ATOM Tokenomics: Phase 1 Outcomes

[mariashaikh]

Summary

Quick intro: I’m a Data Scientist on the research team at Gauntlet and contributed to the development of this report. We appreciate the depth of the review - it engages seriously with the analysis, and on several points it restates limitations the report already documents, which we take as confirmation that the methodology was framed honestly. We address every claim below, but want to flag one framing correction up front, because it recurs throughout the review: the report never claimed to prove executed spot sales. It measures likelihood-weighted, on-chain routing to exchange-associated addresses, explicitly labeled as a directional indicator of market-facing pressure. Read against that stated scope, most of the review’s stronger conclusions and the report’s own conclusions are closer than the review suggests.

Points we agree with

• Two rendered addresses were malformed. The review is correct that the published document contained two typographical corruptions: a z for 2 in the Bybit Reserves 68 address, and an uppercase O for 0 in one HNW address. These were display-layer transcription errors that never touched the underlying analysis, and both are now corrected - we are grateful for the catch.
• A transfer to an exchange is not proof of an executed sale. We agree, and the report states this directly - in the methodology (“a directional, behaviorally grounded sell-pressure indicator”), in the net-flow section, and in the limitations (“All sell-pressure figures are attribution estimates of on-chain routing, not direct observation of executed off-chain market sells”). The review’s list of alternative explanations (consolidation, custody, market-making, OTC settlement) is the same set of caveats the report already carries. This is not a point of disagreement.
• The other_in stream is large and material. The report does not hide this - it devotes a dedicated section (“The central judgment call: the 141.4M other_in stream”) to exactly the point the review raises, quantifies it at 141.4M ATOM, shows a sensitivity table of which wallets would flip if it were counted, and explicitly recommends spot-checking those senders before treating any single wallet as a distributor. We agree it is the single most important interpretive caveat in the net-flow work, which is why it was surfaced prominently.

Where the data resolves the question

1. The malformed addresses did not affect the analysis - the model ran on the correct, checksum-valid addresses. This is verifiable and we have verified it. The corruption existed only in the rendered document, not in the pipeline. Scanning all 110 source CSVs in the analysis run:

   Address	Correct form in source	Malformed form in source
   Bybit Reserves 68 (cosmos17kvae2…)	appears in 66 files	appears in 0 files
   HNW wallet (cosmos1jr9…0…)	appears in 64 files	appears in 0 files

   The pipeline ingested, matched, and computed on the valid addresses throughout. No weighted-sell figure, tier total, or ranking is affected by the display-layer typos.

   This also removes the review’s proposed mechanism for label leakage. The review hypothesizes that malformed addresses prevented entity labels from matching, allowing exchange infrastructure to leak into unclassified cohorts. Because the corrupted strings never existed in the source data, matching occurred against the valid addresses - so the typos cannot be the mechanism. Whether a given wallet is correctly classified is a separate question, one that turns on the completeness of the label set rather than on the rendering errors; we address it directly in Section 3.

2. Aggregate net-flow figures reproduce exactly from source. Independent re-aggregation of the net-flow dataset returns the reported figures to the digit: total out 27,124,539 ATOM, total in (counted) 19,560,883 ATOM, aggregate net 7,563,655 ATOM, split 25 net sellers / 19 net accumulators, and other_in totaling 141,431,329 ATOM. The arithmetic underlying the headline is sound and checkable.

Where the disagreement is interpretive

1. The flagged wallets were unlabeled after a deliberate attribution process, not missed. Entity labeling was part of our mandate, and we treated it as such: we attributed every address the available evidence would support, combining a curated label set with public attribution and on-chain analysis, and we deliberately held unconfirmed addresses as “unclassified” rather than assigning an operator we could not stand behind. The two wallets the review names (cosmos1jr9… and cosmos13dd…) fall in that unconfirmed set - they carry no exchange attribution we could verify to that standard, so we labeled them by balance tier and marked them unclassified. That is a conservative labeling decision, not a label that failed to attach.

   What the review adds is a second attribution source (Range’s clustering) that reaches a different call on these specific wallets. We regard that as genuinely useful, and exactly the input that sharpens attribution over time. It does not, on its own, override our labeling: entity clustering is itself probabilistic (“funded by a Bybit-linked wallet” is strong evidence toward exchange operation, not proof of it), which is why we set a high bar before tagging an address. The right move is to reconcile the two sources wallet-by-wallet and promote any address that clears the bar into a confirmed exchange label, and we would welcome doing that. This is precisely the retail-adjusted, exchange-segmented monitoring view the report already recommends building (see the Upbit #10 caveat in Study 1 and the CEX-dashboard design in Section 4).

