The Mandate Gap

I. Two specimens

Specimen 1: UK PFI hospital approvals, 1997–2010

HM Treasury’s hundreds of Private Finance Initiative approvals between 1997 and 2010 followed a uniform pattern. An NHS Trust would submit a Value for Money assessment built around a Public Sector Comparator — a hypothetical model of what the project would cost under conventional government borrowing. Treasury approved the contract if the PSC showed PFI to be cheaper. The model was the gate.

The approval architecture relied on Value for Money and Public Sector Comparator modelling whose key adjustments — optimism bias, risk transfer, discounting, tax assumptions — were difficult for many contracting authorities to challenge. Two adjustments did most of the work in flipping the comparison toward private finance: an optimism bias uplift applied to the public-sector baseline to penalize it for hypothetical cost overruns, and a risk transfer line that moved theoretical project risk onto the private partner at an asserted value. Because government borrowing is cheaper than private borrowing in nominal terms, the model had to manufacture a counterweight large enough to flip the comparison. The optimism bias and risk transfer adjustments did exactly that.

NHS Trusts lacked the financial-modeling depth to challenge the bespoke algorithms. They accepted the parameters. HM Treasury approved the contracts on the approved comparators. Successive Chancellors cited the cumulative VfM approvals as evidence that PFI was a verified-efficient mechanism, and the approval record became the political shield.

The National Audit Office’s January 2018 report unwound the legitimation. By that point the United Kingdom held over 700 operational PFI and PF2 deals with a capital value of roughly £60 billion, annual charges of £10.3 billion in 2016–17, and future charges extending into the 2040s of approximately £199 billion. The NAO concluded that the optimism-bias and risk-transfer uplifts inside the VfM methodology were “not evidenced,” and that the cumulative effect of the methodology biased outputs systematically toward private financing. The NAO also recorded the Office for Budget Responsibility’s characterization of the structural attraction of PFI as a “fiscal illusion”: projects appeared cheaper because they did not record the full capital value as public debt, even when they cost significantly more over a 25–30 year horizon. On cost, the NAO documented that one PF2 schools group’s financing ran around forty per cent higher than equivalent projects funded by conventional government borrowing; a higher figure — the often-cited seventy per cent — derives from a 2011 Treasury Committee analysis of a privately financed hospital comparison, cited by the NAO but not reproduced as a generalizable NAO finding. The Carillion collapse in 2018 and the Midland Metropolitan hospital project’s freeze made the laundering visible at asset-specific scale.

The citation lasted far longer than the analysis behind it. A formal authority closed on a complex causal mechanism. The mechanism was shaped by parties whose interests aligned with approval. The seat’s bandwidth and technical depth were insufficient to audit the supplied frame. The closure was cited as substantive review for over a decade.

Specimen 2: Microsoft–OpenAI competition reviews, 2023–2026

The UK Competition and Markets Authority, the European Commission, and the US Federal Trade Commission each examined Microsoft’s investment in OpenAI — over $13 billion across an initial 2019 round, the January 2023 $10B announcement, and additional sums in 2021 and October 2024 that remain partly redacted in the public record. Two of the three produced formal closures under merger-control mandates.

The European Commission concluded over the summer of 2024 that the partnership did not constitute a notifiable concentration under the EU Merger Regulation; EVP Margrethe Vestager articulated the position in a 29 September 2024 speech (speech_24_4966). The CMA Phase 1 Decision Group ruled on 5 March 2025 that the partnership did not qualify as a “relevant merger situation” under the Enterprise Act 2002. The CMA found that Microsoft had acquired material influence over OpenAI in 2019 and that the 2024–25 question was whether this had increased to de facto control; it concluded there was no such change. On the November 2023 Altman events, the CMA recorded that Microsoft was not consulted before the dismissal, received less than an hour’s notice, and subsequently supported Altman’s return but was not decisive alone in his reinstatement. The CMA’s full reasoned decision was published 15 April 2025.

The CMA’s own language on the closure is the cleanest summary of why the closure act does not carry what the post-decision debate treats it as carrying. The full decision states that the finding of no relevant merger situation “does not constitute a finding that no competition concerns arise.” The closure was jurisdictional, not substantive.

