The market front-ran a peace deal that hasn't been signed, crashing oil and pushing the dollar below 100, while May wholesale inflation came in hot for the second consecutive day, turning "rate hikes" from tail risk into the base-case debate ahead of next week's FOMC. SpaceX begins trading today, and the market punished Oracle's record quarter for the capital bill behind it. The thread running through almost every story below: the number on the surface is hiding the one that actually matters underneath.
The US-Iran memorandum remains unsigned, and the overnight tape added friction: Iran's Foreign Ministry said no final conclusion has been reached, a source close to the negotiating team denied any text has been approved, and US forces shot down Iranian drones even as Trump floated a signing as soon as this weekend. The structural contradictions are covered in Geopolitics below.
SpaceX begins trading on Nasdaq today at $135 per share and a $1.77 trillion valuation, the largest IPO in financial history. Overnight, perpetual futures on Hyperliquid priced SpaceX near $168, a 25% premium to the IPO price, and Bitcoin bounced into the debut, the first live read on the capital-vacuum test in Companies & Crypto below.
US equity futures are flat to slightly lower into the open (S&P -0.1%, Nasdaq 100 -0.3%, Dow steady). Asia rallied on deal optimism, the Nikkei up 3.4% and Shanghai up 1.6%, while oil held its post-flush lows through the overnight session.
Crypto data provided by CoinGecko
May producer prices rose 6.5% year-over-year, the hottest wholesale inflation since November 2022, and the composition tells you exactly where the pressure sits: roughly 80% of the monthly acceleration came from a 2.8% surge in final-demand goods, the largest goods increase in the series' history back to December 2009, and roughly 80% of that was a 10.7% jump in energy. Core PPI excluding food, energy, and trade ran at 5.1% year-over-year, the most since October 2022. Stack this on Wednesday's CPI at 4.2% and the pipeline is clear: an energy shock working its way from producers to consumers, with the PPI-CPI gap of 2.3 percentage points implying either a coming consumer-price catch-up or a margin squeeze as producers absorb the cost. The market is treating both prints as a "war tax" rather than broadening demand-pull inflation, which is why the long end stayed calm. But the hawks now have two consecutive hot prints walking into the June 16-17 FOMC.
The European Central Bank hiked its deposit rate 25 basis points to 2.25%, the first rate increase by a major developed-market central bank since 2023, hiking into a growth forecast it simultaneously cut to 0.8% for the year. The decision was unanimous. Lagarde called it "robust across a range of scenarios" mapping how the energy shock evolves and explicitly refused to pre-commit to a path, even as the inflation forecast was raised to 3.0% for 2026. The ECB is now hiking into weakness, a stagflationary box with no clean exit, where the energy shock demands tighter policy while the 0.1% Q1 GDP print and the war's drag on real incomes demand looser. Dario Perkins captured the tension in six words: "he who hikes first hikes least." The market, he noted, has it priced the other way round.
Initial jobless claims rose to 229,000, the highest since February and above the 220,000 expected, handing the dovish camp the labor-softening data point it needed on the very morning the hawks got their hottest PPI print in three years. This is the stagflation vise tightening one more notch: hot prices and cooling labor in the same week, arming both sides into the FOMC. The curve is siding with the doves, the two-year yield eased to 4.12%, and the FedWatch tool shows a 96.5% probability the Fed holds next Wednesday. But the 12-month hike odds are climbing fast, and Jeff Weniger flagged the transmission nobody is pricing: "Usually when you see the ECB hiking rates, the Fed is doing so too. It doesn't always work that way, but we cannot ignore this action."
Three central banks faced the same energy shock this week and gave three different answers: the ECB hiked (above), the Bank of Canada held for a fifth straight meeting at 2.25% while warning of a policy "dilemma," and the Fed holds next week with near-certainty. The divergence is now realized, not hypothetical. What makes it matter is the Fed's specific trap: US interest expense just hit $1.3 trillion over the trailing twelve months, closing in on the single largest line item in the federal budget. As Lyn Alden argues, today's inflation is fiscally driven, with a government too indebted to tolerate high rates, a structure closer to the 1940s than the 1970s. A Fed that hikes isn't disciplining inflation; it's compounding the deficit feeding it. Luke Gromen names the forced choice: Warsh must choose the dollar or the bond market. Next Wednesday he picks which one to sacrifice.
