The Iran deal signed Wednesday collapsed before its own ceremony, with Geneva postponed and the ceasefire it was meant to stabilize fraying on the Lebanese border. Strategy's perpetual-preferred stack crosses $10 billion into a hawkish rate regime that raises its funding cost at the worst moment, and enterprises are defecting from commercial AI to open-source models faster than any pricing-power thesis assumed. The thread under the day is which of these changes are one-way doors: a signed deal can evaporate in 48 hours, but the cremation wave hollowing out deathcare, the totaled-car economy rewiring auto insurance, and the migration off commercial AI are structural ratchets that do not swing back.
Crypto data provided by CoinGecko
A counter-consensus trade is forming against the hawkish Fed, and the evidence is not in equities but in the credit plumbing underneath them. Tavi Costa went long gold at $4,200. Andy Constan simultaneously bought TLT and GLD, a pair that only profits together if rates fall while inflation stays elevated, exactly what you would bet if you believed the dots are the hawk peak. The harder data is Jeff Snider's read on private credit: new issuance contracted roughly 40% in a single quarter, the kind of credit-supply collapse that historically leads rate cuts by two to three quarters. Credit markets are already doing the Fed's tightening for it. Today's Take traces where that same contraction surfaces next. The counter-case is Thursday's hot retail and manufacturing data: if private credit issuance declines again next quarter while retail stays strong, the credit market is signaling something the consumer has not felt yet.
The dollar at a 13-month high is not strength; it is stress, and the distinction matters for everything priced in dollars. The consensus reads the dollar's 13-month high as confirmation that the hawkish Fed repriced the currency correctly. Snider's contrarian lens inverts it: the dollar strengthens when global collateral chains tighten, because overseas borrowers must buy dollars to service existing obligations, not because the US economy is winning. If the DXY rise were growth-driven, you would expect US equity outperformance against global peers. Instead, European equities are keeping pace and the Russell 2000, the most dollar-sensitive US index, is the weakest major benchmark since the rate hold. A strengthening dollar that does not pull US risk assets higher is a funding-market signal, not a growth signal. Watch the DXY alongside Russell 2000 relative performance through July: if both rise together, it is growth; if the dollar rises while small-caps lag, it is funding stress masquerading as confidence.
European equities rotated sharply from luxury into energy and healthcare on Thursday, a defensive shift that does not match the buy-the-dip narrative running in US markets. The STOXX 50 closed up 0.4% in a session where the sector leadership told a different story from the index return: LVMH and Kering fell while Shell, TotalEnergies, and Roche led. Defensive rotation in Europe while the US was closed for a holiday is a cross-border timing signal: European desks had one more session to reprice the hawkish FOMC while US desks were dark. If Monday's US open shows a similar defensive rotation, with utilities and healthcare outperforming tech and discretionary, Europe was the early mover and the bounce was the false signal.
Strategy's bitcoin treasury machine has passed $10 billion in perpetual preferred outstanding, and the instrument's design means the blowup everyone prices as imminent may never arrive, which is exactly the kind of risk a perpetual creates. Kyle Soska's teardown shows STRC's preferred is a willingness-to-repay, not ability-to-repay, instrument: no maturity, no covenant, no forced-sale trigger. The risk is not a margin call but a confidence break in the preferred's own market. The carry self-finances only while BTC's expected drift beats the floating funding cost, which now rises with the hawkish rate regime. The part most analysis misses is the equity premium hinge: STRC-style issues accrete value only when sold above NAV, so a sustained discount runs the flywheel in reverse. The 1929 Goldman Sachs Trading Corporation carried the same architecture: leverage pyramided on a rising asset, financing structure amplifying losses toward total when the asset fell. For a treasury company, solvency is a function of market access, not asset value.
Standard Nuclear filed an S-1 to go public the same week a second advanced reactor reached criticality, and the sharp bet in the filing is the fuel, not the machine. The differentiator is HALEU, high-assay low-enriched uranium, and the company is building fuel-cycle vertical integration rather than a single reactor design. HALEU supply, historically dominated by Russia, is the binding constraint on every SMR deployment. Filing to fund a fuel cycle rather than a plant wagers that the bottleneck in the AI-power buildout migrates from reactor design to enrichment, and whoever owns the scarce input owns the margin. NuScale Power's 2022 SPAC is the precedent: first SMR company public, flagship UAMPS project canceled in 2023, stock down roughly 80% before the datacenter-power thesis revived it. NuScale sold the machine; Standard Nuclear sells the fuel, making the binding question whether HALEU stays scarce. If DOE enrichment contracts at Centrus and Urenco scale faster than expected, the vertical-integration premium compresses toward a commodity margin.
