Statecraft

The Strongman Presidency

American democracy faces its current strongman crisis not because of Trump alone but because of a century of structural developments — the rise of the administrative state, deliberate construction of presidential control mechanisms, and a conservative strategic pivot to weaponize executive power against agencies they couldn't dismantle through legislation.

Before the Progressive Era the federal government barely existed as an executive enterprise; most employees were mailmen. Industrialization created demands for expert governance, producing New Deal agencies and, in 1939, the Executive Office of the President — a presidential bureaucracy built to control the broader bureaucracy. From FDR onward, presidents pursued two tools: centralization (pulling decisions into the EOP, NSC, and OMB) and politicization (stacking agencies with loyalists). Project 2025 calls for vastly more of both.

The decisive turn came when 1970s conservatives concluded they couldn't retrench progressive agencies through Congress or courts. Ed Meese's Justice Department developed the unitary executive theory as justification for presidential sabotage of agencies, while popularizing Robert Bork's originalism through the Federalist Society — building the judicial pipeline that produced today's Supreme Court. That Court is not a firewall: it authorized destruction by letting cases "play out" until the EPA, CFPB, and Education Department are gutted beyond repair. The immunity ruling, making a president unaccountable for crimes in office, is the most dangerous Supreme Court decision in American history.

Congress ceased as a check for two reasons: polarization created perpetual gridlock, and unitary executive theory explicitly excludes congressional participation in oversight.

Reform paths exist — court expansion, ending gerrymandering, reversing Citizens United — but require a Democratic trifecta and remain unlikely. Grievances driving strongman politics will outlast Trump; every successor Republican nominee faces identical incentives. Recovery requires the multi-sector coalition — journalists, universities, civic organizations — that the Progressive movement assembled to create the modern state.

What Trump Can Learn From Nixon

Presidents who try to control the bureaucracy through procedural choke-points — Kristi Noem's personal review of every DHS contract over $100,000, Lutnick's identical threshold at Commerce — reliably break the agencies they mean to control. Richard Nathan's book on Nixon's "administrative presidency" shows the pattern is not new.

Nixon entered office with conventional Cabinet picks who quickly "married the natives," captured by agencies and Congressional committees in the iron-triangle dynamic David Truman described in 1958. The White House response was routing around Cabinet secretaries via Ehrlichman's domestic working groups and direct lines to sub-Cabinet officials. That counter-bureaucracy grew to 700 OMB staffers, bottlenecked strategic thinking, created policy churn, and paradoxically made Cabinet secretaries more adversarial rather than less.

By 1972 Nixon pivoted to installing loyal "politician-managers" directly in line positions — demanding jump-and-ask-how-high loyalty, using budget impoundment to cut programs, and assigning Howard Phillips to dismantle the Office of Economic Opportunity without a Reorganization Act.

Watergate ended the experiment early. Nathan's surviving finding: empowering White House staff to overrule Cabinet secretaries destroys the alternative tool of trusted secretaries on a long leash. Running both models simultaneously is what produces the failures visible in Nixon's and Trump's bureaucratic struggles alike. The president must choose.

How to Write the AI Action Plan

Influence in the White House runs on usefulness, not job titles. Dean Ball, who authored the Trump administration's AI Action Plan at OSTP — an office with no budget or formal power — found that being the person others want in the room matters more than rank. Real leverage came from holding the pen: whoever drafts the document adjudicates conflicting agency feedback. Rubio's May 2025 NSC reorganization transferred AI-adjacent processes to OSTP.

The Action Plan is structurally unusual. Most government AI strategy documents offer vague aspiration; this one pairs objectives across three pillars with two to six concrete, agency-executable recommendations each — over 90 total, within existing authority. Rather than circulating the full document, Ball sent each agency only its relevant bullet points, ran ~20–40 parallel mini-clearances, and used AI to pre-simulate pushback. Rollout with five cabinet secretaries, the VP, and the president together gave agencies political cover to act.

Implementation is the harder problem. The White House operates on 10-day time horizons; ideas must arrive fully baked. Key levers: get agency leadership bought in before publication, attach deadlines to executive order provisions, and create forcing functions — a principal's scheduled visit or public event — that manufacture urgency.

On export controls, Ball argues for a global licensing framework: set a capability threshold, define what TSMC output share goes to China below it, and enforce violations through better IC–Commerce Department information sharing and chip location verification.

On AI safety, Republicans are structurally better positioned than Democrats to act on catastrophic-risk concerns. "American AI dominance" maps naturally onto control and alignment; Democratic coalition politics requires routing every concern through dozens of interest groups. The "Deep State" is real — career bureaucrats route high-context memos on two-hour clearance windows — and Trump's second term returned with a plan to pre-empt that pattern.

Three Principles for Running a White House Office

Advisory power gets things done through relationship mechanics, not authority. Tom Kalil's three rules from running Obama's OSTP: assume people won't follow up — end meetings by naming the next concrete step and signal that obstacles are safe to report. Find who owns the paper — the speechwriter, the OMB analyst — because a report changes nothing if you never touched the actual decision document. Make it easy for whoever must act: draft the email your principal needs to send, supply the expert to the agency that can't hire.

How the National Security Strategy Gets Made

The National Security Strategy is less a unilateral presidential decree than a coalition-building document — its value comes from forcing the government to hash out trade-offs in writing. Nadia Schadlow, who led the 2017 NSS under McMaster, explains the mechanics: about 12 interagency meetings in the Eisenhower Executive Office Building, agenda controlled by the principal drafter, papers capped at two or three pages. Whoever drops the first draft controls the template; everyone else reacts. She circulated drafts as PDFs to prevent bureaucratic word-churn — "leverage," "robust," and three-verb strings are the pathologies she names.

Power centers matter concretely. Treasury in 2017 was less hawkish on China than the NSC or DoD because it represented Wall Street investment interests; China hawks had to fight for "strategic competitor" — language that now reads as obvious. Climate change was cut not from interagency battle but because it was not a Trump priority. State consistently pushed to soften language based on foreign-government reception; the NSC's frame was: this document is for the American people.

The 2025 NSS is more confrontational in tone, drops "strategic competitor" for China — Schadlow reads this as leaving negotiating room, not as softening, given the strong language on predatory subsidies and Taiwan — and omits North Korea entirely, which she reads as unresolved internal debate. She is skeptical about Greenland: the strategic goals are achievable without threatening an ally.

Her deeper preoccupation is time as an underappreciated strategic input — permitting a new mine takes roughly 16 years; weapon-system development runs similarly long. She advocates Gantt-chart discipline for national security: map A to Z, surface bottlenecks, make fast-and-good-enough versus slow-and-perfect trade-offs explicit. Reading list for Russia: Adam Ulam and Richard Pipes for the Soviet foundation, Stephen Kotkin for the modern era, Russian classics alongside contemporary authors for cultural texture.

How Bureaucracy is Breaking Government

American government is paralyzed not by too little regulation but by too many procedural constraints layered onto agencies themselves — each well-intentioned individually, collectively crippling.

The Paperwork Reduction Act illustrates the pattern. Passed to reduce form burdens, it created an OIRA review process requiring agencies to get approval before asking more than ten citizens the same question — turning basic user feedback for a Veterans Affairs web form into a six-to-eight-month ordeal. Inside agencies, even that hurdle rarely materializes: lawyers and privacy officers block teams from ever reaching OIRA, citing rules that have been superseded or were never real. The PRA's original coordination goal — spotting that 20 programs collect the same data under different definitions — goes unmet because OIRA's capacity is consumed by regulatory review, which always takes priority.

The broader architecture follows the same logic. FOIA, the Federal Advisory Committee Act, NEPA environmental review, and hard-look judicial review each targeted a real 1960s–70s excess — agencies bulldozing neighborhoods, oil spills, racial abuses. Left-wing judges led; the right followed for its own reasons. The resulting consensus committed governance to judicial oversight as its quality-control mechanism. Today that consensus makes it nearly impossible to permit a solar farm or an interstate transmission line.

The ratchet is one-way because procedural laws create constituencies — environmental groups depend on NEPA even where it blocks renewables — while harms fall on diffuse, unorganized victims: veterans with broken web portals, Medicaid enrollees who can't navigate benefits. Presidents face four-year terms, two-year appointees, and congressional staffers who have never run an agency. George W. Bush's President's Management Agenda was a rare exception; Indiana, Ohio, and Arkansas show what focused state-level executive leadership can accomplish.

Making agencies harder to operate does not make them do less — it makes them do what they do worse.

What Can the Brits Teach Us About State Capacity?

State capacity fails because policy and implementation are structurally severed. Laws descend through hierarchies and nothing feeds back: implementers can't flag perverse outcomes; ministers can't correct course. The UK's Government Digital Service, launched around 2011 under Francis Maude, was built to close that loop. GDS put developers, service designers, user researchers, and policy generalists on equal footing around shared outcomes — replacing the Whitehall tradition where technologists were "below stairs" and requirements got tossed over walls.

The proof is Universal Credit, a consolidation of working-age benefits. A top-down first attempt burned roughly $500 million over three years without a single user test. A hard reset brought in ~30 people who built end-to-end prototypes, tested with small populations, and scaled incrementally. During COVID, demand spiked 12x overnight; Universal Credit absorbed it while changing policy rules daily — the dog that didn't bark.

The American contrast is California's unemployment insurance. Processing a claim legally required 17 years of experience, so 5,000 Deloitte contractors hired for the pandemic backlog couldn't legally touch claims. Fraud logic flagged humans who mistyped middle initials while waving through bot-submitted applications with perfectly formatted stolen identities. The root cause isn't COBOL — it's accumulated mandates and a defensive culture where broken process is the only shield against blame. Legislators add constraints without subtracting them; agencies get hammered during crises and starved between them.

Jennifer Pahlka and Andrew Greenway's prescriptions: hire and exit on merit; fund digital work as products (iterative, experiment-based) not projects (big up-front requirements, outsourced builds); replace bimonthly oversight theater with incremental steering from senior officials; and reduce adversarial legalism — the veto-point density that makes blocking solar and housing easier than building them. USDS proved the model federally, then lost its direct appropriation to a pass-the-hat mechanism that undermined the independence producing results.

How to Build the '90s DOGE

Cutting headcount without simplifying the rules is why government downsizing fails — John Kamensky, Gore's eight-year deputy on the National Performance Review (NPR), applies that diagnosis to both the 1990s initiative and DOGE today.

The NPR ran 1993–2001, the longest federal reform effort in American history. Its bet: ask career civil servants rather than outside consultants what needed fixing, and focus on how government works rather than org charts. Gore's principle: "don't move boxes, fix what's inside them." The project started improvised — Kamensky, Bob Stone (DoD), and Bob Knisely pitched Gore in an accidental job interview, props in hand: a spray-paint can requiring a chemistry-degreed inspector to certify it hadn't expired; a steam trap leaking $50/week while procurement waited on bulk orders to save $10. Gore authorized credit cards for small purchases on the spot.

NPR eliminated over 100 programs and 250,000+ federal jobs. The workforce skewed older, junior hiring froze, and procurement expertise atrophied across two decades. The core failure: Congress cut headcount first and never simplified the rules, so fewer people faced the same maze. Contractors filled functions the government couldn't staff or legally delegate.

One big reinvention bill failed; reforms were threaded into annual appropriations. Wins included scrapping the 10,000-page personnel manual, franchise funds letting agencies competitively buy administrative services from each other, and credit cards for small purchases (CBO valued the rebates at $100 million when Congress tried to ban them). Also abolished: the Board of Tea Experts and Korean War-era mohair subsidies.

Two things NPR never solved: linking performance data to budget data to calculate unit costs, and building cross-agency career paths. Trust in government rose from ~20% to ~40% during the Clinton years. Kamensky's DOGE forecast: trust falls further, Social Security and immigration delivery fractures, and no replacement architecture exists.

There Are Too Many Judicial Injunctions

Federal courts are issuing too many nationwide injunctions against executive action, and a Democratic legal scholar — Michigan Governor Whitmer's former chief counsel Nick Bagley — thinks judges have badly overreached regardless of which party benefits.

Judicial review of agency action is largely a 20th-century invention. Before the Administrative Procedure Act of 1946, courts stayed out of executive operations. The APA created reviewability of "final agency action," and what was once narrow, party-specific relief has grown into nationwide injunctions that let a single district judge freeze federal policy across the country.

Two forces drove the expansion. First, a 1970s bipartisan consensus that courts should "ride herd" on agencies to protect the environment and civil rights. Second, the Federalist Society's campaign for more assertive courts limiting the administrative state — consolidated with the 2024 elimination of Chevron deference. Congressional dysfunction amplified both: a filibuster-paralyzed legislature forced presidents to act unilaterally, giving partisan judges more targets. The nationwide scope has a geographic explanation: most agencies sit in Washington, so D.C. Circuit precedent already applies nationwide; district judges elsewhere absorbed that norm.

Bagley distinguishes two kinds of review. Legal review — did the agency exceed its congressional mandate? — has a legitimate role; the NIH indirect-cost injunction is that kind of case. Arbitrary-and-capricious review is the problem: generalist judges second-guessing policy reasonableness forces agencies into thousand-page defensive records and paralyzes projects before they start.

His advice to Democrats: "Trump is temporary, administrative law is permanent." On reform, the REINS Act faces the paradox that any president who could sign it controls the bureaucracy and won't. Permitting reform is the likelier opening. Near-term Supreme Court involvement will come through the shadow docket — sprint appeals on injunctions — not fully litigated cases, which take two to three years.

How to Rewire City Hall

City halls fail not from bad intentions but from an outdated Fordist operating model — siloed, throughput-measured, built for standardized mass service — while mayors face pandemics, affordability crises, and migration requiring iterative, cross-sector problem-solving. The binding constraint: 50–75% of municipal budgets are fixed costs, leaving a thin slice for data infrastructure and strategic capacity — the first items cut in downturns. Bloomberg Philanthropies' Government Innovation Program has spent 15 years closing that gap, growing from five grants in 2010 to deep technical assistance reaching 900+ cities.

The program's core bet is internal capacity over outsourced expertise. The What Works Cities certification has moved hundreds of US cities up a data-sophistication ladder; 54% of cities over 100,000 are now progressing toward certification. The Bloomberg Harvard City Leadership Initiative trains 40 mayors annually to manage as problem-solvers first. Mayor Tim Kelly of Chattanooga arrived demanding answers; the program taught him to diagnose before acting. When his team traced youth crime to mental health rather than policing, targeted interventions produced a full year of zero murders in one high-crime neighborhood.

Baltimore under Brandon Scott is the anchor case. He built a homegrown COVID contact-tracing operation rather than hire management consultants, outperforming peers on every vaccination metric. A live chief-data-officer dashboard, a declared 15% year-on-year murder reduction target, and group violence reduction strategies drove homicides to 50-year lows — the fastest decline in the country. The same innovation team then tackled police retention, producing a net officer gain for the first time in a decade. None of it is paradigm-breaking; it's disciplined execution of known evidence-based methods, with the mayor's personal credibility staked publicly on the goal.

The smart cities movement failed by leading with solutions rather than problems. AI carries the same risk: mayors see a tool another city deployed and want to import it. The prescription is the same — get data governance clean first, promote staff fluency, experiment on low-risk administrative use cases, and move toward service delivery carefully. Mayors are almost uniquely free of Not Invented Here syndrome; their FOMO is a genuine policy-diffusion engine when paired with technical assistance to adapt, not merely copy, what worked elsewhere.

What’s Wrong with NIH Grants?

The NIH grant system's core dysfunction is hyper-competition: in the 1950s, 60% of applications were funded; today it's 10%. The structural cause is "soft money" — universities hire faculty they cannot pay, requiring researchers to fund their own salaries through grants. Scientists spend roughly 45% of their time on administrative work rather than research, competing for a 10–20% success rate on applications running 100–150 pages each.

The small-project model traces to a Rockefeller Foundation risk-management decision in the 1930s: facing financial pressure, they shifted from large institutional grants to short-term project grants. Alan Gregg, who ran the medical division, predicted it would turn the foundation into "a dispensary of chicken feed." When NIH adopted the same model, no careful policy comparison was made — it became the default. Congressional pressure in 1962, led by Rep. Lawrence Fountain, then layered in mandatory detailed oversight over scientific objections that pre-specifying five-year outcomes makes researchers "bureaucrats and astrologers."

Perverse incentives compound through indirect costs: grants pay not only salaries but overhead bonuses, so universities are financially rewarded for hiring grant-dependent faculty regardless of job prospects. The result is a bloated workforce where the average first independent grant arrives at age 45 — compared to the late 20s for a fully trained physician. Eliminating mandatory retirement in the early 1990s worsened this by keeping funded senior scientists in place, crowding out younger researchers.

The proposed fix is large block grants to universities — one bureaucratic unit replacing thousands — where institutions allocate funds internally and are judged retrospectively on scientific output rather than on promised deliverables. NIH's own intramural program demonstrates the model works: Barney Graham developed the protein-stabilization platform behind the RSV and COVID vaccines there precisely because he could take multi-year risks without annual funding uncertainty.

