The Recognition Rule

A planner's framework for the statistical and accounting recognition of PPP liabilities — why they are "silent," where the rule books disagree, and the recognition gap between a contract's headline fiscal-rule treatment and the economic tail the sovereign actually bears — with a reference implementation on the Austral platform

Abstract

Before a public–private partnership (PPP) delivers a kilometre of road or a day of hospital care, a prior question is settled — usually without the Ministry of Finance in the room: does the contract count as government debt? The answer is not economic but statistical and accounting: it depends on which rule book scores the deal and how the contract allocates risk on paper. And it moves the one number a fiscal rule watches.

This paper is the keystone of the Austral series because it derives the "silent liability" the other papers price. A PPP is silent not because its cost is unknowable but because the recognition rule lets it sit off the recorded stock. Three live rule books answer the recognition question and they disagree by construction: the European statistical standard ESA 2010 (with the Eurostat Manual on Government Deficit and Debt) records a PPP off the government balance sheet whenever the private partner bears construction risk and at least one of availability or demand risk; the accounting standard IPSAS 32 (the public-sector mirror of IFRIC 12) recognises the asset and liability whenever the grantor controls the service and the residual interest — a test most PPPs meet; and the IMF's GFSM 2014 follows economic ownership. The same contract can therefore be off the statistical balance sheet, on the accounting one, and economically sovereign regardless. The United Kingdom's Private Finance Initiative is the textbook case: under IFRS/IFRIC 12 (adopted for central government from 2009-10) nearly all PFI sits on the Whole of Government Accounts, yet under ESA/National Accounts the majority stayed off headline Public Sector Net Debt — the Office for Budget Responsibility put the off-balance-sheet PFI stock at roughly £35 billion, about 2.5% of GDP, at March 2010 (OBR 2011, Box 3.3).

The disagreement has a second layer that is the heart of the matter: the statistical rule is only as good as the jurisdiction that applies it. The European rule is strict — under the EPEC–Eurostat 2016 Guide to the Statistical Treatment of PPPs, any minimum-revenue or minimum-demand guarantee puts the PPP on the government balance sheet, and Portugal's 2010 reclassification of road PPPs into general government shows the rule has teeth. But ESA is European; most of the world's PPPs sit under GFSM 2014, IPSAS, or no binding standard at all — Chile and the emerging markets this series addresses — where that strict guarantee rule is not applied, so the same minimum-revenue guarantee lands off the binding debt measure and is disclosed, if at all, only as a contingent liability. The recognition gap is therefore jurisdiction-dependent: it is widest exactly where the rule is weakest.

The contribution is a reproducible cross-framework recognition scorer: feed a contract's risk-allocation matrix and it returns the on/off classification under all four lenses and computes the recognition gap — the value off the jurisdiction's binding statistical measure yet economically sovereign — for the strict (EU/ESA) and the loose (non-EU/GFSM) regime alike. On three worked contracts the gap is legible: an availability hospital off under the statistical rule but on under IPSAS (the UK-PFI divergence, present in every jurisdiction); a demand-risk toll road with a minimum-revenue guarantee that the EU rule catches but a GFSM/IPSAS jurisdiction does not, so its gap appears only outside the EU; and a government-funded social asset the majority-financing test pulls on the books everywhere. The portfolio recognition gap is 1.2% of GDP under the strict EU rule and 1.8% under the looser non-EU regime — the same contracts, a larger hidden liability where the rule is weaker. The policy lead follows: adopt the strict treatment of guarantees (catch the minimum-revenue guarantee on the balance sheet, as the EU does), disclose the recognition gap alongside the rule-relevant treatment, cap the off-balance-sheet stock, and pre-commit the classification at tender.

Keywords: balance-sheet recognition · ESA 2010 · IPSAS 32 · GFSM 2014 · IFRIC 12 · risks and rewards · control · off-balance-sheet · fiscal commitments and contingent liabilities · PFRAM · public sector net debt · recognition gap · Ministry of Finance.

1. The decision problem — does the PPP count as debt?

A PPP contract is signed. Before it delivers anything, a prior question is settled, and it is settled by a rule, not by a negotiation: will this contract be recorded as government debt? The question sounds like accounting housekeeping. It is the most consequential number in the deal, because it is the number a fiscal rule reads.

A government operating under a debt ceiling, a deficit target, or a Maastricht reference value cares intensely whether a thirty-year availability commitment lands inside or outside the measure the rule binds. Inside, the project consumes scarce fiscal space and competes with every other claim on the budget. Outside, it is invisible to the rule — the investment happens now, the cost is booked later and elsewhere, and the headline debt-to-GDP ratio does not move. The same contract, the same cash flows, the same risk to the taxpayer, can be a line in the debt stock or nothing at all, depending only on how the contract allocates risk on paper and which rule book scores it.

This is the origin of the "silent liability" that the rest of this series prices. The Silent FX Liability (Paper 6) and the Indexation Ceiling (Paper 7) take as given that a multi-currency or indexed PPP obligation sits off the recorded debt stock and surfaces only in the bad state of the world. That silence is not a property of the liability; it is a property of the recognition rule. The liability is perfectly knowable — a competent ministry can price it, and this series shows how — but the rule lets it stay off the books. Understand the recognition rule and you understand why every other paper has something to price that the headline numbers do not show. This is why the paper is the keystone: it is the gate before the pricing.

