Preparing Perpetua for mainnet means being explicit about what has been reviewed, who may review the rest, and what a reviewer is rewarded for finding.

This document is the scope of record. It answers:

1. what counts as in-scope for a security review,
2. how findings are reported,
3. how in-scope findings are rewarded.

The repository holds the artifacts a third-party auditor needs, so a review starts from buildable, pinned, reproducible inputs:

• Pinned toolchain and target — rust-toolchain.toml pins Rust 1.97.1 and the wasm32v1-none WASM target; the contract refuses to build for wasm32-unknown-unknown or with debug assertions (lib.rs guards).
• Reproducible build — script/release.sh and script/provenance.sh produce and verify a SLSA-style wasm manifest (docs/provenance.md).
• Interface of record — docs/ABI.md and contracts/stream/abi/fluxora_stream.json pin the public ABI; docs/audit.md is the drift-checked entrypoint table.
• Invariant suite — the 36 test modules (≈570 tests) assert the I1–I5 invariants after every operation, over random schedules (test::props) and across every entry-point ordering (test::monotonicity). The event byte footprint enforcer invariants (boundary guards, deterministic rejection of degenerate parameters, zero silent wrapping) are pinned by tests/test_event_byte_budget.py.
• Known-limits register — docs/KNOWN-LIMITATIONS.md records what a green suite does not prove, including the residual limits of the event byte footprint enforcer's boundary checks. Reviewers are asked to read it before anything else.

Status: no third-party audit has been completed. Do not claim otherwise. This document lowers the cost of the first one.

• contracts/stream/ — the entire contract: src/accrual.rs (arithmetic), src/lib.rs (all entry points), src/storage.rs (TTL and storage access), src/types.rs, src/error.rs, src/events.rs.
• Fund-safety properties: the I1–I5 invariants in contracts/stream/src/accrual.rs and the pool invariant liability <= contract balance.
• The off-chain math mirror sdk/stream-math.js (must never disagree with accrual.rs).
• The Soroban 16 KiB event byte footprint enforcer script/event-byte-budget.py and its boundary/adversarial guarantees: degenerate parameters (zero-duration cliffs, i128::MAX deposit rates, same-ledger state transitions) are rejected deterministically with explicit error variants before any state is committed or any event is emitted, and no integer arithmetic silently wraps. These invariants are asserted by tests/test_event_byte_budget.py (unit and property-based regression tests) and are part of the invariant suite below.
• Build/provenance tooling in script/ and tools/provenance/ (only insofar as a defect could ship a wrong wasm).

• contracts/factory/, contracts/governance/ — separate products for their own review.
• contracts/archival-probe/ — a throwaway probe, never deployed as product.
• Deploy/infra configuration, key custody, and any frontend.
• Known limits already disclosed in docs/KNOWN-LIMITATIONS.md.

A reviewer accepts this scope by confirming:

1. the release build reproduces byte-for-byte from a fresh checkout (script/release-dry-run.sh), and
2. the invariant suite passes at the pinned toolchain, and
3. ambition of the ABI snapshot (docs/audit.md) matches src/lib.rs.

• Private: email the maintainers (addresses in the GitHub org). Include reproduction, expected vs actual, and impact.
• On-chain: an exploit attempt that provably drains or locks funds is disclosed privately first; there is no public-fine process that worries about front-running.
• Public: anything that is not a fund-safety issue (UI, docs, gas) can be filed straight as a GitHub issue.

Every report is acknowledged within 7 days and answered with a fix plan or a reasoned non-issue within 30.

• Severity scale — Critical (guards broken, funds at risk), High (funds stuck or commutative accounting corruption), Medium (invariant violation without fund loss), Low (gas, UX, informational) — aligned with the settlement model below.
• Rewards — reserved for a campaign the maintainers open before mainnet launch; the amounts and token denom are set in that announcement. Until then this section is a contract, not a funded pool:
  • Critical — top reward tier, awarded at the discretion of the maintainers after a verified impact.
  • High/Medium — tiered, only in an open campaign.
  • Low/informational — acknowledgements in docs/accredited.md once created.
• Eligibility — any original finding; excluded are findings already disclosed in docs/KNOWN-LIMITATIONS.md, docs/audit.md, this file, or developer-found during a scheduled audit. Self-research against the deployed testnet contract is fine; a live mainnet exploit attempt to demonstrate a bounty finding is not eligible and is a crime on most chains.
• Disclosure — bounty-worthy findings are disclosed after a fix ships or the maintainers grant a 90-day embargo.

• docs/audit.md — entrypoint inventory + arithmetic audit (§98).
• docs/dust-theft-proof.md — why micro-top-ups cannot drain the pool (§102).
• docs/KNOWN-LIMITATIONS.md — what a green suite does not prove, including the event byte footprint enforcer's boundary guarantees.
• docs/ABI.md — the interface of record.
• script/event-byte-budget.py — the Soroban 16 KiB event byte footprint enforcer; invariants asserted by tests/test_event_byte_budget.py.