Does Anyone in Government Care About Productivity Growth?

No single office in the U.S. federal government owns productivity growth. Treasury focuses on crises; the CEA, NEC, and OMB operate near the president but none treats long-run science investment as a primary mandate; OSTP sits outside the White House and rarely penetrates core policy. The deeper problem, argues economist Ben Jones, is public and congressional indifference to the link between science and living standards. U.S. public R&D is near a 70-year low as a share of GDP; macro estimates suggest doubling it would generate social returns — in health, income, and national security — exceeding the cost.

The market failure is structural: a new scientific idea published in a journal earns its creator nothing while illuminating the path for everyone else. Private firms fund R&D when private returns are sufficient, but basic science — where direct private returns are zero — gets systematically undersupplied. R&D tax credits help, but they reach only applied market innovation and cannot replace public funding of open science that seeds future applications.

Knowledge accumulation deepens the problem. Jones's "Burden of Knowledge" thesis holds that as fields deepen, researchers specialize more narrowly and reach the frontier later — biomedical PIs receive their first independent NIH grant around age 44–45 on average. Teams now outperform solo researchers on high-impact work in virtually every field, but team authorship makes it harder for young scientists to signal independent merit — worsening a pipeline funding starvation already pinches.

Jones sees AI as the most plausible counterforce: a model trained across the full literature could recombine knowledge across domains in ways narrow experts cannot. Absent that, he favors grants targeting younger scientists, operational A/B experiments within existing programs, and open international collaboration — Fieldhouse and Mertens find DoD R&D generates lower spillovers than civilian funding because weapons findings stay classified.

How the UK Biobank Was Built

Building scientific infrastructure that won't pay off for a decade requires deliberately choosing to accomplish nothing in the short term — and saying so to funders. When Sir Rory Collins faced his ten-year review with the Wellcome Trust and MRC, he told them the project had produced, on that criterion, nothing. That was the correct answer.

UK Biobank recruited 500,000 British adults aged 40–69 between 2006 and 2010, collecting biological samples, questionnaire data, and physical measurements, then linked them to NHS health records for 30-year longitudinal tracking. The age window was deliberate: old enough that disease would appear within 15 years, young enough that risk factors could be measured before disease distorted them. Collins's watchword was "defer." He cancelled an early order for biochemistry analyzers on day one: wait until all 500,000 samples can be assayed simultaneously in randomized order with cheaper future technology, rather than pulling piecemeal case-control subsets. Genotyping the entire cohort — funded by the UK government through Affymetrix — put Biobank on the international map. Researchers who had been meta-analyzing studies of a few thousand people suddenly had half a million in one place.

The decisive design choice was open access with no researcher ownership. Collins published less on Biobank data than almost anyone. Every pound of MRC/Wellcome funding has leveraged roughly £12 of external investment — industry funded exome sequencing, whole-genome sequencing, and proteomic assays; government funded imaging of 100,000 participants. Last year, 5,000 peer-reviewed publications used the data. A parallel China Kadoorie Biobank was set up simultaneously to capture genetic diversity absent from the UK.

Collins's closing argument: the risks of withholding data are systematically underweighted against those of sharing it. Biobank has yielded tools for precision cancer screening and cardiovascular risk stratification — and the value of each such study compounds over decades.

How to Fix Foreign Aid

Foreign aid has always lacked the visible domestic payoff of Social Security or roads. What changed under DOGE wasn't public opinion — it was an administration that didn't share the underlying values. Dean Karlan, USAID's first-ever Chief Economist, argues the institution was genuinely flawed and genuinely worth saving, and that those two facts are rarely held together.

Karlan's office spent 2.5 years as a consulting unit inside a 13,000-person bureaucracy, redirecting $1.7 billion toward higher-effectiveness programs. About 150–170% of USAID funds arrived pre-earmarked by Congress (double-earmarking lets a single dollar satisfy two overlapping mandates), so the real lever was advising in-country mission staff within those constraints. The core method: benchmarking. In refugee livelihoods work, the benchmark is cash transfers — backed by 115 randomized trials — often in a "graduation model" combining a lump sum, income-generating training, and savings access. In primary education, the benchmark shifts to Teaching at the Right Level and structured pedagogy. Proposed innovations must beat the established benchmark, not merely beat doing nothing.

Internal resistance was mainly psychological: consensus among people sharing a goal tends to split the pie. Karlan countered with the full-population frame — 250,000 reachable out of 2 million eligible is a real choice about who gets left out.

DOGE's accounting error was treating spending cuts as savings without counting benefits foregone. Impact evaluations establish whether a program works; audits verify whether a proven program was delivered. DOGE ran neither.

Karlan's prescription for a rebuilt USAID: simpler programs at larger scale; heavy investment in local governance capacity (helping governments collect taxes and regulate banking creates health and education gains that dwarf direct aid); and a bipartisan core explicitly partitioned from ideologically contested margins, so future administrations can toggle the edges without dismantling the whole.

Ten Thoughts on Government Data

Government data is far messier and more incomplete than policymakers assume. Administrative records like DHS's SEVIS have major gaps — departure dates and employer addresses for working international students are absent most of the time — because these systems were built for bureaucratic tracking, not analysis. Errors persist unnoticed for months: in 2024 the US undercounted international students by 200,000. Much "counting" is sampling with built-in assumptions; misuse produced the spurious claim of two million new native-born jobs. Declining survey response rates and AI-generated spam are making imperfect administrative records comparatively more valuable. Unlocking any dataset requires practitioner knowledge — understanding why fields changed, which traces to regulations or memos that only career bureaucrats remember.

What is America's Infrastructure Cost Problem?

American infrastructure is expensive because of four compounding failures — permitting, procurement, personnel, and weak data — and NEPA gets disproportionate attention relative to the others.

Permitting matters most for above-ground linear projects like highways. A Detroit interstate segment illustrates the mechanism: early sections went up cheaply in the 1960s; a later NEPA-challenged segment costs several times more, with sound walls, below-grade construction, and parks built over the highway to settle lawsuits. The average environmental impact statement takes four years; median litigation runs 18 months even when plaintiffs lose. But underground urban projects like New York subways — the world's most expensive per mile — are barely touched by NEPA. Their costs trace to personnel and procurement instead.

Research by Liscow, Cailin Slattery, and Will Nober on California transportation found that a 1% increase in engineer retirements raises costs 4.5%, and improving engineer quality from the 25th to the 75th percentile produces savings three times the hire cost. State DOTs have shrunk workforces for two decades — mostly the skilled positions hardest to defend politically — then outsource at premium rates to the same private-sector engineers. Thin government capacity collapses bidder competition: the typical transportation bid draws three contractors, with almost no outreach to expand that pool.

Procurement is the most neglected driver. Many states must take the lowest-cost bid regardless of track record. European systems weigh quality factors at 75% and price at 25%. Repaving contracts that should be a dozen pages routinely run hundreds, driving out competition.

Better data — basic project cost, components, and timelines — would cost little and is largely missing, preventing rigorous study of which interventions work.

How to Salvage a Transit Project

Highway agencies receive formula funding from Congress and are built to construct; transit agencies are operators by DNA — 90% of budget on daily service — competing for FTA grants one project at a time. That gap explains most of what goes wrong.

Opening cost estimates are almost always fiction — lowballing is rational; a true number kills the conversation. Boston's Big Dig ran from $2 billion to $12 billion. Scope then creeps as stakeholders attach. The Green Line Extension, 4.7 miles into dense Somerville, was never going to cost under a billion — eight bridges had to be rebuilt because abutments left no room for a second track; 24 new trains cost $200 million; a new maintenance facility was needed. Procurement was rushed to lock in federal money before an administration change. By 2015 the project tracked $3–4 billion, with overruns on a transit authority in crisis after four blizzards shut the system down.

Salvage, led by Stephanie Pollack under Governor Baker, worked through hard constraints. A six-month deadline with real cancellation risk forced alignment. The old team was fired. Stations dropped from $50–70 million to simple platforms; all seven stations and 24 trains were kept — frequency is non-negotiable. Contingency was cut: large reserves act as spending targets. Risk retirable pre-contract (moving commuter rail tracks) was resolved first. The Control Board set a hard not-to-exceed price; bidders competed on how many unfunded items they could absorb within it. The winner covered all four, including the bike path, under the ceiling.

The four Ps — planning, permitting, public engagement, procurement — expand to fill decades, not because they're slow but because no go/no-go decision backed by real money has been made. States that decline federal funding to escape its regulatory burden reveal a system working against its own purpose.

How Cheaply Could We Build High-Speed Rail?

High-speed rail on the Northeast Corridor could be built for roughly $18 billion — about six times less than Amtrak's $117 billion Connect NEC estimate — while delivering Boston-to-New York and New York-to-DC trips under two hours each. Transit Costs Project lead author Alon Levy explains why the gap is so large.

The NEC's slowness comes from discrete bottlenecks, mostly sharp curves concentrated in southeastern Connecticut, plus scattered cant-angle restrictions elsewhere. Fix those and you reclaim half an hour cheaply. The biggest item in the Amtrak plan — condemning a Manhattan city block to add 7–12 tracks at Penn Station — is unnecessary: simulation shows the existing station handles doubled frequency. The only real constraint is the one tunnel between New Jersey and Manhattan, which the already-funded Gateway project resolves.

The deeper problem is inter-agency dysfunction. Amtrak and commuter operators — NJ Transit, Metro-North, Connecticut DOT, SEPTA, MBTA — each want separate infrastructure to avoid coordinating schedules. Every request for dedicated tracks is a symptom. The alternative is timetable-first planning: determine how many trains per hour each branch needs, then derive what infrastructure that requires. Germany built this discipline in the 1970s and achieves ridership matching France's despite slower trains.

The political economy is more tractable than it looks. All relevant agencies draw from the same Federal Railroad Administration pot; the FRA has authority to condition grants on coordination. Most states can be satisfied with projects already on their wishlists — New Jersey's three priorities (Portal Bridge, Hunter Flyover, Mid-Line Loop) are exactly what the proposal requires there.

The remaining obstacle is cultural: American rail planning doesn't start from timetables, ignores German engineering standards, and carries 50% soft-cost premiums over European equivalents — driven by low public-sector productivity and competition from an efficient private sector for the same white-collar labor.

When FAFSA Broke, They Called This Guy

Government software projects fail not because the goal was wrong but because Congress hard-codes implementation details, agencies lack technical leadership, and vendors go unaccountable until it's too late. The 2023 FAFSA disaster — a bipartisan simplification that should have cut the form from 100 questions to 36 and pulled IRS data automatically for 17 million annual users — illustrates all three at once.

Jeremy Singer, College Board president, was recruited by the White House in mid-2024 to replicate the healthcare.gov rescue. He diagnosed three compounding problems. First, Lamar Alexander's 2019 bills embedded UI language into statute — a homeless-student disclosure confused every applicant, and the team couldn't legally remove it. Second, Federal Student Aid had no effective CTO. Four vendors, led by General Dynamics IT with COBOL-era engineers, couldn't communicate across systems; getting all of them onto Slack was celebrated as a breakthrough. No one inside the department could hold them accountable for code quality or schedule. Third, political pressure suppressed honest status reporting — vendors were still building core functionality past the October 2023 launch date, and the department didn't know.

Singer's fix: a small cadre of technical experts, staged beta tests (hundreds → thousands → tens of thousands of users), and refusal to launch prematurely. His line — "two moms can't produce a baby in four and a half months" — drove the decision to push to November 2024. That launch succeeded; 1.7 million students became eligible for maximum Pell Grants.

The 2025 GAO audit, he argues, was compliance theater: documentation steps no competent engineer follows during a crisis, which would have made things worse. His prescriptions: write statute as use cases, not UI specs; hire 15–20 permanent senior technical staff rather than outsourcing architecture to Beltway contractors; keep a funded backup vendor as leverage over the primary.

Did the CHIPS "Everything Bagel"...Work?

The CHIPS Program Office allocated $34 billion in semiconductor incentives in under two and a half years, starting with zero staff, processes, or strategy in September 2022. That pace — closing 15–20 major deals in months — rivals what a top private-equity firm like KKR closes in a year under established conditions.

Director Mike Schmidt, CIO Todd Fisher, and COO Sara Meyers built every process from scratch: application portal, evaluation framework, investment committee, legal templates. They had $780 million in administrative funding and Secretary Raimondo as active deal-closer and lead recruiter.

The "Everything Bagel" critique — that the program was loaded with social side-requirements — overstates real friction. Workforce plans including childcare were statutorily required and consumed about 30 minutes per deal. The team deliberately chose a holistic over a scored evaluation framework: rigid point scoring would have made it impossible to close a nationally critical deal if a company fell short on a secondary metric. National and economic security was the explicit first priority; everything else was weighted at discretion.

The actual bottlenecks were commercial and legal. Leading-edge companies requested ~$75 billion collectively; the office awarded ~$25 billion, cutting requests by roughly two-thirds while negotiating milestone schedules simultaneously. Legal structure then consumed weeks in Taipei and DC: companies wanted contractual certainty that funds couldn't be clawed back over a 10-plus-year horizon. Japan's comparable program simply reimburses receipts with no binding contract — a stark contrast.

Davis-Bacon (1931 prevailing-wage law) created genuine friction: retroactively reconstructing wage records for up to 20,000 itinerant subcontractors per site, a problem never addressed at this scale. NEPA was eventually legislated away, but the team had built the compliance machine regardless.

The result: all five companies capable of producing leading-edge chips — TSMC, Samsung, Intel, Micron, SK Hynix — are expanding in the US, which now hosts more of them than any other country. By 2030 the US is on track to produce 20% of global leading-edge logic, up from zero in 2022.

How to Be a Good Intelligence Analyst

The central failure of American intelligence analysis is a communication breakdown between producers and consumers. The core dysfunction is timing: assessments are only actionable in a 2–3-week window. Warn 10 years out, nothing gets done; warn 24 hours out, same result. When surprise hits, the IC is always blamed first.

Resource allocation compounds this. Tunisia had roughly half a full-time analyst at CIA before the Arab Spring. When a self-immolating protester triggered the cascade that toppled Mubarak, the community was blamed for missing it — despite predicting Mubarak's fall for a decade. Timing was unpredictable. Budget politics favor satellites (subcontracts in 43 congressional districts) over human analysts and language training.

Technical collection delivers access, not truth. In every recorded Saddam-generals meeting, the generals lied about WMD programs to survive. Intercepted conversations would have been equally misleading. Political pressure to invade Iraq then distorted community analysis on top of that.

Good analysts are tenacious, obsessive about source quality, and keep score on their own predictions — Johnston's model profession is librarian. Expertise accumulates around 10,000 cases but creates tunnel vision: veterans miss orthogonal threats that novices catch because no one told them it was impossible. The career structure rewards generalist rotation over depth. The State Department's Bureau of Intelligence and Research, where analysts can spend an entire career on one account, is the model worth replicating.

Colin Powell was the exemplary consumer: tell me what you know, what you don't, then what you think. Rumsfeld and Wolfowitz exemplify the failure mode — prolonged success produces certainty of being smarter than everyone, so contrary evidence gets dismissed as noise. Structural fixes: give the DNI real budget authority, enable rapid surge reallocation, and narrow what gets classified — most secrets have a shelf life of days, and over-classification wastes the resources that matter.

How to Predict the Future

Improving forecast accuracy by one percentage point on multi-trillion-dollar decisions like Iraq would be worth tens of billions of dollars — yet society invests almost nothing in human judgment research. Jason Matheny, who directed IARPA 2015–2018 and now runs RAND, explains why.

The ARPA model works through three mechanisms most agencies lack: entrepreneurial program managers with real autonomy, parallel research tournaments, and disciplined defunding of lagging teams. A good IARPA-hard problem sits in the 5–50% success-probability band. Proposals use the Heilmeier questions. Matheny added a second layer Heilmeier omits: how fast could a competitor weaponize this if stolen, would we regret it if leaked, can defensive capabilities run in parallel? Red teaming is chronically skipped because managers resist having investments graded by outside brokers.

Phil Tetlock's 20-year study showed expert political judgment barely beats chance. IARPA's ACE program confirmed that averaging diverse crowd forecasts cancels random and systematic error, and that "superforecasters" — the consistently accurate subset — outperform any single expert. IARPA ran a crowdsourcing platform used by thousands of analysts 2009–2019; it faded without protected budget. The UK's Cosmic Bazaar survives. Robin Hanson's explanation: leaders resist crowdsourcing because it grades their homework — status matters more than accuracy.

On technology asymmetries: encryption and large language models favor democracies; cyberweapons and bioweapons likely do not, because open societies have larger attack surfaces and struggle to distribute countermeasures. The Internet's disinformation problem could have been partially designed away — packet-level attribution, credibility headers — had anyone red-teamed authoritarian use cases first. DNA synthesizers allow smallpox recreation for under $100,000 with no baked-in security. AI is the positive exception: major labs are building guardrails in from the start. The closing prescription: embed conditional forecasting tools at the NSC, where decisions happen under sleep deprivation with no time for coordinated intelligence products.