The FTC’s January 2025 §6(b) staff report does not belong in the same category. A §6(b) study has no enforcement output and does not adjudicate a transaction. Structurally, the §6(b) report is closer to Powerless Intelligence — evaluation capacity producing findings that no procedural primitive converted into binding action.

The CMA and EC closures rest under statutes drafted for the equity-and-control architecture of twentieth-century industrial concentrations. The partnership’s substance — Azure compute credit, board-observer rights, intellectual property licensing under non-profit governance, revenue sharing — is a topology those statutes were not engineered to compute. Frontier-AI capability concentration, compute supply-chain control, safety evaluation, and dual-use proliferation were either out of scope, structurally beyond the authorities’ analytical bandwidth, or both.

Subsequent corporate restructurings have unwound parts of the original exclusivity architecture: a non-binding Memorandum of Understanding on 11 September 2025, a separate OpenAI–AWS $38 billion strategic partnership on 3 November 2025, and Microsoft’s 27 April 2026 amended agreement narrowing Microsoft’s compute exclusivity — Azure retains first-look for new OpenAI products unless Microsoft cannot and chooses not to support the necessary capabilities, and Microsoft’s intellectual-property license becomes non-exclusive. Whether the 2024–2025 regulatory closures should be cited as having reviewed the deal architecture that emerged is the contested question.

What “the regulator cleared this” carries is the closure act. What it does not carry is review of the questions the closure act was not designed to ask.

II. Formal mandate is not operational mandate

The two specimens differ in domain, era, and stakes. They share an architecture.

A formal mandate is the legal right to close a decision: a parliamentary vote, a regulator’s clearance, a cabinet sign-off, a board approval. It is binary and easy to verify.

An operational mandate is the conjunction of four things that have to converge in the same actor before the closure can carry substantive weight:

• Owner — a named institution accountable for the mechanism, not a function distributed across many hands.
• Authority — legal power sufficient to compel the information needed to compute the mechanism, including delay of closure when computation is incomplete.
• Capacity — analytical bandwidth, technical depth, data access, model access, and time to actually compute the mechanism within the closure window.
• Feedback — binding consequence when outcomes diverge from the mechanism the closure was nominally approving.

When the four converge, closure is substantive. When formal mandate is present but operational mandate is absent — owner exists, authority exists on paper, capacity is structurally insufficient, feedback is non-existent or decorative — the closure is laundered: the authority’s assent is taken by the broader system as a substitute for the work the authority did not do.

Three neighboring failure modes sit close enough to be confused with this one and have to be distinguished cleanly.

A Telos Gap — failures that belong to nobody — is the case where the causal object fails to enter any institution’s admissible-object intersection at all. No owner exists. Mandate Gap is the case where an owner is formally designated. The two failures often co-occur — a municipal council approving a hyperscaler datacenter exhibits both Mandate Gap at the closure seat and Telos Gap at the missing national lifecycle owner — but they are distinct primitives.

The Mandate Trap — organizations are selected to preserve their mandate, not to solve the root cause their mandate names — explains why no upstream organization grows into the missing function. The vacancy at the lifecycle layer is real. The Mandate Trap is about why nothing fills it. Mandate Gap is about what happens when the system pretends something has, by pointing at a formal seat that the broader system then treats as the substantive owner.

Powerless Intelligence is the inverse case. There, evaluation capacity exists — an AI safety institute, a central-bank macroprudential office, an FTC §6(b) study — but the response-duty primitive that would convert a finding into binding action is absent. “We know, but cannot act.” Mandate Gap is the mirror: “We must decide, but cannot know enough.”

The relations are an entry diagram for the territory, not the full taxonomy. The actual diagnostic test runs the four-part repair specification end to end.

The bottom-right cell is conditional: absent binding feedback, mandate-plus-capacity produces expert discretion, ritual consultation, or one-shot closure rather than operational mandate. The four-part repair space is the real framework.

III. The mechanism

Five steps, in order. A case qualifies as Mandate Gap when all five hold.

