Search all dialogue chapters, core definitions, Lexicon terms, and key essays.
A statute has a specification, a commitment and a decay. It has never had a carrying value.
In November 2026 Johan observed that policies accumulate without a lifecycle, that the period of borrowing against future earnings has to end, and that specialisation needs governments to cooperate rather than compete. This Part takes the observation literally. Written as an asset, a policy yields a terminal condition that is derived rather than chosen, a deferral gap with a closed form, and one number — the duplicate count — that has to be published before the money is spent.
The observation arrived as a diagnosis of public finance and turned out to be a statement about accounting. It was made in a conversation about inflation, debt and the loyalty of electorates, and it contains one clause that does all the work: an expected lifecycle.
“Many western countries have indulged themselves with lending on future earnings, thinking in tax revenue, which they had to increase to keep up… My idea is that we end this period because it’s unsustainable. Policies need a design, a structure, and an expected lifecycle to be productive and not fall in slack. Can you observe my ideas and find initiative who make governments cooperative instead of competitive. Specialism needs it for it’s vulnerabilities I think.”
Three claims sit inside that paragraph and they are not the same kind of claim. The fiscal one — that the period is unsustainable — is arithmetic and can be checked. The lifecycle one is structural: it asserts that a policy without a stated end behaves like an asset without a depreciation schedule. The third, that cooperation is required for specialisation, is a claim about scale, and it is the one that turns out to be testable against what governments are already doing.
The earlier Parts of this sequence established that loss on a specification change is bimodal rather than graded, and that the speed of recognition, not the size of the shock, is what separates a firm that survives from one that does not. This Part asks the only question that follows: if recognition speed is the variable, what would it take to install it in an institution that has no auditor, no lender and no share price to compel it?
The operator has now been applied at five scales, and it has not needed modification at any of them. That is the strongest evidence for it, because an operator that has to be rebuilt for each new object is a metaphor rather than a model.
| Scale | What carries the specification | What recognition means |
|---|---|---|
| Firm | Book value against recoverable amount | An impairment, compelled by an auditor |
| Farm cell | Housing, contract, quota, skill | Exit, compelled by a lender |
| Development programme | A platform, a tooling line, a launch date | Cancellation, compelled by a cash constraint |
| Policy | A statute and the problem it was written for | Repeal, compelled by nothing at all |
| Specialist | A skill with one buyer | Displacement, compelled by the employer |
In the first three rows something outside the decision-maker compels the recognition: a standard, a covenant, a lender. In the fourth row there is nothing, which is why has to be written into the instrument itself. The fifth row is the one this Part cannot yet handle, and it is the subject of the Branch Point.
The cost of late recognition is measured where it can be. For a firm it is the write-down; for a policy it is the deferral gap derived below; for a person it is larger than either. Displacement loss for high-tenure workers runs at roughly a quarter of annual earnings persistently, and above forty per cent in the year of the event, with earnings still fifteen to twenty per cent below a control group twenty years later. A policy that runs a decade past its threshold is not an accounting error. It is a transfer from people who will never be made whole to people who will not notice.
Let a policy have a fit between zero and one: the share of its design that still matches the world it was written for. Let each euro committed deliver a benefit of phi of that fit, and cost one plus tau to raise and administer, where tau is the deadweight cost of taxation. The surplus per euro is therefore the benefit less the cost.
s* solves phi(s*) = 1 + tau
The policy is worth keeping if and only if the fit is at or above s*.
That threshold is not a political judgement and it is not a number a minister picks. It is the point at which the policy stops covering its own cost plus the deadweight of raising the money, and it can therefore be written into law in advance. This is the difference between a terminal condition and an aspiration: “end when the measured fit falls below s*” is enforceable, and “end when it is no longer needed” cannot be enforced by anyone, because the party who benefits from continuing is the party who would have to find it no longer needed.
The four elements Johan named are not a checklist. They are the four parameters of the operator, and each does a different job. Expected life sets the depreciation rate, and without it the commitment never decays. The review date converts a discretionary decision into a scheduled one, so that recognition happens on a known day rather than whenever an incumbent is brave. The terminal condition is the barrier on the state. The beneficiary count is the political cost of stopping, and it is the only one of the four that works against the other three.
