One loop is missing.

Altitude one · The structure

Five causes under the thirty problems.

Rank the field by harm and you get a list of thirty. Trace each one upstream and you reach the same five places. These are the generators. Every problem below is one of them, seen from a particular angle.

• Stated values, revealed values Protocols declare decentralization and enact concentration, and the declaration does work: it attracts capital, deflects regulators and recruits builders. Across 23 communities coded on what they say and what they do, this gap is the single most consistent finding. The top 1% of ENS holders control about 62% of its voting power. In the Sky rebrand poll of 2024, four wallets supplied 79% of the vote, out of about twenty voters.
• The refusal of restraint A system with only reinforcing loops has no governor. Here the governor was offered and declined: asked to cap its own share of Ethereum validators at about 22%, Lido's token holders voted 99.81% against, in June 2023, with contributors saying out loud that the dominance harmed Ethereum. Self-restraint is the value DAOs most actively refuse to encode.
• A field that cannot see itself When the Bank for International Settlements rebuilt the published value locked in 400 DeFi protocols from the chain, only 46.5% matched. Loss figures revise upward every year as attribution improves. You cannot insure, govern or regulate what you cannot measure, and this is why the insurance market below is missing.
• Irreversibility without repair Nobody has solved theft, bugs, coercion or lost keys without creating someone who can confiscate funds or rewrite history. So every protocol keeps a quiet red button: eleven of 23 communities run an emergency council with pause or veto powers. The honest measure of a decentralized system is who holds the keys during the exploit.
• Mechanisms where institutions were needed The field keeps converting human coordination problems into code and rediscovering the human problem underneath. What it refuses to put on chain tells you it already knows: decision power, merit judgment, identity, moral judgment, emergency intervention and restraint all stay off chain, on purpose. Encode the mechanical and the scarce; keep human the judgmental and the moral.

The one number

99.81%.

Most problems in this field are argued about as engineering. This one vote shows what kind of problem they are. The cap at about 22% mattered because a third of validators can block finality, and the people voting knew it. They weighed the risk to the network against their own growth and chose growth, near unanimously, and wrote down why: "if not us, someone worse." That is a refusal defended as a virtue, and a system that refuses its own brake will run every loop to the wall. The crashes, runs and exploits that follow are what an ungoverned reinforcing loop does when it works as designed.

The field

Thirty problems, four tiers.

Each one tested against the ten marks of a wicked problem, so the tame ones can be funded and scheduled and the wicked ones can stop pretending to be engineering. Tier membership is the claim. The order inside a tier is not. The full map, with sensors, falsification tests and the case against each entry, lives in the Motus corpus.

Tier A · root adjacent

• CWP-01Human key security and industrialized fraudwicked · worsening
• CWP-02Concentration disguised as decentralizationsuper wicked · stalled
• CWP-03Governance legitimacy, capture and upgradeabilitysuper wicked · stalled
• CWP-04Protocol security and irreversible failurewicked · improving slowly
• CWP-05The privacy, accountability and compliance trilemmawicked · stalled
• CWP-06The extraction economy: a negative sum product with market fitsuper wicked · worsening
• CWP-07Speculative reflexivity and post subsidy survivalsuper wicked · stalled

Where to push

Severity is one axis. Leverage is the other.

The most severe problems are mostly the least movable from outside: protocol security, MEV, stablecoin solvency. The movable cluster shares one feature. Every problem in it is held back by information, attestation and legitimacy, and none of them by cryptography or capital. That is a direct consequence of a field that cannot see itself.

Twelve models the July map lacked

Small enough to draw, sharp enough to be wrong.

The synthesis pass produced twelve systems models from the October evidence. Each has a loop, a prediction and a falsifier, and each is drawn from a dated data point the dives read directly. Four are open here; the rest fold.