How Diplomacy Works in Africa

Africa is systematically under-resourced in US diplomacy relative to what national security interests require, while rivals extract growing trade and international-forum alignment through stronger engagement. The core deficit is "care and feeding" — senior-level attention. For too many African countries, the highest-ranking American they ever meet is their local ambassador; the Cote d'Ivoire foreign minister reached Washington repeatedly without meeting anyone above assistant secretary of state until James Baker needed a Security Council vote on the Gulf War.

George W. Bush stands alone in the post-Cold War era for African leader engagement, partly because Condoleezza Rice told him from day one that Africa would take real time. Kennedy devoted 25% of his foreign leader meetings to Africa. Obama had senior Cabinet officials (Susan Rice, Gayle Smith, Samantha Power) who cared, enabling Power Africa and the Ebola military response, though the administration was too preachy. Biden's team did less lecturing but also less engaging — Ukraine and Gaza consumed the bandwidth.

The Niger junta expulsion after the "condescending" US delegation visit illustrates how tone collapses delicate negotiations; a parallel delegation to Gabon's coup leaders, run with more flexibility, produced a two-year return to civilian rule. Engagement with difficult leaders gets blocked by double standards that don't apply elsewhere — the Zimbabwe photo-op choreography at the 2022 Africa Leaders Summit was "kabuki" for a domestic audience that likely didn't notice.

Most of the 60-plus presidential Africa initiatives since the 1990s fail because they're built backward — deliverable first, buy-in never. Durable programs (PEPFAR, Power Africa) came from extended stakeholder work before launch. The Biden-era 21PAS security program, forced on a resistant DOD via Jake Sullivan calling Secretary Austin, died when DOD redirected the money once that political pressure was gone.

Envoys work best with narrow negotiating mandates and real staff; they're mostly fixes for American bureaucratic problems, not African ones. The Africa Directorate folded into the Middle East NSC reduces the continent to a rounding error. The NSC's power is entirely a function of whether agencies believe it speaks for the president — when that belief holds, the coordination layer below cabinet level is where State, DOD, and Treasury hash out competing equities before decisions harden.

How to Rebuild the Arsenal of Democracy

America's defense industrial base fails because DoD spreads innovation money across thousands of small contracts that never reach production scale, while the five primes (Northrop, Lockheed, Raytheon, General Dynamics, Boeing) function as government subsidiaries on cost-plus contracts — all growing in low single digits, none genuinely competing. Trae Stephens (Anduril co-founder, Founders Fund) and Michael Kratsios (former U.S. CTO, now Scale AI) argue the fix requires picking winners from commercial entrants. The Collaborative Combat Aircraft contract — where the Air Force chose Anduril and General Atomics over Lockheed and Northrop — is a rare signal. Stephens's target: half of all DoD 5000-series programs going to tech entrants.

The manufacturing gap is acute: Ukraine showed the U.S. would exhaust munitions in eight days of great-power conflict. Stinger resupply required Raytheon to pull retired engineers back to restart shuttered lines. The barrier isn't cost — the skilled labor no longer exists domestically. Anduril's answer is Arsenal, a planned 5-million-square-foot campus modeled on Tesla's gigafactory: software-driven, modular, pivotable across product types.

The cost economics are damning: a $180,000 Iranian Shahed drone gets countered with $8 million in Patriot missiles. Iran's drone salvos were designed to drain response budgets, not cause damage. Ukraine's preference for cheap Chinese drones over expensive American systems followed the same logic. Attritability now dominates.

Kratsios frames the export side through Huawei: the U.S. had no alternative to offer allies asked to rip out Chinese telecom infrastructure. With AI it does — but the Export-Import Bank (80% Boeing aircraft financing) needs retooling to export software stacks. Chip controls are necessary but leaky: iFlytek accessed Nvidia A100s through a U.S. cloud provider despite the bans.

The underlying constraint is talent: government lacks technical depth for sound procurement decisions, and the fix requires technologists placed in sub-cabinet roles where policy is made.

Leninist Technocracy With Grand Opera Characteristics

China is a "Leninist technocracy with grand opera characteristics" — rationalist by training, campaign-driven by ideology, and prone to Wagnerian collapse when its logic runs to extremes. Dan Wang's *Breakneck* contrasts this with the United States, lawyerly since the Founding, where legal culture shifted after the 1960s from deal-making to litigation. Post-DDT, post-Moses lawyers turned from building railroads to suing the government — which explains why California can't finish high-speed rail and New York pays $2 billion per subway mile.

China's engineering state traces to Deng's overcorrection: by 2002, all nine Standing Committee members held Soviet-style engineering degrees. Since 2022, Xi has promoted military-industrial veterans who managed rockets and fighter-jet programs. The result: 32% of world manufacturing, with chokehold positions in rare earths, solar panels (90%), and key pharmaceuticals. Cadre promotion runs through the Organization Department's nomenklatura dossier — mayors built airports to goose GDP metrics for decades; Xi removed hard metrics, creating ambiguity around incentives.

The One-Child Policy illustrates social engineering's failure mode. Missile scientist Song Jian applied cybernetics math to population trajectories; from 1980–2013, the state carried out over 300 million abortions and hundreds of millions of forced sterilizations by official statistics. Now Beijing harangues women to reproduce — discovering coercion can suppress births but cannot induce them.

Wang's prescription: not engineers, not lawyers — economists, technocratic enough to execute and humanist enough to reject linear extrapolation; scientists on tap, not on top. On growth, he brackets the macro headwinds (debt, deflation, aging, property crisis) and focuses on the top 5% — EVs, AI, advanced manufacturing — which will keep deindustrializing Michigan and Germany regardless. He closes recommending James C. Scott's *The Art of Not Being Governed*, on highland peoples who planted cassava to evade the tax man — and confesses to dreaming of joining them.

Why the Two Parties Operate Differently

The two American parties differ not in ideology alone but in organizational structure: Democratic power flows upward from constituent groups, Republican power flows downward from personal connections to leaders. Political scientist Jo Freeman — attendee of every Democratic convention since 1964 — developed this framework from watching feminists navigate the two parties and finding the rules completely different.

In the Democratic Party, influence requires claiming to represent a group. In the Republican Party, it requires proximity to whoever is on top — Goldwater in 1964, Trump today. Republicans see themselves as insiders even out of power; Democrats as outsiders even when governing, a self-perception tied to the parties' class compositions, though both are shifting.

The 1976 Democratic convention was a turning point for the women's movement: feminists organized a caucus, technically lost their floor fight, but demonstrated mobilizing power. By 2024, the convention's 33 caucuses heard speeches and held receptions; the earlier culture of debating demands had evaporated.

Freeman closes on social movements: they are inherently unstable, never permanent. Once they acquire durable structure, they become interest groups — the fate of labor and feminism both.

How to Fix Crime in New York City

New York City's 1990s crime drop was a change in mission statement: the NYPD decided to care about crime, after two decades consumed by corruption scandals and racial unrest. Peter Moskos reconstructs the mechanisms through oral history with the officers who did it.

The baseline was catastrophic. The 1975 fiscal crisis laid off 5,000 cops in a day; the department retrenched into passivity. By 1981, over 100,000 robberies were reported annually. Processing one arrest consumed up to 24 hours, giving cops perverse incentives to chase overtime on low-level offenses. David Dinkins hired the officers who would fix the problem but gave them no mandate to fight crime.

The transformation began in the subways under Bill Bratton, then spread citywide when Giuliani appointed him NYPD Commissioner in 1994. His de facto second, Jack Maple — a Queens transit cop — asked a question nobody could answer: how many people were shot in 1993? (About 16 per day; nobody was counting.) Maple wrote the strategy on a napkin at Elaine's: timely intelligence, deploy where crime happens, effective tactics, relentless follow-up. CompStat made this real as weekly precinct-commander accountability meetings, where "What are you going to do?" became unacceptable.

August 1994 is the inflection: CompStat began tracking individual officer data and found most officers making zero arrests. Crime fell off a cliff. The squeegee men — only 70 individuals — had already been eliminated under Dinkins by enforcing Traffic Reg 44 every four hours until behavior changed. Bryant Park, Port Authority, and Times Square each required distinct problem-solving. The 80/20 rule ran through everything: small numbers of people and places drove most disorder.

A 2018 directive from de Blasio's homeless czar ended subway loitering enforcement; subway murders went from near-zero to ten in 2022. Public order requires constant political maintenance.

What’s Wrong with Nonprofits?

American nonprofits have lost public trust, and the sector has largely failed to reckon with why. Just over half of Americans express confidence in nonprofits — a yellow flag for institutions once greeted with reflexive goodwill. Volunteerism is declining and the number of donors is shrinking even as total donation dollars remain high. Greg Berman, who built the Center for Justice Innovation from a startup into an $80 million organization, argues this erosion is real and partly self-inflicted.

The backlash is a horseshoe: the left indicts nonprofits for racial disparities at the executive level, below-market wages for frontline workers, and for displacing government's proper service role. The right attacks viewpoint conformity and opacity. Both sides invoke the "nonprofit industrial complex," alleging elite funders perpetuate the sector's own interests over public benefit. Berman concedes truth in both but resists the corruption narrative: nonprofits haven't solved homelessness because nobody knows how, not because they prefer the problem to persist.

A sharper diagnosis: philanthropy shifted from funding discrete service delivery — programs like the Brooklyn Mental Health Court, which measurably reduced incarceration for 200 felony defendants a year — toward funding "intense policy demanders" pursuing transformational campaigns. Foundations chased big bets over proven programs, rewarding advocacy over service. This VC-style logic produced 501(c)(4) political shops, mission creep (the Sierra Club opining on Gaza), and staff cultures where social media made organizational discipline hard to maintain. Yuval Levin's frame applies: staff began treating institutions as platforms for self-positioning rather than molds that shape behavior.

The path back requires mission discipline — avoiding public positions unrelated to core work, using plain language aimed at the full public rather than a progressive in-group, and demonstrating measurable impact. Cross-partisan alliance-building matters too: opposition that reads as another left-right fight damages long-term credibility regardless of the merits.

Why We Don't Build Apartments for Families

American cities stopped building apartments for families because post-2008 financial incentives optimized for a different product. When Fannie Mae and Freddie Mac tightened lending, condo projects became nearly unbankable while rental apartments sailed through — tilting development toward young singles. Average new apartment square footage has fallen to 1994 levels; Austin now sees studios at 20–30% of new units.

The two-bedrooms that do get built are designed for roommates. Equal-sized bedrooms, two baths, no third room — the layout maximizes rent-per-square-foot against the studio baseline. A survey by Bobby Fijan's American Housing Corporation with the Institute for Family Studies found families overwhelmingly prefer the same square footage reconfigured into a larger primary bedroom plus two smaller children's rooms. Bedrooms ranked above yard, price, and space.

Several forces lock in the mismatch. ADA wheelchair turn-radius requirements make bathrooms irreducible below roughly 5×8 feet, so the shift to mandatory two-bath units eats space that pre-war buildings used for a third bedroom. Parking minimums assessed per bedroom rather than per unit directly penalize bedroom count. Institutional operators standardize kitchens across all unit types — the same full-size range and refrigerator in a studio as in a three-bedroom — interchangeability across a 250-unit complex cuts maintenance costs. Studios turn over at roughly double the rate of three-bedrooms, adding vacancy drag.

The deeper driver is finance. Most construction is funded by closed-end private equity with 2–3 year exit targets, making near-term rent maximization the only rational goal. Europe builds through public or fixed-profit vehicles on longer horizons; the comparison fails because the incentive structures are incompatible.

The fixes Fijan favors are subtractive: eliminate per-bedroom parking ratios, permit single-stair buildings (safe with modern sprinklers), and cut permitting friction. Pre-war floor plans — fewer bathrooms, more bedrooms, smaller kitchens — prove the model worked and could again.

Should the Feds Bail Out Chicago?

Chicago carries more pension debt than 43 U.S. states, has a worse credit rating than any other major American city, and 40% of its budget goes to debt and pensions. Seven of the ten worst-funded local pension systems in the country are in Illinois.

The legal architecture made this easy to build. States once treated pensions as a revocable "mere gratuity"; from the 1950s onward they elevated them to constitutionally protected contracts. Illinois adopted the California Rule — locking benefit terms from hire to retirement with no unilateral reduction. This placed pension obligations outside the debt limits governing bonds, making underfunding an attractive way to hide deficits. Governments compound the problem by assuming inflated returns (8–10.5%) to shrink required contributions, even though the legal obligation to pay is absolute. Chicago's funds sit at roughly 18% funded. Illinois legislators in the 1970s debated whether this dynamic would emerge and decided a mandatory savings rule was unnecessary. They were wrong.

The politics reinforce the fiscal failure through universal logrolls. The Illinois General Assembly recently passed — unanimously — a bill adding $11 billion in new first-responder pension liabilities, widening next year's deficit by over $1 billion. Suburban legislators don't block Chicago's requests because Chicago controls theirs. The benefit is immediate; the cost accrues over fifteen years.

When insolvency arrives, three responses exist: bailout, creditor loss, and austerity. A little of each is far less damaging than all of one. Detroit's bankruptcy shows the workable sequence: austerity, then creditor impairment, then a targeted bailout from foundations and the state — paid after the political cost was absorbed, limiting moral hazard. New York's 1975 crisis added mandatory honest accounting as a rescue condition. Both precedents point to the same principle: crisis is the moment to extract structural reform, not just stabilize the balance sheet.

Four Ways to Fix Government HR

More than 20 states have enacted civil service reforms so radical that none are on the table at the federal level — and they work. Texas, Florida, and Georgia moved to at-will employment, eliminated public-sector collective bargaining, and gave managers wide pay discretion. Fewer than 5-10% of managers recall any political influence on hiring, and state capacity metrics hold up.

Federal hiring runs through USA Jobs, where HR specialists keyword-search resumes before the hiring manager sees anyone — omit "manager" from a managerial application and you're screened out. Self-rating as "exceptional" improves hire odds regardless of reality. VA Secretary David Shulkin would recruit strong candidates in person; they vanished into the HR void and he received whoever passed the ranking process instead.

Pay locks onto a 15-grade General Schedule with 10 seniority steps each. Florida managers can broadband top performers 50% above benchmark; federal managers almost nowhere can. Pay compression means the lowest-skilled workers earn up to 40% above private-sector equivalents while post-graduate professionals earn less, structurally blocking elite recruitment.

Firing takes 6-12 months: performance improvement plan, then the Merit Systems Protection Board (employees win 20-30% of cases), then union arbitration, then federal courts. Air traffic controllers caught sleeping kept their positions. Managers assign poor performers to windowless basements and wait for them to quit.

Collective bargaining gained a statutory federal role only with the 1978 Civil Service Reform Act. Trump's redefinition of DoD, VA, and DHS positions as national security-related strips bargaining rights; courts have upheld it so far.

Near-term levers: make OPM an optional resource rather than a 300-page mandate, expand the bipartisan Chance to Compete Act's subject-matter-expert hiring model, and fix reduction-in-force rules so layoffs remove low performers rather than newest hires. A stultified civil service drives work to contractors — 40,000 officers handling $750 billion annually.

How Bureaucracy is Breaking Government

American government is paralyzed not by too little regulation but by too many procedural constraints layered onto agencies themselves — each well-intentioned individually, collectively crippling.

The Paperwork Reduction Act illustrates the pattern. Passed to reduce form burdens, it created an OIRA review process requiring agencies to get approval before asking more than ten citizens the same question — turning basic user feedback for a Veterans Affairs web form into a six-to-eight-month ordeal. Inside agencies, even that hurdle rarely materializes: lawyers and privacy officers block teams from ever reaching OIRA, citing rules that have been superseded or were never real. The PRA's original coordination goal — spotting that 20 programs collect the same data under different definitions — goes unmet because OIRA's capacity is consumed by regulatory review, which always takes priority.

The broader architecture follows the same logic. FOIA, the Federal Advisory Committee Act, NEPA environmental review, and hard-look judicial review each targeted a real 1960s–70s excess — agencies bulldozing neighborhoods, oil spills, racial abuses. Left-wing judges led; the right followed for its own reasons. The resulting consensus committed governance to judicial oversight as its quality-control mechanism. Today that consensus makes it nearly impossible to permit a solar farm or an interstate transmission line.

The ratchet is one-way because procedural laws create constituencies — environmental groups depend on NEPA even where it blocks renewables — while harms fall on diffuse, unorganized victims: veterans with broken web portals, Medicaid enrollees who can't navigate benefits. Presidents face four-year terms, two-year appointees, and congressional staffers who have never run an agency. George W. Bush's President's Management Agenda was a rare exception; Indiana, Ohio, and Arkansas show what focused state-level executive leadership can accomplish.

Making agencies harder to operate does not make them do less — it makes them do what they do worse.

Did the Courts Just Nuke Environmental Review?