The cheapest of the four is the expected life, which costs one clause. It alone turns a permanent commitment into a depreciating one, and it is the precondition for everything else: a policy with no stated life has no schedule against which a review can be a review, rather than a courtesy.
Ending a policy costs an incumbent something specific. The benefits are concentrated on a countable number of people who know who they are, and the saving is diffuse and arrives later. Write the political cost of termination as a function of the number of beneficiaries and the concentration of the benefit, and let beta be the weight the incumbent places on future fiscal saving relative to present political cost. The incumbent then stops at a fit below the social threshold.
s_pol = phi inverse of (1 + tau − C / (beta times b))
where b is the annual commitment and C the political cost of ending.
The distance between the two thresholds is the deferral gap, and with a constant decay rate mu the cost of it has a closed form: the commitment multiplied by the gap in fit, divided by the decay rate. Two terms are observable for a programme with a budget line and a beneficiary register: the commitment and the beneficiary count. The fit, its benefit function and the decay rate are not, and have to be estimated or assumed, which is the limit this Part returns to below. Where they can be bounded, the cost of political delay can be computed and published rather than argued about. Where they cannot, the expression is a frame for the argument rather than a measurement.
A closed form is not a licence to compute, and its conditions belong beside it. The expression holds only if benefit rises with fit, so that the social threshold is attainable; if the inverse exists on the range where the threshold is being sought; if the budget, the political cost, the incumbent weight and the decay rate are all positive; and if the decay is constant rather than path-dependent. Two of these fail quietly. Restrict fit to the unit interval, take a benefit function as plain as one plus fit, and positive inputs can put the social threshold inside the interval while the political threshold falls below zero, which is not a very low threshold but no termination at all. The formula then reports the size of a gap that has been assumed into existence.
One consequence is counter-intuitive enough to be worth stating plainly. The gap depends on the political cost relative to the budget, not on the budget alone. A large programme with diffuse benefits has a small ratio and ends close to its threshold. A small programme with a tight constituency has a large ratio and may never end. Deferral is therefore worst, and cheapest to overlook, in the programmes too small to audit and too concentrated to cut.
This is the standard result about concentrated benefits and diffuse costs, but it now arrives with a magnitude attached and an instruction: it tells you which programmes to examine first, and it is the opposite of the order a budget office normally uses.
The model has a hole in it, and the hole is where most proposals fail. Fit is not observed. It is inferred from outcomes, and the party that commissioned the policy also commissions the evaluation. A policy that buys its own evidence will always report a high fit.
Two devices close the hole. The first is a: before the first appropriation, the minister lodges the objective, the mechanism, the forecast and the terminal condition with an independent custodian, and all four are published. The statement can be amended, but every amendment is a published event with reasons. That single device converts the moving of a goalpost from a drift into a record, and it is the policy analogue of pre-registering an experiment — a discipline this sequence has already applied to its own predictions.
The second is a statutory right of access to the administrative microdata, on the same terms as a supreme audit institution, with obstruction reported by name. An evaluator that receives a ministry’s summary has been handed the ministry’s conclusion.
There is a boundary that most designs miss. The evaluator must not set tau. If it chooses the cost of public funds it chooses the threshold, and its claim to be measuring rather than choosing collapses. The mandate therefore separates the two: the evaluator measures the fit and reports it, and the threshold comes from a tau published under a separate rule. Two verdicts are kept apart — is the policy working, and is it worth its cost — because merging them lets a technical body make a distributional decision without appearing to have made one.
Independence itself is not a property of a charter. It is a property of a market for the evaluator’s output. A reviewer with one client is a consultant; a reviewer with two adversarial clients is a referee. The Congressional Budget Office survives because the majority needs scores for its own bills and the minority uses the same scores to attack them. The Dutch planning bureau survives because parties submit their manifestos for costing and declining to submit is the costly move. The design question is therefore not how to protect the evaluator, but who needs its output badly enough to defend it.