1. A complex causal mechanism requires closure. A procurement contract, a regulatory clearance, a treaty ratification, a strategic infrastructure approval, a funding-model authorization. Closure is the irreversible commitment of public capacity.
2. Formal closure authority is assigned to a specific actor. A parliament, a committee, a cabinet, a regulator, a board, a council. The seat is named. The seat has legal power to approve, reject, or delay.
3. The causal object exceeds the seat’s operational scope. The mechanism depends on cross-sector effects, long horizons, technical models, adversarial counterparty strategy, hidden ledgers, multi-decade lifecycle commitments, or implementation details outside the seat’s normal object-language. The seat can hold the proposal; it cannot independently compute it.
4. No institution with procedural force computes the mechanism before closure. Reports may exist. Expert advisory groups may exist. Journalism may exist. None has procedural force to become a precondition of closure, and none is constructed inside the seat’s own structurally-protected analytical capacity. The seat closes either on supplied frames or on no frame at all.
5. Closure is laundered through the formal mandate. Afterwards, the system says: parliament decided, the regulator cleared it, the committee had hearings, the cabinet approved. The closure becomes the substantive-review answer in subsequent debate. The post-hoc citation does not specify what was actually computed, by whom, with what bandwidth, against what counterfactual. The citation does the substantive work that the closure did not.

The fifth step is what distinguishes Mandate Gap from ordinary capacity failure. Ordinary capacity failure: “they tried to perform their function and lacked resources.” Mandate Gap: “their formal assent was used as proof the function had been performed.” The discriminator is rhetorical and procedural — who is treated as the legitimate owner of the mechanism after closure, and how the closure is invoked when the mechanism comes under question.

The pattern is empirically falsifiable. A case is not Mandate Gap if any of the following hold:

• The closure actor had structurally protected access to relevant models, time, staff, jurisdiction, and a binding expert process — and used it.
• The closure actor could compel missing information and delay closure on inadequacy, and exercised this power when needed.
• The decision record demonstrably shows the mechanism was actually computed by the seat or by a body the seat compelled and subsequently reviewed.
• Post-outcome feedback binds the closure actor or a named operational owner, and the binding is observed in revisions, halts, or recompositions when outcomes diverge.

The Finnish HX fighter procurement supplies a stronger-operational-mandate case. The Finnish Defence Forces ran the HX Challenge — physical testing of each candidate aircraft under Finnish operational conditions at Pirkkala Air Base in early 2020 — before the Finnish Government’s 10 December 2021 decision authorizing the Defence Forces Logistics Command to acquire the Lockheed Martin F-35A. The closure rested on sovereign empirical data the Finnish state generated and owned for the air-system performance dimensions of the choice. The Danish 2016 F-35 closure ran the opposite way: the same procurement choice closed on vendor-influenced evaluation models that subsequently failed the Rigsrevisionen audit, with infrastructure cost overruns attributed to vendor optimism the original closure had absorbed without independent verification.

Same vendor, same aircraft, same strategic motive, different operational mandate. The mechanism is structural, and the difference is the analytical capacity the closure seat invested in before closure.

IV. Five types

The mechanism above identifies the laundering pattern. The types below identify which dimension of operational mandate the closure seat was missing. The categories are recurring across the empirical record, not claimed exhaustive.

Bandwidth gap. The closure window is too short for the seat to compute the mechanism within available time, staff, and competing demands.

TARP, October 2008. The US Congress received a three-page Treasury proposal during a week in which credit markets had effectively frozen. Independent economic modeling of an alternative systemic intervention was structurally unavailable in the time given. The Emergency Economic Stabilization Act passed 3 October 2008. Treasury then used the broad statutory language to pivot from the proposed asset purchases to direct capital injection into banks, citing the rushed passage as democratic mandate for a mechanism Congress had not approved at the level of mechanism.

Model/data-access gap. The seat is formally responsible but the analytical models, data, or technical-verification capacity live with the counterparty.

PFI 1997–2010 (full case in §I). HM Treasury approved hundreds of contracts on Public Sector Comparators whose internal mathematics — optimism bias, risk transfer — the NHS Trusts holding closure authority lacked the financial-modeling depth to audit.

EU Commission clearance of Facebook–WhatsApp, October 2014. The Commission’s clearance rested on Facebook’s representation that automated matching of Facebook accounts to WhatsApp users would not be technically feasible. Two years later WhatsApp updated its terms of service to enable that matching. The Commission’s May 2017 €110 million fine for misleading information left the original 2014 clearance valid. The structural outcome — Facebook’s data architecture extending across the merged platforms — survived the falsification of the technical premise the clearance had rested on.

Jurisdiction-scope gap. The seat owns one slice; the causal object spans jurisdictions, sectors, or generations the seat cannot see.