The Council on Environmental Quality has been issuing binding environmental review regulations for nearly fifty years without statutory authority — and a 2024 D.C. Circuit ruling finally said so, though it may be struck before it changes anything.

NEPA, signed in 1970, was an internal executive-branch directive to consider environmental consequences before approving projects — it never mentioned courts. But the D.C. Circuit enforced it aggressively, and courts treated CEQ's 1978 regulations — issued under a Carter executive order, not statute — as judicially binding. Those rules introduced indirect and cumulative impact requirements that ballooned environmental impact statements to hundreds of pages and multi-year timelines. The Supreme Court, in all 17 NEPA cases heard, ruled for the government and told lower courts to demand less. Lower courts kept demanding more.

In *Marin Audubon Society v. FAA* — a helicopter-tour dispute over San Francisco national parks — Judge Randolph reached past the actual question to invalidate CEQ's rulemaking authority. The move was personal: as Deputy Solicitor General in 1976, he had argued to the Supreme Court that CEQ only issues guidelines, not rules. Neither party had raised the authority question, and both the FAA and environmental petitioners filed en banc petitions to strip that portion of the opinion. A companion Supreme Court case, *Seven County Infrastructure Coalition v. Eagle County*, involves a Utah railroad required to analyze downstream Gulf Coast refinery impacts — showing how far cumulative-effects doctrine had stretched.

All three professors agree CEQ likely never had binding rulemaking authority. The practical effect is uncertain: agencies must still follow CEQ rules as executive directives, and risk-averse agencies will keep writing thick EIS documents regardless. The larger opening is political — a chance to reset NEPA: narrowing reviewable impacts to direct harms, capping timelines, or eliminating injunctions as a remedy.

What Can the Brits Teach Us About State Capacity?

State capacity fails because policy and implementation are structurally severed. Laws descend through hierarchies and nothing feeds back: implementers can't flag perverse outcomes; ministers can't correct course. The UK's Government Digital Service, launched around 2011 under Francis Maude, was built to close that loop. GDS put developers, service designers, user researchers, and policy generalists on equal footing around shared outcomes — replacing the Whitehall tradition where technologists were "below stairs" and requirements got tossed over walls.

The proof is Universal Credit, a consolidation of working-age benefits. A top-down first attempt burned roughly $500 million over three years without a single user test. A hard reset brought in ~30 people who built end-to-end prototypes, tested with small populations, and scaled incrementally. During COVID, demand spiked 12x overnight; Universal Credit absorbed it while changing policy rules daily — the dog that didn't bark.

The American contrast is California's unemployment insurance. Processing a claim legally required 17 years of experience, so 5,000 Deloitte contractors hired for the pandemic backlog couldn't legally touch claims. Fraud logic flagged humans who mistyped middle initials while waving through bot-submitted applications with perfectly formatted stolen identities. The root cause isn't COBOL — it's accumulated mandates and a defensive culture where broken process is the only shield against blame. Legislators add constraints without subtracting them; agencies get hammered during crises and starved between them.

Jennifer Pahlka and Andrew Greenway's prescriptions: hire and exit on merit; fund digital work as products (iterative, experiment-based) not projects (big up-front requirements, outsourced builds); replace bimonthly oversight theater with incremental steering from senior officials; and reduce adversarial legalism — the veto-point density that makes blocking solar and housing easier than building them. USDS proved the model federally, then lost its direct appropriation to a pass-the-hat mechanism that undermined the independence producing results.

How to Beat Megafires

Megafires are solvable — the obstacle is not insufficient suppression capacity but a century of fire exclusion combined with regulatory systems that block the proven remedy: mechanical thinning followed by prescribed fire. That combination reduces wildfire intensity on forested landscapes by 70%; without it, overly dense, drought-stressed forests overwhelm any firefighting force.

The LA fires are a distinct type: wildland-urban-interface blazes that become structure-to-structure conflagrations once they enter neighborhoods. Six of the seven most destructive California fires in a century occurred after 2020. The 2020 fire season alone wiped out 20 years of the state's industrial emissions reductions under cap-and-trade.

Two overlapping failures explain this. First, a federal permitting bottleneck: 40% of Forest Service national-forest staff time goes to planning, assessment, and permitting. Environmental impact statements average five years; even the streamlined "categorical exclusion" takes 180 days. Forests have burned while still under review — the 1999 Six Rivers fire is one documented case. The Clean Air Act compounds this by counting prescribed-burn smoke against air-quality limits while wildfire smoke is exempt, penalizing the safer choice.

Second, agency fragmentation: the Forest Service (under Agriculture's commodity mandate), the National Park Service, Cal Fire, and county and water agencies each manage separate landscapes with no shared risk picture. Some environmental groups add friction — suing to block thinning on philosophical grounds or from distrust dating to the 1980s clear-cutting era.

The proposed fixes are three: landscape-scale treatment ($50 billion over 10 years by the Forest Service's own estimate; current federal spending is roughly 10% of that); community hardening; and real-time ignition detection using technology the Defense Department already operates globally but hasn't deployed for civilian wildfire. A bipartisan Sheehy–Padilla bill would create a unified federal wildland firefighting service within Interior to consolidate what is now split across conflicting agency mandates.

There Are Too Many Judicial Injunctions

Federal courts are issuing too many nationwide injunctions against executive action, and a Democratic legal scholar — Michigan Governor Whitmer's former chief counsel Nick Bagley — thinks judges have badly overreached regardless of which party benefits.

Judicial review of agency action is largely a 20th-century invention. Before the Administrative Procedure Act of 1946, courts stayed out of executive operations. The APA created reviewability of "final agency action," and what was once narrow, party-specific relief has grown into nationwide injunctions that let a single district judge freeze federal policy across the country.

Two forces drove the expansion. First, a 1970s bipartisan consensus that courts should "ride herd" on agencies to protect the environment and civil rights. Second, the Federalist Society's campaign for more assertive courts limiting the administrative state — consolidated with the 2024 elimination of Chevron deference. Congressional dysfunction amplified both: a filibuster-paralyzed legislature forced presidents to act unilaterally, giving partisan judges more targets. The nationwide scope has a geographic explanation: most agencies sit in Washington, so D.C. Circuit precedent already applies nationwide; district judges elsewhere absorbed that norm.

Bagley distinguishes two kinds of review. Legal review — did the agency exceed its congressional mandate? — has a legitimate role; the NIH indirect-cost injunction is that kind of case. Arbitrary-and-capricious review is the problem: generalist judges second-guessing policy reasonableness forces agencies into thousand-page defensive records and paralyzes projects before they start.

His advice to Democrats: "Trump is temporary, administrative law is permanent." On reform, the REINS Act faces the paradox that any president who could sign it controls the bureaucracy and won't. Permitting reform is the likelier opening. Near-term Supreme Court involvement will come through the shadow docket — sprint appeals on injunctions — not fully litigated cases, which take two to three years.

How to Rebuild the Arsenal of Democracy

America's defense industrial base fails because DoD spreads innovation money across thousands of small contracts that never reach production scale, while the five primes (Northrop, Lockheed, Raytheon, General Dynamics, Boeing) function as government subsidiaries on cost-plus contracts — all growing in low single digits, none genuinely competing. Trae Stephens (Anduril co-founder, Founders Fund) and Michael Kratsios (former U.S. CTO, now Scale AI) argue the fix requires picking winners from commercial entrants. The Collaborative Combat Aircraft contract — where the Air Force chose Anduril and General Atomics over Lockheed and Northrop — is a rare signal. Stephens's target: half of all DoD 5000-series programs going to tech entrants.

The manufacturing gap is acute: Ukraine showed the U.S. would exhaust munitions in eight days of great-power conflict. Stinger resupply required Raytheon to pull retired engineers back to restart shuttered lines. The barrier isn't cost — the skilled labor no longer exists domestically. Anduril's answer is Arsenal, a planned 5-million-square-foot campus modeled on Tesla's gigafactory: software-driven, modular, pivotable across product types.

The cost economics are damning: a $180,000 Iranian Shahed drone gets countered with $8 million in Patriot missiles. Iran's drone salvos were designed to drain response budgets, not cause damage. Ukraine's preference for cheap Chinese drones over expensive American systems followed the same logic. Attritability now dominates.

Kratsios frames the export side through Huawei: the U.S. had no alternative to offer allies asked to rip out Chinese telecom infrastructure. With AI it does — but the Export-Import Bank (80% Boeing aircraft financing) needs retooling to export software stacks. Chip controls are necessary but leaky: iFlytek accessed Nvidia A100s through a U.S. cloud provider despite the bans.

The underlying constraint is talent: government lacks technical depth for sound procurement decisions, and the fix requires technologists placed in sub-cabinet roles where policy is made.

One Year of Trump’s Economic Statecraft

Economic statecraft — punitive tools (sanctions, tariffs, export controls) and positive tools (equity, offtake agreements, price floors) — has become the default US foreign-policy instrument; sanctioned entities are up tenfold since 2000. Three Biden-era insiders assess Trump's year-one record.

Venezuela marks a qualitatively new category. Sanctions and ship seizures failed: a ghost fleet outside US jurisdiction routed ~80% of Venezuelan oil to China regardless. Rather than pressure China — which would have wrecked a trade deal — Trump seized Maduro, froze oil revenues in US accounts, and gave Rubio spending discretion. Production runs at ~900,000 barrels/day versus a 2–3 million pre-Chávez peak; recovery needs $100B+ in capex. Energy majors won't commit on a horizon longer than Trump's term, yet the administration presses creditors to fund reconstruction now or forfeit claims.

On tariffs, the effective US rate spiked from 2.5% to 28% post-Liberation Day before retreating to ~18% — highest since Smoot-Hawley. About 80% of $300 billion passed to consumers; the bottom quintile took a ~6–8% real-income hit, four times the top decile's. Manufacturing shed ~72,000 jobs. China ran a $1.2 trillion manufactured-goods surplus, grew exports to Europe by double digits and Africa ~30%, holding 70–80% share in EVs, batteries, solar, drones, and shipbuilding. Singh calls H200 chip licensing a "seismic error": China is bridging to Huawei's chip by 2027, not locking into Nvidia.

The MP Materials deal — DPA Title III equity stake, price floor on neodymium-praseodymium, offtake agreement, multi-billion loan — is the benchmark. The deeper problem: national champions don't fix the pricing benchmarks China controls, leaving Western producers uninvestable.

Three prescriptions: a doctrine limiting punitive-tool use; a Reconstruction Finance Corporation-style strategic investment fund with Senate-confirmed leadership; and a Department of Economic Security consolidating tariffs, sanctions, export controls, and positive tools split across USTR, Treasury, State, and Commerce.

Did the CHIPS "Everything Bagel"...Work?

The CHIPS Program Office allocated $34 billion in semiconductor incentives in under two and a half years, starting with zero staff, processes, or strategy in September 2022. That pace — closing 15–20 major deals in months — rivals what a top private-equity firm like KKR closes in a year under established conditions.

Director Mike Schmidt, CIO Todd Fisher, and COO Sara Meyers built every process from scratch: application portal, evaluation framework, investment committee, legal templates. They had $780 million in administrative funding and Secretary Raimondo as active deal-closer and lead recruiter.

The "Everything Bagel" critique — that the program was loaded with social side-requirements — overstates real friction. Workforce plans including childcare were statutorily required and consumed about 30 minutes per deal. The team deliberately chose a holistic over a scored evaluation framework: rigid point scoring would have made it impossible to close a nationally critical deal if a company fell short on a secondary metric. National and economic security was the explicit first priority; everything else was weighted at discretion.

The actual bottlenecks were commercial and legal. Leading-edge companies requested ~$75 billion collectively; the office awarded ~$25 billion, cutting requests by roughly two-thirds while negotiating milestone schedules simultaneously. Legal structure then consumed weeks in Taipei and DC: companies wanted contractual certainty that funds couldn't be clawed back over a 10-plus-year horizon. Japan's comparable program simply reimburses receipts with no binding contract — a stark contrast.

Davis-Bacon (1931 prevailing-wage law) created genuine friction: retroactively reconstructing wage records for up to 20,000 itinerant subcontractors per site, a problem never addressed at this scale. NEPA was eventually legislated away, but the team had built the compliance machine regardless.

The result: all five companies capable of producing leading-edge chips — TSMC, Samsung, Intel, Micron, SK Hynix — are expanding in the US, which now hosts more of them than any other country. By 2030 the US is on track to produce 20% of global leading-edge logic, up from zero in 2022.

How to Produce a Kamikaze Drone

Getting a new weapon into soldiers' hands requires bypassing acquisition risk aversion through creative legal interpretation, not rule-breaking. Chris Anderson, a former US Army Asymmetric Warfare Group (AWG) program manager, explains how his team shepherded the AeroVironment Switchblade — America's first tactical loitering munition — from concept to Afghanistan deployment around 2010.

The AWG was created by General Schoomaker in 2006 to telescope into the battlefield and force faster adaptation. It paired with the Rapid Equipping Force, which could buy up to $50 million in off-the-shelf gear outside standard procurement timelines. The decisive bureaucratic move was getting the Switchblade classified as a direct-fire weapon by the Maneuver Center of Excellence. Because the operator flying the drone is the same one who pulls the trigger — with the ability to wave off at the last second — it was legally analogous to a scoped rifle, requiring no clearance of fire. That mattered: Taliban mortar teams fired two or three rounds and scattered in under a minute.

The Switchblade weighed under six pounds, launched from a tube, flew for 12-15 minutes, and carried a CL-20 warhead with adjustable tungsten fragmentation — dialing lethal radius from 20 meters down to 2 meters to kill one individual in a group. Three concepts of employment: base defense against counter-battery targets, direct assault with drones loitering overhead as infantry attacked, and reactive maneuver when a patrol was ambushed.

Ukraine confirmed one-way attack drones kill generals and destroy command posts routinely. The cost problem: $25,000 Shaheed drones are being shot down with $500,000 missiles — unsustainable at scale. Testing at Saudi Arabia's Red Sands targets $2,000-per-kill solutions using legacy guns made smart enough to track cheap drones autonomously. The AWG was shuttered in peacetime for budget reasons; the next major war will require recreating it from scratch.

How to Fix a Department's Funding Tools

R&D contracting tools don't work for deployment, and the Department of Energy spent most of its history with no alternative.

DOE was founded in 1977 out of the Atomic Energy Commission and the Arab-oil-embargo-era Federal Energy Administration, then retreated into R&D under Reagan. Commercialization returned only in 2005, when the Energy Policy Act authorized the Loan Programs Office — which went dormant after Solyndra. The Bipartisan Infrastructure Law and Inflation Reduction Act injected $65 billion in grant authority, forcing a reckoning.

Two problems blocked private co-investment. Federal property law gave DOE a running lien on grant-funded projects until equipment fell below $7,500, scaring off bank co-lenders. The Consolidated Appropriations Act of 2023 let the Secretary vest title at project completion, removing the lien while retaining IP protections.

The contractual problem was harder. Seeding a hydrogen or nuclear market requires contracting with developer, offtaker, and fuel supplier on terms that can't be set in advance. Cooperative agreements assume one counterparty; Other Transactions Authority (OTA) allows a blank-sheet contract. Between 2011 and 2014, DOE executed fewer than 20 OT agreements while NASA executed roughly 10,000. Few DOE officers were warranted to sign one.

Congress granted DOE OTA in 2005, but DOE copied DOD's military-prototyping rules — wrong for energy markets. One word, "similar," in the 1977 Organization Act kept the broader authority dormant. A working group launched the day after the IRA was signed found 1977 testimony from John Dingell confirming the authority was modeled on the Space Act and "very, very broad." General Counsel Sam Walsh formalized that in summer 2023; DOE updated its OT regulations in early 2025.

Hydrogen demand-side and Gen III+ nuclear solicitations then invoked OTA for multi-party awards. DOE published a public OT guide and trained more warranted contracting officers — authority without institutional capacity changes nothing.

What’s Wrong with NIH Grants?

The NIH grant system's core dysfunction is hyper-competition: in the 1950s, 60% of applications were funded; today it's 10%. The structural cause is "soft money" — universities hire faculty they cannot pay, requiring researchers to fund their own salaries through grants. Scientists spend roughly 45% of their time on administrative work rather than research, competing for a 10–20% success rate on applications running 100–150 pages each.

The small-project model traces to a Rockefeller Foundation risk-management decision in the 1930s: facing financial pressure, they shifted from large institutional grants to short-term project grants. Alan Gregg, who ran the medical division, predicted it would turn the foundation into "a dispensary of chicken feed." When NIH adopted the same model, no careful policy comparison was made — it became the default. Congressional pressure in 1962, led by Rep. Lawrence Fountain, then layered in mandatory detailed oversight over scientific objections that pre-specifying five-year outcomes makes researchers "bureaucrats and astrologers."

Perverse incentives compound through indirect costs: grants pay not only salaries but overhead bonuses, so universities are financially rewarded for hiring grant-dependent faculty regardless of job prospects. The result is a bloated workforce where the average first independent grant arrives at age 45 — compared to the late 20s for a fully trained physician. Eliminating mandatory retirement in the early 1990s worsened this by keeping funded senior scientists in place, crowding out younger researchers.