Johan’s continuation ·
Johan points to the productive transformation beneath the policy rivalry. Tasks become deeper and narrower along a chain: mining, production, assembly, distribution and sale. At the same time, standardised parts, interfaces and routines allow those tasks to be repeated across many products and markets. These are two dimensions of specialisation, not necessarily vertical integration within one firm. Smith’s argument that the division of labour is limited by the extent of the market supplies the starting point [1].
Where design, tooling and specialist knowledge have large fixed costs, a wider market can spread those costs. A small improvement can become valuable when sold repeatedly worldwide; differentiated product lines can reach more buyers without rebuilding every underlying process. Krugman’s scale-and-variety model explains how larger markets can support both larger production runs and more choice, under its stated assumptions [2].
Variety at the shopfront need not mean diversity underneath. Different models or brands can depend on the same platforms, suppliers and production methods. That is the monoculture risk proposed here, not proof that specialisation always produces monopoly. Boeing and Airbus, cars, agricultural machinery, shipping and retail are starting examples for the separate inquiry; fast food, delivery and car rental extend the question into services. Their competitive structures differ, and not all innovation is incremental.
Buyers are also producers, directly through work or indirectly through the systems that support it. Compared with a household farmer growing food and knitting clothes, exchange permits much greater assortment while increasing dependence on other specialists. This is an analytical contrast, not a claim that past households were wholly self-sufficient or that modern farmers lack specialised expertise.
Mechanisation has a long history, but substantial changes can occur within one human lifetime. The World Bank traces the expansion of international value chains after 1990, while also recording their stalled expansion after 2008 [3]. IFR reported about 1.48 million operational industrial robots worldwide in 2014 and 4.66 million in 2024—more than tripling within ten years [4] [5]. This documents expansion, not universal acceleration or net replacement of human labour.
For CXV, this is a central proposed driver of country and policy competition: governments seek the jobs, tax bases, capabilities and strategic supply attached to scale-dependent production. Rival plans can nevertheless count on the same worldwide buyers. The next section’s duplicate-capacity test therefore asks whether national commitments fit the shared market. Security, resources and domestic politics remain other drivers. The separate task will test when specialisation produces concentration, when it sustains real diversity, and when cooperation improves outcomes; the debt scenarios do not establish those causal claims.
Evidence boundary: the sources support market-scale mechanisms and dated changes; monoculture risk and their role in policy rivalry remain a synthesis to test, not a completed comparison of every industry.
The second half of the observation was that governments compete when specialisation requires them to coordinate. The coordination failure is not a failure of goodwill. It is a failure of visibility: no government can see the others’ capacity plans in a comparable form until the money is committed, and by then the duplication is sunk.
The evidence is not prospective. Solar cell and module facilities already run at roughly half utilisation worldwide. Indian module capacity reached about 233 gigawatts by mid-2026 against domestic demand a fraction of that, and a facility needs fifty to sixty-five per cent utilisation to operate economically. The OECD now links subsidy support directly to low utilisation. Duplication is not a risk that might arrive.
The instrument that follows is a register rather than a treaty: six fields per node, published annually in one format. Node, planned capacity, public support as a single annual figure, expected life, recovery rate, and the number of other registered jurisdictions building the same thing. Publication is a unilateral act, so it can begin without unanimity, and its value rises with each participant without anyone conceding anything.
The design rule is the one this sequence has used throughout: coordinate where the recovery rate is low and compete where it is high. An asset with no second use destroys value when duplicated, so coordination pays. An asset that can be redeployed makes competition efficient, and coordinating it would be a cartel. Getting the rule backwards is how industrial policy becomes collusion.