Municipal closure on hyperscaler-scale datacenter projects (recurring across European jurisdictions, 2022–present). A municipal council holds zoning and land-use authority for the local site but does not own national electricity-market effects, transmission-cost socialization across the wider grid, alternative-use opportunity costs at the same node, heat-system counterfactuals, sovereign-compute implications, or twenty-to-fifty year industrial precedent. Closure happens at the seat that owns the smallest slice of the mechanism, and the broader system cites “the council approved” as if it had reviewed the cross-jurisdictional effects.

Counterparty-asymmetry gap. The seat sees one project; the counterparty sees the portfolio game.

AUKUS Pillar 1, September 2021 (candidate case). The Australian National Security Committee closed on a multi-decade nuclear-propulsion partnership. The Australian Government’s March 2023 official estimate for the SSN-AUKUS pathway over three-plus decades was A$268–368 billion. Workforce-generation, US shipyard-capacity, and IP-transfer constraints became operationally visible after closure. Whether they were adequately computed during closure is contested and rests on records that remain partly classified; AUKUS is offered here as a candidate Mandate Gap, not a proven specimen.

Lifecycle gap. The seat closes on installation; the mechanism’s substance lives in the lifecycle the seat does not own.

Japanese Diet approval of Aegis Ashore, December 2017, cancelled June 2020 (candidate case). The Diet approved the procurement on capability documentation provided through the Foreign Military Sales process. Twenty-nine months later the deployment was cancelled because the SM-3 Block IIA booster separation behavior over Japanese populated areas required additional work estimated at approximately ¥200 billion and a decade. The cancellation was the post-outcome feedback the original closure had not been bound by.

These are the same structural pattern viewed from the angle of which dimension of operational mandate was missing in each case. A condensed typology:

V. The repair

The standard repair instinct is to educate the formal actor into becoming the missing institution. Train councillors. Give parliaments more advisory bandwidth. Build expert committees. Run summits.

That repair keeps the same broken seat in charge. The structural problem is not that the seat lacks information. It is that the seat is being treated as the substantive owner of a mechanism it cannot compute, and the broader system uses that treatment as legitimation. More information into the same seat without rebuilding the operational-mandate structure changes what gets laundered, not whether laundering happens.

The actual repair has three components, in dependency order.

Component one: invest in structural antibodies before the closure event arrives. Antibodies come in two distinct varieties, and the difference matters.

A computation antibody is sovereign analytical capacity, generated before closure, owning the parameters most exposed to counterparty framing. The Finnish HX fighter procurement is the cleanest case. The Finnish Defence Forces required each vendor to deploy candidate aircraft to Pirkkala for physical testing under Finnish operational conditions, and the procurement closure rested on sovereign empirical data the Finnish state generated and owned for the air-system performance dimensions that could be empirically tested under those conditions. HX did not make the Finnish Government omniscient. It placed sovereign empirical testing and operational assessment upstream of political closure, so the closure rested on a state-owned evaluation pipeline rather than on vendor documentation alone.

A temporal antibody is statutory protection of the analytical window against executive manufactured deadlines. The UK Parliament’s Benn Act of September 2019 is the cleaner example. It compelled the Prime Minister to request a third extension from the European Council, structurally insulating the closure from a no-deal deadline the executive had been able to set. The Benn Act did not compute the Withdrawal Agreement mechanism; the eventual ratification came after a general election shifted the deciding body. What it did is preserve time in which independent scrutiny could occur.

The two antibody types address different parts of operational mandate. The computation antibody improves what the state knows. The temporal antibody preserves the time in which what the state knows can be put to use. Both have to be structural — funded, staffed, statutorily insulated — because the closure event is precisely when ad-hoc capacity will not assemble in time.

Component two: complexity-threshold referral rules for cases that exceed the seat’s mandate envelope. Even the best-equipped municipal council cannot become the operational owner of a hyperscaler datacenter’s national grid effects. Even the best-equipped parliamentary budget committee cannot become the operational owner of a fifth-generation fighter’s forty-year ITAR exposure. The repair is to make closure conditional on referral to a capacity-bearing institution when the causal object crosses statutory complexity thresholds — large energy load, public-infrastructure dependence, long-horizon grid effect, monopoly-bargaining counterparty, irreversible land-use commitment, national-security supply dependence.