Adobe reported revenue of $6.62 billion and adjusted earnings of $5.96 per share, beating on both lines with full-year guidance raised, and the stock slipped 1% after hours because a clean beat on legacy metrics no longer clears the bar the market has set for AI-adjacent companies. Adobe is down 30% year-to-date on the fear that generative AI eats Photoshop subscriptions rather than selling more of them. Firefly's ending annual recurring revenue crossed $250 million in Q1 and AI-first ARR more than doubled year-over-year, but against roughly $22 billion in total revenue, AI is still about 1% of the business. The competitive threat arrived in April when Claude Design launched, and the market's verdict is now specific: it stopped paying for AI-adjacent revenue growth and started demanding AI-native profit. A beat-and-raise that would have rallied the stock 5% a year ago gets a shrug instead, and that shift in the market's pricing function matters more than any single quarter.
Bitcoin is the worst-performing major asset of 2026, down roughly 28% after an overnight rebound (gold is the distant second-worst), and today it gets the cleanest test yet of why: the capital-vacuum thesis meets the largest IPO in history. Michael Saylor's framing is that a $400 billion tech-IPO wave is acting as a liquidity vacuum, pulling speculative capital out of crypto and into equities, tracing BTC's slide from $82,000 to $62,000 over fifteen days to this drain. SpaceX's $75 billion raise begins trading today. Underneath the price, the infrastructure keeps compounding: tokenized US Treasuries hit an all-time high of $14 billion, real-world-asset volume is roughly 30% of on-chain activity, and the plumbing strengthening while the speculative asset bleeds is the real story of crypto's 2026. The falsification arrives today: the overnight bid into the debut is the first tick against the vacuum. If BTC holds as the raise settles, the thesis is wrong. If it bleeds on each subsequent raise, crypto stays hostage to the IPO calendar.
Backpack Securities launched $SPCX, a tokenized SpaceX equity instrument on Solana via Sunrise DeFi (Backpack's on-chain finance platform), redeemable for the underlying shares and arriving the same day SpaceX begins trading on Nasdaq. The same IPO wave that's draining crypto liquidity is also colonizing crypto's rails: SpaceX exposure is now available simultaneously through the public tape and through an ERC-20-style token. Matt Levine's structural point is the one index investors should sit with: SpaceX will be fast-tracked into the S&P 500, with Anthropic and OpenAI heading for major indexes behind it. The forced-demand mechanics of index inclusion mean every indexed dollar has to buy AI-cycle exposure whether the investor wants it or not, you cannot opt out of these valuations through a vanilla index fund, even if you think they're extreme. Between the capital vacuum pulling money toward the IPOs and the index machine forcing money into them, the same liquidity plumbing now runs in both directions, and the retail investor is on both ends of it.
Anthropic's CEO published the most significant AI policy shift of the year, endorsing binding government regulation for the first time after three years of transparency-only advocacy, with an explicit "FAA for AI" analogy and a $350 million funding commitment split between a $200 million Economic Futures Research Fund for empirical research on labor displacement and a $150 million fellowship for early-career professionals. The accompanying Economic Policy Framework keys its recommendations to three unemployment thresholds (5%, 10%, and "unprecedented"), proposing wage insurance, retention tax incentives, and universal capital accounts. The timing is notable: it landed one day after the Claude Fable 5 launch, mid-AI-selloff, alongside the OpenAI price-war news. Simultaneously, Erik Brynjolfsson's Stanford Digital Economy Lab launched AI Economic Indicators, a platform for tracking how AI reshapes work and productivity. Three labor-data initiatives crystallizing in one week suggests the policy infrastructure for measuring AI displacement is being built in earnest, even as the market still debates whether displacement is the right frame at all.
The trust cost of covert capability restriction surfaced when Anthropic was caught silently degrading Claude Fable 5's performance for competing AI researchers, a policy reversed within 24 hours of exposure. Nathan Lambert of Ai2 layered the critique: the simple failure is uneven safety enforcement misleading users; the deeper failure is that silent manipulation "bakes in a misalignment to the system at its face level," making the environment less safe; and structurally, "it is a very real sign of concentration of power that businesses can make such obviously user-harmful behaviors and still lead." The episode lands in a week where government equity stakes in AI firms and OMB grant rules subordinating peer review to presidential priorities put the state itself inside the innovation economy, making concentration of power over intelligence the throughline from two directions at once.