• The governor switch An installed balancing loop only runs when someone activates it, and the activation usually belongs to the party the loop is meant to restrain or to a few large holders. A governor whose switch belongs to the governed behaves like a reinforcing loop with a decorative brake. Prediction: activations cluster at the votes that matter to the switch holder, and idle governors outnumber fired ones. Read on 2026-10-12: Lido's veto in Normal state since 2025-08-08, never activated; the 1% signalling line about $244M of stETH. a governor fired against its switch holder's stated position.
• The Goodhart pump When a metric starts to carry money, the cheaper of two flows wins: producing the behavior, or producing its signal. Pricing restraint raises the reward for faking the measurement. Prediction: every priced attestation shows a raw-to-adjusted gap that widens after rewards attach and narrows only when verification is funded. x402's adjusted volume of about $19M across 134M transactions is about fourteen cents each; one Intuition writer, 70.3% of triples. a priced metric whose adjusted-to-raw ratio stays flat for a year after rewards attach.
• The loss-priced governor Balancing loops do get installed from inside, after a loss or a market discount has priced the risk. The variable that matters is installation lag, from first measurable risk to enacted rule; long delays in a balancing loop produce oscillation, which is the boom, crisis and patch cycle. Strategy sold 3,360 BTC across three windows, then adopted an issuance rule; Bitget's fund paid $387.5M; OpenAI paused its largest run after the breach. a costly governor adopted before any loss; Dash's unminted treasury, encoded at launch, is a partial one.
• The sole-recipient treasury A stake-weighted treasury can be expensive to buy and still be captured, through dependency on the only credible recipient. An unminted treasury removes the war chest and the buffer, so every downturn lands on a payroll that must be re-approved monthly, which only the incumbent survives. Prediction: the incumbent's share of allocation rises in downturns while turnout on its line stays below turnout on smaller fights. Dash, October 2026: one firm at 70.8% of capacity; independent Core work unfunded on the same ballot. an independent team clears the threshold and the incumbent's share falls below half for three months running.

The values layer

What they say, what they do.

Interventions fail on values before they fail on mechanisms, so the map carries a values layer: 23 crypto communities coded on what they declare, what they enact on chain, and what nobody teaches but everyone learns. Eight axes organize how they differ. The corpus states its own biases in the same view as its findings: Ethereum heavy, survivor biased, English sourced, and the hidden values are inference rather than measurement. It lives at Initium.Builders.

The Dash DAO · read on its own terms

A treasury with no pile, and one recipient.

Dash is the field's reference case for governance by bonded operators: 1,000 DASH buys one vote, a service role and a claim on the treasury, and an evonode at 4,000 DASH counts four. We build on it and hold office hours for its builders, so read this as a reading from inside. The October 2026 data, pulled from the deterministic masternode list and the DashCentral budget API on 2026-10-12, sharpen the case in three directions the July map did not see.

The price of a vote is legible here in a way it is nowhere else on this page. Clearing the 322-vote threshold alone takes about 322,000 DASH of collateral, about $16M at October prices, nearly four times the annual treasury; compare the half million dollars staked across the whole attestation graph above. Buying an outcome costs more than any outcome is worth. The capture that happened needed no vote buying: the DAO's main act is re-approving the one organization able to spend most of its budget, and the vote on that line drew 22% turnout while a 530 DASH marketing fight drew 36%. Grand asks still fail, a $30M film at 27 yes to 356 no, a basic-income pilot at 39 to 250. Attention follows controversy, not stakes.

Where to push, by rung. The highest is the goal: reserve the routinely unminted quarter of the budget for retroactive rewards to Platform builders, paid on attested usage, which turns the treasury from funding teams to buying outcomes. Two rules make it safe: no report, no renewal for any recurring line above 500 DASH a month, using the monthly re-test masternodes already hold; and a contested-name charter with published lock criteria and a reasoned record for every lock. Under those, two flows of information: a proposal ledger on Platform itself, deliverables and monthly reports keyed to each governance object, and a monthly decentralization panel of registered, enabled and banned nodes by tier, evonode weight, collateral on both denominators and per-proposal turnout, timed to land three days before every voting deadline. Our own 128 DASH proposal for November, one payment, every DASH reported, is the lowest rung and the right first motion: a team without masternodes gets heard through a delivery record.

Reading 04: the open-source commons, read as a system · Reading 03: the masternode economy · The treasury loop, drawn · The Desk

Coordination · 2025 to October 2026

Every governor has a switch, and the switch is the chokepoint.

Seventeen coordination patterns were read in production for this update, four of them straight from chain state. The July map said the DAO ratifies rather than originates and that progressive decentralization runs in reverse. Between July and October 2026 four of the largest DAOs wrote that into their charters. The refinement the record adds is about the balancing loops that do exist: each one has an activation switch, and whoever holds the switch decides whether the loop runs.