Tokenized real-world assets crossed from static holdings to active margin collateral this week, with RWA perps accounting for roughly 30% of all trading volume on Hyperliquid, the largest decentralized perpetual-futures exchange, the composability inflection the RWA thesis has been waiting two years for. Tokenized Treasuries have sat as on-chain PDF files: real assets on rails with under 10% used in lending or leverage. RWA-as-perp-collateral means the tokenized asset is finally load-bearing, earning its T-bill yield while backing leveraged positions, converting a custody novelty into financial infrastructure. The metric that matters is the share of activity that composes, and 30% is a step-change. The 1990s parallel is tri-party repo, when Treasuries stopped being buy-and-hold instruments and became the reusable collateral the shadow-banking system was built on. The counter-risk is that precedent's dark side: a single de-peg or oracle failure cascading through the leverage stacked on RWA collateral relocates yesterday's wrapper risk from the token to the collateral layer.
The enterprise AI procurement decision is flipping from "capability at any cost" to "good-enough at controllable cost," and the repricing is happening faster than any vendor pricing model assumed. Costi Perricos, Deloitte's global generative-AI lead, confirmed what inference bills already showed: companies are reining in usage as costs hit CFO scrutiny, and some are now mandating locally-run open-source models. Corroborating data points are converging. Uber exhausted its 2026 Claude Code budget in four months. Microsoft canceled Copilot licenses for underuse. Thomas Wolf at Hugging Face warned that if enterprises fight open-source safety infrastructure, they will accelerate the flight of commercial users to uncontrolled alternatives. Once a CFO moves AI from an experiment line to a managed cost center, the winner is no longer the best model but the cheapest deployable one. The early-2000s enterprise migration from proprietary Unix to commodity Linux followed the same pattern: once "good enough" open infrastructure existed, IT budgets defected to cut vendor rents and the premium platform's pricing power never recovered.
Three distinct AI governance architectures are crystallizing simultaneously, and the structural divergence between them is now wider than the gap between any two models. Zvi Mowshowitz's analysis of the Fable shutdown maps the terrain: US executive ad-hocracy (Fable frozen by an emergency order with no legislative basis), US Congressional permanence (NDAA section 3252 creating bipartisan statutory framework), and EU codification (the Digital Omnibus Act folding AI oversight into binding statute). Open-weight diffusion, with GLM-5.2 under MIT license at frontier parity and Poolside's Laguna M.1 shipping within 48 hours, sits outside all three, rendering each partially irrelevant to anyone willing to self-host. The 1990s encryption export controls are the precise analogy: the US classified PGP as a munition, the restriction accelerated overseas development, and the policy reversed within a decade. If Fable remains frozen through Q3 while open-weight adoption accelerates, the frameworks constrain compliant actors while the non-compliant build around them.
The open-weights wave is now producing not just models but a full self-hosted stack, and that stack, not any single model's benchmark, is what structurally caps commercial AI pricing. Poolside's Laguna M.1, the second frontier-class open model to ship in 48 hours, supplies the engine; Simon Willison's Datasette Apps, sandboxed-iframe applications backed by client-side SQLite, supply the delivery layer: no API dependency, no per-seat cost, no vendor lock-in. When the model is free and the runtime sits on the client, there is no metered surface left for a premium vendor to charge against, which is the part of the open-weights shift that never shows up in a benchmark table. The counter-case is the AA-Briefcase finding that even frontier models satisfy the full rubric on just 3% of realistic tasks, which says the "good enough" threshold is further away than the release cadence implies.