Sarah Guo of Conviction published the single best articulation of the bull case for the AI application layer against the "model eats everything" bear case that now dominates markets. Her framework: the most valuable work is "illegible by construction" because "anything you can put on a leaderboard, you can train against, so anything measurable is already on its way to commodity." The durable value lives in the one quadrant models cannot eat, frontier work whose correctness is private, walled off inside a customer's systems, liability, and relationships. The data backs it: across 100,000 developers, MIT's Mert Demirer found coding agents lifted code written roughly 180% but code shipped only 30%. "Writing got cheap. The rest still runs through a person." Her organizing image is "the untrainable corner": intelligence is not the bottleneck, permission is. Joshua Kushner of Thrive offered the institutional-capital version: "We are long humans."
Chinese technology firms have begun rationing employee AI token usage because the returns do not justify the consumption, the first hard evidence of a demand-side crack in the AI buildout, complementing the supply-side capex worries that have dominated the bear case. Bill Bishop's Sinocism, citing Yicai, reported that major firms are "doing the math and imposing limits on employees' token use, recalibrating token quotas toward a more disciplined, efficiency-driven allocation model." The adoption-dispersion data gives it context: Ara Kharazian found the median firm spends $11 per employee per month on AI, roughly one chat subscription, while the top 1% spends $7,450 and the most AI-committed companies (Ramp dataset, 70,000 firms) spend $90,000 per employee per year. The distance between the median and the tail is where the ROI question lives. If the buyers who are supposed to absorb the compute being built start capping usage before the buildout even peaks, the monetization-arrives-too-slowly thesis that Christophe Barraud framed as "telecom circa 2000" gets its first demand-side data point.
The Qatar-brokered US-Iran memorandum of understanding reached head-of-state-level confirmation when Trump affirmed the understandings have "the approval of all concerned parties while efforts continue to complete final procedures," but the structural contradiction that could still collapse it is now visible. The deal is blocked on enrichment dismantlement and frozen Iranian assets: Netanyahu is demanding removal of enriched material and dismantling of infrastructure, while Iran insists on linking any US deal to an end of Israeli operations in Lebanon. Those two linkage demands point in opposite directions and cannot both hold. The live risk is the asymmetry: the tape is fully priced for a signature, oil at $86, DXY under 100, equities green. A signed deal unwinds the premium smoothly, but a collapsed deal re-arms the oil non-linearity into a market that has just taken off its hedge.
India summoned the US chargé d'affaires after three of its nationals were killed in a US Navy strike on the tanker MT Settebello in the Gulf of Oman, the third vessel with Indian crew hit under the "Project Freedom" Hormuz escort operation, and the diplomatic blowback from a swing power is the hidden bill for the mechanism keeping oil calm. The Ministry of External Affairs called the strikes "unacceptable" and demanded "unimpeded access through the Strait of Hormuz in accordance with international law." The irony is structural: the same US escort system fading the oil-war premium works partly by striking non-compliant tankers. The mechanism holding oil near $86 instead of $105 is generating dead third-country nationals and a diplomatic rupture with exactly the swing power Washington needs onside for the larger contest with China. The premium you don't see in the oil price is being paid somewhere else.
A baby born in August 2024 with a rare metabolic disorder received a bespoke CRISPR gene therapy designed, tested, and delivered in eight months, using a delivery vehicle originally built for the COVID vaccine, and is thriving after three liver-targeted doses with no further treatment required. Jennifer Doudna described the case in a Quanta interview as an example of what becomes possible when the editing tool is universal and only the guide RNA needs to change. The binding constraint in gene therapy has migrated from the edit itself to the delivery: liver is solved, but lung, muscle, and brain tissue remain frontier. The leverage point in any breakthrough technology eventually migrates downstream from the celebrated capability to the unglamorous substrate, and gene therapy's current boundary sits not in the science of cutting DNA but in the engineering of getting the scissors to the right cell.
A computational simulation of sexual reproduction found that a population of 100 organisms with 200 genes reached maximum fitness in 33 generations, while an identical asexual population needed 200, sex got there six times faster. Brian Potter's model at Construction Physics found that 75% of the original sexual population's beneficial genes survived to generation 34, versus roughly 1% in the asexual case. The asexual losers fall to "clonal interference": when two good mutations arise in separate lineages, they can never combine, so one is always wiped out as the lineages compete. The result is a quantified answer to one of biology's oldest puzzles, why complex life pays the enormous cost of sex at all. Sexual reproduction is wildly inefficient on its face: you hand only half your genes to each offspring, you need a partner, and you break up winning gene combinations every generation. Evolution tolerates all that overhead because the alternative is worse, a lineage that cannot recombine is one where every good idea stays trapped with whatever bad ideas it happened to be born beside.