The proposed fix is large block grants to universities — one bureaucratic unit replacing thousands — where institutions allocate funds internally and are judged retrospectively on scientific output rather than on promised deliverables. NIH's own intramural program demonstrates the model works: Barney Graham developed the protein-stabilization platform behind the RSV and COVID vaccines there precisely because he could take multi-year risks without annual funding uncertainty.

How to Predict the Future

Improving forecast accuracy by one percentage point on multi-trillion-dollar decisions like Iraq would be worth tens of billions of dollars — yet society invests almost nothing in human judgment research. Jason Matheny, who directed IARPA 2015–2018 and now runs RAND, explains why.

The ARPA model works through three mechanisms most agencies lack: entrepreneurial program managers with real autonomy, parallel research tournaments, and disciplined defunding of lagging teams. A good IARPA-hard problem sits in the 5–50% success-probability band. Proposals use the Heilmeier questions. Matheny added a second layer Heilmeier omits: how fast could a competitor weaponize this if stolen, would we regret it if leaked, can defensive capabilities run in parallel? Red teaming is chronically skipped because managers resist having investments graded by outside brokers.

Phil Tetlock's 20-year study showed expert political judgment barely beats chance. IARPA's ACE program confirmed that averaging diverse crowd forecasts cancels random and systematic error, and that "superforecasters" — the consistently accurate subset — outperform any single expert. IARPA ran a crowdsourcing platform used by thousands of analysts 2009–2019; it faded without protected budget. The UK's Cosmic Bazaar survives. Robin Hanson's explanation: leaders resist crowdsourcing because it grades their homework — status matters more than accuracy.

On technology asymmetries: encryption and large language models favor democracies; cyberweapons and bioweapons likely do not, because open societies have larger attack surfaces and struggle to distribute countermeasures. The Internet's disinformation problem could have been partially designed away — packet-level attribution, credibility headers — had anyone red-teamed authoritarian use cases first. DNA synthesizers allow smallpox recreation for under $100,000 with no baked-in security. AI is the positive exception: major labs are building guardrails in from the start. The closing prescription: embed conditional forecasting tools at the NSC, where decisions happen under sleep deprivation with no time for coordinated intelligence products.

How the UK Biobank Was Built

Building scientific infrastructure that won't pay off for a decade requires deliberately choosing to accomplish nothing in the short term — and saying so to funders. When Sir Rory Collins faced his ten-year review with the Wellcome Trust and MRC, he told them the project had produced, on that criterion, nothing. That was the correct answer.

UK Biobank recruited 500,000 British adults aged 40–69 between 2006 and 2010, collecting biological samples, questionnaire data, and physical measurements, then linked them to NHS health records for 30-year longitudinal tracking. The age window was deliberate: old enough that disease would appear within 15 years, young enough that risk factors could be measured before disease distorted them. Collins's watchword was "defer." He cancelled an early order for biochemistry analyzers on day one: wait until all 500,000 samples can be assayed simultaneously in randomized order with cheaper future technology, rather than pulling piecemeal case-control subsets. Genotyping the entire cohort — funded by the UK government through Affymetrix — put Biobank on the international map. Researchers who had been meta-analyzing studies of a few thousand people suddenly had half a million in one place.

The decisive design choice was open access with no researcher ownership. Collins published less on Biobank data than almost anyone. Every pound of MRC/Wellcome funding has leveraged roughly £12 of external investment — industry funded exome sequencing, whole-genome sequencing, and proteomic assays; government funded imaging of 100,000 participants. Last year, 5,000 peer-reviewed publications used the data. A parallel China Kadoorie Biobank was set up simultaneously to capture genetic diversity absent from the UK.

Collins's closing argument: the risks of withholding data are systematically underweighted against those of sharing it. Biobank has yielded tools for precision cancer screening and cardiovascular risk stratification — and the value of each such study compounds over decades.

How to Save Science Funding

The sticker indirect-cost rates that drew outrage when the Trump administration proposed a 15% federal cap are not what universities actually receive — and the institutions hit hardest are those whose research most directly feeds drug development.

Indirect costs cover fixed expenses — lab space, shared equipment, biosafety, regulatory compliance — that can't be tied to any single grant. Universities negotiate rates every two to four years; many exceed 50–70%. Pierre Azoulay (MIT Sloan), Dan Gross (Duke Fuqua), and Bhaven Sampat (Johns Hopkins) assembled data on 350 institutions covering 85% of NIH funding and found that effective rates have held roughly flat for 40 years at around 40% of direct costs. NIH rules exclude training grants and cap which direct costs enter the reimbursement base, explaining most of the gap between sticker and real rates. The administrative share of ICR has been legally capped at 26% of direct costs since 1991 — claims that rising headline rates reflect university bloat are inconsistent with that constraint.

The administration's "most-favored nation" argument — foundations pay 15% and universities accept it, so federal rates should match — misreads the picture. Foundation grants often reclassify overhead items as direct costs, and foundations fund only 6% of university R&D against the federal government's 55%; they cannot substitute at scale.

Most damaging to the 15% proposal: institutions facing the largest cuts generate the most pharmaceutical patents. Ozempic traces to Joel Habener's GLP-1 research at Massachusetts General, NIH-funded from 1979 to 2008. Cutting fixed-cost infrastructure doesn't spare commercially irrelevant science — it hits exactly what the private sector later builds on. Greater transparency into how indirect funds are spent remains the most defensible reform path.

Does Anyone in Government Care About Productivity Growth?

No single office in the U.S. federal government owns productivity growth. Treasury focuses on crises; the CEA, NEC, and OMB operate near the president but none treats long-run science investment as a primary mandate; OSTP sits outside the White House and rarely penetrates core policy. The deeper problem, argues economist Ben Jones, is public and congressional indifference to the link between science and living standards. U.S. public R&D is near a 70-year low as a share of GDP; macro estimates suggest doubling it would generate social returns — in health, income, and national security — exceeding the cost.

The market failure is structural: a new scientific idea published in a journal earns its creator nothing while illuminating the path for everyone else. Private firms fund R&D when private returns are sufficient, but basic science — where direct private returns are zero — gets systematically undersupplied. R&D tax credits help, but they reach only applied market innovation and cannot replace public funding of open science that seeds future applications.

Knowledge accumulation deepens the problem. Jones's "Burden of Knowledge" thesis holds that as fields deepen, researchers specialize more narrowly and reach the frontier later — biomedical PIs receive their first independent NIH grant around age 44–45 on average. Teams now outperform solo researchers on high-impact work in virtually every field, but team authorship makes it harder for young scientists to signal independent merit — worsening a pipeline funding starvation already pinches.

Jones sees AI as the most plausible counterforce: a model trained across the full literature could recombine knowledge across domains in ways narrow experts cannot. Absent that, he favors grants targeting younger scientists, operational A/B experiments within existing programs, and open international collaboration — Fieldhouse and Mertens find DoD R&D generates lower spillovers than civilian funding because weapons findings stay classified.

How to Bring Down Healthcare Costs

American healthcare's core problem is affordability, worsened by demographics: a fertility rate of 1.6 means an aging population will demand more Medicare spending precisely as the tax base shrinks. The US has plateaued at 20% of GDP on healthcare for a decade — not obviously excessive, since healthcare may be a good people spend proportionally more on as income rises — but demographic shift plus embedded inefficiency makes the trajectory unsustainable.

Anup Malani, the first Chief Economist at CMS ($2 trillion in annual federal health spending), maps the levers. In Medicaid, states don't fully internalize costs because the federal government covers the majority share — a dinner-party incentive to over-order. In traditional Medicare, fee-for-service produced cost-plus billing, then prospective pay, then Accountable Care Organizations shifting risk to providers. In Medicare Advantage, the government pays risk-adjusted premiums to private insurers — but the formula is manipulable: insurers can inflate a patient's risk score via extra diagnostic tests this year, then collect higher premiums next year. The administration is tightening the formula while raising base reimbursement rates, on the theory that genuine population costs are rising even as fraudulent inflation is removed.

On fraud, the shift is from retrospective False Claims Act suits to ex ante detection — flagging providers billing for 5,000 hours per year before payment goes out.

On drug prices, cutting US payments doesn't mechanically reduce R&D: savings may come from Pharmacy Benefit Manager margins, or R&D burden shifts to foreign payers. Operation Warp Speed is the model for advance market commitments when the government knows what it wants; patents serve when it doesn't.

The dynamic cost-bending priority is health data interoperability — common identifiers, encryption standards, machine-readable records — so AI can work across claims and EHR data, enabling AI scribes, cross-hospital record sharing, and LLM-assisted diagnosis.

How to Fix Crime in New York City

New York City's 1990s crime drop was a change in mission statement: the NYPD decided to care about crime, after two decades consumed by corruption scandals and racial unrest. Peter Moskos reconstructs the mechanisms through oral history with the officers who did it.

The baseline was catastrophic. The 1975 fiscal crisis laid off 5,000 cops in a day; the department retrenched into passivity. By 1981, over 100,000 robberies were reported annually. Processing one arrest consumed up to 24 hours, giving cops perverse incentives to chase overtime on low-level offenses. David Dinkins hired the officers who would fix the problem but gave them no mandate to fight crime.

The transformation began in the subways under Bill Bratton, then spread citywide when Giuliani appointed him NYPD Commissioner in 1994. His de facto second, Jack Maple — a Queens transit cop — asked a question nobody could answer: how many people were shot in 1993? (About 16 per day; nobody was counting.) Maple wrote the strategy on a napkin at Elaine's: timely intelligence, deploy where crime happens, effective tactics, relentless follow-up. CompStat made this real as weekly precinct-commander accountability meetings, where "What are you going to do?" became unacceptable.

August 1994 is the inflection: CompStat began tracking individual officer data and found most officers making zero arrests. Crime fell off a cliff. The squeegee men — only 70 individuals — had already been eliminated under Dinkins by enforcing Traffic Reg 44 every four hours until behavior changed. Bryant Park, Port Authority, and Times Square each required distinct problem-solving. The 80/20 rule ran through everything: small numbers of people and places drove most disorder.

A 2018 directive from de Blasio's homeless czar ended subway loitering enforcement; subway murders went from near-zero to ten in 2022. Public order requires constant political maintenance.

How to Fix Foreign Aid

Foreign aid has always lacked the visible domestic payoff of Social Security or roads. What changed under DOGE wasn't public opinion — it was an administration that didn't share the underlying values. Dean Karlan, USAID's first-ever Chief Economist, argues the institution was genuinely flawed and genuinely worth saving, and that those two facts are rarely held together.

Karlan's office spent 2.5 years as a consulting unit inside a 13,000-person bureaucracy, redirecting $1.7 billion toward higher-effectiveness programs. About 150–170% of USAID funds arrived pre-earmarked by Congress (double-earmarking lets a single dollar satisfy two overlapping mandates), so the real lever was advising in-country mission staff within those constraints. The core method: benchmarking. In refugee livelihoods work, the benchmark is cash transfers — backed by 115 randomized trials — often in a "graduation model" combining a lump sum, income-generating training, and savings access. In primary education, the benchmark shifts to Teaching at the Right Level and structured pedagogy. Proposed innovations must beat the established benchmark, not merely beat doing nothing.

Internal resistance was mainly psychological: consensus among people sharing a goal tends to split the pie. Karlan countered with the full-population frame — 250,000 reachable out of 2 million eligible is a real choice about who gets left out.

DOGE's accounting error was treating spending cuts as savings without counting benefits foregone. Impact evaluations establish whether a program works; audits verify whether a proven program was delivered. DOGE ran neither.

Karlan's prescription for a rebuilt USAID: simpler programs at larger scale; heavy investment in local governance capacity (helping governments collect taxes and regulate banking creates health and education gains that dwarf direct aid); and a bipartisan core explicitly partitioned from ideologically contested margins, so future administrations can toggle the edges without dismantling the whole.

What’s Wrong with Nonprofits?

American nonprofits have lost public trust, and the sector has largely failed to reckon with why. Just over half of Americans express confidence in nonprofits — a yellow flag for institutions once greeted with reflexive goodwill. Volunteerism is declining and the number of donors is shrinking even as total donation dollars remain high. Greg Berman, who built the Center for Justice Innovation from a startup into an $80 million organization, argues this erosion is real and partly self-inflicted.

The backlash is a horseshoe: the left indicts nonprofits for racial disparities at the executive level, below-market wages for frontline workers, and for displacing government's proper service role. The right attacks viewpoint conformity and opacity. Both sides invoke the "nonprofit industrial complex," alleging elite funders perpetuate the sector's own interests over public benefit. Berman concedes truth in both but resists the corruption narrative: nonprofits haven't solved homelessness because nobody knows how, not because they prefer the problem to persist.

A sharper diagnosis: philanthropy shifted from funding discrete service delivery — programs like the Brooklyn Mental Health Court, which measurably reduced incarceration for 200 felony defendants a year — toward funding "intense policy demanders" pursuing transformational campaigns. Foundations chased big bets over proven programs, rewarding advocacy over service. This VC-style logic produced 501(c)(4) political shops, mission creep (the Sierra Club opining on Gaza), and staff cultures where social media made organizational discipline hard to maintain. Yuval Levin's frame applies: staff began treating institutions as platforms for self-positioning rather than molds that shape behavior.

The path back requires mission discipline — avoiding public positions unrelated to core work, using plain language aimed at the full public rather than a progressive in-group, and demonstrating measurable impact. Cross-partisan alliance-building matters too: opposition that reads as another left-right fight damages long-term credibility regardless of the merits.

How to Fix Risk Assessment in Child Welfare

Human screeners are poor at risk estimation, yet child welfare agencies make 40+ high-stakes triage calls a day on fragmentary information. Allegheny County addresses this with the Allegheny Family Screening Tool (AFST), a machine learning algorithm at CPS intake predicting whether a judge will order child removal within two years. It draws on 25 years of integrated county records and scores cases alongside human judgment; staff retain override authority.

The early fear was algorithmic racial bias. Researcher Katherine Rittenhouse found the opposite: AFST eliminates roughly 75% of the black-white gap in removal rates among screened-in cases. Predicted removal risk also correlates with hospitalization for maltreatment injuries.

The integration traces to merging child welfare, behavioral health, housing, and aging under one department rather than siloed agencies — a decision made 25 years ago. Internal analytics, data science, and engineering teams replaced vendor dependency.

The same data revealed that 5,000 people subject to involuntary psychiatric commitment — 2% of Medicaid enrollees — account for 25% of behavioral health spending and face an 8% one-year mortality rate. Research by Pim Welle, Natalia Emanuel, and Valentin Bolotnyy shows marginal involuntary commitments worsen outcomes. Medication adherence sits at 20% before and after hospitalization; a Stanford Impact Labs trial is testing financial incentives for long-acting injectable antipsychotics.

The $350 million NIH HEALing Communities Study found null results using best-available opioid strategies. No new addiction drugs have been approved for alcohol or opioid use disorder in 20 years; none exist for stimulants. Jutca is partnering with CASPR on Phase III semaglutide (GLP-1) trials for alcohol use disorder and a Pittsburgh trial delivering GLP-1 drugs to people leaving jail — a population with a 40% re-incarceration rate within one year.

Most jurisdictions could build comparable capacity from existing resources; the real constraint is leadership culture.

What Trump Can Learn From Nixon

Presidents who try to control the bureaucracy through procedural choke-points — Kristi Noem's personal review of every DHS contract over $100,000, Lutnick's identical threshold at Commerce — reliably break the agencies they mean to control. Richard Nathan's book on Nixon's "administrative presidency" shows the pattern is not new.

Nixon entered office with conventional Cabinet picks who quickly "married the natives," captured by agencies and Congressional committees in the iron-triangle dynamic David Truman described in 1958. The White House response was routing around Cabinet secretaries via Ehrlichman's domestic working groups and direct lines to sub-Cabinet officials. That counter-bureaucracy grew to 700 OMB staffers, bottlenecked strategic thinking, created policy churn, and paradoxically made Cabinet secretaries more adversarial rather than less.

By 1972 Nixon pivoted to installing loyal "politician-managers" directly in line positions — demanding jump-and-ask-how-high loyalty, using budget impoundment to cut programs, and assigning Howard Phillips to dismantle the Office of Economic Opportunity without a Reorganization Act.

Watergate ended the experiment early. Nathan's surviving finding: empowering White House staff to overrule Cabinet secretaries destroys the alternative tool of trusted secretaries on a long leash. Running both models simultaneously is what produces the failures visible in Nixon's and Trump's bureaucratic struggles alike. The president must choose.

How to Write the AI Action Plan

Influence in the White House runs on usefulness, not job titles. Dean Ball, who authored the Trump administration's AI Action Plan at OSTP — an office with no budget or formal power — found that being the person others want in the room matters more than rank. Real leverage came from holding the pen: whoever drafts the document adjudicates conflicting agency feedback. Rubio's May 2025 NSC reorganization transferred AI-adjacent processes to OSTP.