The Iran nuclear MOU signed Wednesday has effectively collapsed before its scheduled Geneva ceremony, a 48-hour shelf life that reveals less about Iran than about the structural fragility of summit diplomacy in a multi-front environment. The Geneva signing event was postponed to a conditional meeting, CENTCOM lifted its naval blockade of Iranian oil even as the diplomatic framework underneath it dissolved, and the ceasefire between Israel and Hezbollah that the deal was meant to stabilize began fraying on the Lebanese border. Zeihan's read is blunt: the US never had the relationship infrastructure to sustain a negotiated outcome, and the downgrade from ceremony to conditional meeting is the tell that neither side treated the MOU as binding. The structural risk has migrated from "will they sign" to "what happens to the oil risk premium that leaked out over the past week now that the deal is unraveling." If crude holds steady despite the collapse, the market is saying Iranian barrels were never coming off the risk ledger to begin with.
ChinaTalk's analysis reframes the Chinese AI conversation from a capability race to a labor-displacement anxiety crisis, where the binding political constraint is not compute access but the regime's management of mass technological unemployment. The framing shifts the usual lens: China's AI challenge is not building better models but maintaining the social contract when those models automate the administrative, educational, and white-collar roles that absorb hundreds of millions of workers. The regime's legitimacy rests on employment stability, and the AI automation wave threatens exactly the tier of work, processing, teaching, middle-management, that the CCP expanded over three decades to keep urban populations employed and politically quiescent. If Chinese GDP growth decelerates while AI-linked unemployment rises in tier-2 and tier-3 cities, the regime faces a Polanyi double movement: the technology generates wealth while destroying the social arrangements that distribute it.
Rajagopalan's analysis of India's structural alignment with the US frames it as absence of alternative rather than affirmative choice, a distinction that produces a durable but fragile partnership. India's alignment is not ideological but geometric: it faces a nuclear-armed adversary to the north, depends on Gulf energy it cannot secure alone, and has a demographic dividend that requires foreign capital and technology to convert into growth. The partnership's durability comes from the absence of any other patron who can supply all three: security, energy access, and technology transfer. Its fragility comes from the same source. A partnership held together by the absence of alternatives generates compliance without enthusiasm, and compliance without enthusiasm breaks the moment an alternative appears. The analogy is Turkey's NATO membership: durable for decades precisely because Ankara had no other security architecture, and increasingly strained the moment Russia offered a transactional alternative.
NASA's James Webb Space Telescope detected methane on interstellar comet 3I/ATLAS, the first time methane has been directly observed on an object from another star system, suggesting 3I/ATLAS was born in a chemically distinct environment with no known analog in our solar system. The JWST team used the Mid-Infrared Instrument to observe the comet in December 2025 as it traveled back out of the solar system after perihelion. The methane was undetectable on approach, buried beneath surface ice, and appeared only when the comet's icy surface sublimated under solar heating. The comet also carries exceptionally high carbon dioxide levels, a volatile composition unlike any solar-system comet, implying that the chemical diversity of extrasolar systems extends beyond orbital architecture into the specific molecular inventory that planets and moons inherit from their birth environments.
Geoscientists discovered that several well-studied subglacial basins beneath East Antarctica, including the Wilkes basin, the Aurora basin, and the basin holding Lake Vostok, are actually segments of a single fan-shaped geological structure formed by rotational extension of the continental crust. The finding, published in Nature Geoscience, means that features analyzed individually for decades were parts of one interconnected system all along, hidden beneath more than three kilometers of ice. The structure likely formed through multiple tectonic episodes during the breakup of Gondwana and the later separation of Antarctica from Australia. The reframe matters for ice-sheet modeling: the shape of bedrock beneath the ice determines how ice flows, and a connected basin system channels ice differently than isolated depressions do, which means the stability projections for East Antarctica's most vulnerable regions may need revision.
Physicists discovered that grains of rice violate standard compression physics: they weaken under rapid compression but remain stronger when force is applied slowly, the opposite of what most granular materials do. Conventional granular materials, such as sand and gravel, lock up and resist force faster when compressed quickly. Rice does the reverse. The team traced the mechanism to the grain's elongated geometry: under slow compression, rice grains have time to align and distribute force efficiently, while under rapid compression the grains jam in random orientations that create stress concentrations and fracture. The finding challenges the assumption that granular behavior is geometry-agnostic and suggests that industrial grain handling, food processing, and silo design are using compressive models built for spherical particles on decidedly non-spherical materials.