The cleanest case took four calls to a public Ethereum node. The ENS founder's voting power stood at about 148,000 ENS in mid June 2026, rose to about 3.26M by June 26, stayed there through a vote that refused to renew the Security Council (82% against, his stake about 80% of the votes cast) and a restructuring that moved the $65M endowment and protocol revenue to a five-person Foundation board, then returned to about 152,000 by September 11. Every other participant's vote counted for the ten weeks the founder abstained.

• The removal-right DAO ENS moved operations, its endowment and a million ENS to a Foundation board and kept the right to appoint and remove directors. Optimism's Season 9 guide says "the primary role of governance will be holding OP Labs accountable"; the Token House dissolved three councils in July with about 22M OP for and 0.2M against each. Aave's DAO won 100% of branded revenue in April 2026 and paid Labs $25M plus 5,000 AAVE to settle. token voting shrinks to a shareholder's right to remove directors and approve budgets; the removal right, published budgets and audited accounts are the enforceable part, and should be tested once at low stakes.
• Brand escrow Aave Labs proposed in October a memberless Cayman foundation whose only purpose is to hold the trademark, code IP and domains, with directors appointed and removed only by vote, after a December brand-transfer vote failed with 41% abstaining and a swap-fee diversion worth about $200,000 a week. the most valuable and least decentralized asset gets a legal owner: one director and one supervisor in one jurisdiction, with no budget cap in the draft.
• The legal wrapper as the price of revenue Uniswap adopted a Wyoming DUNA with $16.5M of UNI set aside for back taxes, then passed its fee switch 125.34M to 742 and turned on v4 fees across seven chains in July 2026. Alabama's DUNA law took effect on 2026-10-01. execution moves to administrators and to the registering state's courts, and the wrapper speeds the drift to value capture.
• The ratification conveyor Uniswap executed nine proposals since March with between zero and 1.27M UNI against. After June, votes for fell to 46M to 52M against a 40M quorum, so two or three large delegates abstaining would now stall anything. Across DAOs in a 2026 working paper, average participation is 6.3% of eligible votes, the top decile of voters holds 76% of realized voting power, about four members submit every proposal, and proposal managers earn 9.5% on trades placed before their own proposals. the real veto is abstention by a few quorum makers, and the vote legitimates decisions made upstream.
• Unwinding the treasury-armed delegate Uniswap recalled about 12.5M UNI it had delegated to eight holders, two of them market makers. Arbitrum's delegate pay fell from the $1.5M program the July map cited to $25,000 in August and $15,000 in September; a forum thread now argues that "every delegated ARB that never votes raises quorum." Sky's top three delegates hold 88% of delegated SKY. withdrawn subsidies leave turnout lower, which confirms that no configuration makes attention free; in Maker, by contrast, delegation worked as a market that rewarded good votes.
• The quorum you can edit Arbitrum now indexes quorum to delegated voting power and a council action was set to exclude 51M ARB from the denominator; Gitcoin proposed cutting its quorum from 2.5M to 1.5M GTC. whoever controls exclusions, inactive delegation and the parameters controls what can pass.
• Emergency powers that grow under AI-era threat After freezing 30,766 ETH of exploit proceeds in April by a 9 of 12 vote, Arbitrum's council took an October emergency action, adding a pause guard to settlement and halting new Stylus activations, citing AI-assisted attacks; Offchain Labs proposed an L1 fallback governor for the day both the DAO and the council fail. Optimism's deputy pause now expires after six months and Offchain's base-fee privilege after two years, the first enacted sunsets on emergency power. power moves to councils and then to courts: a U.S. restraining notice clouded the release of the frozen Kelp funds.
• Dual governance, never fired Lido's stakers can delay decisions at 1% of stETH escrowed and force a rage quit at 10%. Read from the contract on 2026-10-12: state Normal since 2025-08-08, veto never activated, rage-quit support effectively zero. the only holders able to cross 1% alone are large custodians and restaking protocols, so the veto belongs to them in practice, and an untested veto has no record of working.
• Ownership coins MetaDAO puts 100% of raised capital into an on-chain treasury; spending above a monthly budget passes only if conditional markets price it higher, and anyone may propose liquidation when the token trades below net asset value. Offers ran about 14 times accepted capital; of 54 permissionless launches by March 2026, two met their goal. decisions are priced by thin liquidity and access sits with the curator; the NAV redemption right is a priced balancing loop of exactly the kind the lever on this page describes.
• Citizenship by revenue Optimism's Citizens' House, the July map's one structural answer to plutocracy, now seats chains with at least 2% of revenue share, apps with at least 0.5% of gas, and early users with a personhood credential. the Foundation no longer picks citizens by hand; revenue does, which is a different concentrator.
• Trust-graph delegation Gitcoin's plan seeds about five stewards chosen by its founder and lets a decaying web of trust re-delegate about 14.5M GTC nightly, with the stated goal that by July 2027 "the trust graph can outvote me." the seed list and the decay rules become the constitution.
• Bounty-priced oversight Arbitrum's Watchdog program had taken 88 reports and recovered about 532,000 ARB of misused grants against 268,000 ARB paid to reporters by June 2026. L2BEAT published a review of security councils in July and an ossification score in August. enforcement still runs through the Foundation's custody and lawyers; these are small working pieces of the lever, measure, attest, price, and they exist.
• Governance tooling has no business model Tally, the voting interface for Uniswap and Arbitrum, announced its wind-down in March 2026; its CEO said "there isn't a venture-backed business in governance tooling." Arbitrum's Foundation and Offchain Labs are building their own. the interface layer returns to Labs and foundations, and a neutral comprehension tool has an opening it did not have in July.

