Skip to content

task.md

Latest commit

 

History

History
64 lines (58 loc) · 3.73 KB

task.md

File metadata and controls

64 lines (58 loc) · 3.73 KB
Raw

QuorumProof production-readiness pass — resume prompt

Continue the QuorumProof production-readiness pass. Last commit on main is 45497af (wires monte_carlo_sweep into real tests). History: 55a179a (jurisdiction modeling) → 0d31908 (real get_slice_attack_cost_estimate stake-weighted math + fixed the challenge_attestation/vote_on_challenge panic-on-real-challenge bug) → 45497af (this session).

Status of the three items from last resume:

  • (a) DONE. Full cargo test --target x86_64-unknown-linux-gnu --lib -p quorum_proof confirmed green: 251 passed, 0 failed. 0d31908's fixes are solid.
  • (b) DONE, with a finding surfaced, not silently fixed. Added two tests in economic_security_tests.rs that call monte_carlo_sweep for real. One test documents a real discovery: the off-chain attack-cost model (simulate_corrupt_existing/simulate_sybil/concentration_risk_score) keys off base_weight and corruption_price only — it never reads AttestorAgent::effective_weight() — so sweeping different reputation distributions through monte_carlo_sweep currently produces byte-identical results every trial. The "variance" the function's own doc comment claims to measure is a no-op today. This is a real gap versus the on-chain contract, which does weight consensus by reputation via get_effective_weight (lib.rs) when reputation_weighting_enabled is set. Not fixed — deciding how the off-chain simulator should incorporate reputation (mirror get_effective_weight exactly? apply it only to concentration_risk_score? something else?) is a design call, not a mechanical fix. Flag to user next session if not already discussed.
  • (c) Still undecided — same disabled feature_tests/doc_tests situation as before (~1600 lines in lib.rs, 144 compile errors if re-enabled, likely a multi-hour rewrite). Not started; still needs a user decision on whether it's in scope.

Context from the original audit (15-item list, 4 picked)

  1. BBS+ selective-disclosure credentials (contracts/bbs_plus_v1) — DONE. Real signing/verification/selective-disclosure/revocation, 45/45 tests passing in 3 build configs incl. wasm32-unknown-unknown. Still not wired into quorum_proof/sbt_registry, still excluded from the root workspace/CI, and has had no external cryptographic audit.
  2. api-server test suite actually passes at runtime — DONE (npm test run, 5 stale tests fixed).
  3. Light-client bridge gap (crossChainBridge.ts trusting RPC provider) — DONE. Real BLS sync-committee verification (Altair light-client protocol, SSZ Merkle proofs) for Ethereum mainnet/Sepolia.
  4. Jurisdiction modeling + economic-security simulation wiring — DONE (mechanically complete; one design gap flagged, see item (b) above):
    • Jurisdiction modeling in api-server/src/searchIndex.ts — pushed in 55a179a. Hierarchical ISO 3166-1/3166-2 + supranational group (EU) codes, indexed and filterable/facetable, 23 new tests.
    • Economic-security simulation wiring — get_slice_attack_cost_estimate now does real stake-weighted greedy set-cover instead of hardcoded placeholder constants (0d31908), with 5 new integration tests driving real create_slice/attest/challenge_attestation/ vote_on_challenge/deposit_attestor_stake calls. monte_carlo_sweep (off-chain Monte Carlo variance sweep in simulation_agent_based.rs) is now wired into real tests (45497af) but exposed that reputation currently has zero effect on its output — see (b) above for the open design question.

Remaining 11 audit items not yet started

(Not detailed here — re-derive from the original audit if needed, or ask the user for the list again.)