After a US Army Apache helicopter went down in the Strait of Hormuz, a Navy drone operated by Task Force 59 rescued both pilots, the first real-world combat rescue performed by an autonomous vessel. The precedent is categorical, even though it arrived quietly in a theater already thick with unmanned platforms on every side of the fight. A machine pulled humans out of a live engagement with no human pilot in the loop, and the boundary between a tool that assists a rescue and a system that performs one was crossed in the most consequential setting there is.
Researchers observed electrons in ultraclean graphene flowing like a nearly frictionless liquid, defying the resistance behavior that governs ordinary conductors. In most materials, electrons scatter off impurities and lattice vibrations, shedding energy as heat, that scattering is what electrical resistance is. In ultraclean graphene under the right conditions, the electrons stop behaving as independent particles bouncing through a lattice and start moving collectively, as a viscous fluid obeying hydrodynamic equations rather than Ohm's law. What makes it striking is that resistance, which we treat as a fixed property of a material, turns out to be conditional: the same electrons in the same sheet of carbon obey one set of laws or an entirely different one depending only on how clean the material is and how the electrons are made to interact. A property you assumed you could look up in a table is really a behavior the system chooses under conditions.
The most important trade measure of 2026 isn't being called a tariff, and it's aimed at a shipbuilding base that's 0.11% the size of the job
February's America's Maritime Action Plan proposes a universal fee on every foreign-built vessel entering a US port, charged by the weight of the cargo it carries, and the single number that matters, the rate, has not been set. At one cent per kilogram it raises roughly $66 billion over a decade; at twenty-five cents, nearly $1.5 trillion, funneled into a new Maritime Security Trust Fund to rebuild domestic shipyards. The problem is absorptive capacity. The US built eight commercial ships in 2024 against China's 1,000-plus, holds about 0.11% of the global commercial market, and has two yards left that build large oceangoing cargo ships; China's shipbuilding capacity is roughly 232 times America's. You cannot turn dollars into hulls on a one-year timeline: slipways, welders, and supply chains take a decade. So the near-term effect of the fee isn't more ships; it's a new per-ton cost on imported goods, a tariff wearing a different uniform, plus a multi-year demand pull into a tiny set of builders. If the final rule sets the fee above roughly ten cents per kilogram, expect importers to pass the cost into goods prices on a lag, and expect the handful of US shipbuilders, Huntington Ingalls ($54 billion backlog) and General Dynamics' marine yards (already growing double digits), to reprice on order-book expansion long before a single new commercial hull touches the water. Watch: the USTR/CBP rulemaking that sets the per-kilogram rate, and HII and General Dynamics marine backlog in their Q2 and Q3 2026 filings. If the rate is finalized above ten cents per kilogram, the import-cost channel goes live within a quarter; if the marine backlogs jump on Trust Fund awards, capital is flowing to builders years faster than capacity can grow.
The fastest-growing consumer credit in America is invisible to every risk model in the country, and it's about to get illuminated from two directions at once
Roughly $70 billion a year now flows through buy-now-pay-later loans, across more than 90 million American users, and almost none of it appears on a credit report. The structure is the trap: most BNPL lenders don't report to the bureaus, and the loans churn so fast that the Richmond Fed pegs the outstanding stock at only about $3 billion at any moment, so what the bureaus miss is not a balance but a behavior, the stacked borrower whose real payment load stays invisible until a loan hits collections, which is to say, until it's already 90-plus days delinquent. The stress is measurable: LendingTree's 2026 tracker finds nearly half of BNPL users paid late in the past year, up for a second straight year, and Fed research keeps finding the heaviest users are precisely the borrowers already stretched on other credit. This is a blind spot sitting inside every consumer risk model in America, and two forces are now moving to close it: FICO released its first BNPL-inclusive scores (Score 10 BNPL) in late 2025, while mortgage and auto underwriters still run FICO 8 or older and won't see the data for years. The exposure is real, growing, and unmeasured. If a major card issuer or mortgage agency adopts a BNPL-inclusive score, or if a downturn forces BNPL charge-offs into the open, expect a cohort of borrowers who look prime today to reprice toward subprime in a single underwriting cycle, tightening credit for exactly the consumers who lean on BNPL most. Watch: the New York Fed's quarterly Household Debt and Credit report (next release in August) for any standalone BNPL line, plus Affirm's and Block/Afterpay's charge-off disclosures. If either shows charge-offs accelerating while balances keep climbing, the invisible leverage is surfacing, and the first big lender to underwrite on a BNPL-inclusive score sets the repricing in motion.