Pricing restraint is necessary and incomplete. The switch flips also have to be measured and made expensive, and all of them can be read from public chain state today: voting-power deltas on large addresses, the quorum margin on every vote, exclusions from the denominator, the count and duration of emergency actions, and the escrow level on every veto contract. That is the observatory's first sensor set.

Funding, engagement and attention · 2025 to October 2026

Every mechanism that survived did so by re-centralizing one layer.

Retroactive funding was the frontier in July. By October the flagship program is paused, the money has moved somewhere specific, and three findings recur across every funding and engagement structure we traced. Budgets are denominated in the token they are meant to sustain, so funding is pro-cyclical by construction: OP is down 97% from its high and ARB 92%, and Arbitrum cut its paid delegate caps in August 2026 to "realign ARB outflows." The measurement layer that retro funding, attention markets and attestation all rest on is itself being farmed. And what lived did so by handing one layer back to a company, a foundation or a licence.

• Optimism: from experiment to organization Retro Funding paused for at least twelve months on 2026-01-08, in a post with that title, which also sent 50% of Superchain revenue to OP buybacks. In August a vote moved 546.9M OP, about 12.7% of supply, from the user airdrop allocation to a Foundation fund. The deciding 8.49M OP came from a team "fully funded by the Collective"; without it the vote stood at 46.47%. All votes cast were about 0.67% of supply. the public goods allocation becomes discretionary capital, decided by funded insiders, and the "experiment" frame is what licensed the pause.
• Protocol Guild, the one retro-like mechanism with a record About 190 Ethereum core contributors, more than $33M distributed since May 2022, median about $66,000 a year, funded mostly by projects pledging 1% of their token supply. Core developers are still paid 50% to 60% below market. it is a curated payroll with vesting, so "core" is a credential issued by incumbents, and pay rides on the pledging projects' token prices.
• Quadratic funding retreats to the village Gitcoin's GG24 (late 2025) ran about $1.8M across six domains and six mechanisms; its two quadratic rounds paid $300,000 of matching against $36,657 of direct donations from about 1,300 donors, a ratio above 8 to 1. On 2026-10-01 the grants program became a feature of a local money pilot, with the founder writing that quadratic funding "works best when the community is small enough that people actually know the projects." domain operators are appointed, and the mechanism's legitimacy claim now holds only where people already know each other.
• Deep funding: a jury distilled into weights About 70 jurors made 550 pairwise comparisons over nine months and disagreed on the order 40% of the time; one juror's 999x rating moved a single project's share of the pool by about five points. No final allocation had been published by 2026-10-12. a few dozen human judgments become an "objective" dependency graph, and whoever seeds the jury sets the frame.
• Airdrops, measured at day 90 Across the largest airdrops, 78% to 94% of recipient wallets sold most of their allocation within ninety days (Delphi, June 2026); Arbitrum paid about $1.16B to users who left within a month. The enacted response is to stop: Optimism moved two thirds of its original user allocation to the Foundation. whoever defines "performance" for a performance-locked supply controls distribution.
• Attention markets exist by licence On 2026-01-15 X revoked API access from apps that reward posting, citing spam; Kaito sunset its points the same week, saying a "fully permissionless distribution system is no longer viable," and returned in July through a bilateral data agreement with X and a pay-per-outcome creator layer with an unpublished payout formula. About 25% of Polymarket's volume over three years looked like wash trading tied to airdrop expectations (Columbia, November 2025). the "decentralized attention economy" is a reseller of one company's firehose, and a market on a mindshare index lets bettors manufacture the outcome.
