From Gatekeeper to Orchestrator — The Ministry of Finance as Leader of the PPP Programme

A re-partition of analytical responsibility in PPP governance, with empirical evidence from 50 countries and 12 canonical doctrine documents, implemented on the Austral Platform

Author: David Duarte Arancibia, Austral Intelligence Date: May 2026  ·  Revised: 28 September 2026 (see Errata, Appendix E) Audience: Ministers of Finance and their fiscal-risk teams, MDB and DFI advisory units, PPP units, sector regulators, supreme audit institutions, rating-agency sovereign teams Companion software: Austral platform — https://platform.austral-intelligence.com

Abstract

The prevailing doctrine of public-private partnership (PPP) governance assigns the Ministry of Finance (MoF) the role of gatekeeper: a defensive fiscal-affordability veto exercised after a contracting authority and its transaction advisors have done the analytical work. The doctrine runs through the World Bank's PPP Reference Guide v3 — which uses the literal word "gatekeeper" of the Ministry of Finance's own role² — the APMG PPP Certification Program Guide, the OECD Principles for Public Governance of Public-Private Partnerships, and the European PPP Expertise Centre's Establishing and Reforming PPP Units. Across the twelve canonical doctrine documents we score in this paper, none prescribes that the MoF should run the analytical chain end-to-end. Four prescribe a co-leadership role at the methodology level (PFRAM, OECD, Irwin 2007, IADB); five prescribe a fiscal-affordability gate (WB v3, APMG, EPEC's 2014 Establishing and Reforming PPP Units, EPEC's 2021 Guide to Guidance, EBRD); three are silent on the MoF role (GIH, UNECE, UNCITRAL).

This paper makes the case for an orchestrator role. The MoF does not hold — and should not pretend to hold — the contracting authority's technical knowledge of engineering, demand, environmental risk, or sector operations. But it does hold (or can cheaply acquire) what the contracting authority cannot easily obtain: the country's cost of capital, the price of contingent liabilities, the menu of available multilateral guarantees, the FX path, the standing fiscal envelope. The orchestrator partition assigns each side what it knows best. The contracting authority owns the project's technical half; the MoF owns the project's financial half and the programme-level methodology that binds the technical half to the country's fiscal capacity. We give that partition a formal statement, derive the five analytical decisions it implies (bankability, risk allocation, value-for-money, third-party guarantees, FCCL & fiscal envelope), and show that the partition becomes implementable once the information cost on the MoF side collapses.

We test the orchestrator hypothesis against three independent bodies of evidence. First, the twelve canonical doctrine documents above. Second, institutional briefs for fifty countries — sixteen Latin American, eight African, nine Asian, four Middle-Eastern, ten European, three Anglosphere — scored on the same four-level rubric (NONE, GATE, CO-LEAD, LEAD). Seven countries reach the LEAD cell on observable evidence: Uruguay, Portugal, South Africa, Egypt, Korea, India, New Zealand. Twenty-nine score CO-LEAD, fourteen score GATE, none score NONE. Third, two World Bank empirical datasets — Procuring Infrastructure PPP 2018 and its 2023 successor Benchmarking Infrastructure Development. LEAD countries score 15 to 26 points higher than GATE countries on the sub-score that, in each edition, carries the fiscal-orchestration weight, significant at p < 0.05 in both years. On the World Bank PPI Database (1990 to 2024), LEAD countries' PPP cancellation rate is 1.0% versus 5.2% for CO-LEAD and 13.5% for GATE; LEAD versus CO-LEAD is marginally significant at p = 0.065 with effect size r = 0.41. The signal is detected through measurement instruments that the canonical doctrine itself designed to look elsewhere, which makes its persistence informative.

The paper concludes that the orchestrator partition is implementable on existing analytical tooling. The Austral platform — already used to implement the two prior frameworks in this series, the PPP-feasibility sizing framework³⁶ and the third-party guarantee decision framework³⁷ — closes the information asymmetry that justified the gatekeeper frame in the first place. Bankability, risk allocation, value-for-money, guarantee selection, and FCCL-ceiling enforcement run as five MoF-owned analytical steps, each tamper-evidently audit-chained, with the contracting authority operating inside the resulting envelope. The paper closes with a chapter on lessons learned across the seven LEAD cases and the three documented regression episodes (Canada 2018 — the federal PPP function, with provincial agencies still CO-LEAD; Pakistan 2021 → 2024 — statutory removal followed by partial recovery via the Finance Division's Risk Management Unit; and the World Bank doctrine itself between 2007 and 2017). A qualitative case for the orchestrator partition — built on the information-cost first principles, the portfolio-view argument, the audit-defensibility argument, the stress-moment argument, and the MDB-relationship leverage argument — is set out in §13.6 for the reader who discounts the empirical signal heavily; that case stands without the empirical evidence.