Scientists discovered a new Amazonian spider species whose body mimics a parasitic fungus so convincingly that it deters predators by looking like something already infected and dying. The mimicry strategy inverts the usual logic: most arachnid disguises copy plants or debris to hide, while this spider imitates a pathogen to make itself look like something no predator wants to eat. The fungal appearance extends to color, texture, and the asymmetric bulges that parasitic infections produce on insect bodies. The discovery adds a new category to the mimicry taxonomy, where the mimic signals not "I am not here" but "I am already destroyed," exploiting the predator's evolved avoidance of diseased prey rather than its failure to detect prey at all.
The cremation wave is hollowing out deathcare's per-unit economics at the exact moment the US mortality curve steepens, and the only operator positioned to profit from both trends is the industry's sole national consolidator.
Deathcare is walking into a demographic tailwind and a margin headwind at the same time, and the second one is the part nobody is pricing. The tailwind is real: the baby-boom cohort is entering its highest-mortality decades, US deaths are projected to exceed births around 2030, and annual deaths keep climbing toward roughly 3.6 million by the late 2030s. More deaths should mean more revenue. It will not, because how Americans dispose of their dead is changing structurally. The US cremation rate hit 63.4% in 2025 and the NFDA projects it reaches 82.3% by 2045, with burial collapsing from 31.6% to 13% and cremation outpacing burial six-to-one, a ratio the country has never seen. A cremation throws off a fraction of a traditional burial's revenue: no casket, no vault, no plot, often no embalming. Per-death economics are deflating even as the death count inflates. That arithmetic quietly breaks the sub-scale, burial-dependent independent funeral home and the casket supply chain, while it rewards the one operator with the scale, preneed backlog, and premium-cremation memorialization to defend revenue per call and buy up the distressed independents: Service Corporation International (SCI), the industry's only national consolidator. The exposed side is the casket and burial-hardware makers like Matthews International (MATW) and the thousands of single-location operators whose model assumed a burial mix that is disappearing. If SCI keeps growing core revenue per service even as its core services-performed volume slips, exactly the split it printed in late 2025 (revenue per service +3.2%, services performed -1.9%), the consolidate-and-upsell defense is working and the cremation squeeze is landing on everyone below it. Watch: SCI's quarterly core average revenue per service versus core funeral services performed, and Matthews International's memorialization (casket) segment volumes. If revenue per service keeps rising while industry casket volumes keep falling, the cremation shift has moved from a slow consumer preference into a permanent rewiring of who in deathcare makes money: the scaled consolidator, not the local funeral home.
The car has quietly become too complex to repair, and that one fact is rewiring auto insurance toward the salvage auction and away from the body shop.
A structural shift is running through auto insurance that has nothing to do with how often people crash and everything to do with what a car now is. In 2025 a record 23.1% of all auto claims, nearly one in four, ended with the insurer totaling the vehicle rather than repairing it, the highest share in the industry's history. The cause is not worse accidents; it is that the modern car is a rolling computer. Advanced driver-assistance sensors, cameras, and the calibration they demand now appear on better than 31% of repair estimates (up from 24% a year earlier), the average repair bill has climbed to roughly $4,800, and more than 70% of total-loss vehicles are seven years or older: cars whose market value is low but whose repair cost has exploded, so even a moderate fender-bender tips "fix it" into "write it off." This resets the loss-cost base permanently rather than cyclically. Every model year adds more electronics, so the total-loss share ratchets up and does not come back down. The money follows the wreck. Salvage and claims-data businesses are the structural winners: Copart (CPRT), which auctions the totaled cars, already grew global volume 8% on record total-loss frequency, and CCC Intelligent Solutions (CCCS), whose software prices the repair-versus-total decision, gets more valuable as that decision gets harder. The squeezed side is the carriers who cannot reprice fast enough, Progressive (PGR) and Allstate (ALL) win or lose on how quickly their rates track the new loss base, and downstream, the drivers priced out entirely. As premiums chase repair costs, more motorists drop coverage, and the uninsured-motorist claims everyone else pays keep rising. If total-loss frequency pushes past 24-25% in 2026 while Copart's salvage volume keeps compounding double digits, the repair-to-replace shift is confirmed structural, not cyclical, and the profit pool in auto insurance keeps migrating from the body shop to the auction lot. Watch: CCC's quarterly Crash Course total-loss-frequency reading and Copart's unit-volume growth. If total-loss share sets another record while salvage volume rises and average premiums climb, the car's complexity has become a permanent tax on the whole system, paid by carriers in loss costs and by drivers in premiums, and collected by the salvage and claims-tech layer in between.