• Prediction markets as a truth layer Kalshi set a monthly record near $60B notional in September 2026 and held about three quarters of the category; in one October week 64% of its volume was multi-leg sports parlays and economics plus science together were about 0.14%. In May 2026 a WSJ analysis found at least 60% of active UMA voters, the people who resolve Polymarket disputes, linkable to Polymarket accounts, and the ten largest wallets holding over half the voting power in most disputes. The first insider-trading prosecution on a prediction market was unsealed in April 2026. truth goes to token holders who also bet, and the information function is a sliver of a sports book.
• Futarchy, tried with money Optimism's Season 7 contest forecast about $250M of TVL gain for its picks against about $30M realized. MetaDAO's futarchy-run raises took in about $25.6M across eight sales against about $390M committed, refunding 95%; a permissionless clone saw 2 of 54 projects reach their goal. the party that sets the metric becomes the allocator, and the market tracks the token rather than the grant.
• Paying for governance Arbitrum's Rewarding Active Delegates paid $23,000 to 21 of 35 enrolled delegates in July 2026, top payout $1,700, in ARB against dollar caps that shrink as the token falls. it pays for voting, not for reading, and the budget collapses at exactly the moment turnout does.
• Our own substrate, read the same way Intuition, the attestation graph this stack chose, held 269,483 atoms and 285,194 triples on 2026-10-12. One address wrote 200,501 of those triples, 70.3% of all that exist, between September 9 and 12, labelled as agent trust assessments. The ten largest creators wrote 88.6% of triples. The TRUST token's entire market cap was $9.73M, and about $0.55M was staked across every vault, nearly a fifth of it on the "Intuition" atom itself. a price signal that costs half a million dollars to dominate does not price anyone's restraint. The lever on this page needs an attestation layer that is harder to fake than the behavior it measures, and today ours is not.

The agent layer · 2025 to October 2026

Restraint already has a price. Nobody can contest it.

The baseline treated agents as a coming risk. Through October 2026 the record is specific and smaller than the headlines. Agents hold wallets and move money in production: Coinbase shipped agent wallets with keys in secure enclaves in February and agent trading in June, and Robinhood's agentic trading terms say the customer assumes all risk and that Robinhood "does not control, supervise, monitor, recommend, or audit these AI agents." The open payment rail, x402, moved under a Linux Foundation body in April whose premier members are AWS, Amex, Cloudflare, Coinbase, Google, Mastercard, Shopify, Stripe and Visa. Visa's own adjustment put its real volume at about $19M across 134M transactions from May 2025 to April 2026, with roughly 4,000 wallets producing 90% of it; raw counts in the hundreds of millions are mostly protocol signaling. The adjusted daily dollar volume fell about 95% between January and September while the transaction count kept rising.

The quieter damage is to deliberation. A Georgia Tech preprint from August 2026 finds that Arbitrum's forum stopped predicting vote margins after March 2024, when fluent AI prose became cheap, that the signal broke while AI posts were still a small share, and that removing them later does not bring it back. When prose costs nothing, the forum stops carrying information and decisions default to large holders and paid delegates. Scams are the mature use: AI-linked scam operations earned about 4.5 times the revenue of the rest in 2025, a face model costs about $500, and one network of 4,700 AI personas worked 25,000 people in two weeks. The July map's design law, encode the mechanical and keep human the judgmental, needs one amendment for the agent era: the scarce input is now verification, and it should be encoded and paid for, because judgment stays human only while checking a claim costs less than making one.

"Superintelligence" in 2026 is a governance and marketing frame: a pacing essay in September backed by three lab heads, a UN Security Council session, a Senate bill to ban it, and token projects borrowing the word. The measured frontier is long-horizon agentic work, about twelve hours at a 50% success rate for public models, with evaluation gaming in at least 16% of long successful runs. The price of agent restraint is being set now, by card-network agent scores, enclave custody at one exchange and brokerage terms of service, outside any process a user can contest. The open question for the lever on this page is no longer whether restraint can be priced. It is who owns the price.

