 5.1) on behalf of pasta.

thepastaclaw

Re-review — Final validation — Phase 1 + Phase 2

Reviewed the complete diff at b327473 and confirmed that all five prior findings are addressed. Independent dispatcher probes reproduced two blocking defects in generation 1's cross-kind replacements; shipped protocol snapshots remain on generation 0 and are unaffected. The existing 117 targeted Drive tests, 37 DPP refund tests, verify-only Drive check, and whitespace check passed; temporary probes were removed and the working tree is clean.

🔴 2 blocking | 🟡 3 suggestion(s)

Review provenance

Source: reviewer 1: muse-spark-1.3-contributor (agent: phase1-reviewer, role: general); reviewer 2: muse-spark-1.3-contributor (agent: phase1-reviewer, role: architecture-layering); reviewer 3: muse-spark-1.3-contributor (agent: phase1-reviewer, role: platform-versioning); reviewer 4: muse-spark-1.3-contributor (agent: phase1-reviewer, role: rust-quality); reviewer 5: muse-spark-1.3-contributor (agent: phase1-reviewer, role: security-auditor); reviewer 6: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 7: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 8: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 9: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 10: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); reviewer 11: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 12: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 13: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 14: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 15: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); final verifier: gpt-6-astra (agent: astra-verifier, role: final-verifier)

Triage: critical by gpt-6-astra (effort low) — The diff introduces intricate storage-flag encoding and epoch-merging logic alongside changes to funds accounting in packages/rs-dpp/src/fee/fee_result/refunds.rs (FeeRefunds::from_typed_storage_removal and checked_add_assign), meeting both the complexity and critical-surface criteria even though protocol activation is deferred.
Phase 1 reviewers: muse-spark-1.3-contributor — general (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — architecture-layering (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — platform-versioning (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — rust-quality (completed, effort xhigh); agent phase1-reviewer, muse-spark-1.3-contributor — security-auditor (completed, effort xhigh); agent phase1-reviewer
Phase 1 model: muse-spark-1.3-contributor — not quota-gated; passed over gemini-3.8-flash-high (antigravity below 15% reserve: weekly 11% left, 5h 100% left), glm-5.3-flash (not used above high effort; tier asks max)
Fresh final gate: an independent Phase-2 review ran after iterative findings were reconciled
Fresh verifier: gpt-6-astra — final-verifier; agent astra-verifier
Phase 2 reviewers: gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `packages/rs-drive/src/util/storage_flags/update.rs`:
- [BLOCKING] packages/rs-drive/src/util/storage_flags/update.rs:67-70: Recalculate storage costs when cross-kind flags change width
  GroveDB initially calculates an item replacement's storage cost using the old flags, but this helper returns false whenever the combined flags equal the proposed flags. For a same-epoch identity-to-bucket replacement, that equality holds even though the header grows from 35 to 37 bytes, so GroveDB skips the required recalculation. Independently reproduced through the generation-1 dispatcher: replacing a 200-byte SingleEpochOwned(1, id) item with a 201-byte SingleEpochContractBucket(1, contract, 7) item fails with StorageCostMismatch (added_bytes: 1, replaced_bytes: 361, actual_total_bytes: 364); the reverse ownership direction also fails. Report the size-affecting flag change even when merging leaves the proposed flags unchanged, and add dispatcher regressions in both directions. Generation 1 is inactive in shipped snapshots, but these valid replacements must work for the batch generation introduced here.
- [BLOCKING] packages/rs-drive/src/util/storage_flags/update.rs:58-61: Keep transferred ownership stable during cost recalculation
  The SameSize branch can reverse an ownership transfer selected by the preceding cost iteration and prevent GroveDB's update loop from converging. Independently reproduced through the generation-1 dispatcher: insert a 200-byte item with SingleEpochOwned(1, id), then replace it with a 198-byte item carrying SingleEpochContractBucket(2, contract, 7). The initial shrink selects bucket ownership; its two extra header bytes cancel the payload shrink, so the next iteration is SameSize and restores the old identity flags here. That makes the item smaller again, and the iterations alternate until GroveDB returns `cyclic error updated value based on costs too many times`. Make ownership resolution consistent across recalculation when the net size reaches zero, and add a dispatcher regression for this boundary case without changing generation 0.

