Notes / Findings

Finding (2026-09-18): the DAO-via-bridge sub-model and the cross-chain multisig grid, quantified

Findings2026-09-18Chain: Base, Arbitrum

Source: data/finding_2026-09-18-cross-chain-dao-bridge-and-multisig-grid.md in the repository, rendered here without changes. Source file

Requested by a tracked GitHub Projects item bundling two related cross-ecosystem questions. Pure investigation -- no scorer file (chains/*/scorers.py, scripts/lib/scorers.py) is modified by this finding. A change to any of these formulas needs this project's own adversarial-review discipline before committing, per the project's review rules.

Part 1: "external DAO governance via bridge" -- 4 real instances, only 1 shares code#

chains/arbitrum-ecosystem/scorers.py:60-69 already names this exact gap in a comment: "Not built this pass; each scorer below discloses its own slice of the gap individually." Confirmed, with exact numbers:

Aave (Arbitrum and Base) is the exact same BGD-Labs Aave-Governance-V3 relay shape, byte-identically duplicated across the two ecosystem files, and lands on the same numbers (65/100/50) -- consistent with itself, just not with a shared helper. Compound-Base (75/100/65) and Uniswap-Base (80/100/75) are each independently hardcoded to DIFFERENT numbers for what the tracker item calls "the same" confirmed-L1-DAO- via-bridge authority shape.

The one place this pattern DOES have a shared helper is Robinhood Chain's own Uniswap stack: _score_uniswap_bridge_alias_root() (scripts/lib/scorers.py:124-174), called by 4 scorers, returns a consistent (80, 100, 75) confirmed / (5, 0, 0) degraded. Coincidentally the SAME numbers as Base's independently-hardcoded Uniswap scorer -- by explicit stated convention (each docstring says so), not shared code.

Open question: is the 65 vs 75 vs 80 admin_key spread justified by a real difference in each target's authority chain (a defensible per-target judgment call, the way this project scores most things), or is it accidental drift from writing each scorer independently without cross-checking the others? This finding does not decide that -- tracing each target's real authority chain closely enough to know is the next step, not attempted here.

Part 2: the 2-of-3 multisig grid -- one divergence is deliberate and documented, one is genuinely unexplained#

The tracker item's own framing ("2-of-3 = 39 Hyperliquid vs 35 Safe") is accurate for multisigScore, not compositeScore -- worth correcting before anyone reasons from it. Three independently verified formulas for a confirmed 2-of-3:

Important correction to the tracker item's implicit framing: Hyperliquid's divergence from the Safe formula is not an oversight -- chains/hyperliquid/METHODOLOGY.md:415-423 documents the actual reason IN THE METHODOLOGY ITSELF: the Safe formula 15k+5(n-k) rewards a larger n at equal k, which contradicts HIP-3's own approved ordering rule (a larger n at equal k should score WEAKER, since any one of more keys being compromised is a bigger attack surface). The doc states the tradeoff explicitly: "a 2-of-3 scores 39 here against 35 under the Robinhood Chain Safe formula" -- a conscious, justified, already-disclosed choice, not a gap.

Solana's 57 is the real outlier, and IS closer to "ad hoc" as the tracker item says: its constants (15k + 40k/n) trace to a one-off calibration against a real 5-of-7 Realms council (chains/solana/METHODOLOGY.md:379-389), not a stated first-principles ordering argument the way Hyperliquid's is. At the composite level the practical gap is smaller than the raw multisig numbers suggest (32 vs 31 vs 35 -- a 4-point spread, not the ~20-point spread the raw multisigScore numbers alone would imply), since admin_key and the 0.4/0.3/0.3 weighting compress it.

What this does NOT do#

• Does not change any published score in any ecosystem.
• Does not modify any scorer file.
• Does not decide whether the Arbitrum/Base DAO-via-bridge scorers should be unified into one shared helper (Part 1), or whether Solana's multisig formula should be re-derived to match one of the other two ecosystems' stated ordering principle (Part 2) -- both are real design questions for an adversarial-review pass, not settled here.

Not an audit, not investment advice. All notes · Back to the dashboard.