The Action Plan is structurally unusual. Most government AI strategy documents offer vague aspiration; this one pairs objectives across three pillars with two to six concrete, agency-executable recommendations each — over 90 total, within existing authority. Rather than circulating the full document, Ball sent each agency only its relevant bullet points, ran ~20–40 parallel mini-clearances, and used AI to pre-simulate pushback. Rollout with five cabinet secretaries, the VP, and the president together gave agencies political cover to act.

Implementation is the harder problem. The White House operates on 10-day time horizons; ideas must arrive fully baked. Key levers: get agency leadership bought in before publication, attach deadlines to executive order provisions, and create forcing functions — a principal's scheduled visit or public event — that manufacture urgency.

On export controls, Ball argues for a global licensing framework: set a capability threshold, define what TSMC output share goes to China below it, and enforce violations through better IC–Commerce Department information sharing and chip location verification.

On AI safety, Republicans are structurally better positioned than Democrats to act on catastrophic-risk concerns. "American AI dominance" maps naturally onto control and alignment; Democratic coalition politics requires routing every concern through dozens of interest groups. The "Deep State" is real — career bureaucrats route high-context memos on two-hour clearance windows — and Trump's second term returned with a plan to pre-empt that pattern.

The Strongman Presidency

American democracy faces its current strongman crisis not because of Trump alone but because of a century of structural developments — the rise of the administrative state, deliberate construction of presidential control mechanisms, and a conservative strategic pivot to weaponize executive power against agencies they couldn't dismantle through legislation.

Before the Progressive Era the federal government barely existed as an executive enterprise; most employees were mailmen. Industrialization created demands for expert governance, producing New Deal agencies and, in 1939, the Executive Office of the President — a presidential bureaucracy built to control the broader bureaucracy. From FDR onward, presidents pursued two tools: centralization (pulling decisions into the EOP, NSC, and OMB) and politicization (stacking agencies with loyalists). Project 2025 calls for vastly more of both.

The decisive turn came when 1970s conservatives concluded they couldn't retrench progressive agencies through Congress or courts. Ed Meese's Justice Department developed the unitary executive theory as justification for presidential sabotage of agencies, while popularizing Robert Bork's originalism through the Federalist Society — building the judicial pipeline that produced today's Supreme Court. That Court is not a firewall: it authorized destruction by letting cases "play out" until the EPA, CFPB, and Education Department are gutted beyond repair. The immunity ruling, making a president unaccountable for crimes in office, is the most dangerous Supreme Court decision in American history.

Congress ceased as a check for two reasons: polarization created perpetual gridlock, and unitary executive theory explicitly excludes congressional participation in oversight.

Reform paths exist — court expansion, ending gerrymandering, reversing Citizens United — but require a Democratic trifecta and remain unlikely. Grievances driving strongman politics will outlast Trump; every successor Republican nominee faces identical incentives. Recovery requires the multi-sector coalition — journalists, universities, civic organizations — that the Progressive movement assembled to create the modern state.

How the National Security Strategy Gets Made

The National Security Strategy is less a unilateral presidential decree than a coalition-building document — its value comes from forcing the government to hash out trade-offs in writing. Nadia Schadlow, who led the 2017 NSS under McMaster, explains the mechanics: about 12 interagency meetings in the Eisenhower Executive Office Building, agenda controlled by the principal drafter, papers capped at two or three pages. Whoever drops the first draft controls the template; everyone else reacts. She circulated drafts as PDFs to prevent bureaucratic word-churn — "leverage," "robust," and three-verb strings are the pathologies she names.

Power centers matter concretely. Treasury in 2017 was less hawkish on China than the NSC or DoD because it represented Wall Street investment interests; China hawks had to fight for "strategic competitor" — language that now reads as obvious. Climate change was cut not from interagency battle but because it was not a Trump priority. State consistently pushed to soften language based on foreign-government reception; the NSC's frame was: this document is for the American people.

The 2025 NSS is more confrontational in tone, drops "strategic competitor" for China — Schadlow reads this as leaving negotiating room, not as softening, given the strong language on predatory subsidies and Taiwan — and omits North Korea entirely, which she reads as unresolved internal debate. She is skeptical about Greenland: the strategic goals are achievable without threatening an ally.

Her deeper preoccupation is time as an underappreciated strategic input — permitting a new mine takes roughly 16 years; weapon-system development runs similarly long. She advocates Gantt-chart discipline for national security: map A to Z, surface bottlenecks, make fast-and-good-enough versus slow-and-perfect trade-offs explicit. Reading list for Russia: Adam Ulam and Richard Pipes for the Soviet foundation, Stephen Kotkin for the modern era, Russian classics alongside contemporary authors for cultural texture.

Three Principles for Running a White House Office

Advisory power gets things done through relationship mechanics, not authority. Tom Kalil's three rules from running Obama's OSTP: assume people won't follow up — end meetings by naming the next concrete step and signal that obstacles are safe to report. Find who owns the paper — the speechwriter, the OMB analyst — because a report changes nothing if you never touched the actual decision document. Make it easy for whoever must act: draft the email your principal needs to send, supply the expert to the agency that can't hire.

How Cheaply Could We Build High-Speed Rail?

High-speed rail on the Northeast Corridor could be built for roughly $18 billion — about six times less than Amtrak's $117 billion Connect NEC estimate — while delivering Boston-to-New York and New York-to-DC trips under two hours each. Transit Costs Project lead author Alon Levy explains why the gap is so large.

The NEC's slowness comes from discrete bottlenecks, mostly sharp curves concentrated in southeastern Connecticut, plus scattered cant-angle restrictions elsewhere. Fix those and you reclaim half an hour cheaply. The biggest item in the Amtrak plan — condemning a Manhattan city block to add 7–12 tracks at Penn Station — is unnecessary: simulation shows the existing station handles doubled frequency. The only real constraint is the one tunnel between New Jersey and Manhattan, which the already-funded Gateway project resolves.

The deeper problem is inter-agency dysfunction. Amtrak and commuter operators — NJ Transit, Metro-North, Connecticut DOT, SEPTA, MBTA — each want separate infrastructure to avoid coordinating schedules. Every request for dedicated tracks is a symptom. The alternative is timetable-first planning: determine how many trains per hour each branch needs, then derive what infrastructure that requires. Germany built this discipline in the 1970s and achieves ridership matching France's despite slower trains.

The political economy is more tractable than it looks. All relevant agencies draw from the same Federal Railroad Administration pot; the FRA has authority to condition grants on coordination. Most states can be satisfied with projects already on their wishlists — New Jersey's three priorities (Portal Bridge, Hunter Flyover, Mid-Line Loop) are exactly what the proposal requires there.

The remaining obstacle is cultural: American rail planning doesn't start from timetables, ignores German engineering standards, and carries 50% soft-cost premiums over European equivalents — driven by low public-sector productivity and competition from an efficient private sector for the same white-collar labor.

How to Salvage a Transit Project

Highway agencies receive formula funding from Congress and are built to construct; transit agencies are operators by DNA — 90% of budget on daily service — competing for FTA grants one project at a time. That gap explains most of what goes wrong.

Opening cost estimates are almost always fiction — lowballing is rational; a true number kills the conversation. Boston's Big Dig ran from $2 billion to $12 billion. Scope then creeps as stakeholders attach. The Green Line Extension, 4.7 miles into dense Somerville, was never going to cost under a billion — eight bridges had to be rebuilt because abutments left no room for a second track; 24 new trains cost $200 million; a new maintenance facility was needed. Procurement was rushed to lock in federal money before an administration change. By 2015 the project tracked $3–4 billion, with overruns on a transit authority in crisis after four blizzards shut the system down.

Salvage, led by Stephanie Pollack under Governor Baker, worked through hard constraints. A six-month deadline with real cancellation risk forced alignment. The old team was fired. Stations dropped from $50–70 million to simple platforms; all seven stations and 24 trains were kept — frequency is non-negotiable. Contingency was cut: large reserves act as spending targets. Risk retirable pre-contract (moving commuter rail tracks) was resolved first. The Control Board set a hard not-to-exceed price; bidders competed on how many unfunded items they could absorb within it. The winner covered all four, including the bike path, under the ceiling.

The four Ps — planning, permitting, public engagement, procurement — expand to fill decades, not because they're slow but because no go/no-go decision backed by real money has been made. States that decline federal funding to escape its regulatory burden reveal a system working against its own purpose.

What is America's Infrastructure Cost Problem?

American infrastructure is expensive because of four compounding failures — permitting, procurement, personnel, and weak data — and NEPA gets disproportionate attention relative to the others.

Permitting matters most for above-ground linear projects like highways. A Detroit interstate segment illustrates the mechanism: early sections went up cheaply in the 1960s; a later NEPA-challenged segment costs several times more, with sound walls, below-grade construction, and parks built over the highway to settle lawsuits. The average environmental impact statement takes four years; median litigation runs 18 months even when plaintiffs lose. But underground urban projects like New York subways — the world's most expensive per mile — are barely touched by NEPA. Their costs trace to personnel and procurement instead.

Research by Liscow, Cailin Slattery, and Will Nober on California transportation found that a 1% increase in engineer retirements raises costs 4.5%, and improving engineer quality from the 25th to the 75th percentile produces savings three times the hire cost. State DOTs have shrunk workforces for two decades — mostly the skilled positions hardest to defend politically — then outsource at premium rates to the same private-sector engineers. Thin government capacity collapses bidder competition: the typical transportation bid draws three contractors, with almost no outreach to expand that pool.

Procurement is the most neglected driver. Many states must take the lowest-cost bid regardless of track record. European systems weigh quality factors at 75% and price at 25%. Repaving contracts that should be a dozen pages routinely run hundreds, driving out competition.

Better data — basic project cost, components, and timelines — would cost little and is largely missing, preventing rigorous study of which interventions work.

How to Save DC's Metro

Washington Metro reversed post-COVID decline by treating safety as a non-negotiable floor and pursuing frequency, crime reduction, and fare enforcement simultaneously — the insight being that these reinforce rather than trade off.

Randy Clarke arrived in summer 2022 to a grounded 7000-series fleet and six homicides in his first year. He restored trains, drove crime to a seven-year low, and installed taller fare barriers that cut rail evasion 82–85%. The enforcement logic is asymmetric: nearly every serious offender fare evades, so gate sweeps double as warrant dragnets. Revenue recovered tens of millions, though isolating the gate effect from fare increases and ridership growth is impossible. The system has now logged 50 consecutive months of ridership growth.

Metro's $5 billion budget splits evenly between operations and capital, with zero capital going to new construction — every dollar toward State of Good Repair. Clarke argues that no major new project is the harder management problem: you must execute across a thousand small things rather than one big one. The 2019 capital package of $500 million was never indexed to inflation, and expiring debt service will cut the capital program by $600–800 million in 2029. The DMV Moves regional task force is pushing a revolving bond program and indexed baseline funding before that cliff arrives.

New-line expansion has been displaced by converting to CBTC (communication-based train control) with platform screen doors and full automation — like the Dulles Airport people mover — resolving the Rosslyn bottleneck where Blue, Orange, and Silver lines share tracks. Target: 98–99% reliability at lower operating cost. Extensions remain contingent on density; single-family alignments are disqualifying.

The "transit-industrial complex" — the lasagna problem — is risk-aversion compounded across DC's oversight stack (GAO, FTA, USDOT, NTSB, three state legislatures, an OIG). A bus shelter redesigned to survive truck strikes ballooned from $80K to $450K. Clarke's response: push staff to weigh the cost of inaction, treat minor audit findings as healthy, and shift procurement toward outcomes rather than specifications.

How the Federal Transit Administration Works

The FTA funds and regulates roughly 900 transit agencies — from five-van rural operations to the NYC subway — and their variation is the central fact of federal transit policy. No two are funded alike: rural agencies may get 80–90% of costs from formula grants; large urban systems piece together revenue from local taxes and appropriations. When Cleveland lost population in a census, its allocation fell while the administration was touring cities promising increases.

Peter Rogoff spent 22 years as the top Democratic staffer on the Senate Appropriations Committee, ran the FTA from 2009 to 2014, then led Sound Transit in Seattle. His signature issue was State of Good Repair: agencies replaced components on fixed schedules rather than actual wear cycles while Congress funded expansions in systems that couldn't maintain what they had. Rogoff required Muni to show a maintenance plan before receiving expansion money, and won a dedicated maintenance grant category from Congress. Transit boards resist this — stacked with officials competing for ribbon-cuttings while infrastructure deteriorates.

The TIGER Grant program, designed in the 2009 Recovery Act, broke from formula distribution by letting the Secretary make large discretionary grants to any mode — ports, freight rail, bike paths, transit — bypassing the 80/20 highway-transit split and Senate committee rivalries that fragment transportation money. Streetcar grants proved a mistake: city departments lacked rail expertise, and Cincinnati's mayor had signed a non-compliant contract before federal funds could legally flow.

On high US transit costs: NEPA invites litigation, local governments treat permits as leverage for unrelated demands, and agencies lack European authorities' power to commandeer utility infrastructure. Faster disbursement isn't the fix — knowing a grantee's real capacity before the check goes out is. Timing governs policy as much as argument: a stalled OMB reversal cleared in one meeting because the director was distracted by a personal scandal.

Four Ways to Fix Government HR

More than 20 states have enacted civil service reforms so radical that none are on the table at the federal level — and they work. Texas, Florida, and Georgia moved to at-will employment, eliminated public-sector collective bargaining, and gave managers wide pay discretion. Fewer than 5-10% of managers recall any political influence on hiring, and state capacity metrics hold up.

Federal hiring runs through USA Jobs, where HR specialists keyword-search resumes before the hiring manager sees anyone — omit "manager" from a managerial application and you're screened out. Self-rating as "exceptional" improves hire odds regardless of reality. VA Secretary David Shulkin would recruit strong candidates in person; they vanished into the HR void and he received whoever passed the ranking process instead.

Pay locks onto a 15-grade General Schedule with 10 seniority steps each. Florida managers can broadband top performers 50% above benchmark; federal managers almost nowhere can. Pay compression means the lowest-skilled workers earn up to 40% above private-sector equivalents while post-graduate professionals earn less, structurally blocking elite recruitment.

Firing takes 6-12 months: performance improvement plan, then the Merit Systems Protection Board (employees win 20-30% of cases), then union arbitration, then federal courts. Air traffic controllers caught sleeping kept their positions. Managers assign poor performers to windowless basements and wait for them to quit.

Collective bargaining gained a statutory federal role only with the 1978 Civil Service Reform Act. Trump's redefinition of DoD, VA, and DHS positions as national security-related strips bargaining rights; courts have upheld it so far.

Near-term levers: make OPM an optional resource rather than a 300-page mandate, expand the bipartisan Chance to Compete Act's subject-matter-expert hiring model, and fix reduction-in-force rules so layoffs remove low performers rather than newest hires. A stultified civil service drives work to contractors — 40,000 officers handling $750 billion annually.

99.8% of Federal Employees Get Good Performance Reviews. Why?

Federal performance management is structurally broken: 99.8% of employees receive ratings of "fully successful" or higher, matching the 0.2% who are formally removed annually — while private-sector healthy organizations run 5–10% involuntary turnover. Scott Kupor, Trump's OPM Director and former Andreessen Horowitz managing partner, frames this as a failure of organizational design, not individual bad actors.

On the 2025 headcount reduction: of ~300,000 employees who left, only ~25,000 (under 9%) were fired or RIF'd. The rest took voluntary separation or deferred resignation. Rehires at the IRS, National Weather Service, FDA, Forest Service, and HHS reflected agencies cutting without understanding what those roles actually did — a known byproduct of large-org restructuring, not a unique government failure.

The contractor economy is the hidden fiscal problem: two to three times as many contractors as FTEs, costing $750 billion annually. The Clinton-Gore 1990s headcount cuts drove this substitution — shrinking official FTE counts while inflating contractor rolls. Kupor argues the right answer may be more FTEs and fewer contractors, not fewer of both.

On hiring reform: the 45-year-old Luevano consent decree, which blocked objective testing after a 1981 disparate-impact finding, has been terminated. OPM is buying off-the-shelf private-sector assessments for generic roles and building government-specific tests only where necessary (procurement officers). The 10-to-15-page resume and self-attestation culture both emerged as substitutes for banned tests; two-page resumes and LinkedIn-style submissions are the target endpoint.

The demographic problem is acute: 7% of federal workers are under 30 versus 22% in the private sector. Kupor's pitch reframes the value proposition from lifetime employment — which he calls false advertising — to a two-to-three-year high-impact tour with easy re-entry to the private sector. Techforce.gov launched shortly after this conversation to formalize that pathway.

Merit vs. Tenure: Reforming Federal Firing

The federal civil service firing system is broken in a specific, diagnosable way: it ranks employees for Reduction In Force (RIF) by tenure and length of service first, with veterans preference layered on top, and performance added last as a minor adjustment. A long-tenured poor performer survives over a short-tenured high performer, leaving agencies no practical way to prioritize the work that matters.