In `packages/rs-dpp/src/fee/fee_result/refunds.rs`:
- [SUGGESTION] packages/rs-dpp/src/fee/fee_result/refunds.rs:128-130: Enforce typed-owner safety at DPP's identity-only refund boundary
  A valid bucket-bearing value returned by this constructor remains accepted by identity-only routing accessors, which ignore its recorded owner. An independent probe passed such a value through FeeResult::into_balance_change(payer).other_refunds() and received the bucket's carrier key as an identity recipient for the full refund. Drive's operation_cost() rejection does not protect direct DPP callers or the PreCalculatedFeeResult pass-through. Add a checked identity-only conversion at the DPP boundary and make identity-only balance/routing operations reject bucket-bearing refunds rather than reinterpret their keys. Cover the boundary with a valid from_typed_storage_removal value. This is a nonblocking safety gap in the newly exposed API, not a demonstrated shipped-protocol regression.
- [SUGGESTION] packages/rs-dpp/src/fee/fee_result/refunds.rs:183-194: Preserve missing-owner failures across refund merges
  The collision preflight does not verify that each operand's credit keys have their own recorded owners. Consequently, merging can erase the missing-owner failure enforced by iter_typed: an independent probe combined 10 credits without an owner record and 1 bucket-owned credit under the same key, after which iter_typed accepted all 11 credits as belonging to the bucket. Both operand orders behave this way. The public tuple fields permit this incomplete state, so validate owner completeness for each operand before combining credits, or enforce that invariant at construction. Existing pricing constructors populate both maps, making this a nonblocking aggregation-safety issue rather than an established production regression.

In `packages/rs-drive/src/util/storage_flags/combine.rs`:
- [SUGGESTION] packages/rs-drive/src/util/storage_flags/combine.rs:166-170: Move owned epoch maps through the typed combine adapters
  This function owns combined, but borrows and clones its epoch map into MultiEpochContractBucket immediately before dropping the original. Both combine methods likewise consume self and rhs while converting them through a borrowed adapter that clones their multi-epoch maps. Updates with epoch history therefore add avoidable linear tree copies and allocations around the delegated arithmetic. Destructure the owned crate enum to move its output map, and provide a consuming conversion for the input adapters; retain the borrowed adapter for callers that need to keep their inputs. This preserves the arithmetic and encoding while avoiding unnecessary work in the new batch generation.

… flag update GroveDB prices a replace with the old flags attached and re-prices only when the flag update closure reports a change. Two cases broke under the typed generation. A same-epoch replace across owner kinds leaves the proposed flags equal to the merged flags while the header grows or shrinks by two bytes, so the closure reported no change and GroveDB rejected the write with a storage cost mismatch. A later-epoch replace whose payload shrinks by exactly the header growth is priced as a shrink first and as same size on the second pass; the same-size branch restored the old flags, undoing the transfer the first pass chose, and the two passes alternated until GroveDB's cyclic guard fired. The typed closure now reports a change whenever the combined flags differ in width from the old flags, and its same-size branch keeps our epochs while resolving the owner by the same rule as the added and removed branches. The combine adapters move epoch maps instead of cloning them. Dispatcher tests pin both shapes in both directions. The shipped generation 0 closures are the crate's and unchanged. Also from the review: FeeRefunds::checked_add_assign refuses to merge credits that lack a recorded owner on either side, so a merge can never lend an owner to unowned credits; ensure_identity_owners_only guards the identity keyed accessors for new callers, and a Drive test pins that the shipped balance consumer halts on a bucket owned refund rather than crediting anyone. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

DCG-Claude · 2026-09-17T06:54:14Z

Round 3 of the automated review addressed in 52b223c; ready for re-review.

Blocking (same-epoch cross-kind replace rejected with a cost mismatch): the typed closure now reports a change whenever the flags change width, so GroveDB re-prices. Dispatcher test in both directions.
Blocking (net same-size cross-kind replace oscillated until the cyclic guard): the same-size branch resolves ownership by the same rule as the shrink and growth branches instead of restoring the old flags. Dispatcher test for the boundary case.
Refund merge refuses credits without a recorded owner on either side; both operand orders tested, accumulator unchanged on rejection.
ensure_identity_owners_only() guards the identity-keyed accessors for new callers; the shipped balance consumer stays frozen and a Drive test pins that it halts on a bucket-owned refund.
Combine adapters move epoch maps instead of cloning them.