And the lifecycle requirement applies to the register itself, where Europe supplies a case that cuts both ways. The state aid frameworks of the past five years did expire and were replaced on schedule: the pandemic framework was wound down, the Temporary Crisis Framework of March 2022 was replaced that October, and the Temporary Crisis and Transition Framework was withdrawn in June 2025 and replaced by the Clean Industrial Deal framework. The terminal conditions worked. What did not happen is termination. Each expiry was followed by a successor carrying the same function under a new name, which is the specification change this Part describes, applied to the instrument rather than to the policy. A regime that renews itself under a new title every two years has a terminal condition on paper and none in effect, and the inattention sits in the succession rather than in the expiry.
Correction, 11 November 2026: the earlier comparison mixed US federal/public-debt measures and federal revenue with EU member general-government accounts. Its interest burdens therefore used different government boundaries. The earlier fiscal comparison and its affordability conclusions are withdrawn, rather than carried over to new numbers.
Hover, focus or tap to inspect the correction. Tap to keep it open; press Escape to close. Without JavaScript, expand this explanation normally.
The detailed source limitations, frozen inputs and model downloads appear beside the recalculated tables below.
The recalculation uses IMF October 2026 general-government accounts on both sides: US federal, state and local government; EU27 member central, regional/state and local government and social security funds. Net interest is derived as primary balance minus overall balance, not gross interest expenditure, and is divided by total general-government revenue, including social contributions. The EU27 is a GDP-weighted statistical aggregate, not a joint budget; separate EU-institution liabilities are not included in this member-government aggregate. Matching sectors does not make debt definitions identical: US pension liabilities, valuation and accrual adjustments can differ from EU accounts. Consolidation and aggregation require the source definitions, not simply adding headline debts. The baseline and scenario tables below show the recalculated values without turning those accounting differences into a policy verdict.
The revised model is model/cxv-debt-paths.py, with scenario output in model/cxv-debt-paths-output.csv and matched inputs and annual paths in the downloads below. It runs from 2025 through 2050, with 26 annual observations. Common nominal GDP growth is held at four per cent; the effective net interest cost rises by one percentage point through a six-year rolling adjustment. Revenue and baseline primary balance ratios to GDP are held constant, and stock-flow adjustments are set to zero. Additional annual deficits of 0.3 or 1.0 percentage points of GDP start in 2026; the first 2025 row is unchanged across scenarios. The next-step stabilising balance is pb* = (r_h − g) times d_h / (1 + g), where r_h is the effective net cost proxy, not a market issuance rate. A positive primary balance is a surplus; a positive adjustment means tightening. The stabilising primary balance can be negative when the assumed net cost is below growth. This is reproducible, and conditional on the stated inputs and assumptions, not a real-world forecast or a cost estimate for a particular policy.
Source vintage: IMF October 2026 WEO — 2025 baseline; mixed actual/estimate. The 2025 baselines below use general-government accounts on both sides. The six scenario rows are conditional calculations, not a real-world forecast.
The scenario export contains all six recalculated rows. The annual export contains all 156 jurisdiction/scenario/year observations, not just the chart selection.
Net interest is derived as primary balance minus overall balance, not gross interest expenditure. Net interest/revenue means this net-interest amount divided by total general-government revenue, including social contributions, on the same sector basis; the ratio is expressed as a percentage.
On narrow screens, scroll each table horizontally. Keyboard users can focus a table region and use the left and right arrow keys.
| Jurisdiction | Gross debt | Total revenue | Overall balance | Primary balance | Derived net interest |
|---|---|---|---|---|---|
| United States general government | 124.4 | 30.5 | -7.1 | -3.9 | 3.2 |
| EU-27 member general governments (GDP-weighted; not pooled) | 82.6 | 46.2 | -3.1 | -1.5 | 1.6 |
United States: federal, state and local general government. EU27: member central, regional/state and local government and social security funds. The EU27 is a GDP-weighted statistical aggregate, not a joint budget; separate EU-institution liabilities are outside this member-government aggregate. Consolidation and aggregation are not simply adding headline debts. US pension liabilities, debt valuation and accrual adjustments can differ from EU accounting; matching sectors does not remove those caveats.