OPM director Scott Kupor and senior advisor Noah Peters trace why the system became so rigid. RIF regulations evolved over 80 years, untouched for 40. Agencies must build elaborate spreadsheets integrating multiple HR systems, and errors cascade. "Bump and retreat" lets displaced employees claim lower-graded slots, multiplying complexity. CTAP/ICTAP gives recently RIF-ed employees preference for new openings even when managers know them as poor performers. The result: the 2025 workforce reduction relied on the Deferred Resignation Program — eight months' severance for roughly 300,000 departures — with no ability to target who left.

OPM's proposed reform inverts the priority order. Performance ratings averaged over three years — up to 7 points per year for a top score, 21 maximum — become the primary ranking factor. Veterans preference is preserved and made more generous: 5 points for disabled veterans out of a 26-point maximum (roughly 20%), versus 5–10 points out of 100 in hiring. Tenure becomes a tiebreaker. Opponents are mainly long-tenured underperformers and unions that prefer categorical seniority.

For durability, Peters points to DOD's 2017 performance-first RIF system, enacted under Obama and kept through subsequent administrations because tenure-based rules couldn't be justified. Legislative codification is ideal but unlikely given congressional priorities. The broader OPM agenda includes cutting contractor-heavy staffing (OPM reduced its own contractor count by two-thirds) and targeting a third of new hires from early-career pipelines, using centralized job descriptions and Handshake to reach college students.

What Can We Learn From Estonia?

Estonia built the world's most digitally advanced government from necessity: 1.3 million people, no natural resources, corruption risk in a thin post-Soviet bureaucracy, and Finland — a country Estonia had matched in living standards before WWII — now decades ahead. That gap forced leapfrogging rather than gradual catch-up.

Two pillars underpin the system. The e-ID card gives every citizen a secure digital identity for all government services, built through a bank consortium to contain costs. X-Road is the data exchange layer enforcing the "once-only principle": submit your address once and no agency asks again. Citizens can see every government access to their records — less privacy risk than transparency tool, backed by genuine rule of law. Former President Ilves captured the corruption logic: "You can't bribe a computer." Ukraine's Prozorro open contracting system applied the same idea to procurement.

The payoff is concrete. Estonian taxes take five minutes; the government pre-populates everything. A newborn is linked to parents' digital identities, triggering applicable benefits without parental filing. Estonia carries the lowest debt-to-GDP ratio in Europe.

Origins matter for exportability. Western vendors quoted $20 million for systems Estonia couldn't afford, forcing domestic public-private partnerships that became a tech export engine. Digital IDs were mandatory from day one — inducing supermarkets, banks, and transit to build on top of them. Opt-in at 20% would have killed the ecosystem.

Three transferable lessons: do politically costly things early (Laar's shock therapy), invest long-term (Tiger Leap connected every school to the web in the 1990s), and make foundational platforms mandatory. On DOGE: the core mistake was not bringing the civil service into the reform coalition. The structural fixes worth prioritizing — easier hiring and firing, outcome-defined agency mandates, AI applied first to VA healthcare and housing permitting — matter more than copying Estonia's specific technology.

How to Build the '90s DOGE

Cutting headcount without simplifying the rules is why government downsizing fails — John Kamensky, Gore's eight-year deputy on the National Performance Review (NPR), applies that diagnosis to both the 1990s initiative and DOGE today.

The NPR ran 1993–2001, the longest federal reform effort in American history. Its bet: ask career civil servants rather than outside consultants what needed fixing, and focus on how government works rather than org charts. Gore's principle: "don't move boxes, fix what's inside them." The project started improvised — Kamensky, Bob Stone (DoD), and Bob Knisely pitched Gore in an accidental job interview, props in hand: a spray-paint can requiring a chemistry-degreed inspector to certify it hadn't expired; a steam trap leaking $50/week while procurement waited on bulk orders to save $10. Gore authorized credit cards for small purchases on the spot.

NPR eliminated over 100 programs and 250,000+ federal jobs. The workforce skewed older, junior hiring froze, and procurement expertise atrophied across two decades. The core failure: Congress cut headcount first and never simplified the rules, so fewer people faced the same maze. Contractors filled functions the government couldn't staff or legally delegate.

One big reinvention bill failed; reforms were threaded into annual appropriations. Wins included scrapping the 10,000-page personnel manual, franchise funds letting agencies competitively buy administrative services from each other, and credit cards for small purchases (CBO valued the rebates at $100 million when Congress tried to ban them). Also abolished: the Board of Tea Experts and Korean War-era mohair subsidies.

Two things NPR never solved: linking performance data to budget data to calculate unit costs, and building cross-agency career paths. Trust in government rose from ~20% to ~40% during the Clinton years. Kamensky's DOGE forecast: trust falls further, Social Security and immigration delivery fractures, and no replacement architecture exists.

How to Run a $5 Trillion Payment System

The federal government's payment system is designed never to fail, and that conservatism is exactly what makes it hard to reform. The Bureau of the Fiscal Service processes 1.4 billion transactions and $5 trillion annually — Social Security, tax refunds, vendor payments — with access controls so strict that each login is limited to a single purpose to prevent data extraction.

A cross-agency tiger team built data-matching pipelines between BFS, the Department of Labor, and HHS Medicaid rolls, running payment files against multi-state fraud patterns, FinCEN suspicious activity reports, and bank "know your customer" data. Fraud prevention jumped from $650 million to $7.2 billion in year one. Nineteen additional datasets could prevent another $30 billion, but OIRA approval for even one proven pilot took ten months.

The GAO counted $236 billion in improper payments in FY2023. Lebryk distinguishes improper (missing documentation, reconciled later) from fraudulent, estimating actual fraud is probably below the private-sector baseline of 2–5%. During COVID, the Do Not Pay screening system was switched off to speed disbursement; an IG report found $1–2 billion in preventable fraud from that choice alone. PPP paid corporations that didn't exist; the employee retention tax credit had the same failure.

The fix: score every payment like a credit card transaction, mandate that states running federally funded programs report data back to Treasury, and centralize fraud services at BFS so agencies — which lack the funding and expertise to prioritize fraud — can outsource the function.

DOGE focused on broad payment stoppages rather than targeted detection. Lebryk's analogy is a leveraged buyout that completes only the cutting phase — no business plan, no operational experts, no equity incentives to retain staff. "Vision without execution is hallucination."

50 Thoughts on DOGE

DOGE's core problem is that Elon Musk's Twitter feedback loops prime him to see waste and enemies rather than the regulatory and structural failures that actually constrain federal performance. The result is an organization fixated on two brute metrics — headcount and dollar spend — that are highly vulnerable to Goodhart's Law.

The firing strategy has backfired on its own terms. Civil-service rules (Title V, Part 351 CFR) force DOGE to cut probationary employees first: the younger, less-entrenched, more technically capable people. At the Bureau of Industry and Security, among those cut were staffers best positioned to enforce AI chip export controls against China. The Presidential Management Fellowship was also shuttered, while no effort to hire better replacements has materialized.

The claimed savings figures are largely artifacts of misread contracting data or already-dead contracts. The same data-parsing errors have persisted two months without correction. Long-run education studies costing almost nothing have been canceled; PEPFAR-linked programs disrupted. USAID's shuttering is the clearest failure — lifesaving aid halted, driven jointly by DOGE and Russ Vought's impoundment theory, with Musk's public statements about it demonstrably inaccurate. The CFPB and Education Department moves largely reflect Vought's unitary-executive agenda rather than Elon's initiative.

Genuine upsides exist — IRS data modernization, technical talent like Joe Gebbia, pressure on Beltway contractors. But Singapore-style state capacity requires paying civil servants more and freeing them from procedural burdens, not punishing them. The medium-term pivot point is a forthcoming SCOTUS fight over presidential impoundment authority — a ruling for the administration would fundamentally undermine Congress's ability to pass enforceable budgets.

Should the Feds Bail Out Chicago?

Chicago carries more pension debt than 43 U.S. states, has a worse credit rating than any other major American city, and 40% of its budget goes to debt and pensions. Seven of the ten worst-funded local pension systems in the country are in Illinois.

The legal architecture made this easy to build. States once treated pensions as a revocable "mere gratuity"; from the 1950s onward they elevated them to constitutionally protected contracts. Illinois adopted the California Rule — locking benefit terms from hire to retirement with no unilateral reduction. This placed pension obligations outside the debt limits governing bonds, making underfunding an attractive way to hide deficits. Governments compound the problem by assuming inflated returns (8–10.5%) to shrink required contributions, even though the legal obligation to pay is absolute. Chicago's funds sit at roughly 18% funded. Illinois legislators in the 1970s debated whether this dynamic would emerge and decided a mandatory savings rule was unnecessary. They were wrong.

The politics reinforce the fiscal failure through universal logrolls. The Illinois General Assembly recently passed — unanimously — a bill adding $11 billion in new first-responder pension liabilities, widening next year's deficit by over $1 billion. Suburban legislators don't block Chicago's requests because Chicago controls theirs. The benefit is immediate; the cost accrues over fifteen years.

When insolvency arrives, three responses exist: bailout, creditor loss, and austerity. A little of each is far less damaging than all of one. Detroit's bankruptcy shows the workable sequence: austerity, then creditor impairment, then a targeted bailout from foundations and the state — paid after the political cost was absorbed, limiting moral hazard. New York's 1975 crisis added mandatory honest accounting as a rescue condition. Both precedents point to the same principle: crisis is the moment to extract structural reform, not just stabilize the balance sheet.

How to Rewire City Hall

City halls fail not from bad intentions but from an outdated Fordist operating model — siloed, throughput-measured, built for standardized mass service — while mayors face pandemics, affordability crises, and migration requiring iterative, cross-sector problem-solving. The binding constraint: 50–75% of municipal budgets are fixed costs, leaving a thin slice for data infrastructure and strategic capacity — the first items cut in downturns. Bloomberg Philanthropies' Government Innovation Program has spent 15 years closing that gap, growing from five grants in 2010 to deep technical assistance reaching 900+ cities.

The program's core bet is internal capacity over outsourced expertise. The What Works Cities certification has moved hundreds of US cities up a data-sophistication ladder; 54% of cities over 100,000 are now progressing toward certification. The Bloomberg Harvard City Leadership Initiative trains 40 mayors annually to manage as problem-solvers first. Mayor Tim Kelly of Chattanooga arrived demanding answers; the program taught him to diagnose before acting. When his team traced youth crime to mental health rather than policing, targeted interventions produced a full year of zero murders in one high-crime neighborhood.

Baltimore under Brandon Scott is the anchor case. He built a homegrown COVID contact-tracing operation rather than hire management consultants, outperforming peers on every vaccination metric. A live chief-data-officer dashboard, a declared 15% year-on-year murder reduction target, and group violence reduction strategies drove homicides to 50-year lows — the fastest decline in the country. The same innovation team then tackled police retention, producing a net officer gain for the first time in a decade. None of it is paradigm-breaking; it's disciplined execution of known evidence-based methods, with the mayor's personal credibility staked publicly on the goal.

The smart cities movement failed by leading with solutions rather than problems. AI carries the same risk: mayors see a tool another city deployed and want to import it. The prescription is the same — get data governance clean first, promote staff fluency, experiment on low-risk administrative use cases, and move toward service delivery carefully. Mayors are almost uniquely free of Not Invented Here syndrome; their FOMO is a genuine policy-diffusion engine when paired with technical assistance to adapt, not merely copy, what worked elsewhere.

How to Run New York City

Running a city of 8.5 million — 12,000 tons of daily trash, a billion gallons of water, 300,000 employees — requires deliberate management architecture. Maria Torres-Springer, first deputy mayor under Eric Adams after senior roles under Bloomberg and de Blasio, draws the contrast: Bloomberg used data before politics and delegated to agency heads; de Blasio organized around explicit North Stars (universal pre-K, affordable housing) and a decision memo matrix clarifying scope and sign-offs; Adams prioritized constituent services and accessibility.

Three structural lessons Torres-Springer treats as non-negotiable: minimize direct reports to the mayor to cut silos; keep information and mayoral access abundant rather than scarce (scarcity produces game-playing); model low-drama culture from the top. Bloomberg's open bullpen — still in place under later administrations — created velocity by making disputes visible.

City of Yes for Housing Opportunity was the most pro-housing zoning reform in city history: legalizing ADUs, re-legalizing transit-oriented two-to-five-story buildings, relaxing parking minimums, enabling office conversions. Projected yield is 80,000 units over 15 years. Passage required a broad coalition, $5 billion in committed funding ($1 billion from the governor), and state action lifting an artificial density cap.

NYCHA's year-plus apartment turnover reflects $80 billion in assessed capital needs on stock averaging 50 years old, braided financing, and federal monitor oversight — not simple management failure. The contractor payment system carries the same pathology: anti-fraud layers so thick that city reliability as a payer became a deterrent.

The most decisive unglamorous work is permitting and procurement reform: a unique business identifier so firms stop re-entering details across a dozen agencies, online portals, automated payments. Each vacant day is another family in shelter.

Forward priorities: maintain the public safety and transit foundations businesses weigh when locating, diversify beyond healthcare, and connect workforce systems to emerging sectors like life sciences and AI.

Is the Senate Fixing Housing Policy?

The ROAD to Housing Act — 27 bills — cleared the Senate Banking Committee 24-0 in July 2025, the committee's first bipartisan housing markup in over a decade, then passed the full Senate in October. Tim Scott and Elizabeth Warren co-sponsored. The central argument: subsidizing demand while constraining supply cannot work, and the regulatory problem is local zoning, not federal environmental law — a distinction that had to be established before both parties could agree.

The bill works three levers. Regulatory reform: a Nixon-era requirement that manufactured homes carry a permanent steel chassis (unused once installed, adding $5,000–$10,000 to a $90,000–$120,000 unit) is eliminated; FHA multifamily is cut out of NEPA review that never applied to FHA single-family or Fannie/Freddie multifamily. Incentives: the Build Now Act redirects 10% of the $3B+ annual Community Development Block Grant based on per-unit production in high-cost jurisdictions — federal dollars chasing heads rather than flowing agnostically; the Build More Housing Near Transit Act ranks transit capital-grant applicants (from a $4.6B+ annual pool) by whether they zone for housing near new stations; a $200M Innovation Fund covers infrastructure matching costs for jurisdictions that reform zoning and hit unit targets. Program fixes: Section 8 pre-approval lists and remote inspection (over half of vouchers go unleased within 90 days), income-certification paperwork reuse, small-dollar mortgage reform at the CFPB, and permanent HUD disaster-recovery authorization.

The political shift had two causes: shortage spreading to Montana, the South, and Atlanta made deregulation a Republican constituent issue; and Warren replacing Senator Brown as ranking member, moving the Democratic caucus from skepticism to co-authorship. Niskanen Center analysts supplied a previously obscure sub-county census permitting dataset that made Build Now's unit-based measurement administratively viable. Full bill enacted by end of 2026: roughly 20%. Significant chunks in some form: higher.

Why We Don't Build Apartments for Families

American cities stopped building apartments for families because post-2008 financial incentives optimized for a different product. When Fannie Mae and Freddie Mac tightened lending, condo projects became nearly unbankable while rental apartments sailed through — tilting development toward young singles. Average new apartment square footage has fallen to 1994 levels; Austin now sees studios at 20–30% of new units.

The two-bedrooms that do get built are designed for roommates. Equal-sized bedrooms, two baths, no third room — the layout maximizes rent-per-square-foot against the studio baseline. A survey by Bobby Fijan's American Housing Corporation with the Institute for Family Studies found families overwhelmingly prefer the same square footage reconfigured into a larger primary bedroom plus two smaller children's rooms. Bedrooms ranked above yard, price, and space.

Several forces lock in the mismatch. ADA wheelchair turn-radius requirements make bathrooms irreducible below roughly 5×8 feet, so the shift to mandatory two-bath units eats space that pre-war buildings used for a third bedroom. Parking minimums assessed per bedroom rather than per unit directly penalize bedroom count. Institutional operators standardize kitchens across all unit types — the same full-size range and refrigerator in a studio as in a three-bedroom — interchangeability across a 250-unit complex cuts maintenance costs. Studios turn over at roughly double the rate of three-bedrooms, adding vacancy drag.

The deeper driver is finance. Most construction is funded by closed-end private equity with 2–3 year exit targets, making near-term rent maximization the only rational goal. Europe builds through public or fixed-profit vehicles on longer horizons; the comparison fails because the incentive structures are incompatible.

The fixes Fijan favors are subtractive: eliminate per-bedroom parking ratios, permit single-stair buildings (safe with modern sprinklers), and cut permitting friction. Pre-war floor plans — fewer bathrooms, more bedrooms, smaller kitchens — prove the model worked and could again.

How to Build a Flying Ebola Hospital

The US pandemic apparatus talks but doesn't coordinate, and doesn't learn — the same failure that let CDC deploy broken COVID tests in 2020 repeated with monkeypox assays. CDC is a research organization running itself as a response authority: each outbreak activates a siloed pathogen group from scratch, with no doctrine, no standing command structure, no cross-event memory.