2. On the eight “largest sellers” attribution. This does not overturn the report, for two reasons. First, the report already flagged the router pattern on these wallets: seven of the eight appear in its other_in sensitivity table, with the text warning that a wallet sending ~600K to a CEX while receiving ~11M from unlabeled sources is likely a router, not a seller. The review is restating our own caveat. Second, on the two wallets it leads with (its highest-confidence, highest-value calls), the data does not support the claim: cosmos1djc… (“Coinbase consolidation, high confidence”) has just 34,480 ATOM of unlabeled inbound against 1.49M out, so the consolidation read is the weakest of the eight, not the strongest; and cosmos1u7h… (1.70M, the review’s largest) has inbound roughly equal to outflow, consistent with either reading. Its evidence in both cases is upstream-funding clustering, which is directional, not proof of exchange operation. Either way the finding holds: sell-routing is concentrated in a few high-throughput addresses, whether holders or exchange rails, and that is the decision-relevant result. Confirmed exchange wallets belong in the exchange-segmented view the report already recommends, reported as a holder-only and a total figure side by side.

3. On CEX-to-CEX movement. The review’s underlying principle is correct, and we agree with it: ATOM moving between an exchange’s own wallets is internal plumbing, not a holder selling, and it should not be counted as market-facing pressure. Its example is cosmos1djc…, which the review reads as a Coinbase consolidation wallet, receiving from the cex-cosmoshub family and sending back to it.

   In our data, that wallet’s counted inflow from labeled exchange wallets (cex_in) is 61,906 ATOM, and its unlabeled inflow (other_in) is 34,480 ATOM, against 1.49M sent out. There are two ways to read that:

   • If the review is inferring large exchange inflow from upstream clustering, the data does not bear it out: this wallet’s confirmed exchange-sourced inflow is under 62K, and a consolidation wallet is defined by the large exchange inflow it forwards onward. On this reading, the “high-confidence Coinbase consolidation” attribution is the weakest of the eight, not the strongest.
   • If instead the inflows the review points to (cex-cosmoshub-10 and 11) are wallets our label set did not tag as exchange, then those transfers landed in our other_in bucket rather than cex_in. In that case the review is not exposing a scoring error but a label-completeness gap, the same one we address above in point 1.

4. On the data the analysis doesn’t include. The review lists eight categories the report omits. It is worth being clear about why those items are absent: six of the eight are not available to us, to the reviewer, or to any on-chain analyst. They are off-chain data that live inside exchanges and private venues and are not observable on-chain by anyone. The reviewer’s own tracing is entirely on-chain and runs into the same wall. Listing them as gaps describes the boundary of on-chain analysis in general, not a shortcoming specific to this report, which is exactly why the report states plainly that its figures are estimates of on-chain routing, not observations of executed off-chain sales.

5. The report already distinguishes routing concentration from beneficial-seller identification. The review’s core concern - that concentrated routing to exchanges is not the same as a small number of beneficial holders selling - is a distinction the report itself draws. It classifies wallets into likely distributors, potential routers, and accumulators; flags the router pattern explicitly and states that address-level interpretation should not label every positive-net wallet as discretionary selling until other_in senders are checked. The concentration finding - that observable ATOM sell-routing is dominated by a small set of exchange-bound rails - is one the review also endorses. On this, the report and the review agree.

6. On Prop 848. The report labels the 10.25% figure as a share of the event’s top-95% liquid cohort - the metric name in every table is “% of token supply” computed on that cohort, and the methodology note states this explicitly and flags that the cohort/supply equivalence is looser for the older 2022 events. The report also states plainly that a 7-day window establishes repositioning around the event, and frames all event figures as attribution estimates rather than proof that every transfer was a discretionary sale caused by the proposal. The finding - that a binding governance outcome produced the largest normalized reaction in the set - stands on the cross-event ranking, which is robust to the cohort-vs-supply denominator because it is applied consistently across all seven events.

What we are not claiming

To be precise about scope, so there is no ambiguity:

• We do not claim to observe executed spot sales, execution prices, or realized quantities. On-chain routing is the best available proxy; it is not exchange-internal ground truth.
• We do not claim to fully explain ATOM’s market price. Price is set by order-book imbalance and liquidity, which are off-chain and outside the dataset. The report addresses supply-side, on-chain, market-facing flow - one input to price, not the whole.

What the analysis does establish

• Observable ATOM sell-routing is heavily concentrated in centralized-exchange rails - a finding the review independently endorses.
• The concentration finding survives the entire critique. The review disputes what the top wallets are - it does not dispute that a small number of addresses carry the large majority of observable sell-routing. Whether they are large holders or exchange infrastructure, monitoring the same short list captures the same flow - so the operational conclusion stands untouched.
• Inflation is not equivalent to selling: most withdrawn rewards are re-staked or held, while Cosmos’s liquid-reward footprint per unit of supply is structurally larger than peers’.
• Governance finality, not market-wide shocks, produced the largest normalized event reaction in the observed set.

These conclusions are consistent with the review’s own defensible conclusion. The remaining distance is largely about how prominently to foreground caveats the report already contains, and about reconciling two entity-label sources which we are glad to take up.

On the review’s two verdict statements specifically, to leave no ambiguity: we did not claim to have identified 44 independent sellers - the 44 are the top-outflow addresses, some of which are plausibly exchange-operated, which is why the report segments them into distributors, routers, and accumulators rather than presenting them as 44 discrete holders. And we did not claim that 7.56M ATOM was executed as spot sales - 7.56M is counted net on-chain outflow, which the report explicitly frames as an upper bound on distribution, not realized selling. On both points the review restates limits the report already states, and we are glad to make that framing even more prominent in the next revision.