Stress Displacement: in a market with no mark-to-market price discovery, risk cannot show up in prices, so it displaces into the system's open valves, first new issuance, then redemption terms, finally a forced repricing. The calm you observe is a function of how rarely the asset is priced, not of how little risk it carries.
This week, private credit's plumbing started backing up. New US direct-lending issuance fell roughly 40% in a quarter, about $75 billion to $45 billion by May per PitchBook data via Jeffrey Snider, while reported NAVs barely moved. Markets & Macro reads that same contraction as a rate-cut signal; the deeper tell is the mechanism behind it. Sponsors will not sell at today's clearing prices, because a sale admits yesterday's marks were fiction. One investor called it constipation. Exactly right, and exactly the point.
Both sides watch the wrong variable. Bulls cite low volatility and near-zero defaults as proof of stability; bears wait for NAV markdowns to vindicate the "shadow-banking time bomb." Both watch prices, in a market built so that stress can never appear in them. A roughly $1.7 trillion asset class marked quarterly to model shows serene volatility whether it is healthy or hemorrhaging; the calm is manufactured by the cadence, not earned by the credit. Stress does not vanish because no one marks it. It moves, and issuance is the first place it surfaced. The next valves are mechanical: PIK interest rising, more IOUs booked instead of impairments, then redemption gates.
Projection: issuance does not reclaim its roughly $75 billion quarterly pace through 2026; PIK income and non-accruals climb across the business-development companies; a prominent private-credit vehicle gates redemptions before Q1 2027, all while headline NAVs fall less than 10%. The early warning for any unmarked asset is always a flow, never a price. SVB's held-to-maturity book showed no loss until a deposit run forced the sale that realized it.
Where this might be wrong: the drop may be plumbing, not stress. Direct lending tracks LBO and M&A volume, and with sponsors frozen by a wide valuation bid-ask, new loans dry up mechanically, a dealflow pause rather than a default signal. Frozen markets can stay frozen for years: commercial real estate has ground down rather than snapped since 2023, and "private credit is a time bomb" has been wrong since 2022 precisely because locked-up capital has no forced seller to light the fuse. Patient money can amortize losses quietly for a decade, which makes the mark-to-model fiction stabilizing rather than fragile. And if the fade-the-hawk camp in Markets & Macro is right and the Fed cuts, refinancing pressure eases and issuance rebounds, and the whole signal evaporates as a rate artifact. Falsification: if direct-lending issuance is back above roughly $70 billion a quarter by Q4 2026, with PIK and non-accruals flat and no fund gating, this was constipation, not hemorrhage, and Stress Displacement mistook a slow market for a sick one.
"In Africa, when an old person dies, a whole library burns down."
— Amadou Hampate Ba
You know the person. The colleague who understands why the system was built that way and never wrote it down. The grandparent who carried stories nobody else thought to record. The mentor who told you the one thing that changed how you thought about your work, whose number you have not called in months. They are a library, and libraries do not send reminders that they are closing. The knowledge they hold is not the kind you can look up later, because it was never written down. It lives in the way they pause before answering, the warnings they give from experience no book contains, the pattern recognition they built across decades that will not survive them. The bias is to assume the library stays open. It does not. It closes without announcement, and what was in it is gone.
Call or visit that person this week. Ask them the one question you have been meaning to ask, and record the answer, in writing or voice, somewhere it will outlast the conversation. The library is open now. It will not always be.
For roughly two thousand years, European naturalists arranged all of life on a single ladder. The scala naturae, the great chain of being, ran from rocks at the bottom up through plants, insects, fish, mammals, humans, and angels, each rung strictly above the last. It was not a casual metaphor; it was the organizing framework of Western biology, and it felt like the deepest truth about nature, that creation had an up and a down and every creature held a rank. Then Darwin replaced the ladder with a branching bush. On the tree of life there is no highest organism. A tapeworm is not a degraded mammal; it is exquisitely adapted to a different problem. The ladder had forced a single axis, more complex to less complex, onto a reality that was actually arrayed across thousands of independent axes at once. Nothing about the animals changed. What changed was the recognition that the ranking had been imported by the observer, not read off the world.