Keywords: Ministry of Finance, public-private partnership, fiscal commitments and contingent liabilities (FCCL), value for money, bankability, third-party guarantees, institutional design, audit-evidence.

Feedback & discussion. This paper is published openly so that practitioners — Ministers of Finance and their teams, sector regulators, PPP units, MDB and DFI advisors, rating-agency sovereign teams, academics — can challenge it, refine it, and put it to work on real programmes. We welcome corrections, counter-examples, and proposed extensions on any section.

• Comment in-page at https://austral-intelligence.com/research (open comments on every section)
• Email the author: [email protected]
• Live demo or walkthrough on request

Table of contents

1. The decision problem — the gatekeeper paradigm
2. Formal statement of the role-partition problem
3. Stage A — Bankability
4. Stage B — Risk allocation
5. Stage C — Value for Money
6. Stage D — Third-party guarantees
7. Stage E — FCCL and fiscal envelope
8. Evidence — literature scoring, PPI, Benchmarking Infrastructure Development
9. Comparator cases — seven orchestrators, four gatekeeper failures, three design-space cases
10. Architecture — how the Austral platform closes the asymmetry
11. Walkthrough — reproduce in 10 minutes
12. Worked example — a synthetic frontier-market pipeline
13. Lessons learned — what makes orchestration stick
14. Limitations and roadmap

Appendices A, B, C, D, E follow the page break.

1. The decision problem — the gatekeeper paradigm

Why this matters: the role assigned to the Ministry of Finance in PPP governance shapes who runs the analysis, who owns the methodology, who carries the verdict to the contracting authority and the sponsor, and ultimately whose error the country bears when a project goes wrong. The current assignment is wrong, and the cost of the wrong assignment is large enough to be measured.

1.1 The doctrine in its own words

The prevailing doctrine is consistent across the canonical sources. The World Bank's PPP Reference Guide v3 — the most widely circulated reference manual in the field, used as a teaching reference by the APMG PPP Certification programme⁴ — describes the MoF's role as one of fiscal affordability assessment, contingent-liability disclosure, and budget integration. The contracting authority and its transaction advisors hold the analytical pen on bankability, value-for-money, and risk allocation; the MoF receives summaries, comments on fiscal implications, and confirms or refuses on affordability grounds. This is the gatekeeper frame in textbook form³.

The Reference Guide also gives the doctrine its sharpest articulation, and gives it in the literal word. Setting out where a country should site its PPP unit, it observes that "many countries do establish their central PPP Unit in the Ministry of Finance, to better fulfil its role of gatekeeper"². The field's most widely circulated reference manual thus names the ministry's role in a single word, and that is the diagnostic label we adopt.

EPEC, the European PPP Expertise Centre at the European Investment Bank, uses the same word but attaches it to the unit rather than to the ministry. Its 2014 paper Establishing and Reforming PPP Units records that a PPP unit "may not have the authority to issue an approval but is required to review proposals and provide recommendations to a separate committee or act as a 'gatekeeper' prior to the submission of proposals to an approval committee", and that in some jurisdictions "PPP Units are the gatekeepers to funding for PPP projects"¹. The distinction sharpens this paper's argument rather than softening it: what EPEC documents — a unit that carries the technical review without holding the authority to decide — is exactly the split between analytical function and decision right that the partition rule of §2 is built on.