Who decides? Not, in the first instance, the Ministry of Finance. The recognition of a PPP is settled at the intersection of three communities that rarely sit at the same table: the national statisticians, who apply the European System of Accounts (ESA 2010) and the Eurostat Manual on Government Deficit and Debt to produce the government debt and deficit figures the fiscal rule binds; the government accountants, who apply IPSAS (here IPSAS 32) or its IFRS analogue (IFRIC 12) to produce the accrual accounts; and the fiscal-risk unit of the ministry itself, whose Fiscal Commitments and Contingent Liabilities (FCCL) mandate is to know what the State is actually exposed to, recognised or not. These three answer different questions, on different bases, and — the central fact of this paper — they routinely give different answers about the same contract.

The United Kingdom makes the divergence concrete. When central government moved to International Financial Reporting Standards from 2009-10, IFRIC 12 brought nearly all Private Finance Initiative (PFI) assets onto the Whole of Government Accounts balance sheet. Yet the National Accounts measure that the fiscal rules actually target — Public Sector Net Debt — kept the majority of PFI off the books, and the implementation of ESA 2010 in 2014, which reclassified some £130 billion of other items onto the measure, did not pull PFI on. The same roughly seven hundred contracts, with a capital value near £60 billion and future unitary-charge commitments near £199 billion running to the 2040s, were simultaneously on the accountant's balance sheet and off the statistician's debt number (NAO 2018; ONS 2018; OBR 2017, ¶7.67). A ministry that read only the fiscal-rule number saw a fraction of what it owed.

The decision problem of this paper is therefore not "what does the PPP cost" — that is the rest of the series — but "is the cost recognised, where, and by whose rule," and what a finance ministry should do when the rule that governs its fiscal-space arithmetic systematically records less than the State actually bears.

2. The recognition question and the four lenses

Recognition is not one question with one answer; it is the same contract read through up to four lenses, each with its own purpose, its own test, and its own number that it moves. A ministry that wants to know its true position has to read all four — and, crucially, to read the gaps between them.

Lens 1 — ESA 2010 / Eurostat (the statistical, fiscal-rule lens). The European System of Accounts 2010, operationalised for PPPs by the Eurostat Manual on Government Deficit and Debt (MGDD) and the 2016 Eurostat–EPEC Guide to the Statistical Treatment of PPPs, produces the general-government debt and deficit figures that fiscal rules bind. Its test is risks and rewards: who, on the balance of the contract, bears the economic risks of the asset. This is the lens that decides whether a PPP consumes fiscal space, and therefore the lens a government most wants to manage. It is also the lens the Austral platform did not previously implement, and the one this paper's engine adds.

Lens 2 — IPSAS 32 (the accounting lens). IPSAS 32, Service Concession Arrangements: Grantor — the public-sector mirror of IFRIC 12 — governs the grantor's accrual accounts. Its test is control: the grantor recognises the service-concession asset and a corresponding liability when it controls or regulates what services the operator must provide with the asset, to whom, and at what price, and controls any significant residual interest at the end of the arrangement. Because the entire rationale of a PPP is that the government specifies and controls a public service, the control test is met by most PPPs — so IPSAS 32 tends to put more on the government's books than the ESA risks-and-rewards test does.

Lens 3 — GFSM 2014 (the IMF statistical lens). The IMF's Government Finance Statistics Manual 2014 records on the basis of economic ownership: the unit that bears the risks and reaps the rewards of an asset is its economic owner, regardless of legal title. In substance it is close to the ESA risks-and-rewards test, and this paper scores it with the same substantive gate and the same secondary provisions. What it does not carry is the EU's guarantee layer: as applied outside the EU, a sub-majority minimum-revenue guarantee does not change who owns the asset, and is recorded — where it is recorded at all — as a separate contingent liability. That single omission is what opens the recognition gap of §§5–6. This paper scores the GFSM lens with the ESA substantive tree minus the guarantee layer — a proxy for GFSM as applied outside the EU, not a restatement of the Manual's own criterion, which is a preponderance-of-risks-and-rewards judgement in which financing, guarantees and end-of-contract residual arrangements weigh alongside the three risk categories.

Lens 4 — the economic-tail / FCCL lens (the substance test). Independent of any statistical or accounting standard, the Fiscal Commitments and Contingent Liabilities view — the basis of the IMF–World Bank PFRAM and of this series — asks a blunter question: does the sovereign bear the material downside? A minimum-revenue guarantee, an exchange-rate guarantee, a debt guarantee, or a termination clause that repays the partner's lenders puts the tail on the State whatever the other three lenses say. This lens recognises nothing on a balance sheet; it measures exposure. It is the benchmark against which the recognition gap is defined.

The point of reading all four is not pedantry. The fiscal rule binds Lens 1; the published accounts show Lens 2; the IMF Article IV looks through Lens 3; and the actual risk to the taxpayer lives in Lens 4. Where they agree, recognition is clean. Where they disagree — and §§4–5 show the disagreement is structural and exploitable — the difference is precisely the liability that is silent.

3. The ESA-2010 risks-and-rewards test

Because Lens 1 is the one the fiscal rule binds, it is the one worth stating exactly. The ESA 2010 / MGDD test is a decision tree with a primary gate and a set of secondary provisions that can override it. The Austral engine encodes it verbatim; the full tree is in Appendix B, and the substance is here.