The IMF classifies these series as general government, but country coverage is not identical. Croatia and Malta list fewer separate subsectors; Latvia explicitly excludes some reclassified units and accounts payable. Bulgaria, Estonia, Latvia and Romania have cash-basis tags, unlike the accrual tags elsewhere. Debt valuation and instrument definitions also vary. The EU total uses national ratios weighted by nominal GDP at market exchange rates; it does not additionally consolidate cross-member claims or include a separate EU-institution balance sheet.
The 2025 US inputs are IMF-reported actuals subject to revision. Austria’s fiscal inputs remain IMF estimates and Bulgaria’s are preliminary. This is a source-vintage and broad-sector correction, not a claim that every national accounting difference has been eliminated.
IMF WEO dataset and country metadata| Jurisdiction | Scenario | 2025 net interest/revenue (%) | 2050 net interest/revenue (%) | 2050 gross debt/GDP (%) | Next-step stabilising primary balance 2050 (% GDP) | Adjustment 2050 (percentage points GDP) | First year net interest reaches at least the selected 25% revenue threshold |
|---|---|---|---|---|---|---|---|
| United States general government | Baseline balance held constant | 10.4 | 21.7 | 197.3 | -0.9 | 3.0 | Not reached through 2050 |
| United States general government | Additional annual deficit 0.3pp GDP | 10.4 | 22.4 | 204.3 | -0.9 | 3.3 | Not reached through 2050 |
| United States general government | Additional annual deficit 1.0pp GDP | 10.4 | 24.2 | 220.8 | -1.0 | 3.9 | Not reached through 2050 |
| EU-27 member general governments (GDP-weighted; not pooled) | Baseline balance held constant | 3.4 | 5.6 | 94.0 | -1.0 | 0.5 | Not reached through 2050 |
| EU-27 member general governments (GDP-weighted; not pooled) | Additional annual deficit 0.3pp GDP | 3.4 | 6.0 | 100.5 | -1.1 | 0.8 | Not reached through 2050 |
| EU-27 member general governments (GDP-weighted; not pooled) | Additional annual deficit 1.0pp GDP | 3.4 | 6.9 | 115.8 | -1.2 | 1.3 | Not reached through 2050 |
Assumptions: Common nominal GDP growth of 4% annually; the effective net interest cost rises by 1 percentage point toward its assumed endpoint through a six-year rolling adjustment. Revenue/GDP and the baseline primary balance/GDP are held constant, with zero stock-flow adjustment. The scenarios are “Baseline balance held constant”, “Additional annual deficit 0.3pp GDP” and “Additional annual deficit 1.0pp GDP”. Additional deficits start in 2026 and lower the primary balance by the stated amount each year; the first 2025 row is unchanged across scenarios. These are deficit sensitivities, not estimates of specific industrial-policy costs.
No crossing: A dash (—) means net interest does not reach the selected 25% of revenue threshold during 2025–2050; it makes no claim about later years.
Sign convention: A positive primary balance means a surplus; a positive adjustment means fiscal tightening. A negative next-step stabilising primary balance permits a primary deficit under the assumed net cost and growth rates; it is not a sign error. Adjustment is measured in percentage points of GDP.
Selected diagnostic: The 25% revenue threshold is a selected diagnostic threshold, not a legal or validated affordability ceiling.
Interpretation limit: These figures are conditional scenarios, not forecasts and not proof that cooperation is necessary. Net interest is an accounting-derived cost proxy, not a gross coupon or a market borrowing rate. This model does not capture policy responses, member-state financing differences or future stock-flow adjustments.
Frozen conditional scenarios, not forecasts: all 156 annual rows from the corrected model, 2025–2050. Filtering changes visibility only, not model parameters or the common scale.
Debt is gross general-government debt. Net interest is primary balance minus overall balance, not gross interest expenditure; net interest/revenue divides that amount by total general-government revenue, including social contributions. EU27 is a GDP-weighted member-government statistical aggregate, not a common budget. Sector matching does not eliminate the accounting differences described in the accompanying tables.
Net debt subtracts financial assets in debt instruments from gross debt; it is not net interest. Net-debt paths are unavailable on the full EU27 basis. Switching to net debt shows the source gap, not substitute curves. Net interest/revenue remains available independently of this switch.