The APMG PPP Certification Program Guide — chapter 5, on the public sector — frames the MoF as the fiscal affordability custodian, with the PPP unit (in the certification's worked examples, typically sited outside the MoF) running the analytical chain. The 2017 / 2018 EBRD PPP Laws Assessment benchmarks national legal frameworks on whether the MoF has a binding fiscal-affordability opinion at the right gateway, not on whether the MoF leads the analysis. The OECD Principles for Public Governance of Public-Private Partnerships⁷ are slightly stronger: Principle 3 names the MoF and Principle 5 makes affordability a Treasury concern, but the principles still treat the MoF as a participant in a CA-led process, not as the process's analytical owner. The Global Infrastructure Hub's PPP Risk Allocation Tool and the UNECE Standard on a Zero Tolerance Approach to Corruption in PPP Procurement and the UNCITRAL Model Legislative Provisions on Public-Private Partnerships are silent on the MoF role altogether — they treat the public-sector counterparty as a unitary actor.

The IMF / World Bank PFRAM 2.0 user guide (PPP Fiscal Risk Assessment Model) and the Irwin work that anchors the FCCL methodology both go further than the WB Reference Guide. PFRAM is, in its own framing, a tool for the Ministry of Finance to compute fiscal commitments and contingent liabilities under candidate PPP contracts⁵. Irwin's Government Guarantees: Allocating and Valuing Risk in Privately Financed Infrastructure Projects⁶ makes the analytical case for MoF ownership of guarantee valuation methodology — explicitly treating the MoF as the appropriate analytical home of the contingent-liability question. Both are CO-LEAD documents in our scoring. The IADB's LAC fiscal-PPP guidance similarly treats the MoF as the analytical home of fiscal-risk modelling.

1.2 The doctrine has regressed

The most striking finding in the literature pass is internal to the World Bank itself. Irwin's 2007 Government Guarantees prescribes MoF ownership of the analytical chain on contingent liabilities — a CO-LEAD position. The World Bank PPP Reference Guide v3 (2017) prescribes MoF ownership of the fiscal affordability gate — a GATE position. The doctrine has walked back, between 2007 and 2017, an analytical responsibility that had earlier been assigned to the MoF. The 2007 prescription was not implemented; the 2017 doctrine codifies the under-implementation as the design.

This regression is one of three documented in this paper, with different durabilities. Canada dissolved PPP Canada in 2018, contracting the federal PPP function from CO-LEAD to GATE — and the federal position has persisted to the time of writing. Pakistan moved its PPP Authority out of the Finance Division in 2021 by statute, dropping the country to GATE; the Finance Division walked the change back in 2024 by establishing a Risk Management Unit with a binding FCCL review function, restoring the country to CO-LEAD even though the P3A itself remains housed outside the Ministry of Finance. The doctrine and the practice can both retreat — and, in Pakistan's case, the practice can also recover. §13.2 takes up the durability distinction.

1.3 Three modes of practitioner response, all unsatisfactory

Practitioners working under the gatekeeper doctrine respond to its limitations in one of three modes. Heuristic deference: the MoF defers to the contracting authority's advisor on bankability, accepts the headline fiscal-cost estimate, and applies a fiscal-affordability ratio test against the country's budget envelope. Adversarial review: the MoF retains its own advisor to re-do the analysis adversarially, raising procurement costs and slowing decisions without producing a methodologically aligned output. Withholding the veto: the MoF, lacking the tooling to defend a refusal in detail, accepts projects it suspects are over-leveraged because refusing on fiscal grounds requires a model the MoF does not have.

All three modes share a structural defect: the MoF cannot cheaply observe what it would need to challenge the contracting authority's analysis. The doctrine encodes that information asymmetry as the design choice. The argument of this paper is that the asymmetry has collapsed — analytical tooling that runs the same canonical financial model across every project, with audit-evidence, costs the MoF not noticeably more than reading a one-page summary — and that with the asymmetry collapsed the design choice should be revisited.

1.4 What this paper does not do

This paper does not argue that the MoF should run sectoral engineering, demand modelling, environmental assessment, or operations-and-maintenance design. Those tasks belong with the contracting authority, where the technical information is. The paper's argument is partition-specific: the financial half of the analytical chain belongs with the MoF when the MoF has the tooling to run it, because the MoF holds the financial information that the contracting authority does not. The two halves communicate through a programme-level methodology that the MoF sets and publishes.

The paper also does not argue that doctrine reform is the leading lever. All seven LEAD countries we identify have reached LEAD through national institutional choices — statute, decree, statutory delegation, sustained ministerial conviction, accumulated executive convention — that outran the international doctrine; the doctrine prescribes the LEAD cell nowhere. The paper's case is that doctrine should catch up to those countries, not that it should push others. Doctrine is the slow lever; the fast lever is the analytical tooling that makes the orchestrator partition operable inside any individual MoF.