Lens	Standard	Test	What it governs / the number it moves
Statistical (EU)	ESA 2010 + MGDD	risks and rewards	General-government debt & deficit — the fiscal-rule number
Accounting	IPSAS 32 / IFRIC 12	control	The grantor's accrual balance sheet (e.g. Whole of Government Accounts)
Statistical (IMF)	GFSM 2014 (non-EU proxy)	economic ownership	GFS public-sector debt and the Public Sector Balance Sheet
Economic / FCCL	PFRAM / this series	does the sovereign bear the tail?	The State's true contingent exposure (recognised or not)

The primary gate. The asset is recorded off the government balance sheet — that is, on the private partner's books — if and only if the partner bears the construction risk and at least one of the two operating risks, availability risk or demand risk. The three risks are standard. Construction risk covers late delivery, cost overrun and sub-standard build. Availability risk is the risk that the asset, once built, fails to meet the contracted standard, exposing the partner to payment deductions. Demand risk is the risk that usage — traffic, patients, throughput — falls short. If the government bears construction risk, or bears both operating risks, the asset is recorded on the government balance sheet.

The weak-deduction caveat. A transfer of availability risk is only real if the payment mechanism actually puts the partner's revenue at stake. Where the deduction regime is weak or absent — the partner is paid substantially regardless of performance — the MGDD treats the availability-risk transfer as ineffective, and the primary gate is not passed on availability alone. The engine encodes this: a nominal availability transfer with a weak deduction regime does not count.

The secondary provisions. Even when the primary gate is passed, a set of contract features re-consolidates the asset onto the government balance sheet, because each one means the State has taken back the substance of the risk it appeared to transfer. Any one of the following flips the classification from off to on:

The joint test — the provisions are read together, not one at a time. The first two provisions are not two independent gates to be cleared separately, and a contract engineered to pass each of them on its own does not thereby pass. The MGDD is explicit: "both tests for majority financing and guarantees in relation to the capital expenditure of the PPP project must be undertaken jointly. It might well be the case, in PPP contracts, that government would provide a minority of the total capital expenditure, but would then guarantee a major part of the remaining project finance … In this case, if the combined effect of government support would represent more than a majority of capital expenditure, this would lead to the conclusion that a majority of risks would be incurred with government" (MGDD §6.4.3.4 ¶69; identical in the 2019 and 2022 editions, and restated for concessions inside the construction-risk analysis at 2022 §6.3.1.5.3 ¶36). The arithmetic follows the text: the guarantee is scored on the remaining project finance, so a State that funds 40% of the capital cost and guarantees 40% of the rest is supporting 0.40 + 0.40 × 0.60 = 64% of capex — and the asset returns to its balance sheet even though neither limb reached a majority alone. The threshold is the source's own: more than a majority, a strict inequality, so exactly 50% combined does not consolidate. Because the test resolves who bears the majority of risks, it belongs to the substantive tree that ESA and GFSM share — it is not part of the EU-only guarantee layer described next.

The guarantee rule — the EU goes further. The 2016 Eurostat–EPEC Guide to the Statistical Treatment of PPPs tightened the test in a way that matters for this paper: it provides that any form of minimum-revenue or minimum-demand guarantee — whether a stand-alone undertaking or built into the payment mechanism — by itself puts the PPP on the government balance sheet, and that a government commitment financing 50% or more of the capital expenditure does the same (EPEC–Eurostat 2016, Themes 4.10.3 and 14.4–14.5). This is decisive. A minimum-revenue guarantee is the classic way a State backstops a demand-risk concession — it is the mechanism the rest of this series prices — and under the strict EU rule it is caught: the asset is consolidated, the liability recognised. The European statistical rule, properly applied, does not miss the guarantee.

The tree is therefore not weak. It takes more than a clever headline risk allocation to keep an asset off the books: the secondary provisions catch the obvious claw-backs, and the EU guarantee rule catches the minimum-revenue backstop on top. What lets the liability stay silent is not a loophole in a correctly-applied rule — it is that most jurisdictions do not apply this rule at all, which is the subject of §§4–5 and the gap the rest of the paper turns on.

The output of this lens, for any contract, is a class (on or off) and the deciding test — the specific branch that settled it — so that every classification is auditable rather than asserted. That audit trail is what lets a ministry, a statistician and an auditor look at the same contract and agree on why it was classified, even where the lenses disagree on whether.

4. Where the rule books disagree — control versus risks and rewards

The cleanest divergence is between the two lenses a government reads most often: the statistical ESA test (Lens 1) and the accounting IPSAS 32 test (Lens 2). They are built on different questions, so they disagree by construction, not by error.

ESA asks who bears the risk. IPSAS asks who controls the service. For the standard availability PPP these point opposite ways. Consider an availability hospital: the partner designs, builds and maintains it, bears the construction risk and — under a genuine deduction regime — the availability risk, while the government pays a unitary charge for the service whether or not patients come, and so bears the demand risk. Under ESA, the partner bears construction plus one operating risk: the asset is recorded off the government balance sheet. Under IPSAS 32, the government plainly controls what services the hospital provides, to whom and at what price, and controls the residual interest at hand-back: the asset and a matching liability are recorded on the grantor's accounts. Same contract; opposite answers; both correct under their own standard.

This is not a hypothesis. It is exactly the United Kingdom's PFI experience, run at national scale. IFRIC 12 — the IFRS rule whose public-sector mirror is IPSAS 32 — put PFI on the Whole of Government Accounts from 2009-10 on the control logic above, while ESA kept it off Public Sector Net Debt on the risks-and-rewards logic. The divergence was not a transitional glitch to be reconciled away; it persisted for a decade and survived the 2014 ESA implementation. The control lens and the risks-and-rewards lens were simply answering different questions about the same buildings.