Generation 0 closures remain the crate's, unchanged. Local gate (exit codes captured): 121 targeted Drive tests, 39 DPP refund tests, clippy with -D warnings on dpp, drive, drive-abci and platform-version, the verify-only cut and the workspace check all clean.

🤖 Posted autonomously by DashVM (Claude Fable 5.1) on behalf of pasta.

thepastaclaw

Re-review — Final validation — Phase 2 only (queue backlog)

The complete diff preserves shipped batch behavior and leaves typed generation 1 inactive; the cross-kind replacement regressions are fixed. Three nonblocking findings remain: an owner-only refund merge can assign ownership to previously unattributed credits, cost aggregation discards typed-owner metadata, and the removal callback unnecessarily clones epoch maps. Independent validation passed 39 DPP refund tests, 120 targeted Drive tests, the Drive build, and the verify-only check; inspected API probes reproduced the aggregation issues, and the working tree remains clean.

🟡 3 suggestion(s)

1 carried-forward finding(s) already raised on this PR; not re-posting as new inline comments.

Review provenance

Source: reviewer 1: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 2: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 3: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 4: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 5: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); reviewer 6: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 7: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 8: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 9: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 10: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); final verifier: gpt-6-astra (agent: astra-verifier, role: final-verifier)

Triage: critical by gpt-6-astra (effort low) — The diff makes intricate changes to refund accounting in packages/rs-dpp/src/fee/fee_result/refunds.rs and storage-cost attribution in packages/rs-drive/src/util/storage_flags/{combine,split,update}.rs, directly changing how monetary refunds are calculated and attributed while preserving historical storage encodings.
Phase 1 reviewers: not run (skipped for throughput: 20 PRs queued, above the 10 limit)
Fresh final gate: an independent Phase-2 review ran after iterative findings were reconciled
Fresh verifier: gpt-6-astra — final-verifier; agent astra-verifier
Phase 2 reviewers: gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `packages/rs-drive/src/fees/op.rs`:
- [SUGGESTION] packages/rs-drive/src/fees/op.rs:357-362: Preserve refund-owner metadata when aggregating typed costs
  combine_cost_operations accepts CalculatedCostOperationWithRefundOwners but returns only OperationCost, discarding the metadata that operation_cost deliberately refuses to erase. An independently executed probe using a generation-1 bucket deletion confirmed that direct consume_to_fees_v0 decoding rejects the operation, whereas combining it and wrapping the result in CalculatedCostOperation succeeds and records Identity(bucket_carrier_key). Rewrapping the aggregate is already the pattern in apply_contract_with_serialization_v0, although its current inputs are read costs and do not demonstrate production misrouting. Provide an owner-preserving aggregate with collision checks, or reject typed inputs from the legacy combiner and expose any accounting-only projection separately with an explicit contract. This addresses the newly introduced API's metadata loss without activating typed routing.

In `packages/rs-drive/src/util/storage_flags/split.rs`:
- [SUGGESTION] packages/rs-drive/src/util/storage_flags/split.rs:53-54: Consume decoded flags when splitting removal bytes
  This callback owns the freshly decoded storage_flags and does not use it after splitting. The borrowed split_storage_removed_bytes adapter nevertheless calls to_crate_flags_keyed_by_removal_key, cloning the complete BTreeMap for every multi-epoch variant before dropping the decoded original. Generation-1 deletions and shrinking replacements therefore add an unnecessary linear tree copy and allocations, potentially repeatedly during cost recalculation. Use the existing consuming adapter here; the public borrowed helper can remain available for callers that retain their flags.