2. Formal statement of the role-partition problem

Why this matters: a precise statement of the partition problem makes the disagreement with the gatekeeper doctrine concrete. Without it the paper would be a position essay; with it the paper can derive testable consequences and check them against evidence.

2.1 Setting

A PPP programme is a portfolio of candidate projects, each with two specifications: a technical specification (engineering design, demand profile, environmental risk, operations-and-maintenance plan, performance regime) and a financial specification (capex schedule, opex stack, contract length, revenue stack, tariff or availability-payment regime, debt and equity structure). Two state actors observe these specifications: a sector contracting authority (CA) and the Ministry of Finance (MoF). Their observation costs are asymmetric and the asymmetry has structure.

2.2 The two-dimensional information asymmetry

The asymmetry is two-dimensional, not one-dimensional. A unitary "the public sector knows best" framing collapses the table to a single column and erases the partition the paper is about. A unitary "the MoF should be cautious" framing erases the same partition from the other side.

2.3 The objective function

A PPP programme exists to deliver a stream of services at the lowest all-in society cost subject to fiscal sustainability and to delivery feasibility:

\[\min_{\text{contract}} \; \mathbb{E}\!\left[\sum_{t=0}^{T} \frac{C_t^{\text{users}} + C_t^{\text{state}} + L_t^{\text{contingent}}}{(1+r)^t}\right]\]

subject to bankability (\(\text{Min ADSCR} \geq \tau_{\text{ADSCR}}\), \(\text{IRR}_e \geq K_e\)), value-for-money (\(\text{PPP NPV} \leq \text{PSC NPV}\)), and fiscal-affordability (\(\sum_t \text{PV}(C_t^{\text{state}} + L_t^{\text{contingent}}) \leq F^{\text{envelope}}\)).

The benevolent central planner that minimises this objective does not care which actor inside the state runs which step. It cares only that each step is run with the lowest-cost information for that step.

2.4 The partition rule

The partition rule follows directly from the asymmetry table and the objective function:

Partition rule. For each analytical step \(s\) in the chain, assign \(s\) to the actor whose marginal information cost on the variables that \(s\) requires is lower. If \(s\) requires both technical and financial inputs, the actor with the financial-input advantage sets the methodology and runs the verdict; the actor with the technical-input advantage produces the project.

Under this rule, the bankability decision belongs with the MoF (it requires cost of capital, FX, debt menu — all on the MoF side), not the CA. Risk allocation belongs with the MoF in design (it requires risk-pricing across the portfolio) and with the CA in instantiation (it requires sector judgement). Value-for-money belongs with the MoF (it requires the public-sector comparator, which is a fiscal-cost computation). The third-party guarantee decision belongs with the MoF (the menu and pricing are MoF information). FCCL aggregation belongs with the MoF by definition.

2.5 The five steps where the MoF leads under the partition rule

The contracting authority retains the technical half of the chain — engineering, demand, ESG, O&M, performance regime, sector regulation — and operates each project inside the MoF-set methodology and envelope.

2.6 The partition is implementable when the MoF's information cost on the financial half is low

The gatekeeper doctrine made sense in a world where the MoF could observe the financial half of a PPP only by hiring an external advisor for each project, repeating the contracting authority's spreadsheet work at high marginal cost. In that world the MoF rationally settled for a one-page summary and a fiscal-affordability test against a pre-set ceiling. The doctrine encoded the high observation cost as a design choice.

Once the MoF can run the same canonical financial model on every project at near-zero marginal cost — once the sizing of the third-party guarantee, the PSC vs PPP comparison, the FCCL aggregation, and the bankability test all run inside one platform with one audit chain — the high-observation-cost premise no longer holds. The partition rule then re-allocates the financial half to the MoF, and the doctrine should follow. The argument of this paper is that it has not.

3. Stage A — Bankability

Why this matters: the bankability decision determines whether the project closes at all and, if it does, on what terms. Today the contracting authority and its arrangers run that analysis; the MoF sees a one-page summary. Under the orchestrator partition the MoF runs a canonical financial model alongside the arranger's model and reconciles to it on the four or five outputs that govern the verdict, with the contracting authority operating inside the MoF-set discount rate, MoF-set DSCR floor, and MoF-set debt menu.