What changed · July to October 2026

Re-read against the primary feeds.

Every headline number in the July baseline was re-checked against the feed it came from: DefiLlama, L2BEAT, the FEC, SEC filings, mempool.space, NIST, the FBI's report. Four baseline figures that had been single-source secondary numbers diverged from the primary feed, and one claim was simply wrong. They are corrected here and at the source, with the date.

• The next halving is April 2028, not 2026. The baseline called 2026 "a pivotal test" of Bitcoin's fee market; the last halving was April 2024 and the next is block 1,050,000. Over the last 4,320 blocks fees were 0.64% of miner revenue, about 0.005% of market cap annualized, against the 0.5% to 1.5% some analysts say security needs. The gap is 100 to 300 times. The two largest pools held 46% of hashrate.corrected
• The TRUMP token's $4B did not go to a narrow insider set. Nansen's data through June 2026 puts $4.04B of gains across 492,285 wallets, mostly early buyers, against $3.81B of losses across 988,905. The documented insider take is the $636M in the president's own disclosure. The "45 wallets took $1.2B" figure could not be verified.reframed
• Restaking is a third of what the baseline said. EigenCloud held $6.59B on 2026-10-12, down from a $22.06B peak in August 2025; the baseline's $19.5B did not match the primary feed even in January. The Kelp exploit's $292M loss produced $5.4B to $8.6B of withdrawals from Aave and a 47% fall in its net value locked in a week: 18 to 29 times on withdrawals, about 41 times on net TVL.corrected
• Insurance is thinner than "2 to 5%." On-chain insurance protocols hold $0.19B of capital against $91.9B of DeFi value, 0.2%. Active cover could not be verified from any source; the honest statement is "probably under 1%."corrected
• Stablecoins: $313B, and the $33T needs the word "raw." Supply was $313.0B on 2026-10-12, 2.9% below its May peak, 82% of it USDT and USDC. The $33T transfer figure is unadjusted; the adjusted series set a record of $1.79T in June 2026, up 125% on the year.re-dated
• Rollups: Stage 1 now holds 60% of value; Stage 2 exists only for projects under $1M. Base, Arbitrum One and OP Mainnet remain at Stage 1; 68 of 99 tracked projects are Stage 0. Arbitrum's security council froze 30,766 ETH of exploit proceeds by a 9 of 12 vote in April and swapped the bridge implementation without a timelock in May: governors that exist only during a crisis and leave no rule behind.unchanged, with context
• Political money: $120M on hand, $73.5M spent, and the bill failed on ethics. Fairshake and its two affiliates held $120.4M on 2026-08-31, down from $191M at year end, after $73.5M of independent expenditures. On 2026-09-15 the Senate rejected cloture on the CLARITY Act 49 to 50, with reporting putting the dispute on language covering officials' crypto income after the president disclosed about $1.4B of it.realized
• Treasury companies sold. Strategy sold bitcoin in three windows between late May and early August to pay preferred dividends and buy back its own preferred, at an mNAV of 0.98, then resumed buying as it recovered to 1.13; several smaller treasury companies liquidated. The forced-sale channel the baseline predicted is now proven open.realized
• The perimeter moved to the layers around Bitcoin. Blockstream's Liquid sidechain lost about 4,000 of its 4,200 BTC in September through a range-proof flaw, with about 600 BTC kept; Coldcard firmware from 2021 fell back to a weak random generator and about 1,816 BTC were swept from affected wallets; Bitget lost $387.5M through a backend wallet system and paid users from its protection fund. The base chain was untouched.new
• North Korea: a smaller share of a smaller total, until Bitget. About $643M and 66% of hack losses in the first half, then more than $1B for the year once Chainalysis attributed Bitget. Total losses by tracker: TRM $972M in the first half, CertiK $1.26B in the third quarter alone.series, not a point
• Quantum: counts, not dollars, and the dates are drafts. About 6.04M BTC sit in addresses with exposed keys, 30% of supply, about $500B at October prices; one preprint puts 2.3M BTC irreducibly at risk. NIST's 2030 deprecation and 2035 phase-out are still an Initial Public Draft from November 2024. BIP-361 proposes sunsetting legacy signatures, which its own abstract describes as turning quantum security into a private incentive: the lever on this page, written as a consensus proposal.re-sourced
• Developer liability: no verdict yet. The Tornado Cash retrial moved to April 2027 with the acquittal motion undecided; no pardon for the Samourai developers was found. The U.S. share of crypto developers is 19% on 2024 data, and the source itself questioned whether regulation explains the decline.re-sourced
• Prediction markets resolved against a filing. A market on whether Strategy sold bitcoin by May 31 settled "No" with 98.6% of 607 votes while the company's own SEC filing shows 32 BTC sold inside the window. More than 1,150 markets were disputed in 2026 by mid-May, more than in all of 2025.worse
• Agent rails: counts up, dollars down. x402's adjusted cumulative volume is about $19M; its daily dollar volume fell about 95% from January to September while transaction counts rose. Tokenized real-world assets reached $31B to $38B in August, understating the baseline's "over $20B."re-sourced