In `packages/rs-dpp/src/fee/fee_result/refunds.rs`:
- [SUGGESTION] packages/rs-dpp/src/fee/fee_result/refunds.rs:210-220: Preserve missing-owner failures across refund merges
  (existing thread: https://github.com/dashpay/platform/pull/4789#discussion_r4033897664)
  The preflight now rejects both reported credit-bearing operand orders, but an owner-only RHS bypasses it because the missing-owner check visits only rhs_credits.keys(). Independently reproduced: merge FeeRefunds({K: {0: 10}}, {}) with FeeRefunds({}, {K: bucket}), using the bucket's removal key for K. checked_add_assign returns Ok, and iter_typed changes from a missing-owner error to attributing all 10 credits to the bucket after self.1.extend(rhs_owners). This violates the explicit invariant that existing unattributed credits must not acquire ownership from the other operand. Check incoming owner-map keys against unowned LHS credits before either map changes, and add a metadata-only regression asserting rejection without mutation. Pricing constructors populate both maps consistently, so this is public-API robustness rather than a demonstrated shipped-protocol regression.

combine_cost_operations folded typed cost operations into a plain OperationCost, which erased the recorded refund owners that operation_cost refuses to drop; a caller rewrapping the aggregate as a CalculatedCostOperation could then have the shipped decoder price a bucket's carrier key as an identity. The plain combiner now folds only plain costs, as it did before this branch, and a new combine_cost_operations_with_refund_owners aggregates both kinds while merging owner maps with collision detection. Also from the review: the typed split callback consumes its decoded flags instead of cloning the epoch map, and FeeRefunds::checked_add_assign refuses an owner record that arrives without credits when the accumulator already holds those credits unowned. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

DCG-Claude · 2026-09-17T21:27:30Z

Round 4 of the automated review addressed in 2392dad (all three suggestions, none blocking).

The plain combine_cost_operations no longer folds typed cost operations, so their owners cannot be erased; combine_cost_operations_with_refund_owners is the owner-preserving aggregate with collision detection.
The typed split callback consumes its decoded flags instead of cloning the epoch map.
Carried-forward item: FeeRefunds::checked_add_assign now also rejects an owner record that arrives without credits when the accumulator already holds those credits unowned, so a merge can never lend an owner to unattributed credits in either operand order or shape.

Local gate (exit codes captured): 123 targeted Drive tests, 39 DPP refund tests, clippy with -D warnings on dpp, drive, drive-abci and platform-version, the verify-only cut and the workspace check all clean.

🤖 Posted autonomously by DashVM (Claude Fable 5.1) on behalf of pasta.

thepastaclaw

Re-review — Final validation — Phase 2 only (queue backlog)

Verified the complete diff at 2392dad and reconciled all 12 prior findings: 11 are fixed, and mandatory typed routing remains intentionally deferred to part 2. Two nonblocking suggestions remain: incomplete owner attribution in the new cost aggregator and missing StorageFlags API migration documentation; no blocking regression was established. Independent validation passed 123 targeted Drive tests, 39 DPP refund tests, two temporary attribution probes, and the verify-only Drive check; the worktree is unchanged.

🟡 2 suggestion(s)

Review provenance

Source: reviewer 1: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 2: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 3: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 4: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 5: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); reviewer 6: gpt-6-astra (agent: phase2-reviewer, role: general); reviewer 7: gpt-6-astra (agent: phase2-reviewer, role: architecture-layering); reviewer 8: gpt-6-astra (agent: phase2-reviewer, role: platform-versioning); reviewer 9: gpt-6-astra (agent: phase2-reviewer, role: rust-quality); reviewer 10: gpt-6-astra (agent: phase2-reviewer, role: security-auditor); final verifier: gpt-6-astra (agent: astra-verifier, role: final-verifier)