3.1 What changes

Under the gatekeeper partition, the MoF receives — late in the project cycle — a memo stating that the senior debt is X bps over the local benchmark at Y years tenor with a Z minimum DSCR covenant, that the equity hurdle is H%, that the project clears or doesn't, and that the fiscal effect is F USD over T years. The MoF cannot easily challenge any of those numbers without retaining an advisor of its own.

Under the orchestrator partition, the MoF runs a canonical financial model alongside the arranger's model and reconciles to it. "Alongside" matters: the arranger's model carries the sponsor-side detail (specific debt-syndicate terms, drawdown mechanics, equity-bridge facilities, hedging instruments) that the MoF need not replicate, but the MoF's parallel run is built on one schema, one engine, one audit chain³⁶ — the same canonical financial-model engine across every project in the country's pipeline. Reconciliation between the two is at the level of the four or five bankability outputs that govern the verdict (Min ADSCR, equity IRR, equity NPV, project IRR, feasibility flag) and the discount-rate / DSCR-floor / debt-menu parameters that the MoF sets. The MoF sets the public discount rate for the country (a fiscal-policy variable that should not differ project-to-project), publishes it, and applies it to every project. The MoF sets the DSCR floor at which it will accept a senior debt tranche as bankable on a project of a given sector and a given country-risk class. The MoF sets the eligible debt menu — which lender categories and tenor bands the country considers prudent. The contracting authority and its arranger then compose a project that clears those constraints; the MoF's parallel canonical run is what the verdict is defended on.

This is what the LEAD cases actually do. Korea's PIMAC runs the Preliminary Feasibility Study and the Value-for-Money Test on its own canonical methodology and reviews the bidder's model against them; it does not produce the arranger's financial-close model. South Africa's GTAC reviews and certifies against the PPP Manual methodology at four Treasury Approval gateways; the arranger still produces the underlying model. Portugal's UTAP gives mandatory prior opinions; the contracting authority and sponsor still produce the underlying model. The orchestrator pattern is "MoF owns the canonical methodology and runs the verdict on its parallel canonical model"; it is not "MoF takes over the arranger's model end-to-end." The distinction matters operationally: the orchestrator partition asks for analytical parity with the arranger, not replacement of the arranger.

3.2 Why the MoF has the comparative advantage

The cost of capital is a country variable, not a project variable. The country's sovereign spread, the FX volatility, the macroeconomic outlook, the IMF-programme alignment — all of these enter the bankability calculation and all of them sit on the MoF side of the asymmetry table. A contracting authority's advisor that prices senior debt at 600 basis points over the local benchmark for a frontier-market energy project is making, implicitly, a sovereign-risk judgement that the MoF has more information about than the advisor does. The advisor's price is consistent with the lender market the advisor speaks to; the MoF can know, in addition, whether a multilateral guarantee or a sovereign exchange has shifted the relevant curve.

The DSCR floor is a fiscal-prudence variable. A 1.30× DSCR covenant against a 1.40× covenant changes the project's tariff, the equity IRR, and the bankability flip. Whether the country is willing to accept the lower covenant is a question about its tolerance for refinancing risk and its expected FX path — both MoF-side variables.

3.3 The audit-evidence requirement

A MoF that runs the bankability analysis on every project must be able to defend its verdict to the contracting authority, to the supreme audit institution, to the IMF Article IV mission, and (where applicable) to a parliamentary committee. The verdict must be reproducible bit-for-bit from the saved inputs, must carry a tamper-evident audit-chain identifier on every exported page, and must be re-runnable by any auditor against the same inputs. The Austral platform implements this requirement directly: every run_model call writes an immutable entry to the workspace activity log via an append-only, SHA-256 hash-chained audit trail; every PDF, DOCX, and XLSX export references the audit-chain ID at the bottom of every page³⁶.

3.4 The contracting authority is not displaced

The MoF runs the canonical bankability model and issues the verdict; the CA still owns the project. The CA chooses the engineering specification, the procurement strategy, the contract structure, the performance regime, the ESG plan. The CA's advisor still produces the financial structure and the arranger's own close model. The MoF's role is to verify that the structure clears the country-set bankability constraints — using the MoF's canonical methodology, reconciled against the arranger's outputs at the verdict level. If the structure does not clear, the CA and its advisor restructure; if it does, the project proceeds. The relationship is not adversarial; it is methodologically aligned.