Because the control test is met by most PPPs — controlling the service is why a government does a PPP — IPSAS 32 systematically recognises more PPP assets than the risks-and-rewards test. So the accounting books are the most inclusive lens and the statistical measure the least, and the measure a government most wants to look good on is precisely the one that recognises the least. The availability-hospital gap is, in this sense, jurisdiction-independent: wherever the accounting standard is control-based and the binding statistical standard risk-based, the asset sits on the one and off the other, and the UK ran exactly that divergence at national scale.

But there is a second axis of disagreement, and it is the one this paper turns on: the statistical standard itself comes in two strengths. The European rule — ESA plus the EPEC–Eurostat guarantee provisions of §3 — is strict: it catches the minimum-revenue guarantee and consolidates the asset. The IMF's GFSM 2014 follows economic ownership but, as most non-EU jurisdictions apply it, does not treat a sub-majority revenue guarantee as a change in the asset's ownership — the asset stays with the partner and the guarantee is booked, if at all, as a separate contingent liability. So the same demand-risk concession with a minimum-revenue guarantee is consolidated under the EU rule and left off under the looser one: a divergence not between accounting and statistics, but between a strict statistical regime and a loose one. Which statistical answer binds a given ministry is a fact about its jurisdiction, not about the contract.

The recognition gap therefore has two drivers: the control-versus-risk divergence that is present everywhere, and the strict-versus-loose statistical divergence that depends on where the project lives. The second is where the silent liability of the rest of this series actually resides — and §5 follows it home.

5. Where the gap lives — the jurisdiction, not the loophole

The natural suspicion, given §4, is that off-balance-sheet PPPs are an engineering trick: structure the deal to fail the statistical test and pocket the fiscal space. For the binding EU rule that suspicion is largely wrong, and it is important to say so plainly. The European rule is comprehensive — the risks-and-rewards primary gate, the secondary claw-back provisions, and, decisively, the EPEC–Eurostat rule that any minimum-revenue or minimum-demand guarantee consolidates the asset. The classic way a State backstops a demand-risk concession, the minimum-revenue guarantee, is exactly what the EU rule catches. Portugal proves the point: in its 2010 statistical revision a number of road and transport PPPs were reclassified into general government, with a deficit-increasing impact of 0.5% of GDP for that year (§9). Under a correctly-applied EU rule, the guarantee does not let the liability hide.

So the recognition gap is not, in the main, a loophole in a well-applied rule. It is that most of the world's PPPs are not scored by that rule. ESA is European. Outside the EU, the binding debt measure follows GFSM 2014 or a local statistical convention, and the accrual accounts follow IPSAS or national practice — none of which carries the EPEC–Eurostat provision that a minimum-revenue guarantee consolidates the asset. In those jurisdictions the demand-risk concession passes the economic-ownership test (the partner bears construction and demand), the guarantee is treated as a contingent liability rather than a change of ownership, and the asset sits off the binding debt measure with the guarantee disclosed — if the jurisdiction discloses contingent liabilities at all — as a separate, off-debt line. That is precisely how Chile treats it: its concession minimum-income guarantees are reported in the contingent-liabilities report, not in the debt stock (§9). The liability is not hidden by cleverness; it is left off by a rule that never reached for it.

This is the configuration the recognition scorer is built to expose, and its answer depends on the jurisdiction's binding measure. Run a demand-risk toll road with a minimum-revenue guarantee through all four lenses: under the EU rule it is consolidated — on the balance sheet, no gap; under a GFSM/IPSAS regime it is off both the statistical and the accounting books while the economic-tail lens registers the sovereign's exposure — a recognition gap that exists in Santiago or Lagos but not in Lisbon, for the very same contract. The Silent FX Liability (Paper 6) is the same configuration with the exchange rate, the Indexation Ceiling (Paper 7) with an indexation clause: a sovereign-written guarantee that the binding measure, as applied where the project lives, does not consolidate. The recognition gap is the size of what that un-reaching rule leaves off — and the remedy, developed in §11, is less to police a loophole than to adopt the rule that already closes it.

Two honest qualifications carry into the worked cases. First, the scorer alleges no gaming: it measures the gap between a contract's binding-measure treatment and its economic substance, and lets a ministry see where the two part company — a gap that is mostly a fact about the jurisdiction's rule, not the drafter's cunning. Second, where the rules already bite they bite correctly: the government-funded asset of §8 is pulled on the books by the majority-financing provision under every regime, and carries no gap. The claim is narrow and sharp: a common, legal structure — a demand-risk concession backstopped by a minimum-revenue guarantee — produces a sovereign liability that the binding rule, as applied where the project lives, may not record, and that a ministry can and should measure.

6. The recognition gap

The instrument that measures it is the recognition gap: the value that sits off the jurisdiction's binding statistical balance sheet yet is economically sovereign.

Define it per contract, against the binding statistical measure \(B\) — ESA where the jurisdiction is in the EU, the GFSM-basic or local convention where it is not. With the accounting class \(A\) under IPSAS 32 and the economic class \(E\) under the FCCL tail test, the contract diverges when the binding measure records it off but a substance lens records it on:

\[\text{diverge} \;=\; (B = \text{off}) \;\wedge\; \bigl(A = \text{on} \;\vee\; E = \text{on}\bigr).\]

The gap is the magnitude that divergence leaves unrecorded — the asset's capital value where the accounting lens would recognise it, or the contingent tail where the economic lens catches it, whichever is larger:

\[G \;=\; \begin{cases} \max\bigl(\,K\cdot\mathbf{1}[A=\text{on}],\; T\cdot\mathbf{1}[E=\text{on}]\,\bigr) & \text{if diverge} \\ 0 & \text{otherwise,} \end{cases}\]

where \(K\) is the asset's capital value and \(T\) the sovereign's maximum contingent exposure, both as a share of GDP. Because the binding measure \(B\) differs by jurisdiction, so does the gap: the same contract can carry a gap under a loose regime and none under a strict one. The portfolio recognition gap is the sum \(G_{\text{portfolio}} = \sum_i G_i\), reported for each regime.