+0.3pp / +1.0pp mean additional annual deficit in percentage points of GDP from 2026; all scenarios share the 2025 baseline.
The full default chart and 2050 values are shown without JavaScript. Interactive controls enable only when JavaScript is available; otherwise use the accompanying native tables and full annual-path CSV. On narrow screens, scroll the chart horizontally; keyboard users can focus its region and scroll with arrow keys. Numeric cards remain below the chart, outside the scrolling area.
Gross debt / GDP (%) · All three scenarios · fixed 0–250% axis for both regions
Unrounded numeric values from the frozen annual CSV (% of GDP). These digits preserve the model output, not forecast precision.
Reproduce these six rows from the matched general-government inputs and corrected model. No Python code is executed by this webpage.
Source definitions: IMF Fiscal Monitor; Eurostat government finance statistics; US Treasury federal-debt definitions.
The policy cards and argument that follow are the Part's interpretation, not a theorem proved by these debt paths. The selected 25 per cent revenue threshold is a diagnostic choice, not a legal ceiling or an independently validated affordability limit. The calculation does not establish that cooperation is the only possible strategy. The earlier categorical US and EU affordability claims are not reinstated by recalculation.
The calculation covers federal, state and local accounts, not the federal budget alone. It tests how additional deficits affect debt and net interest under fixed assumptions; it does not establish that the United States cannot afford a subsidy race or that cooperation is its only affordable option.
The EU27 member-government aggregate is not a common treasury and cannot show that fiscal room is available to every member or transferable between them. Member-state financing conditions and decision procedures remain distinct; these scenarios establish neither affordable subsidy competition nor an inability to coordinate.
Cooperation remains a proposal for reducing duplication, not the only affordable US strategy or the only feasible EU strategy demonstrated by these accounts. A selected interest/revenue threshold is not an affordability test, and an aggregate debt path cannot prove that twenty-seven decisions cannot be coordinated. The register must be tested against alternatives: whether it reduces duplicate capacity and fiscal cost without sacrificing useful competition. A state that coordinated its own capacity without loss would still challenge the need for cross-border coordination in that case.
If the model is right, a state that recognises early should outperform one that defers — and there is a case that looks like the demonstration. The Limburg mines closed between 1965 and 1974, and the same region then carried a chemical cluster and a university. KLM merged with Air France in 2004 while still profitable, which is recognition before the market repriced and recovery captured in the price rather than destroyed in a write-down.
But the same country supplies the counter-example, and the counter-example is fiscal before it is anything else. The Netherlands levies a fifteen per cent withholding tax on dividends that the United Kingdom does not, and Shell and Unilever lobbied for its abolition for the better part of two decades. Unilever’s first attempt to unify in Rotterdam collapsed in 2018 on exactly that tax. When a GreenLeft amendment proposed a conditional exit levy in 2020, Unilever put its own exposure at eleven billion euros and said the bill would reverse the unification; the Council of State called the proposal irresponsible, and the company left anyway. Shell followed in January 2022, after the last attempt at abolition failed. Two departures, one fiscal cause, and twenty years in which the state could have settled it and did not.
What the firms published on the way out is more useful to this argument than the exits themselves, because it is the instrument of this Part being used by a company against its own workforce. Unilever’s completion notice of 30 November 2020 states that there will be no change to the operations, locations, activities or staffing levels in either country as a result of unification. Shell’s Netherlands recruitment pages, still running in 2026, state that even after the headquarters moved to London in January 2022 the commitment to the Netherlands remains very strong, that The Hague still houses the headquarters of three business divisions and that Rotterdam remains a trading, marketing and sales hub — alongside 9,380 Dutch employees in March 2024, the Pernis refinery, the Moerdijk chemicals park, the Amsterdam transition campus and 6.5 billion euros invested in low-carbon projects in the country.