4. Stage B — Risk allocation

Why this matters: risk allocation determines who bears the cost when a contracting risk materialises. Under the gatekeeper doctrine each contracting authority's advisor proposes a project-by-project allocation, drawing on the GIH risk-allocation tool⁸ as a reference. Under the orchestrator partition the MoF maintains a standing per-country risk-allocation policy, and the contracting authority instantiates each project against that policy.

4.1 The standing policy

A risk-allocation policy specifies, for each named risk in a thirteen-dimension taxonomy (construction overrun, completion delay, demand volatility, FX devaluation, regulatory change, force majeure, expropriation, breach by counterpart, environmental liability, archaeological discovery, change in law, tariff inflation indexation, contract termination), the country's standing position on the four allocation modes (CA, SPV, shared with named split, insured to a third party). The policy is derived from the country's FCCL appetite, its FX capacity, its history of payment under the relevant contracts, its political economy, and its statutory constraints (e.g. constitutional restrictions on sovereign guarantees).

The standing policy is the MoF's analytical product. The risk allocation on a given project then becomes a question of whether the project fits the policy — a fast and replicable check — rather than a question of whether the project's allocation is wise on its own terms — a slow, expensive, and hard-to-defend judgement.

4.2 Why the MoF has the comparative advantage on the policy

Risk allocation is a portfolio-level question. A country that consistently allocates demand risk to the SPV will see a higher tariff but a smaller FCCL exposure; a country that consistently retains demand risk will see a lower tariff but a larger contingent liability. The trade-off is fiscal-strategic, not sector-tactical. It belongs with the MoF.

A contracting authority's advisor working in isolation tends toward the allocation that closes its project, not the allocation that fits the country's portfolio. The same advisor working under a published per-country policy can do its sector work efficiently — the allocation question is now answered.

4.3 The thirteen-dimension search

When the standing policy needs to be set or revised — typically once per government cycle, or once per major shift in country fundamentals — the MoF runs the thirteen-dimension allocation search across the candidate pipeline. The search produces a Pareto front of (FCCL exposure, expected tariff) outcomes, and the MoF picks the standing position from that front³⁶. The contracting authorities then operate inside it.

4.4 A worked sketch — energy IPP allocation under a standing policy

Consider a 1,000 MW hydroelectric IPP with a 25-year power purchase agreement (PPA). Under a defensible standing policy for a frontier-market country: (a) construction-overrun risk on the SPV with a contingency-cap at 12% of capex, above which the SPV may invoke a force-majeure carve-out and the country bears the residual; (b) hydrology risk on the SPV in the first ten operating years and shared 50/50 with the offtaker thereafter, recognising that multi-decadal climate uncertainty makes a longer SPV exposure economically inefficient to price; (c) FX risk on the offtaker via a dollar-indexed PPA tariff capped at the country's USD-bond-curve indexation, with a partial pass-through above the cap; (d) termination-payment risk shared, with the SPV bearing the first 25% of debt-service interruption and the offtaker bearing the remainder up to a published cap. The CA's advisor structures the project to fit the policy; the MoF's verdict on each risk line is reproducible from the saved policy document and the project's cash-flow model.

A second project — a 90 km motorway corridor under an availability-payment regime — is checked against a different cell of the same policy: construction risk on the SPV with the same cap, demand risk on the offtaker via the availability payment by construction, FX risk on the offtaker if the availability payment is dollar-indexed (or shared with a partial pass-through if it is in local currency), and termination risk symmetrically on the offtaker. Both projects are analysed against the same standing policy, and the contracting authorities — in different sector ministries — never see the policy as a moving target.

5. Stage C — Value for Money

Why this matters: the value-for-money (VfM) test is the formal answer to the question would the country be better off doing this project as a public works contract. The Public Sector Comparator (PSC) is the cost the country would bear if it built and operated the project with public capital and conventional procurement; the PPP NPV is the cost under the proposed contract. Today the contracting authority's advisor produces both numbers and the verdict; under the orchestrator partition the MoF owns both and sets the methodology.