On the three worked contracts of §8, the gap is legible — and splits cleanly by jurisdiction:

Contract	ESA (EU)	IPSAS 32	GFSM (non-EU)	economic tail	gap, EU	gap, non-EU
A. Availability hospital (UK-PFI style)	off	on	off	off	1.2	1.2
B. Demand toll road + minimum-revenue guarantee	on	off	off	on	0.0	0.6
C. Government-funded social asset	on	on	on	off	0.0	0.0

Contract A is the UK-PFI divergence of §4: off the statistical measure, on the accounting books, a gap equal to its capital value — and because it is driven by the control-versus-risk split, not by a guarantee, it carries in every jurisdiction. Contract B is the jurisdictional case of §5: the EU rule consolidates the minimum-revenue guarantee, so there is no gap in the EU; a GFSM/IPSAS jurisdiction does not, so the asset sits off both its statistical and its accounting books while the sovereign bears the tail — a gap equal to the guarantee exposure, present outside the EU only. Contract C is the control case: the majority-financing provision records it on the books under every regime, so there is no gap. Across the portfolio the recognition gap is 1.2% of GDP under the strict EU rule and 1.8% under the looser non-EU regime — the same three contracts, a larger hidden liability where the rule is weaker.

That magnitude is not a modelling artefact of stylised contracts; it is the order of magnitude the real cases show. The Office for Budget Responsibility put the United Kingdom's off-balance-sheet PFI stock at roughly £35 billion, about 2.5% of GDP, at March 2010 — a national recognition gap of exactly this kind, on exactly this scale (OBR 2011, Box 3.3). The recognition gap is, in other words, a measurable quantity that a finance ministry can compute contract by contract and report in aggregate.

The policy object follows directly, and it is developed in §11: the recognition gap should be a disclosed line, reported alongside the rule-relevant treatment, so that the fiscal-rule debt number is read next to the substance it omits. The remaining sections give the engine that computes it (§7), run the three contracts through it in full (§8), anchor the magnitudes in the UK, Eurostat and Chilean records (§9), situate the contribution in the literature (§10), and set out the limitations and the policy lead (§11).

7. Architecture and reproduce in ten minutes

The scorer is one small module with a deliberate design principle: encode the statistical leg and the divergence engine; reuse everything else. The recognition question spans three communities, and the platform already speaks two of their languages. The IPSAS 32 control test ships in the platform's accounting module; the GFSM 2014 Public Sector Balance Sheet and the PFRAM contingent-liability importer ship in the government-finance-statistics module; the minimum-revenue, exchange-rate and debt guarantees that drive the economic-tail lens ship in the guarantees module. What did not exist — and what this paper adds — is the ESA 2010 risks-and-rewards leg and the cross-framework divergence engine that runs a contract through all four lenses and computes the recognition gap.

The core is a pure function of a single input: a contract's risk-allocation matrix. The matrix names, for each primary risk (construction, availability, demand), who bears it; the secondary ESA provisions (government financing share, debt-guarantee share, debt-repaying termination, below-market residual, the strength of the availability-deduction regime); the IPSAS control inputs (control of services/price/recipients; control of the residual interest); and the economic-tail inputs (expected and maximum contingent exposure), together with the asset's capital value. Each of the four classifiers returns a class and the deciding test; the divergence engine combines them into the recognition gap of §6. Because the core is pure and dependency-free, the worked examples re-run anywhere, and every figure in §§6 and 8 is reproducible from a seed-free script.

On the platform the engine is exposed as POST /workspaces/{wid}/recognition/classify, EDITOR-gated and audit-chained: the request carries the risk-allocation matrix, the response carries the four classifications, their deciding tests, the divergence flag, the recognition gap, and an audit block recording the input hash and the framework versions. The classification is therefore tamper-evident — a statistician, an auditor and the ministry can re-run the same contract and obtain the same answer with the same stated reasons.

Reproduce in ten minutes. Three commands reproduce the paper's recognition results: python3 recognition_core.py runs the golden self-check over the full ESA decision tree; python3 worked_examples.py scores the three contracts and writes the recognition table of §6 to out/worked_examples.json; python3 -m pytest test_recognition.py runs the fifteen tests that pin every ESA branch, each secondary provision, the EU/non-EU guarantee split, and the jurisdiction-dependent gap. The deciding-test strings in §8 are lifted verbatim from that output.

8. Worked examples — three contracts, three mechanisms

The three contracts are not a sample; they are chosen so that each isolates one way the recognition question can go, and together they span the result space. Magnitudes are illustrative shares of GDP, chosen to make the gap legible, not estimates of any specific programme.

A — the availability hospital (the accounting–statistical divergence). The partner designs, builds and maintains the hospital, bearing construction risk and, under a strong deduction regime, availability risk; the government pays a unitary charge for the service regardless of patient numbers, and so bears demand risk. The ESA leg passes the primary gate on partner bears construction + availability/demand and reaches none of the secondary provisions: off. The IPSAS leg finds the grantor controlling the service, its price and recipients, and the residual interest at hand-back: on. The economic-tail leg finds no material contingent exposure — an availability payment is a direct, certain commitment, not a guarantee — so the tail is off. The contract diverges (statistical off, accounting on); the recognition gap equals the asset's capital value, 1.2% of GDP. This is the United Kingdom's PFI, contract-for-contract: on the accountant's books, off the fiscal rule's.