The rule does work independent of whether the receiving capacity-bearing institution exists yet. The rule classifies decisions and makes the absence of an adequate referral target visible as a separate problem to solve. The mechanism-realism literature has discussed candidate architectures — Mechanism Authority, Mechanism Vehicle, the Fourth Branch — at length elsewhere. The point here is that the referral rule does work before the target institution exists. It changes who is responsible for the missing computation from “no one” to “the legislature that has not yet built the target.”

Component three: binding feedback against the closure act. The fifth step of the laundering mechanism — closure cited later as substantive review — works in part because the closure is one-shot. Once Treasury approved the PFI contract, the approval was permanent ammunition for the next PFI contract. Once the CMA cleared Microsoft–OpenAI, the clearance was ammunition for the next AI-partnership clearance under analogous reasoning.

The repair is statutory feedback that binds the closure act to subsequent outcomes. When the mechanism’s downstream effects materially diverge from the closure’s stated rationale — when NAO post-hoc finds PSC adjustments not evidenced, when Aegis Ashore cancels two years after closure, when General Court rules vaccine-contract redactions illegal post-procurement — the closure act loses its citation force. The seat that approved is required to publicly account for the divergence within a bounded review window, and the citation in subsequent debate becomes legally constrained by what the post-closure record actually shows.

This is the most institutionally expensive component of the repair. It requires statutory architecture most parliamentary systems do not currently have. Without binding feedback, antibodies and referral rules become best-effort norms that the next high-pressure closure will route around.

VI. What this is not

Democratic closure remains legitimate when the closure authority has invested in its own analytical capacity ahead of time and structurally protects the right to delay closure on inadequacy. The Finnish HX procurement and the Benn Act show the repair is feasible. The argument is against treating a formal closure as substantive review when the operational mandate was never assembled, not against democratic decision-making as such.

The repair is not a technocratic transfer either. The referral rule in component two requires that closure events above a complexity threshold trigger an analytical procedure whose output enters the decision frame before closure. The substantive choice remains with the political authority. What changes is the analytical floor the closure has to clear.

The mechanism does not require corruption. Honest authorities working in good faith under structurally insufficient operational mandate produce the same laundered-closure outcome as cynical ones. The Bundestag Budget Committee genuinely faced a NATO nuclear-sharing deadline. The European Commission genuinely had to procure vaccines under pandemic pressure. The CMA genuinely had to evaluate the Microsoft–OpenAI partnership against statutes drafted before frontier-AI capability concentration existed as a category. The architecture, not the intent, is the mechanism.

VII. Close

A civilization fails when nobody owns a real harm. It also fails when someone is formally assigned to own a real harm and the system treats that formal assignment as the substantive review. The first failure is the Telos Gap. The second is the Mandate Gap. Both are present in current public-policy architecture across infrastructure procurement, financial-stability response, defense acquisition, treaty ratification, and platform-competition oversight. Neither is resolved by exhortation. Both require structural reallocation of analytical capacity and feedback authority to seats that can actually hold them, and making the absence of that allocation visible as a discrete repair before the next closure event.

The diagnostic question, applied to any institutional closure: what was actually computed, by whom, under what authority, with what bandwidth, against what counterfactual — and what consequence followed when reality diverged? Where the answer is “the closure act itself,” the closure was laundered. Where the answer names independent computation that the seat owned, compelled, or was statutorily bound by, the closure was substantive. Mandate Gap is the place in modern governance where this distinction is most consequential and least visible.

See also: The Telos Gap (failures that belong to nobody — the no-owner case) · The Mandate Trap (why downstream organs cannot grow into the upstream lifecycle owner) · Powerless Intelligence (the inverse case — capacity without authority) · Implementation Ledger (post-closure sibling; conditional closure with preauthorized re-entry as the execution-trace primitive) · What Bureaucracy Is (the positive synthesis: bureaucracy as the runtime of traceable discretion; the mandate gap is one shape of failure at the assign function) · The Stack and The Layer Walk (the broader failure taxonomy this essay sits inside; Mandate Gap anchors at Layer 8 Decision with cross-listing to Layer 6 Compilation, Layer 7 Computation/adoption, and Layer 9 Execution) · Mechanism Authority and The Fourth Branch (institutional repair candidates for the referral-target component)