Apply the four-part test and neither sentence is a commitment. Unilever’s is bounded by a transaction: it states what unification will not change, not what the company will maintain, and a sentence true on one day has been read ever since as a standing promise. Shell’s is a statement of present fact with an indefinite horizon. Neither carries a headcount floor, a review date, a terminal condition or a beneficiary count, and no published document supplies one — the absence is the finding rather than a gap in the search. Both are the form this Part criticises in a statute, adopted voluntarily by firms that had every reason to understand what they were doing. The workforce ends up in the fifth row of the table: exposed to a decision whose state it cannot observe, secured by a sentence with no barrier on it.
So the symmetry test is narrower than a story of engineered departures, and it cuts in both directions. The state lost an argument it had two decades to settle, and the firms left with a promise carrying none of the four elements that would have made it checkable. Applied in one direction the four traces — price against a commercial valuation, timing against the operating record, migration of patents and headcount, the behaviour of the acquirer afterwards — describe ordinary fiscal defeats. Applied in the other, the promises are the same deferral this Part describes, made by the party that was leaving. What the Dutch employees needed was a published plan with a horizon, a floor and a review date. What they received was a sentence in a press release and a page in a recruitment brochure. The discriminator between conversion and departure remains whether the infrastructure was redeployed in the same place within a decade: the mines converted, and the headquarters took the capability with them.
The deeper point is structural and slightly deflationary for the national reading. Recognition speed is a function of the size of the buffer, not of virtue. A large economy can run a strategic industry at a loss for a decade because it has an internal market and a state balance sheet to amortise it. A small open economy cannot, so its choice is between exiting early at a decent price and exiting late at zero. The Dutch pattern is a constraint handled well, which is a real achievement and a different one from foresight.
The comparison changes what the Dutch case can be used for. A state facing a corporate migration holds four instruments, and they do different work: remove the tax that triggers the move, keep the tax and rebuild the refund machinery, charge the company for leaving, or make leaving unnecessary. The Netherlands argued about the first for twenty years, reached for the third in 2020, and is now subject to the fourth like every other member state.
Removal is the instrument that decided both cases, and it is the one the Dutch debate could not name honestly. The United Kingdom withholds nothing on dividends. Ireland withholds twenty-five per cent and relieves eligible non-resident holders, but on conditions: a holder resident in a treaty or European Union jurisdiction, satisfying the beneficial-ownership tests, and in some cases recovering the tax by refund rather than receiving it free of charge. Its corporation tax is twelve and a half per cent on trading income, twenty-five per cent on specified non-trading income, and fifteen per cent at the floor for large groups under the Pillar Two rules. The relevant contrast is not the Irish rate. Both jurisdictions tax a dividend paid to the shareholder, and Ireland relieves the holder rather than the company, which is the party a listed company is trying to keep. Abolition was expensive in revenue and cheap in political cover, which is why it failed twice in The Hague.
The second instrument accepts the rate and attacks the friction, and it is where Europe has actually legislated. The FASTER Directive, adopted in December 2024, replaces more than 450 national reclaim forms with a single digital residence certificate, relief at source and a refund within fifty days, with transposition due by the end of 2028 and application from 2030. Its stated justification is a refund system slow enough to be worth avoiding and opaque enough to be worth abusing: 150 billion euros lost to Cum/Ex and Cum/Cum between 2000 and 2020. The Netherlands has taken a narrower version of the same turn since 2024, keeping the fifteen per cent and adding a conditional 25.8 per cent on dividends paid into low-tax jurisdictions. The base was narrowed and the rate was defended.
The third instrument is the one the Netherlands chose, and its legal effect is narrower than its name. National Grid Indus held in 2011 that an exit charge on unrealised gains is compatible with freedom of establishment, but that requiring payment at the moment of migration is disproportionate. Polbud then extended establishment to a conversion that moves only a registered office, and Article 5 of the anti-avoidance directive codified the compromise: the gain is assessed when the assets leave, payment is deferred for eligible transfers within the Union or the EEA, and the deferred amount may be settled in five annual instalments, with interest and guarantees where a member state requires them, and with acceleration on later disposal or on a transfer outside the EEA.