B — the demand-risk toll road with a minimum-revenue guarantee (the jurisdictional case). The concessionaire builds and operates a toll road, bearing construction risk and genuine demand risk, sets tolls in a merchant-style structure under which the government does not control the service in the IPSAS sense, and is backstopped by a minimum-revenue guarantee. Here the recognition answer depends on the binding measure. Under the EU rule, the EPEC–Eurostat provision that any minimum-revenue guarantee consolidates the asset is decisive: the ESA leg returns on — the guarantee is caught, the asset recognised, no gap. Under a non-EU regime — GFSM economic ownership or a local convention — the guarantee is a contingent liability, not a change of ownership: the statistical leg returns off, the IPSAS leg finds no control and returns off, and only the economic-tail leg registers the State's exposure (the guarantee's maximum draw, 0.6% of GDP) as on. So the same contract carries a recognition gap of 0.6% of GDP outside the EU and zero inside it. This is the silent liability of the rest of this series — and the proof that its silence is a property of the jurisdiction's rule, not of the contract.

C — the government-funded social asset (the secondary provision doing its job). The partner builds and bears availability risk, and the government bears demand, but the government also finances the majority of the capital cost. The ESA leg passes the primary gate — and is then pulled back by the majority-financing secondary provision: on. The GFSM leg agrees on economic ownership: on under the non-EU regime too. The IPSAS leg recognises it on control; with no guarantee, the tail leg is off. The binding statistical measure is on under either regime, so there is no divergence and no gap, in the EU or outside it. C is the control case — the statistical test catching a claw-back it was designed to catch — and it matters because it shows the scorer does not manufacture gaps: where the rules already bite, the gap is zero.

Across the three, the portfolio recognition gap is 1.2% of GDP under the EU rule and 1.8% outside it. The shape is the lesson: contract A's gap is everywhere (an accounting-versus-statistical split no guarantee rule touches), contract B's gap appears only where the statistical rule is loose, and contract C carries none because the rules already bite. A ministry's exposure to the silent liability is, in large part, a function of which rule book scores its debt.

9. Empirical anchor — UK PFI, Eurostat reclassifications, Chile

The worked magnitudes are illustrative, but the mechanism and its scale are not. Three real records anchor the paper.

The United Kingdom — the recognition gap at national scale. The UK ran the accounting–statistical divergence of contract A across an entire programme. Some seven hundred PFI and PF2 deals, with a capital value near £60 billion and future unitary-charge commitments near £199 billion to the 2040s, carried annual charges of £10.3 billion in 2016-17 (National Audit Office, PFI and PF2, HC 718, 2018). The move to IFRS in 2009-10 brought nearly all of them onto the Whole of Government Accounts under IFRIC 12, while the National Accounts kept the majority off Public Sector Net Debt — the Office for National Statistics put roughly £6 billion of PFI inside PSND against some £28 billion outside it at March 2018, and the Office for Budget Responsibility put the off-balance-sheet stock at just under £35 billion, about 2.5% of GDP, at March 2010 (ONS 2018; OBR 2011, Box 3.3). The recognition rule can shape the design — though the sources stop short of attributing motive to particular contracts. The OBR lists off-balance-sheet financing among the fiscal illusions, the IMF's term (Irwin 2016) that the OBR adopts for accounting treatments that move a measured aggregate without changing the true fiscal position, and it flags "where accounting rules drive policy decisions" as an issue for government to address (OBR 2017, ¶¶7.65–7.66, 7.70). On PFI it records that the deals "were widely used in the early 2000s when the 'sustainable investment rule' target of keeping PSND below 40 per cent of GDP was subject to limited headroom" — while declining to go further: "It is not for us to comment on the motivation behind these decisions, but it is possible to see why people might believe that their statistical treatment may have played a part" (¶7.67). The Treasury Select Committee was less reticent, recommending in 2011 that PFI be brought on balance sheet so it could not be used to circumvent departmental limits. PF2, and PFI for future projects, was abolished in the 2018 Budget — a policy that recognised the gap by closing the instrument that produced it. (The frequently quoted "£300 billion" figure conflates the £199 billion of future charges with charges already paid; the clean, comparable figures are the ~£60 billion capital value and the ~£35 billion off-PSND stock.)

The European Union — the rule with teeth. Eurostat's test is exactly the §3 tree, strengthened by the EPEC–Eurostat guarantee provisions, and Eurostat is candid about why it matters: the joint Guide to the Statistical Treatment of PPPs (2016) states that the decision to procure as a PPP "is often influenced by expectations as to their statistical treatment (i.e. impact on government debt and deficit figures)." The rule bites. Eurostat maintains a standard supplementary table of PPP liabilities recorded off-balance-sheet of government by member state, and reclassifications occur: in Portugal's 2010 statistical revision, a number of public-private partnerships were reclassified into the general government sector, with a deficit-increasing impact of 0.5% of GDP for that year (a further 0.5% came from reclassified public transport corporations), as documented in the subsequent EU–IMF adjustment programme (European Commission, Economic Adjustment Programme for Portugal, 2011, paragraph 12 and Table 3). The reclassification did exactly what the recognition gap measures — it moved an economically sovereign exposure from off the statistical balance sheet to on it — and it shows the EU rule does not, in the end, let a backstopped PPP hide. The recognition gap of §§5–6 is what remains outside that rule's reach.