Eric Van Gieson — DTRA and DARPA veteran — argues the fix requires three things. First, a permanent pandemic response authority modeled on FEMA, rather than the ad hoc structure Bob Kadlec improvised as Operation Warp Speed. Second, standing IDIQ contracts with diagnostics and PPE manufacturers: South Korea had millions of tests per day in February 2020; the US wasn't at scale until April or May — the "invisible graveyard" of excess deaths from that lag. Third, systematic screening of the 10,000–20,000 on-market generics against known pathogens using tools like Lawrence Livermore's GUIDE platform; every pandemic since 2009 has surfaced a useful existing drug, but only after the fact.

Van Gieson built the Containerized Biocontainment System — a shipping-container-footprint flying ICU with HEPA filtration and battery backup — from a napkin sketch with Will Walters, Paul Allen Foundation funding, fabricated in six weeks, and later acquired by Air Mobility Command (it moved Diamond Princess patients in 2020). His team also placed diagnostic labs beside West African Ebola treatment units to catch malaria patients who would otherwise enter Ebola wards unscreened — the diagnose-first sequence the US keeps inverting by prioritizing vaccines over rapid testing.

On EcoHealth/DARPA: the proposal crossed his desk, raised significant red flags, and DARPA rejected it. The deeper failure is interagency inconsistency on what dual-use research is safe to fund — the solution is domestic research infrastructure with unified standards, not foreign subcontractors beyond US oversight.

How Diplomacy Works in Africa

Africa is systematically under-resourced in US diplomacy relative to what national security interests require, while rivals extract growing trade and international-forum alignment through stronger engagement. The core deficit is "care and feeding" — senior-level attention. For too many African countries, the highest-ranking American they ever meet is their local ambassador; the Cote d'Ivoire foreign minister reached Washington repeatedly without meeting anyone above assistant secretary of state until James Baker needed a Security Council vote on the Gulf War.

George W. Bush stands alone in the post-Cold War era for African leader engagement, partly because Condoleezza Rice told him from day one that Africa would take real time. Kennedy devoted 25% of his foreign leader meetings to Africa. Obama had senior Cabinet officials (Susan Rice, Gayle Smith, Samantha Power) who cared, enabling Power Africa and the Ebola military response, though the administration was too preachy. Biden's team did less lecturing but also less engaging — Ukraine and Gaza consumed the bandwidth.

The Niger junta expulsion after the "condescending" US delegation visit illustrates how tone collapses delicate negotiations; a parallel delegation to Gabon's coup leaders, run with more flexibility, produced a two-year return to civilian rule. Engagement with difficult leaders gets blocked by double standards that don't apply elsewhere — the Zimbabwe photo-op choreography at the 2022 Africa Leaders Summit was "kabuki" for a domestic audience that likely didn't notice.

Most of the 60-plus presidential Africa initiatives since the 1990s fail because they're built backward — deliverable first, buy-in never. Durable programs (PEPFAR, Power Africa) came from extended stakeholder work before launch. The Biden-era 21PAS security program, forced on a resistant DOD via Jake Sullivan calling Secretary Austin, died when DOD redirected the money once that political pressure was gone.

Envoys work best with narrow negotiating mandates and real staff; they're mostly fixes for American bureaucratic problems, not African ones. The Africa Directorate folded into the Middle East NSC reduces the continent to a rounding error. The NSC's power is entirely a function of whether agencies believe it speaks for the president — when that belief holds, the coordination layer below cabinet level is where State, DOD, and Treasury hash out competing equities before decisions harden.

How to Be a Good Intelligence Analyst

The central failure of American intelligence analysis is a communication breakdown between producers and consumers. The core dysfunction is timing: assessments are only actionable in a 2–3-week window. Warn 10 years out, nothing gets done; warn 24 hours out, same result. When surprise hits, the IC is always blamed first.

Resource allocation compounds this. Tunisia had roughly half a full-time analyst at CIA before the Arab Spring. When a self-immolating protester triggered the cascade that toppled Mubarak, the community was blamed for missing it — despite predicting Mubarak's fall for a decade. Timing was unpredictable. Budget politics favor satellites (subcontracts in 43 congressional districts) over human analysts and language training.

Technical collection delivers access, not truth. In every recorded Saddam-generals meeting, the generals lied about WMD programs to survive. Intercepted conversations would have been equally misleading. Political pressure to invade Iraq then distorted community analysis on top of that.

Good analysts are tenacious, obsessive about source quality, and keep score on their own predictions — Johnston's model profession is librarian. Expertise accumulates around 10,000 cases but creates tunnel vision: veterans miss orthogonal threats that novices catch because no one told them it was impossible. The career structure rewards generalist rotation over depth. The State Department's Bureau of Intelligence and Research, where analysts can spend an entire career on one account, is the model worth replicating.

Colin Powell was the exemplary consumer: tell me what you know, what you don't, then what you think. Rumsfeld and Wolfowitz exemplify the failure mode — prolonged success produces certainty of being smarter than everyone, so contrary evidence gets dismissed as noise. Structural fixes: give the DNI real budget authority, enable rapid surge reallocation, and narrow what gets classified — most secrets have a shelf life of days, and over-classification wastes the resources that matter.

How to Run a Private Military Company

Wagner was never a pure private military company — it was an entrepreneurial vehicle Yevgeny Prigozhin used to sell the Kremlin a dream of Russian great-power reach while running a side business empire. An ex-convict who built his way into Putin's circle through catering contracts, Prigozhin stood up a mercenary force in 2014 using the Donbas crisis as cover, drawing from the failed Slavonic Corps and Cossack networks. Fighters mixed ideology and paycheck and saw no contradiction.

The model diverges sharply from Blackwater's. Blackwater embedded within one client government, performed passive security — convoys, fixed-site protection — and had no offensive mandate. Executive Outcomes, a 1990s South African firm, ran full offensive operations but without a state patron. Wagner fused both: offensive combat plus diplomacy (the 2019 Khartoum Accords in CAR), gold and diamond mining, a vodka brewery in Bangui, and troll farms. Mining cash flow in an active civil war was a nightmare, but the pattern was always throw everything at the wall.

Kremlin leash length tracked Russian institutional density. In Syria and Ukraine, the GRU, FSB, and MOD were all present, constraining Prigozhin. In CAR, with only two embassy staff, Wagner became the Russian state. Prigozhin presented himself as a Kremlin envoy while operating on his own capital, hoping for retroactive subsidies.

After his mutiny and death, Russia's PMCs reoriented toward MOD-subordinate structures, though oligarch-sponsored units persist as feudal loyalty signals to win future state contracts.

Russia may yet retain Hmeimim air base and Tartus naval port in Syria: Ahmed al-Sharaa needs hard currency and has few takers. In Africa, multipolarity has already arrived — African elites are practiced at playing great powers against each other, and Wagner helped manufacture the competition narrative it then profited from. Niger illustrates the failure mode: maximum demands, minimum inducements, no carrot.

Leninist Technocracy With Grand Opera Characteristics

China is a "Leninist technocracy with grand opera characteristics" — rationalist by training, campaign-driven by ideology, and prone to Wagnerian collapse when its logic runs to extremes. Dan Wang's *Breakneck* contrasts this with the United States, lawyerly since the Founding, where legal culture shifted after the 1960s from deal-making to litigation. Post-DDT, post-Moses lawyers turned from building railroads to suing the government — which explains why California can't finish high-speed rail and New York pays $2 billion per subway mile.

China's engineering state traces to Deng's overcorrection: by 2002, all nine Standing Committee members held Soviet-style engineering degrees. Since 2022, Xi has promoted military-industrial veterans who managed rockets and fighter-jet programs. The result: 32% of world manufacturing, with chokehold positions in rare earths, solar panels (90%), and key pharmaceuticals. Cadre promotion runs through the Organization Department's nomenklatura dossier — mayors built airports to goose GDP metrics for decades; Xi removed hard metrics, creating ambiguity around incentives.

The One-Child Policy illustrates social engineering's failure mode. Missile scientist Song Jian applied cybernetics math to population trajectories; from 1980–2013, the state carried out over 300 million abortions and hundreds of millions of forced sterilizations by official statistics. Now Beijing harangues women to reproduce — discovering coercion can suppress births but cannot induce them.

Wang's prescription: not engineers, not lawyers — economists, technocratic enough to execute and humanist enough to reject linear extrapolation; scientists on tap, not on top. On growth, he brackets the macro headwinds (debt, deflation, aging, property crisis) and focuses on the top 5% — EVs, AI, advanced manufacturing — which will keep deindustrializing Michigan and Germany regardless. He closes recommending James C. Scott's *The Art of Not Being Governed*, on highland peoples who planted cassava to evade the tax man — and confesses to dreaming of joining them.

How a Congressional Office Actually Works

A House office of roughly ten people runs as a small business on a ~$2 million annual Members' Representational Allowance (raised 20% in 2022 for staff retention), with salaries consuming up to 90% of it. The most valuable resource is the member's time, and the scheduler — often dismissed as administrative — controls it more directly than anyone else, gaining more insight into the principal's thinking than most legislative work does.

Staff divides into four functions. Legislative (four to five people: a director, assistants each covering six to eight policy areas, a correspondent) handles bill introductions, committee hearings, vote recommendations, and constituent mail. Communications (one to three) translates wonky legislation into constituent-facing messaging and books press. The chief manages relationships and staffs unofficial events like fundraisers. Scheduler and chief are the only two people who see both official and unofficial calendars.

A week in session layers votes (6:30 pm, ~25 minutes notice, sometimes a two-hour series for appropriations amendments), committee hearings, and caucus standing lunches — Progressive Caucus, New Democrat Coalition, Blue Dog Coalition each hold regular ones. Members in competitive seats spend 10–15 hours a week on fundraising call time alone, which never fully stops.

The first filter for meeting requests is district nexus: a Los Angeles conference attendee doesn't get face time with a San Diego member. The Congressional Research Service is the most underrated resource — call with any question, receive a custom briefing or written report. The Chief Administrative Officer has worked to professionalize career tracks previously dependent on individual mentorship.

For a freshman wanting everything — signature legislation, earned media, sharp committee hearings — the honest answer is choose. Members who spread across every issue accomplish less than those who match their own background to a focused lane and empower a team to multiply them.

Why the Two Parties Operate Differently

The two American parties differ not in ideology alone but in organizational structure: Democratic power flows upward from constituent groups, Republican power flows downward from personal connections to leaders. Political scientist Jo Freeman — attendee of every Democratic convention since 1964 — developed this framework from watching feminists navigate the two parties and finding the rules completely different.

In the Democratic Party, influence requires claiming to represent a group. In the Republican Party, it requires proximity to whoever is on top — Goldwater in 1964, Trump today. Republicans see themselves as insiders even out of power; Democrats as outsiders even when governing, a self-perception tied to the parties' class compositions, though both are shifting.

The 1976 Democratic convention was a turning point for the women's movement: feminists organized a caucus, technically lost their floor fight, but demonstrated mobilizing power. By 2024, the convention's 33 caucuses heard speeches and held receptions; the earlier culture of debating demands had evaporated.

Freeman closes on social movements: they are inherently unstable, never permanent. Once they acquire durable structure, they become interest groups — the fate of labor and feminism both.

How to Budget for the SEC

The SEC collects substantial revenue from filing fees, penalties, and disgorgements but cannot keep any of it — everything goes to the Treasury while the agency returns hat-in-hand to Congress for its annual appropriation. This makes the SEC perpetually dependent on the appropriations cycle, unlike self-funded agencies such as the Federal Reserve or the CFPB (Elizabeth Warren's deliberate design to insulate it from congressional budget leverage).

The budget cycle runs through OMB first: the SEC negotiates a request, OMB folds it into the president's unified budget, and appropriators hold hearings. Those hearings feature a ritual trap — appropriators ask chairmen whether the president's request is enough, fishing for a public complaint that would let Congress override the White House. Agency heads who take the bait get fired; a former Army Corps of Engineers chief lost his job practically the next day for saying exactly that.

Earmarks create their own distortions. When Senator Schumer added $20 million for SEC enforcement post-financial crisis, the agency faced a hiring math problem: because salaries accrue as people are hired across the year, spending $20 million in year one required roughly $40 million of headcount on the books, inflating the baseline Congress would expect the following year. Congressional staff called monthly to check hiring pipelines — the pressure was to spend visibly, not wisely.

The Madoff scandal broke in December 2008 against this backdrop. The SEC had examiners in Madoff's shop repeatedly over fifteen years, received credible tips, and missed returns that were mathematically implausible in their consistency. Examination is fractional — too many investment advisors, too few examiners to cover each annually.

*Loper Bright*'s overturning of Chevron deference and the major-questions doctrine from *West Virginia v. EPA* now require agencies to trace every significant rulemaking to explicit congressional authorization. Gensler-era SEC climate-disclosure and investment-advisor rules pushed past that line. Future administrations will likely write narrower, better-anchored rules rather than invite litigation on contested authority.

When FAFSA Broke, They Called This Guy

Government software projects fail not because the goal was wrong but because Congress hard-codes implementation details, agencies lack technical leadership, and vendors go unaccountable until it's too late. The 2023 FAFSA disaster — a bipartisan simplification that should have cut the form from 100 questions to 36 and pulled IRS data automatically for 17 million annual users — illustrates all three at once.

Jeremy Singer, College Board president, was recruited by the White House in mid-2024 to replicate the healthcare.gov rescue. He diagnosed three compounding problems. First, Lamar Alexander's 2019 bills embedded UI language into statute — a homeless-student disclosure confused every applicant, and the team couldn't legally remove it. Second, Federal Student Aid had no effective CTO. Four vendors, led by General Dynamics IT with COBOL-era engineers, couldn't communicate across systems; getting all of them onto Slack was celebrated as a breakthrough. No one inside the department could hold them accountable for code quality or schedule. Third, political pressure suppressed honest status reporting — vendors were still building core functionality past the October 2023 launch date, and the department didn't know.

Singer's fix: a small cadre of technical experts, staged beta tests (hundreds → thousands → tens of thousands of users), and refusal to launch prematurely. His line — "two moms can't produce a baby in four and a half months" — drove the decision to push to November 2024. That launch succeeded; 1.7 million students became eligible for maximum Pell Grants.

The 2025 GAO audit, he argues, was compliance theater: documentation steps no competent engineer follows during a crisis, which would have made things worse. His prescriptions: write statute as use cases, not UI specs; hire 15–20 permanent senior technical staff rather than outsourcing architecture to Beltway contractors; keep a funded backup vendor as leverage over the primary.

Ten Thoughts on Government Data

Government data is far messier and more incomplete than policymakers assume. Administrative records like DHS's SEVIS have major gaps — departure dates and employer addresses for working international students are absent most of the time — because these systems were built for bureaucratic tracking, not analysis. Errors persist unnoticed for months: in 2024 the US undercounted international students by 200,000. Much "counting" is sampling with built-in assumptions; misuse produced the spurious claim of two million new native-born jobs. Declining survey response rates and AI-generated spam are making imperfect administrative records comparatively more valuable. Unlocking any dataset requires practitioner knowledge — understanding why fields changed, which traces to regulations or memos that only career bureaucrats remember.

How to Run the Treasury Department

The Democratic Party's problem wasn't messaging — it was implementation. Wally Adeyemo, Biden's Deputy Treasury Secretary, diagnoses two failures: execution infrastructure that never existed, and policy gaps on housing and drug costs left unaddressed.

Treasury's systems hadn't been upgraded since the 1960s and still ran COBOL. The Advanced Child Tax Credit required a new monthly-payment system from scratch because the IRS only processes annual returns. IRS call-answer rates were 13% when Adeyemo arrived; he raised them to 85%. His biggest regret: not creating a permanent cross-cutting implementation unit from day one — every crisis (ARP's $1.9 trillion, the IRA, Russia sanctions) got its own ad hoc team.

Direct File — free IRS filing built internally with 18F staff, not consultants — cut average filing time from 13 hours to just over one, saving users roughly $200. First-year rollout: 140,000 people; without Trump administration promotion, 300,000 used it the next year at above-85% satisfaction. Now slated for elimination. The CBO scores only FTE headcount for revenue projections, not technology — blocking Congress from funding IRS tech Adeyemo estimates returns $10 per dollar.

Russia sanctions worked best multilaterally — allies control other convertible currencies, foreclosing substitution. They've compounded damage (Russian inflation far above global norms) but haven't changed Kremlin behavior because China, Iran, and North Korea backstop Russian supply chains. Adeyemo's Foreign Affairs piece argues Europe must carry more financial burden as U.S. support withdraws.

Housing is the generational crisis the administration treated incrementally. Adeyemo now works with 15 D.C. churches building on donated land and finds it near-impossible due to permitting and local politics. His prescription: condition non-housing federal funds — not just LIHTC — on zoning reform. "Housing that is affordable" versus "affordable housing" is the key distinction: middle-income cost crowding requires removing restrictive zoning covenants, not more subsidy dollars.