This is the difference between a ladder and a spectrum, and confusing the two is one of the most common errors in how we think. A ladder collapses many independent dimensions into one ordinal axis and ranks everything on it: better and worse, higher and lower, more and less advanced. A spectrum keeps the dimensions separate and accepts that two things can differ without one standing above the other. The move from ladder to spectrum is not relativism, the claim that nothing can be ranked. It is the narrower recognition that ranking is a choice of axis, and that the axis is usually invisible to the person doing the ranking. When you reduce a multidimensional thing to a single number and sort by it, the number starts to feel like a property of the thing. It is not. It is a projection, and like any projection it throws away every dimension except the one you chose to keep.
The uncomfortable part is that the ladder feels like rigor. Ranking looks like clear thinking, decisive, quantified, defensible, while "it depends which dimension you mean" sounds like equivocation. But the confident ranking is frequently the one that has discarded the most information. A single "best hospital" list collapses surgical outcomes, infection rates, cost, and bedside care into one ordinal that hides which of those the patient in front of you should weight. A leaderboard that ranks AI models by one benchmark erases that they are strong at different kinds of work. A development index that sorts countries from advanced to backward smuggles in one civilization's axis as if it were the axis. In each case the ranking is not so much wrong as it is a decision disguised as a measurement, and the disguise is the danger: the reader inherits the ranker's choice of axis without ever seeing that a choice was made.
The diagnostic is a single question, asked whenever you catch yourself ranking: what am I collapsing into this one number, and who chose the weights? If the honest answer is that several genuinely independent qualities have been flattened onto one axis, you are on a ladder, and the move is to pull the dimensions back apart and ask which one actually decides the matter in front of you. Sometimes a single axis is exactly right; when you truly care about one thing and one thing only, a ladder is the correct tool, and pretending otherwise is its own failure. But the error runs heavily in one direction. We reach for ladders because they end the discomfort of holding several incommensurable things in mind at once, and the relief of a clean ranking is often the feeling of having stopped thinking one step too early. The most confident "this is plainly better than that" is the place to look hardest for the axis you did not notice you had picked.
Start with the most confident ranking you made this week.
Stewart Brand, cataloguing what makes a civilization durable, noticed that a resilient system is not one thing moving at one speed. It is several layers moving at very different speeds, and the spread between them is the point. He named six, fastest to slowest: fashion, commerce, infrastructure, governance, culture, and nature, each roughly an order of magnitude slower than the one above it. The fast layers are where novelty lives; they experiment, fail cheaply, and grab all the attention. The slow layers are where memory and stability live; they move grudgingly and hold the real power. His compression of it: "Fast learns, slow remembers. Fast proposes, slow disposes." The counterintuitive part is that the system stays healthy not when the layers agree but precisely because they do not. The friction between a fast layer straining forward and a slow layer holding it back is what keeps the whole structure from either ossifying or flying apart.
The reframe is that most failures we blame on bad execution are really layer mismatches, a change introduced at the wrong speed for the layer it is trying to move. We instinctively treat a system as a single object that is either "working" or "broken," when it is actually a stack of clocks, and the trouble almost always comes from one clock being forced to run at another's tempo. A fast-layer fix bolted onto a slow-layer problem, a new tool dropped on a broken culture, a reorg meant to rebuild eroded trust, a policy tweak aimed at decayed infrastructure, gets visibly rejected, and we call it resistance. The opposite error is just as common and far more expensive: a slow-layer change sold as a fast one, promised on a quarterly timeline when it actually moves at the speed of culture or infrastructure, which is how confident plans quietly slip by years.
So when something keeps failing or snapping back despite real effort, stop asking "who dropped the ball?" and ask "which layer is this actually in, and which layer am I colliding with?" Locate the change on the stack: is it fashion-fast or culture-slow? Then match your intervention's patience to that layer instead of to your deadline. When the two do not match, you have found the failure before it happens. The structure is everywhere once you can see the speeds: in software, configuration changes by the hour while the data schema beneath it changes by the year; in markets, prices tick by the second, market structure shifts over years, and regulation slower still; in an organization, slogans change overnight, incentives over quarters, and the actual culture slowest of all; even inside one person, mood is the fast layer, routines the middle, and identity the slow one that remembers. Test it this week on a single stuck problem: name the layer, name its real clock speed, and notice whether the fix you were about to apply runs at the same speed, or whether you were about to push on a slow layer with a fast hand.