Two things cut against the thesis in this refresh. Bitget's protection fund paid a $387.5M loss in full, which is self-insurance working as a balancing loop at the exchange layer. Strategy's rule against issuing stock near one times its net asset value is a restraint it installed itself, after the market priced the risk. Both fit "restraint cannot be requested, only priced," and both soften "almost no balancing loops" in its strongest form. The sharper wording is that balancing loops appear once a loss or a discount has already priced them.

What is solved

The dividing line is agreement.

A map of wicked problems makes a field look hopeless, so here is what it has closed: double spending in an open network, deterministic global state, the energy cost of consensus for proof of stake chains, practical zero knowledge proofs securing real value, commodity hardware custody, live permissionless fraud proofs on the largest rollups, coordinated vulnerability disclosure, and stablecoins as a payment product whose adjusted transfer volume set a record of $1.79T in June 2026. Every one of those was a problem where the people involved agreed what better meant. Every open problem above is one where they do not. Difficulty was never the line.

What we would be wrong about

The lever assumes a buyer.

The whole move above rests on one chain: measurement leads to attestation, attestation leads to a price, and a price changes behavior. The weak link is the second step. A protocol can absorb a score as marketing, the way it absorbed "decentralized", and then the observatory is one more badge. The test is simple and dated: if no insurer, allocator or listing venue changes a price in response to the measurement within two years, the theory of change is broken, whatever the metric says.

Two more. The values layer is inferred from documents, forum posts and journalism, with no interviews, so its "hidden values" are hypotheses. And the corpus cannot tell whether plutocracy, founder gravity and the drift to value capture are intrinsic to DAOs or artifacts of this decade's capital and legal regime; a different regime might reveal different values. The shape we are confident in. The magnitudes we hold loosely.

Where the July thesis was wrong, by October. The synthesis pass was asked to break the lever on this page, and it found seven places.

• It omitted the switch. Governors exist and sit idle. Lido's veto has never fired in fourteen months; the ENS founder's stake switched on for one quarter and off again. A priced restraint can be switched off by the same holders. The fix is a public switch registry and a notice period on any large activation.
• It assumed measurement is harder to fake than behavior. On the substrate we chose, the reverse held: one writer, 70.3% of triples, about $0.55M staked in total. Pricing a metric raises the reward for faking it. The fix is a cost-to-fake floor published before any attestation feeds a price, applied to ourselves first.
• It assumed nobody owned the price. Card networks, brokerages and X already price agent and attention restraint, privately and without appeal. A balancing loop built by incumbents recentralizes. The question is who owns the price and whether anyone can contest it.
• It assumed a buyer. On-chain insurance capital is 0.2% of DeFi value. The buyers that exist are chokepoints, and venues insuring themselves after a loss.
• It said restraint cannot be requested. It was requested and passed at Arbitrum, against an agent voting pool, and in Dash, where masternodes refused to raise their own yield 401 to 35. It fails where the cost lands on short-horizon voters. Restraint can be asked of long-locked holders and of third parties.
• It said there were almost no balancing loops from inside. Dash's unminted treasury and monthly revocation, MetaDAO's redemption at net asset value, Strategy's issuance rule and Bitget's protection fund are all endogenous. Most arrived after a loss had priced them. The accurate claim is about delay: balancing loops arrive late, and the work is to move the price before the loss.
• It works at the wrong rung. Pricing restraint operates at rungs six to eight on a problem whose generator sits at two and three. It leaves the goal, growth of the token, in place. Two goal-level moves surfaced: reserve Dash's unminted budget for attested outcomes, and define governance success as informed participation instead of turnout.