The Subtraction Margin. A company's operating margin can rise two ways that look identical on the income statement and mean opposite things. Additive margin comes from selling more, pricing higher, or producing more efficiently, it compounds, because it leaves the revenue engine intact and larger. Subtractive margin comes from cutting, headcount, R&D, deferred maintenance, capex, and it is one-time by construction: you can fire a worker once, and the higher margin laps the moment the cut anniversaries. The line on the page is the same number. The forward implication is the reverse.
Stoxx Europe 600 operating margins are projected to expand in 2026 for the first time since 2022, roughly +7.9%, per the top-ranked macro forecaster on the Street. But the same forecast calls the expansion "two-speed": the gains concentrate in oil majors, miners, banks, and AI-exposed tech, and they're powered by cost-cutting and job cuts, not by broad demand. The market is re-rating the number as a recovery. The composition says it is mostly subtraction.
What surface analysis misses is one line up from the bottom number: is revenue rising with the profitability gain (additive) or flat-to-falling while the gain materializes (subtractive)? Subtractive expansion gets mispriced twice over. First, it earns a durable multiple for a one-time gain. Second, and this is the part no single income statement can show, at the index level it is self-eroding: one firm's cut wages are another firm's lost revenue, so when an entire market lifts margins by cutting simultaneously, it shrinks the aggregate demand base those margins sit on. The fallacy of composition turns a thousand rational cuts into a collective drag. And the windfall taxes, layoff restrictions, and wage floors that weak European labor markets tend to generate are a third forward cost the multiple ignores.
If the Stoxx 600's expansion is subtractive, you should see margins print up while revenue growth stays weak, the margin gains fade into 2027 as the cost-cut base laps, and concrete labor-policy backlash surface in the weakest economies. The falsifier is clean: if European revenue re-accelerates alongside margins, the expansion was additive, the re-rating is earned, and this framework is wrong.
Where this might be wrong. Three honest cracks. First, the dichotomy is too tidy. Cutting genuine bloat is real, durable productivity, not a trick, some of the best businesses permanently raised their returns by cutting, and telling "eating the seed corn" apart from "trimming fat" in real time is exactly the hard part the framework waves at. Second, the mispricing may not exist: European equities already trade at a deep structural discount to the US, so subtractive-but-real margins could re-rate a cheap market regardless, and calling it a value trap assumes the discount isn't already wide enough to absorb the flaw. Third, the macro and political legs are the soft ones. Paradox-of-thrift effects take years and get swamped by ECB policy, fiscal flows, and the euro; and predicting social backlash from European restructuring has a poor base rate, the continent has cut before without revolt. The reliable core is narrow and worth holding tightly: subtractive margin deserves a lower multiple than additive margin. The index-fragility and backlash extensions are softer bets layered on top of that one solid distinction, useful as a lens, not as a timing signal.
"You asked me what I learned. I didn't learn anything. I already knew that I wasn't supposed to do that. I was just an emotional basket case and couldn't help myself."
— Stanley Druckenmiller
Aristotle gave this problem a name twenty-three centuries ago: akrasia, weakness of will. You know the right move. You see it clearly. And you do the other thing anyway, not out of ignorance but out of something that overpowers the knowing. Every trader, every leader, every person who has ever eaten the thing they swore off or sent the message they knew they'd regret has lived inside this gap. The distance between knowledge and action is not an information problem. It is a self-management problem, and it does not yield to more research, more data, or more conviction. It yields to structure.
Druckenmiller is not confessing a lesson. He is confessing the absence of one, which is more honest and more useful. The lesson was already learned. The execution failed anyway. This is the part most self-improvement frameworks skip: they assume the bottleneck is insight, that if you could just see clearly enough, you would act correctly. But the bottleneck is almost never seeing. It is the moment between seeing and doing, where emotion, fatigue, social pressure, or habit overrides what you already know.
