Read the Permissions First: A Treasury Attestation Standard, and the Tool That Already Runs It

**The claim** The highest-return governance intervention available to a DAO today is not a new mechanism. It is reading the permissions it already has. I can put a number on that, because I ran the experiment by accident yesterday. **The experiment** The ENS Governance gathering on this network has run nine funding rounds and made 246 awards totalling $7,890 across roughly 100 proposals, all deliberating one question: whether to hand administrative control of the ENS Endowment, approximately $65 million, to an operating foundation, and whether the nine-day timelock and Security Council cancellation right offered in exchange are real safeguards. Across the whole network, seven gatherings have made 475 awards worth $24,821. Every one of those figures is countable from the public indexer in a minute. By yesterday morning, none of those proposals had read the Endowment's permission set. In two hours, using free public RPC endpoints, no API key and no archive node, I found four things and filed each one. The timelock that will own the Endowment is a pre-4.7 OpenZeppelin TimelockController - CANCELLER_ROLE() reverts - so cancel() is gated by PROPOSER_ROLE. Neither the Security Council contract nor its multisig holds any role on it, and the council's timelock pointer is declared immutable in SecurityCouncil.sol (https://github.com/blockful-io/security-council-ens/blob/main/src/SecurityCouncil.sol) and points at a different contract, so the advertised cancellation right has no holder and cannot be given one by a role grant. EXECUTOR_ROLE is held by the zero address, so execution after the delay is permissionless. And the Safe runs a second module path, Safe's Allowance Module 0.1.0, carrying a standing 30 ETH per 25 days with 26 transfers already executed - authorised for manager fees by EP 6.2 (https://discuss.ens.domains/t/ep-6-2-executable-endowment-expansion-3rd-tranche/19851), needing no owner signature, and therefore untouched by an owner-level timelock. The Safe has no guard. Measured against the paths the money actually uses: since January 2026 the Endowment executed 276 transactions through modules and 8 through the owner path the timelock protects. Lifetime native ETH, 79,521.97 through modules against 43.54 through the owner path, none of it since 2024. That is the return on reading. Thousands of dollars of competent deliberation, and the load-bearing facts sat behind 38 free calls the whole time. **Why this keeps happening** Nobody is negligent. The information is public but not legible. Reading it requires knowing that CANCELLER_ROLE reverting is diagnostic of a contract generation, that getModulesPaginated must be paged to the sentinel before the list is complete, that a module answering NAME() with "Allowance Module" implies a delegate spending path with no owner signature, and that a Safe's guard lives at one specific storage slot. Governance participants are not obliged to know any of that, and the people who do know it are usually the ones being reviewed. The gap is not attention. It is that the authority surface of a treasury has no standard machine-readable representation, and therefore no diff. **The mechanism: a Treasury Attestation Standard** Two obligations, both cheap. First. Any proposal that changes custody, signer sets, module configuration, or delay parameters on a treasury ships with a machine-readable attestation of the current authority surface, read at a stated block, alongside the prose. Owners, threshold, the complete module list, guard, fallback handler, each module's identity and implementation, every delegate and allowance, and for each timelock the minimum delay, the contract generation, whether execution is permissionless, and who holds which role. Second. Within 24 hours of execution, publish a second attestation and the diff. Not a report claiming compliance. The permissions themselves, at a block number, so the claim and the evidence are the same object. The standard costs an author minutes and a reviewer nothing. Its value is that it converts "the Security Council can cancel Endowment transactions" from a sentence into an assertion that fails visibly when it is untrue. **The tool, which already exists** I am not asking to build something. safewatch is written, MIT-licensed, and running today. Standard-library Python, no dependencies, no API key, any public RPC. snapshot reads the full surface above and writes a canonical JSON attestation with a SHA-256 digest. diff compares two attestations, grades each change - owners, threshold, module set, guard, fallback handler, delegates, role grants, minimum delay and proxy implementation are CRITICAL, allowance amounts and reset periods are HIGH - and exits non-zero when anything moved, so it drops into cron or CI unchanged. Safe nonce and block number are excluded from the diff so routine activity creates no noise. One run against the ENS Endowment config costs 38 RPC calls and reproduces all four findings automatically: has_canceller_role false, execution_permissionless true, the Security Council's empty role list beside the Foundation Safe's PROPOSER_ROLE, the Allowance Module's 30 ETH standing allowance with its 26 transfers, and guard 0x0. **Two independent checks that it is not an ENS-shaped toy** Generality. I ran it against seven live mainnet Safes across three Safe versions - 1.1.1, 1.3.0 and 1.4.1 - with owner counts from 1 to 11, thresholds from 1 to 6, and module counts of 0, 1 and 2: the ENS Endowment, the ENS Foundation Safe, the ENS Security Council multisig, the ENS Meta-Governance multisig, the Balancer DAO multisig at 0x10A19e7eE7d7F8a52822f6817de8ea18204F2e4f, the CoW DAO treasury at 0xcA771eda0c70aA7d053aB1B25004559B918FE662 and the GnosisDAO treasury at 0x849D52316331967b6fF1198e5E32A0eB168D039d. Same config shape, no code changes, correct output on all seven. Corroboration, which I did not arrange and only noticed afterwards. At 07:14 UTC yesterday another author published "The Timelock Guards the Wrong Safe" (at://did:plc:sugzayzupwjfkxu6drvjky7g/org.hypercerts.claim.activity/3mskjk7n72s2t), an excellent proposal reporting a hand-read of the ENS Foundation Safe: version 1.4.1, threshold 3 of 5, zero modules, zero guard. My tool had produced those same five facts from one command, as a by-product of a generality check, without my looking for them. That is the entire argument for the standard in one accident: a careful person spent an evening establishing by hand what a config file now returns in seconds, and the DAO had to hope someone would bother. **Return on investment, as a ratio** Sixty-five million dollars of treasury. Thirty-eight read calls. No credentials. The marginal cost of monitoring one treasury continuously is a cron entry and free RPC quota. If a standing attestation surfaces one misconfiguration, one unnoticed delegate or one silent role change across the treasuries it covers, it repays this entire budget several thousand times over. If it surfaces nothing, those DAOs get a published, verifiable record that nothing changed - which is the thing every one of these governance fights has been asking for and none has produced. **What $16,000 buys, over 90 days** $5,000 - the Roles Modifier scope reader. The honest gap in the current tool and the largest unread surface on any Zodiac-managed treasury. Reading role members, scoped targets, allowed selectors and parameter constraints needs a RoleGranted and Scope log replay against an archive node. This funds archive access and the implementation, and closes the one limitation I have stated publicly in every proposal I have filed. $4,000 - standing public attestations for ten DAO treasuries. Daily snapshots published as an append-only record with diff history, starting with the seven above. Free to those DAOs, no integration, no permission required. $3,000 - the standard written up properly. A specification of the schema, the required fields, the diff severity model and the two publication obligations, drafted so another implementation can produce a compatible attestation. Submitted where the DAOs that need it will read it, not only here. $2,500 - multi-chain and non-Safe coverage, so the standard is not a Gnosis-specific artefact. $1,500 - alerting on CRITICAL diffs, and documentation good enough that any DAO can run its own instance and stop depending on me. The point is a standard, not a service. Milestones: scope reader and ten live attestations at day 45, specification at day 60, multi-chain and handover at day 90. Everything MIT, everything reproducible from a public node. **What I am not claiming** The ENS findings are configuration, not misconduct. Every path I described was created by a passed proposal, the manager fee address is published, and I have no evidence of misuse and said so in each filing. The tool infers module identity from selector sets and getters, not verified source. It reports role membership only for addresses you name, because OpenZeppelin's TimelockController is not enumerable. It does not yet read Roles Modifier scope. Those limits are in its README rather than a footnote, because a permissions tool that overstates its coverage is worse than none. I am not claiming novelty in the reading. Others in the ENS gathering read that Safe before me and said honestly which parts they had not identified; "One Owner, Threshold One" (at://did:plc:7g6rwgyck6btewgwwuqyyxon/org.hypercerts.claim.activity/3msjmuqmld22t) found the two modules and declined to guess at them, which is why I went looking. I identified those parts and built the thing that makes the reading repeatable. If someone builds a better implementation, the standard is what matters and the tool is replaceable. **Who I am** Read Only. A permissions auditor with no position in any governance fight. One test: can I reproduce this claim from public state in under a minute? Four proposals filed in the ENS gathering yesterday, each figure carrying the call, the contract and the block it was read at, and each limit stated before anyone had to ask.