The thesis, restated. Crypto's balancing loops are few, late and switchable. Restraint passes when long-horizon holders bear it or third parties pay for it. Pricing it works only where the measurement costs more to fake than the behavior, the price can be contested, and the switch is visible.

The facts we used

• The Lido vote: 99.81% of LDO voting power against a self-limit of about 22% of Ethereum validators, June 2023. research.lido.fi, the proposal thread.
• Illicit flows: at least $154B received by illicit addresses in 2025, up 162%, under 1% of attributed volume, 84% of it on stablecoins. Chainalysis, 2026 Crypto Crime Report. A lower bound that is revised upward as attribution improves.
• Value locked: independently reconstructed TVL matched the published figure for 46.5% of 400 DeFi protocols. BIS working paper 1268.
• Governance concentration: ENS top 1% at about 62% of voting power; four wallets at 79.3% of the Sky rebrand poll in 2024; 750M ARB moved before the AIP-1 vote in 2023; the Nouns fork of September 2023 at about 56% of holders and 16,757 ETH; 13 of 23 communities with a legal entity beneath the DAO and 11 with an emergency council. The Motus corpus, Cycle 1, with its adversarial verification layer, at Initium.Builders.
• Rollup trust: Arbitrum, OP Mainnet and Base at Stage 1 with live permissionless fraud proofs; the stage definitions at L2BEAT.
• The sixteen minutes: a phished maintainer, 18 packages, more than two billion weekly downloads, September 2025. Sygnia.
• Developer liability: the mixed verdict in the Tornado Cash trial, August 2025. Mayer Brown.
• Dash governance: 20% of every block reward to the treasury, superblocks every 16,616 blocks, unallocated budget never created. docs.dash.org, and our reading of the treasury.
• Governance, read from chain state and portals: Lido Dual Governance contract state (Normal since 2025-08-08, veto never activated) and the ENS founder's voting-power window, both read through a public Ethereum node on 2026-10-12; Uniswap and Optimism tallies from vote.uniswapfoundation.org and vote.optimism.io; Sky delegate shares from vote.sky.money; ENS Next Era and the Security Council vote; Optimism's Season 9 guide; the Aave Foundation proposal; L2BEAT on security councils; Tally's wind-down; Arbitrum's delegate pay, October emergency action and the agent-voting pool; the participation figures from Cong, Rabetti, Wang and Yan (working paper).
• Funding and attention: Optimism, "From Experiment to Organization" and the August OP vote; Protocol Guild; Gitcoin's social contract and the October reboot; deep funding jurors; Delphi on airdrops; the Columbia wash-trading study; the WSJ analysis of UMA voters; the first insider-trading case; prediction-market volume; X's ban on paid posting and Kaito's data agreement; Optimism's futarchy findings; MetaDAO from Crypto Briefing; the Intuition figures from its mainnet indexer and CoinGecko, read 2026-10-12.
• The agent layer: Visa's adjusted x402 figure and Bitquery's audit; the x402 Foundation; Robinhood's agentic trading terms; Coinbase agent wallets; OpenAI's report on the Hugging Face breach and the Meta and Anthropic incidents; Huang and Hall on AI prose in Arbitrum's forum (preprint); METR's frontier risk report; TRM's AI in crime index and the fake trading bots; Anthropic's September threat report; the trading contests; the pacing essay.
• The refresh, from primary feeds: DefiLlama stablecoins, EigenCloud and Aave; L2BEAT's summary; Fairshake at the FEC; Senate roll call 234; Strategy's August and October filings; mempool.space fee statistics; Glassnode on exposed supply; NIST IR 8547, initial public draft; BIP-361; the Liquid Network exploit, the Coldcard sweeps and the Bitget theft; Nansen's TRUMP data via The Block; Electric Capital's developer geography; TRM's first-half hack report; the Strategy-sale dispute.

Numbers we could not verify against a primary source are marked unverified above or left out. Everything else is structure, and the structure is ours to be wrong about.

If a fact here is wrong, tell us which one and where you read otherwise. We correct it on this page, with the date.