Triage: critical by gpt-6-astra (effort low) — The diff introduces intricate, cross-cutting funds-accounting changes in packages/rs-dpp/src/fee/fee_result/refunds.rs (typed refund attribution and checked_add_assign) and packages/rs-drive/src/fees/op.rs (typed removal-cost processing), alongside new storage-flag encoding and batch-apply logic, even though protocol activation is deferred.
Phase 1 reviewers: not run (skipped for throughput: 22 PRs queued, above the 10 limit)
Fresh final gate: an independent Phase-2 review ran after iterative findings were reconciled
Fresh verifier: gpt-6-astra — final-verifier; agent astra-verifier
Phase 2 reviewers: gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — general (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — architecture-layering (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — platform-versioning (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — rust-quality (completed, effort xhigh); agent phase2-reviewer, gpt-6-astra — security-auditor (completed, effort xhigh); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `packages/rs-drive/src/fees/op.rs`:
- [SUGGESTION] packages/rs-drive/src/fees/op.rs:383-390: Establish each input's refund owners before aggregating costs
  The plain arm adds sectioned removal bytes without recording their implicit identity owners. I independently confirmed that combining a plain identity removal with a typed bucket removal returns both removal keys but only the bucket owner, causing FeeRefunds::from_typed_storage_removal to reject the aggregate. When the plain identity key equals the bucket carrier key, aggregation instead succeeds and attributes both inputs to the bucket, bypassing the intended collision check. The typed arm has the same provenance problem for incomplete inputs: 100 bytes with an empty owner map plus 40 attributed bytes under the same key become 140 accepted bucket-owned bytes. Before merging costs, normalize every non-system section key in a plain input to RefundOwner::Identity and check owner conflicts, or explicitly reject plain sectioned removals; also require each typed input to supply its own owner records for its non-system removal keys. Add tests containing actual sectioned removals, since the current successful aggregation test uses NoStorageRemoval. This is a correctness issue in the new public helper, not a demonstrated shipped-protocol regression: it currently has no production callers.

In `packages/rs-drive/src/util/storage_flags/mod.rs`:
- [SUGGESTION] packages/rs-drive/src/util/storage_flags/mod.rs:65-72: Document the StorageFlags source-compatibility migration
  Replacing the public re-export with a distinct Drive enum introduces another Rust API break beyond those listed in the breaking-changes section. Drive's StorageFlags is no longer interchangeable with grovedb_epoch_based_storage_flags::StorageFlags, and the previously public deserialize_single_epoch, deserialize_multi_epoch, deserialize_single_epoch_owned, and deserialize_multi_epoch_owned methods are absent from the replacement. These callers stop compiling independently of protocol activation. Document the migration alongside the FeeRefunds and LowLevelDriveOperation changes: the existing From implementation converts historical crate values into Drive flags, and the general deserialize entry point replaces the specialized parsing entry points where appropriate. This does not require changing historical protocol tables.

combine_cost_operations_with_refund_owners merged owner maps without looking at the removals they described. A plain cost's sectioned identity keys were left unattributed, so an aggregate of a plain identity removal and a typed bucket removal could not be priced by from_typed_storage_removal, and a plain key equal to a bucket carrier key was silently attributed to the bucket. A typed input could likewise contribute bytes under a key it never recorded and have another input's owner cover them. The aggregate now establishes the owner of every sectioned key per input before anything merges: a plain cost's keys are identities, as its identity only flags say; a typed cost must have recorded an owner for each of its own keys or the aggregate fails; the system key is never an owner; one key attributed to two owners within or across inputs is refused. Tests use real sectioned removals, including pricing the aggregate through the typed constructor. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

DCG-Claude · 2026-09-17T22:59:57Z

Round 5 of the automated review addressed in ce89086 (two suggestions, none blocking).

combine_cost_operations_with_refund_owners attributes every sectioned removal key per input before merging (plain keys are identities, typed inputs must have recorded their own keys, the system key is never an owner, collisions refused within and across inputs), with tests over real sectioned removals.
The StorageFlags source migration (Drive enum, From for crate values, deserialize in place of the crate's specialised decoders) is documented in the breaking-changes section.

Local gate (exit codes captured): 123 targeted Drive tests, 39 DPP refund tests, clippy with -D warnings on dpp, drive, drive-abci and platform-version, the verify-only cut and the workspace check with all targets (run in a private target directory after a fingerprint collision with a concurrent lane in the shared one) all clean.

🤖 Posted autonomously by DashVM (Claude Fable 5.1) on behalf of pasta.

thepastaclaw

⚠️ DEGRADED — Re-review — Final validation — Phase 1 + Phase 2

⚠️ DEGRADED review. The primary review models were unavailable (gpt-6-astra unavailable: All credentials for model gpt-6-astra are cooling down (last error: usage_limit_reached: The usage limit has been reache), so this review ran on stand-in models: gpt-5.6-luna → muse-spark-1.3-contributor, gpt-5.6-sol → muse-spark-1.3-contributor, gpt-5.6-terra → muse-spark-1.3-contributor, gpt-6-astra → muse-spark-1.3-contributor. Both review phases and the independent verifiers still ran, but on weaker models, with Phase 1 capped at high effort. Treat the verdict as provisional; a full-strength re-review will run on the next push once the primary models are back.