Chile — the economic-tail lens, already operational. Chile is where the FCCL lens is institutionalised, and the author's home ground. The Dirección de Presupuestos (DIPRES) publishes an annual Informe de Pasivos Contingentes — the 2025 edition (December 2025), produced by its Public Finance Division and its Contingent-Liabilities-and-Concessions sub-department — that discloses the State's contingent liabilities as both a stock and an annual budgetary flow, each as a share of GDP. The concession minimum-revenue guarantee (Garantía de Ingresos Mínimos del Sistema de Concesiones, the ingresos mínimos garantizados that are precisely the contract-B mechanism) is reported at a stock of 0.06% of GDP in 2025 — down from 0.09% in 2024 and 0.08% in 2023 — with an annual flow of 0.005% of GDP on a probabilistic estimate; it is committed in inflation-indexed units (the Unidad de Fomento), so the stock rises with inflation, and its activation tracks economic activity through road traffic — the demand channel exactly. Total reported contingent liabilities stand at 8.92% of GDP in 2025 (down from 9.58% in 2024 and 10.47% in 2023), of which the concessions also contribute a controversias (dispute) line of 0.10% (DIPRES 2025). That report is the economic-tail lens of this paper made operational by a real ministry: it records, as a disclosed line, the guarantee substance that the statistical treatment of the concession assets keeps out of the headline debt. Chile is therefore the positive template for the policy lead of §11 — a ministry that already publishes the tail — and pairing the disclosed guarantee exposure with the concessions' off-balance-sheet statistical treatment yields the recognition gap directly, on a programme the author administered.

10. The recognition question in the literature

The recognition question sits at the intersection of four literatures, each of which supplies one piece of the picture and none of which assembles the whole. This paper's contribution is the assembly: a single contract read through every live rule book and the economic-substance lens together, with the divergence made computable.

The rule books themselves. The statistical and accounting standards define the tests this paper runs but do not run them against each other. The risks-and-rewards treatment originates in Eurostat's decision of 11 February 2004 (Eurostat 2004), is codified in the Manual on Government Deficit and Debt (Eurostat 2022) and made operational for practitioners by the joint Eurostat–EPEC Guide to the Statistical Treatment of PPPs (EPEC–Eurostat 2016); the control treatment is set by IPSAS 32 (IPSASB 2011), the public-sector mirror of IFRIC 12; and economic ownership by the IMF's GFSM 2014. Each is authoritative within its own perimeter and silent on the others — the standards do not, and are not meant to, tell a ministry where they disagree.

Why off-balance-sheet is sought. A public-finance literature establishes the motive. Engel, Fischer and Galetovic (2013, 2014) show that a PPP does not relax the government budget constraint — the case for PPPs is bundling and efficiency, not fiscal space — so a PPP that appears to create fiscal room is an accounting effect, not an economic one. Maskin and Tirole (2008) show that public accounting and spending rules shape officials' incentives in exactly the direction of cost-understatement. Irwin (2012) catalogues the "accounting devices and fiscal illusions" — PPPs among them — by which governments defer the recognition of investment, and Koen and van den Noord (2005) document the "fiscal gimmickry," including PPP off-balance-sheet structures, used to meet euro-area targets; the IMF's Fiscal Affairs Department had made the same point at the policy level (IMF 2004; Funke, Irwin and Rial 2013). This literature argues the motive convincingly — but qualitatively. It diagnoses why governments route around the rule; it does not turn the routing into a per-contract, computable quantity.

The accounting boundary. Closest to this paper is the UK accounting-boundary literature, which examined the on/off decision directly. Heald (2003) showed how the value-for-money test and the balance-sheet decision became entangled, creating an incentive to manufacture risk transfer; Heald and Georgiou (2011) argued that UK PFI assets were pushed off the public balance sheet for fiscal-rule reasons rather than economic substance; Hodges and Mellett (1999) showed, on NHS PFI, how the choice of reporting entity could remove economic substance from public accounts. This work is the direct ancestor of the recognition gap — but it is critical and descriptive rather than instrumental, it centres on a single jurisdiction and the IFRS/national-accounts pair, and it predates both IPSAS 32 and any cross-framework treatment.

The contingent tail. A fourth literature measures the substance the statistical rule omits. Polackova's (1998) fiscal-risk matrix and Brixi and Schick's Government at Risk (2002) framed implicit and contingent liabilities as the largest hidden fiscal threat; Irwin's Government Guarantees (2007) priced the guarantee tail; Schwartz, Corbacho and Funke (2008) tied PPP fiscal risk to fiscal space; and the IMF–World Bank PFRAM (2016; 2019) and the IMF's fiscal-risk guidance (IMF 2016) made contingent-liability assessment a standard instrument. This is the economic-tail lens of this paper, and the basis of the series. But it asks how large is the exposure, not is it recognised, and where do the rule books diverge; it measures the tail without locating it on — or off — a balance sheet.

The gap this paper fills. Each strand does one thing: the standards define the tests separately, the public-finance literature argues the motive qualitatively, the accounting-boundary literature critiques one rule book in one jurisdiction, and the contingent-liability literature measures the tail without the recognition boundary. None runs a single contract through all the live statistical and accounting rule books and the economic-substance lens at once, and none turns the divergence into a reproducible, contract-level recognition gap that a ministry can compute and disclose. That operationalisation — the cross-framework scorer, the gap as a quantity, and the gap as a reported line — is what this paper adds.

