Notes / Corrections
Source: data/correction_2026-09-17-drift-protocol-adversarial-review.md in the repository, rendered here without changes. Source file
Same-day follow-up to the Drift Protocol addition (third retroactive backtest, data/backtest_2026-09-17-drift-protocol-security-council- compromise.md, plus a new live scorer, chains/solana/scorers.py:: score_drift_protocol, and chains/solana/data/methodology_test_2026-09-17- drift-protocol.md). Before committing, this addition went through an adversarial review workflow (3 independent reviewers -- byte-layout decode correctness for the new read_governance_v2, independent re-derivation of every on-chain fact claimed in the backtest, and scorer-code robustness -- each finding piped to an independent verifier), matching this project's established "no shortcuts" review discipline. Result: 3 findings, 3 confirmed, 0 refuted. The byte-layout dimension returned zero findings (read_governance_v2 independently re-verified correct against two real accounts).
none_means_renounced=True. score_drift_protocol's program-upgrade path called _resolve_squads_v4 without this flag -- the one call site in the file that didn't, unlike the structurally identical calls in score_kamino_lend (klend_sq, scope_upgrade_sq). Without it, a renounced (None) upgrade authority would score as a false MISMATCH (20/20/0, with a factually wrong "degrading" note) instead of the safest band (100/100/100) -- an inverted risk signal for the one state that's actually safest, the exact bug class this project's own code already guards against elsewhere. Drift's authority is not currently renounced, so the published score (compositeScore = 48) is unaffected -- verified by re-running the live dry-run after the fix and confirming it's unchanged. A new regression test (test_renounced_program_authority_is_not_scored_as_a_mismatch) directly exercises the renounced case via monkeypatching, since the real chain state can't be used to test it (Drift's authority isn't renounced today).compositeScore = ... = 48 -- but that formula, applied to those exact inputs, evaluates to 53, not 48 (independently confirmed via the project's own _composite() function). 48 IS the correct, live-verified score for Drift Protocol as a whole, but only once METHODOLOGY.md 6.2's "minimum over every full-power path" rule folds in a SECOND path (Drift DAO's own Realms governance, multisigScore=70, lower than this path's 87) -- a step the backtest file's own formula box never showed, even though the companion methodology-test file already did this combination correctly. Fixed: the backtest file now shows both numbers explicitly (53 for this path alone, 48 combined) with the min() step spelled out, instead of a bare, arithmetically-false = 48.VaultTransactionCreate, ProposalCreate, omitting a third instruction actually present in that same transaction's own logs: ProposalApprove. This mattered beyond completeness: the multisig's 2-of-5 threshold needs two approvals, and the two-transaction narrative as originally written showed only one (ProposalApprove in Tx 2 alone could not have satisfied a 2-of-5 threshold by itself). Fixed: the transcript now includes all three Tx 1 instructions, with an added note explaining the real mechanism (Tx 1's signer creates the proposal AND immediately self-approves it; Tx 2, one second later, casts the second approval from a different signer and executes once threshold is met).python3 -m unittest discover -s scripts/lib/tests: 107/107 pass (92 pre-existing + 15 new Drift-Protocol tests from the original addition, including the new renounced-authority regression test).python3 scripts/validate_all_scorers.py --ecosystem solana: 4/4 targets pass, OVERALL: PASS.score_drift_protocol after the fix: adminKeyScore=65, multisigScore=70, timelockScore=2, compositeScore=48 -- unchanged from before the fix, since Drift's authority isn't in the renounced state the bug only affected.python3 -m unittest discover -s scripts/lib/tests
python3 scripts/validate_all_scorers.py --ecosystem solana
cd chains/solana && python3 -c "import sys; sys.path.insert(0,'.'); from scorers import score_drift_protocol; import json; print(json.dumps(score_drift_protocol('https://api.mainnet-beta.solana.com'), indent=2, default=list))"Not an audit, not investment advice. All notes ยท Back to the dashboard.