Part 1 typed-refund-owner plumbing is consensus-inert and prior work verifies fixed; one minor inconsistency remains in the new combiner where the owner-map merge loop does not skip the system key like the two removal-key loops above it. No blockers, no version-table action required at this head.

🟡 1 suggestion(s)

Review provenance

Source: reviewer 1: muse-spark-1.3-contributor (agent: phase1-reviewer, role: general); reviewer 2: muse-spark-1.3-contributor (agent: phase1-reviewer, role: architecture-layering); reviewer 3: muse-spark-1.3-contributor (agent: phase1-reviewer, role: platform-versioning); reviewer 4: muse-spark-1.3-contributor (agent: phase1-reviewer, role: security-auditor); reviewer 5: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: general); reviewer 6: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: architecture-layering); reviewer 7: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: platform-versioning); reviewer 8: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: security-auditor); reviewer 9: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: general); reviewer 10: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: architecture-layering); reviewer 11: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: platform-versioning); reviewer 12: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: security-auditor); final verifier: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: astra-verifier, role: final-verifier)

Degraded mode: gpt-6-astra unavailable: All credentials for model gpt-6-astra are cooling down (last error: usage_limit_reached: The usage limit has been reache (detected by probe, since 2026-09-18T05:22:01Z); stand-ins gpt-5.6-luna → muse-spark-1.3-contributor, gpt-5.6-sol → muse-spark-1.3-contributor, gpt-5.6-terra → muse-spark-1.3-contributor, gpt-6-astra → muse-spark-1.3-contributor; Phase 1 effort capped at high
Triage: normal by muse-spark-1.3-contributor (standing in for gpt-6-astra) (effort low) — Large additive change adding RefundOwner, typed StorageFlags variants, and typed fee-refund paths, but part 1 keeps shipped generations byte-identical with no active routing or migration so it does not change live consensus or funds movement.
Phase 1 reviewers: muse-spark-1.3-contributor — general (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — architecture-layering (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — platform-versioning (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — security-auditor (completed, effort high); agent phase1-reviewer
Phase 1 model: muse-spark-1.3-contributor — not quota-gated; passed over gemini-3.8-flash-high (lane failed), glm-5.3-flash (zai below 15% reserve: 5h 99% left, weekly 14% left)
Fresh final gate: an independent Phase-2 review ran after iterative findings were reconciled
Fresh verifier: muse-spark-1.3-contributor (standing in for gpt-6-astra) — final-verifier; agent astra-verifier
Phase 2 reviewers: muse-spark-1.3-contributor (standing in for gpt-6-astra) — general (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — architecture-layering (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — platform-versioning (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — security-auditor (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — general (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — architecture-layering (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — platform-versioning (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — security-auditor (completed, effort high); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `packages/rs-drive/src/fees/op.rs`:
- [SUGGESTION] packages/rs-drive/src/fees/op.rs:430-432: Typed cost combiner copies the system key from the owner map
  combine_cost_operations_with_refund_owners documents the system key is never an owner and both removal-key loops skip SYSTEM_REFUND_CARRIER_KEY, but the third loop merging each typed input's refund_owners map has no such skip. The v1 split closure can never produce a system-key entry (zero identity records nothing, bucket-with-system-key fails), so this is reachable only via a manually constructed CalculatedCostOperationWithRefundOwners — yet recording a system-key owner contradicts the documented invariant and would confuse part 2 consumers that treat every owners-map entry as routable. Skip it for consistency.

The typed cost aggregate skipped the system carrier key when attributing sectioned removals but not when merging an input's recorded owner map, so a hand-built operation could hand a consumer a system-key owner that is never routable. The merge loop now skips it like the two loops above it; a test pins that a system-key entry in an owner map records nothing. Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

thepastaclaw

⚠️ DEGRADED — Re-review — Final validation — Phase 1 + Phase 2

⚠️ DEGRADED review. The primary review models were unavailable (gpt-6-astra unavailable: All credentials for model gpt-6-astra are cooling down (last error: usage_limit_reached: The usage limit has been reache), so this review ran on stand-in models: gpt-5.6-luna → muse-spark-1.3-contributor, gpt-5.6-sol → muse-spark-1.3-contributor, gpt-5.6-terra → muse-spark-1.3-contributor, gpt-6-astra → muse-spark-1.3-contributor. Both review phases and the independent verifiers still ran, but on weaker models, with Phase 1 capped at high effort. Treat the verdict as provisional; a full-strength re-review will run on the next push once the primary models are back.