What remains is a claim on value rather than a barrier at the border. The charge is assessed whether or not anything is sold, it is collected later, and what a company discounts is the later collection. Whether that deferral changes a relocation decision is an economic question the case law does not answer and this Part does not measure. The honest statement is narrower: the instrument was built to protect the tax base, its immediate effect on movement is attenuated by design, and nothing here establishes that the effect is zero.
The fourth instrument is company law, and no member state can use it alone. The Commission proposed a twenty-eighth regime in March 2026, offering registration within forty-eight hours at a maximum of one hundred euros, on the premise that firms migrate partly because operating through twenty-seven subsidiaries is expensive. The same logic produced the common substance test for shell entities, which never achieved consensus: the Unshell proposal was formally discontinued in June 2025 and replaced by a simplification package amending the parent-subsidiary and interest-royalty directives instead. Europe’s answer in 2026 is to harmonise the paperwork and abandon the substance test, leaving rate competition untouched because unanimity makes it untouchable.
Returned to the four-part test, the comparison locates the Dutch failure precisely. What The Hague lacked was never an instrument. It was a review date. The dividend tax was debated, shelved and debated again across four cabinets with no scheduled moment at which the question would be settled either way, and two companies eventually settled it. The charge that was then legislated arrived after the traffic had passed, which is a terminal condition applied at the exit rather than a lifecycle instrument applied before it. The retention statements the firms published carried the same defect from the other side, with no floor, no review date and no beneficiary count. A state that had scheduled the decision would have had something to negotiate with. A state that deferred it had a promise to accept.
This does not make the tax the whole cause, and it should not be read as a case for abolition. Shell’s move also collapsed a dual share structure and aligned its listing with an Anglo-American shareholder base; Unilever’s ended a ninety-year arrangement that had become an administrative liability. A state that had abolished the tax in 2015 might still have lost both, and that counterfactual cannot be run. What can be said is narrower and survives it: of the four instruments available, the one the Netherlands reached for arrives after the decision, and the one it refused for twenty years is the one the departing companies named.
Four limits, stated rather than buried.
The falsifier is concrete. A programme with a declared life, a scheduled review and a measurable terminal condition that nevertheless ran a decade past its threshold without political contest, in a jurisdiction with a functioning evaluator, would show that the four-part test is decorative. Several such cases would end the argument.
Part CXIII — The Instrument of Correction asked what would count as evidence that a correction happened, and answered with a standard, a register and a sense organ. This Part supplies the object the instrument should be pointed at: a commitment with a specification and no carrying value, where the absence of an impairment test is the defect rather than the absence of a signal.
Part CXIV — The Employment Problem found that a specification which does not run is indistinguishable from one that does not exist, and named the host who must carry it. The evaluator described here is that host for a different object, and the sealed statement is the first-year publication that makes the host’s duty checkable.
Part CXI — The Correction Function found that durability belongs to a named reviser and that the fifth link kept failing because nobody held the duty, the money and the term of office. A review clause is that finding written as a clause: it gives the reviser a date, and it costs the incumbent nothing to honour because the default is termination.
Part CX — The Immunity Counterparty required independence to be operational rather than declared. The boundary drawn here — the evaluator measures the fit and must not set the cost of public funds — is that requirement applied to a threshold, and it is the same reason a central bank is given an instrument rather than a preference.
Measured findings, structural derivations and this Part’s own definitions are kept distinct here. Every figure printed on the page is attributed below and carries a URL. The only projection of the Part’s own is the debt comparison, provided as model/cxv-debt-paths.py with matched inputs and output files, and labelled as conditional wherever it appears. The source definitions and material caveats are stated alongside it.
Part CXIV predicted that Part CXV would be The Red Team, a protocol for attacking the instrument itself. Johan’s question of 10 November 2026 redirected the sequence, and the Red Team remains open rather than dropped. The question this Part cannot answer is the one its own table exposes: the fifth row.
A note on sequencing: Part CXIV named The Red Team for this numeral. The redirect is recorded on this page and corrected on the earlier page in the same commit, because a Branch Point is a claim about the site and has to stay true.