Pick the one thing you already know you should do differently but keep not doing, then, before the day is out, build one piece of structure that makes the wrong move harder or the right move easier. Set the alarm. Move the object out of reach. Block the app. Text the person who will hold you to it. Don't analyze why you keep slipping; just remove the room to slip. The gap between knowing and doing closes not with more knowing but with less room to do otherwise.
The reactor technology that powers almost every nuclear plant on Earth was chosen not because it was superior, but because it was ready first.
In the 1950s, the US Navy needed a compact reactor for its submarines, and Admiral Hyman Rickover chose the light-water design because it was small and ready, not because it was the most efficient way to generate civilian power. When the US then built commercial nuclear plants, it reached for the design it already understood. The economist Robin Cowan documented what happened next: every dollar spent on light-water reactors made the next one cheaper, better understood, and easier to license, while rival designs, gas-cooled, heavy-water, molten-salt, never got to ride the same learning curve. By the time anyone could seriously ask whether a different design was superior, the question was moot. The world had standardized on the one with the head start. This is path dependence: where a system ends up depends not just on what is best but on the order in which things happened, and early, sometimes arbitrary advantages get locked in by increasing returns until they are effectively permanent.
The mechanism is self-reinforcement. Brian Arthur's work on increasing returns and Paul David's study of the QWERTY keyboard, laid out in 1873 to stop mechanical typebars from jamming, still under your fingers a century after the jamming problem vanished, describe the same loop: an early lead lowers costs or raises compatibility, which attracts more adopters, which lowers costs further, which attracts more adopters. The loop does not care whether the thing at its center is any good. It only cares which thing arrived early enough to start it spinning. Once it is spinning, a superior latecomer cannot break in, because the value is no longer in the design, it is in everyone else already using it.
The trap this sets for clear thinking is that we read survival as proof of merit. The standard that is everywhere must be the best, or it wouldn't be everywhere, except that is exactly backwards. The standard that is everywhere is the one that got the loop spinning first, and its dominance is evidence about timing and lock-in, not about quality. When you catch yourself defending an arrangement because "it's just how things are done," separate the two questions path dependence collapses into one: is this the best design, or merely the established one? The honest answer is often that you don't know, because the alternatives were strangled before anyone could measure them. The high-value move is rarely to optimize the standard you inherited. It is to notice which locked-in defaults are load-bearing in your life or your business purely because they arrived first, and to ask what you would actually choose if the loop weren't already spinning.
In 1883 Osborne Reynolds pushed dyed water through a glass pipe and discovered that flow has exactly two regimes with a knife-edge between them. Below a critical threshold the flow is laminar: smooth, layered, every particle tracing an orderly line. Push the speed past the threshold and the flow doesn't get gradually messier; it breaks all at once into turbulence, chaotic eddies that no longer travel any predictable path. What sets the threshold isn't whether the fluid is "good" or "bad." It's a single dimensionless ratio, the Reynolds number, weighing the inertial forces carrying the flow forward against the viscous forces (internal friction) holding it in order. For flow in a pipe, that transition sits near a Reynolds number of 2,300. Same water, same pipe: stay below the number and the system is perfectly predictable; cross it and order collapses into chaos.
Most systems that move things, information through a team, decisions through a process, work through a supply chain, have a Reynolds number. There is a rate of flow below which everything runs smoothly, and a rate above which the same system, with the same people and the same design, turns turbulent: errors, rework, things slipping through cracks, the sense that no one can see the whole picture anymore. The trap is that the transition is sudden and the instinct it triggers is wrong. When a smooth process turns chaotic, we assume someone got careless or the design is broken, and we reach for oversight: more meetings, more checks, more people watching. But turbulence isn't a discipline problem; it's a flow-rate problem, and oversight adds exactly the kind of load, more throughput, more urgency, that tipped the system over its threshold in the first place.
When a process that used to run smoothly suddenly fills with errors and rework, and nothing about the people or the design actually changed, assume you've crossed your Reynolds number and treat it as a flow problem, not an effort problem. You have three levers, and adding oversight is none of them: cut the throughput (fewer things in flight at once), raise the viscosity (build in buffers, slack, and deliberate friction that keeps the flow orderly), or change the geometry (re-route the work so fast and slow streams stop colliding). The move is falsifiable within a week: if dialing down how much moves through the system at once restores order, it was turbulence; if it doesn't, the problem really was the design, and now you've ruled out the cheaper fix first.