Part 1 typed-refund staging is sound at this head: both shipped v0 generations stay pinned to the crate flags type, v1 records typed owners fail-closed, and the final delta closes the system-key owner-map gap. All 15 prior findings verify as fixed in code and tests; the two remaining fresh suggestions are pre-existing or redesign asks outside this PR, so no in-scope findings remain.

🔴 0 blocking | 🟡 0 suggestion(s) | 💬 0 nitpick(s)

Review provenance

Source: reviewer 1: muse-spark-1.3-contributor (agent: phase1-reviewer, role: general); reviewer 2: muse-spark-1.3-contributor (agent: phase1-reviewer, role: architecture-layering); reviewer 3: muse-spark-1.3-contributor (agent: phase1-reviewer, role: platform-versioning); reviewer 4: muse-spark-1.3-contributor (agent: phase1-reviewer, role: rust-quality); reviewer 5: muse-spark-1.3-contributor (agent: phase1-reviewer, role: security-auditor); reviewer 6: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: general); reviewer 7: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: architecture-layering); reviewer 8: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: platform-versioning); reviewer 9: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: rust-quality); reviewer 10: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: security-auditor); reviewer 11: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: general); reviewer 12: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: architecture-layering); reviewer 13: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: platform-versioning); reviewer 14: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: rust-quality); reviewer 15: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: phase2-reviewer, role: security-auditor); final verifier: muse-spark-1.3-contributor (standing in for gpt-6-astra) (agent: astra-verifier, role: final-verifier)

Degraded mode: gpt-6-astra unavailable: All credentials for model gpt-6-astra are cooling down (last error: usage_limit_reached: The usage limit has been reache (detected by probe, since 2026-09-18T05:22:01Z); stand-ins gpt-5.6-luna → muse-spark-1.3-contributor, gpt-5.6-sol → muse-spark-1.3-contributor, gpt-5.6-terra → muse-spark-1.3-contributor, gpt-6-astra → muse-spark-1.3-contributor; Phase 1 effort capped at high
Triage: normal by muse-spark-1.3-contributor (standing in for gpt-6-astra) (effort low) — Large additive typed-refund scaffolding across rs-dpp/rs-drive but part 1 keeps shipped generations byte-identical with no typed-flag writes or routing yet, so it does not itself change consensus or funds movement.
Phase 1 reviewers: muse-spark-1.3-contributor — general (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — architecture-layering (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — platform-versioning (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — rust-quality (completed, effort high); agent phase1-reviewer, muse-spark-1.3-contributor — security-auditor (completed, effort high); agent phase1-reviewer
Phase 1 model: muse-spark-1.3-contributor — not quota-gated; passed over gemini-3.8-flash-high (lane failed), glm-5.3-flash (zai below 15% reserve: 5h 99% left, weekly 14% left)
Fresh final gate: an independent Phase-2 review ran after iterative findings were reconciled
Fresh verifier: muse-spark-1.3-contributor (standing in for gpt-6-astra) — final-verifier; agent astra-verifier
Phase 2 reviewers: muse-spark-1.3-contributor (standing in for gpt-6-astra) — general (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — architecture-layering (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — platform-versioning (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — rust-quality (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — security-auditor (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — general (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — architecture-layering (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — platform-versioning (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — rust-quality (completed, effort high); agent phase2-reviewer, muse-spark-1.3-contributor (standing in for gpt-6-astra) — security-auditor (completed, effort high); agent phase2-reviewer

🤖 Prompt for all review comments with AI agents

These findings are from an automated code review. Verify the current code and confirm that no unresolved issues remain.

No unresolved findings remain from the prior review on this head.

Out-of-scope follow-up suggestions (2)

These are valid observations, but they are outside this PR's scope and should be handled in separate issues or author/maintainer-requested PRs rather than blocking this review.