11. Limitations and the policy lead

The policy lead. The recognition gap is a measurable quantity, so it can be governed — and the reframe points to the highest-leverage move first. Four controls, in order of leverage:

The cheapest fiscal discipline a ministry has is to count what it owes on the same basis it owes it. The recognition rule is what decides whether the count is honest; this paper makes the rule — and the gap it leaves — auditable.

Appendices

Appendix A — Reproducible API call

POST /workspaces/{wid}/recognition/classify
Content-Type: application/json

{
  "name": "Demand-risk toll road + MRG",
  "construction_risk": "PARTNER",
  "availability_risk": "PARTNER",
  "demand_risk": "PARTNER",
  "availability_deduction_regime": "STRONG",
  "min_revenue_guarantee": true,
  "govt_controls_services_price_who": false,
  "govt_controls_residual_interest": false,
  "max_contingent_pct_gdp": 0.6,
  "capex_pct_gdp": 0.8,
  "headline": "non-EU"
}

{
  "headline_regime": "GFSM-basic (non-EU)",
  "esa2010":      { "class": "ON",  "deciding_test": "EPEC-Eurostat 2016: minimum-revenue/demand guarantee → on balance sheet" },
  "ipsas32":      { "class": "OFF", "deciding_test": "no control of services and/or residual interest" },
  "gfsm2014":     { "class": "OFF", "deciding_test": "economic owner: private partner bears construction + operating risk" },
  "economic_tail":{ "class": "ON",  "deciding_test": "sovereign bears a material contingent tail", "max_contingent_pct_gdp": 0.6 },
  "binding_class": "OFF",
  "divergence": true,
  "recognition_gap_pct_gdp": 0.6,
  "audit": { "inputs_hash": "…", "framework_versions": { "esa": "2010/MGDD-2019+EPEC-2016", "ipsas": "32", "gfsm": "2014" } }
}

The strict EU rule (esa2010) would consolidate the guarantee — class ON — so under headline: "EU" the binding class is ON and the gap is zero; under a non-EU binding measure the same contract diverges and the gap is 0.6% of GDP.

Appendix B — The ESA-2010 decision tree in full

classify_esa2010(ra):
    # primary gate
    partner_construction        = ra.construction_risk == PARTNER
    partner_availability_effective = (ra.availability_risk == PARTNER
                                      and ra.availability_deduction_regime == STRONG)
    partner_demand              = ra.demand_risk == PARTNER
    partner_operating           = partner_availability_effective or partner_demand

    if not (partner_construction and partner_operating):
        return ON   # government bears construction and/or both operating risks

    # secondary provisions — any one re-consolidates ON
    if ra.govt_capex_funding_share  >= 0.5:  return ON   # government finances 50% or more of capex
    if ra.govt_debt_guarantee_share  > 0.5:  return ON   # majority of debt guaranteed

    # the JOINT test (MGDD §6.4.3.4 ¶69): the two above are read TOGETHER, with the
    # guarantee scored on the project finance government did not itself provide
    combined = (ra.govt_capex_funding_share
                + ra.govt_debt_guarantee_share * (1 - ra.govt_capex_funding_share))
    if combined > 0.5:                       return ON   # more than a majority of capex, jointly

    if ra.termination_reimburses_debt:       return ON   # termination repays partner debt
    if ra.residual_value_to_govt_below_mkt:  return ON   # below-market residual to govt

    # EPEC-Eurostat 2016 guarantee rule (the strict EU layer)
    if ra.min_revenue_guarantee:             return ON   # any min-revenue/demand guarantee -> ON

    return OFF      # partner bears construction + availability/demand, no claw-back

The weak-deduction caveat is encoded in partner_availability_effective: a nominal availability transfer with a WEAK or NONE deduction regime does not count as the partner bearing availability risk. The GFSM-basic (non-EU) classifier shares the primary gate and all five secondary provisions — the joint test included — but omits the EPEC–Eurostat guarantee rule, which is exactly why a minimum-revenue guarantee consolidates the asset in the EU yet leaves it off the binding measure elsewhere. The joint test sits in the shared tree because its conclusion is that "a majority of risks would be incurred with government," which is the economic-ownership question GFSM asks; the EU-only layer is the any-guarantee rule, whose logic — any amount consolidates, regardless of size — is of a different kind from a majority-of-capex computation.

Appendix C — Cross-framework mapping

Appendix D — Glossary

Appendix E — References

	ESA 2010 + EPEC (EU)	IPSAS 32	GFSM 2014 (non-EU proxy)	Economic / FCCL
Basis	statistical (national accounts)	accrual accounting	statistical (GFS)	economic substance
Test	risks and rewards + the guarantee rule	control	economic ownership	does the sovereign bear the tail?
Recognises ON when	govt bears construction, or both operating risks, or a secondary provision triggers — including financing and guarantees jointly exceeding a majority of capex (MGDD ¶69) — or any minimum-revenue/demand guarantee	govt controls service + residual	govt is the economic owner in substance: the same primary gate and the same five secondary provisions, joint test included, but no any-guarantee rule	material contingent tail on the State
Catches a minimum-revenue guarantee?	yes (EPEC–Eurostat 2016)	only via control	no (treated as a separate contingent liability)	yes (it is the tail)
Number it moves	general-govt debt/deficit (the fiscal rule)	grantor accrual balance sheet	GFS public-sector debt / PSBS	true contingent exposure

Numbers in §§6 and 8 reproduce from recognition_core.py / worked_examples.py (golden self-check + 13 tests); empirical anchors are primary-sourced in EMPIRICAL_ANCHORS.md; citations verified 2026-06-22.

The Recognition Rule — When a PPP Lands on the Government Balance Sheet (and How the Line Is Gamed)