Some of what this API publishes is about **where a subnet's money goes**: how much emission the owner receives, whether an owner-run validator keeps what it earns, whether a treasury allocation matches what a subnet says it takes. Those are the most useful figures here and the easiest to get wrong in a way that harms someone. A wrong revenue number is an error. **"This team is quietly taking 60%" is not retractable once an agent has quoted it.** This page is the standard we hold ourselves to. It is published rather than internal so you can check us against it, and so a subnet operator who thinks we have it wrong knows exactly which rule to point at. ## 1. A take rate does not tell you what an operator keeps [#1-a-take-rate-does-not-tell-you-what-an-operator-keeps] What a validator operator actually retains is: ``` kept = take × dividends + (1 − take) × dividends × (operator_stake ÷ total_stake_on_hotkey) ``` A validator with an 18% take and no self-stake keeps **18%**. The same validator, fully self-staked, keeps **100%**. The take is published on chain and tells you almost nothing on its own — the self-stake fraction is the decisive term. So we do not report "the owner takes X%" from a take rate. Take-rate-only analysis is worse than none, because it looks rigorous. ## 2. A large nominator is not evidence of anything [#2-a-large-nominator-is-not-evidence-of-anything] Concentrated delegation behind an owner-run validator has at least four innocent explanations before "hidden team wallet": * a **custodial exchange** staking on behalf of many users * a **delegation service** * a single **unaffiliated whale** * a **DAO or treasury** with no relationship to the subnet Every one produces the identical on-chain shape. The default verdict for an unlabelled coldkey is `unresolved` — never `owner`. ## 3. Affiliation needs linkable evidence, not correlation [#3-affiliation-needs-linkable-evidence-not-correlation] A coldkey moves above `unresolved` only on evidence you can check: | Evidence | What it is | | ---------------- | ---------------------------------------------------------------------------------------------------- | | `funding-path` | a transfer from the owner coldkey, cited to the extrinsic | | `chain-identity` | an on-chain identity naming the team | | `self-declared` | the subnet's own docs, repo or site — pinned to a commit SHA, because a branch moves under the claim | | `key-rotation` | a coldkey swap or hotkey rotation linking the two | | `known-entity` | a labelled third party, which supports `third-party` rather than `affiliated` | Timing correlation, similar stake sizes and "registered in the same block" are **reviewer hints, not publishable evidence**. They are exactly the reasoning that turns a coincidence into an allegation. This is enforced in the schema, not by convention: a verdict above `unresolved` **cannot be serialised without an evidence entry**, and an evidence entry without a `source_url` or an `extrinsic_hash` is refused. `owner` is the one exemption, because it comes from `SubtensorModule.SubnetOwner` and the chain read is itself the citation. ## 4. Disclosed is not dishonest [#4-disclosed-is-not-dishonest] A treasury allocation written into a subnet's own public repository is a **disclosed business model**, not a deception. Publishing it as a scandal would be a misrepresentation in the other direction. The signal worth serving is **divergence between what a subnet declares and what the chain shows** — and the honest headline for most subnets is *"declared take matches observed take."* If a surface cannot produce that boring answer, it is not a measurement. ## 5. Right of reply [#5-right-of-reply] `operator-attested` is already a first-class provenance class here: displayed, and never summed into a headline. Subnet teams get the same deal on wallets. Declare your affiliated coldkeys and your treasury policy and we publish the declaration beside the measurement, showing where the two agree. A team that declares gets a cleaner card. A team that stays silent gets `unresolved` — **a neutral fact, not an allegation.** ## 6. We publish the mechanism, never the motive [#6-we-publish-the-mechanism-never-the-motive] | We do not say | We say | | ------------------------------------------ | -------------------------------------------------------------------------------------------- | | "The owner secretly takes 60% of emission" | "60% of emission is attributed to UIDs held by the owner coldkey and one unresolved coldkey" | | "This subnet is a scam" | "98% of registered miner UIDs recorded zero emission over the last 31 days" | | "The team dumped tokens" | "44,186 TAO unstaked and 88,284 staked by the owner coldkey over 30 days; net accumulating" | | "Hidden 10% tax on miners" | "A 10% treasury allocation is present in the subnet's published source at \" | Each pair is the same underlying number. The right-hand column is defensible, sourceable, and more useful to someone deciding where to put capital. ## Two things these surfaces cannot see [#two-things-these-surfaces-cannot-see] **Root delegation.** Root (netuid 0) stake is TAO-denominated with a `tao_weight` of 0.18, and root stakers earn dividends in alpha. It is a capture path that is **invisible on a subnet's own metagraph** — a subnet-scoped measurement cannot see it, and says so rather than implying its absence. **Per-subnet take.** `take` is stored globally per hotkey, not per subnet, so a validator operating on six subnets reports the same take on all six. A `null` take means no `Delegates` entry — **not 0%**. Childkey takes dilute it further, are read live, and have no history at all. ## If you think we have this wrong [#if-you-think-we-have-this-wrong] Open an issue naming the rule and the row. A claim that cannot survive that conversation should not have been published, and we would rather correct it than defend it.