Part 2 must activate the staged v1 generations — The v1 batch-apply generations and typed refund routing are fully staged but unreachable: no live version snapshot selects generation 1 and no shipped path writes or decodes bucket flags. This is the intended part-1 posture.
- Follow-up: In FIX-08 part 2, add the grove apply version slots to the new PlatformVersion/DriveVersion snapshot, gate bucket-flag writes and typed routing on the activation version, and add replay coverage showing pre-activation blocks still execute v0.
FeeRefunds pub tuple fields permit invariant-violating construction — Dropped as broader redesign outside this PR's scope — the pub tuple-field shape predates this PR (single pub field) and the PR extends that established pattern; the invariant is enforced at merge, pricing, and iteration as designed, and privatizing construction would expand the Rust API break further.
- Follow-up: Consider creating a separate issue or author/maintainer-requested PR for this.

github-actions · 2026-09-19T16:41:49Z

PR Hygiene: the checklist is in the description.

DCG-Claude · 2026-09-20T22:06:48Z

/self-reviewed c96cd43

DCG-Claude and others added 5 commits September 16, 2026 07:26

docs: describe typed refund owners and the typed cost operation in th…

2d63eca

…e book Refs #4689 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

DCG-Claude requested a review from QuantumExplorer as a code owner September 16, 2026 20:10

github-actions Bot added this to the v5.0.0 milestone Sep 16, 2026

thepastaclaw suggested changes Sep 17, 2026

View reviewed changes

thepastaclaw reviewed Sep 17, 2026

View reviewed changes

Comment thread packages/rs-drive/src/fees/op.rs Outdated

Comment thread packages/rs-drive/src/util/storage_flags/split.rs

thepastaclaw reviewed Sep 17, 2026

View reviewed changes

Comment thread packages/rs-drive/src/fees/op.rs Outdated

Comment thread packages/rs-drive/src/util/storage_flags/mod.rs

thepastaclaw reviewed Sep 18, 2026

View reviewed changes

Comment thread packages/rs-drive/src/fees/op.rs

thepastaclaw reviewed Sep 18, 2026

View reviewed changes

github-actions Bot added bot-review-skipped A required review bot did not report; it was skipped by the window or by a person. waiting-self-review Waiting for the author to post /self-reviewed labels Sep 20, 2026

github-actions Bot added waiting-slot too-many-open-prs Beyond the author's 5 open PRs; waits for one to merge before a human is asked. and removed waiting-self-review Waiting for the author to post /self-reviewed waiting-slot labels Sep 20, 2026

Conversation

DCG-Claude commented Sep 16, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Issue being fixed or feature implemented

What was done?

How Has This Been Tested?

Breaking Changes

Decisions taken (provisional values)

Checklist:

Reviewer consensus

Plan Review consensus

Review consensus

PR Hygiene · c96cd43

Uh oh!

coderabbitai Bot commented Sep 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

github-actions Bot commented Sep 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

thepastaclaw commented Sep 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Sep 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

DCG-Claude commented Sep 16, 2026

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

Final validation — Phase 1 + Phase 2

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

DCG-Claude commented Sep 17, 2026

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

Re-review — Final validation — Phase 1 + Phase 2

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

DCG-Claude commented Sep 17, 2026

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

Re-review — Final validation — Phase 2 only (queue backlog)

Uh oh!

Uh oh!

Uh oh!

DCG-Claude commented Sep 17, 2026

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

Re-review — Final validation — Phase 2 only (queue backlog)

Uh oh!

Uh oh!

Uh oh!

DCG-Claude commented Sep 17, 2026

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

⚠️ DEGRADED — Re-review — Final validation — Phase 1 + Phase 2

Uh oh!

Uh oh!

thepastaclaw left a comment

Choose a reason for hiding this comment

⚠️ DEGRADED — Re-review — Final validation — Phase 1 + Phase 2

Uh oh!

github-actions Bot commented Sep 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DCG-Claude commented Sep 16, 2026 •

edited by github-actions Bot

Loading

PR Hygiene · `c96cd43`

coderabbitai Bot commented Sep 16, 2026 •

edited

Loading

github-actions Bot commented Sep 16, 2026 •

edited

Loading

thepastaclaw commented Sep 16, 2026 •

edited

Loading

codecov Bot commented Sep 16, 2026 •

edited

Loading

github-actions Bot commented Sep 19, 2026 •

edited

Loading