5.1 The PSC is a fiscal-cost computation

The PSC is, line by line, a fiscal-cost computation: capex over the construction period, opex over the operating period, lifecycle replacement costs, residual value, retained risks priced at the country's risk-pricing parameters, and a discount rate. Every input on that list is a variable the MoF either holds directly (discount rate, cost of public-sector capital) or has comparative advantage on (risk pricing, lifecycle inflation). None of the inputs is a sector-technical input that the contracting authority has comparative advantage on, except for the underlying capex and opex profiles — which the CA produces, but which the MoF then prices.

The contracting authority's advisor that runs the PSC implicitly chooses the discount rate, the cost-of-capital adjustment between the two delivery modes, and the risk-pricing curves. Each of those choices is contestable on fiscal-policy grounds. Under the orchestrator partition the MoF chooses them once, publishes them, and applies them to every project; the CA produces the underlying capex and opex.

5.2 The threshold is also fiscal-policy

The country's VfM threshold — how much PSC-cost-saving the PPP must deliver to be preferred to public delivery — is a fiscal-policy choice. Some countries set it at zero (PPP must be at least as cheap); some at 5% (PPP must be at least 5% cheaper); some refuse a numeric threshold and substitute a qualitative weighting framework (UK Green Book⁹). All three are defensible policy positions. None is the contracting authority's choice to make.

5.3 What the CA still owns

The CA owns the project specification (capex, opex, performance regime), the procurement strategy, the contract design, the supplier-relationship management. The MoF takes those inputs and runs the PSC against them; the CA does not run a parallel PSC of its own.

This is the place where the partition is hardest to enforce in practice. Most VfM assessments today are run by transaction advisors retained by the CA, with the MoF receiving the result as a one-line VfM differential. The orchestrator partition requires the methodology, the discount rate, and the threshold to be fixed by the MoF and applied by the MoF. The CA's advisor remains useful — it produces the underlying cost stack — but the verdict is not its to issue.

6. Stage D — Third-party guarantees

Why this matters: a third-party guarantee can move a project from "unbankable in the current market" to "investible". It can also be expensive and may be taken when it should not be, or rejected when it should be taken, depending on whose perspective runs the analysis. Under the orchestrator partition the MoF runs the take-or-reject decision for every candidate guarantee, on every project, against the country-set discount rate.

6.1 The decision framework

The third-party guarantee decision was the subject of the prior paper in this series³⁷. The framework is summarised as follows. The financial model is run twice — once at market terms (Run A), once at guaranteed terms (Run B with the spread reduction, tenor extension, and DSCR-covenant relaxation that the guarantee buys). The differential equity NPV minus the NPV of the guarantee fee determines the take-or-reject verdict. The framework distinguishes between the sponsor-view differential (which underweights the value of the guarantee) and the planner-view society saving (which captures the lower equity hurdle that the guarantee unlocks via the equilibrium-pricing mechanism of Engel, Fischer, and Galetovic³²). The planner-view is the right comparison for a benevolent central planner; the sponsor-view is the right comparison for the sponsor.

6.2 Why the MoF has the comparative advantage

The menu of available guarantees — MIGA PRI and NHSFO, IBRD and IDA Project-Based Guarantees, IFC and ADB Partial Credit Guarantees, OECD-Arrangement ECAs, private PRI providers — is a piece of MoF-side information. The MoF speaks to MIGA, the World Bank Treasury, the IFC, and the home-country ECAs as part of its standing relationships. The contracting authority does not. The pricing of each guarantee is also MoF-side information, and changes faster than a CA's advisor can track.

The framework's planner view — the society-saving comparison — is by construction a MoF computation. The CA's advisor, which is paid by a sponsor or by the CA itself, has no business running a society-view computation; it has business running a sponsor-view or CA-view computation. Confusing the two has been a major source of the under-uptake of multilateral guarantees in countries that would benefit from them.

6.3 The programme-level orchestration

When the MoF runs the guarantee decision for every project in the country's pipeline against the same discount rate, with the same audit chain, on the same fee tables and the same counterfactual-term tables, the country's relationship with the multilateral guarantor moves from project-level to programme-level. The MoF can negotiate a master arrangement with MIGA or the World Bank Treasury covering the country's pipeline rather than negotiating one project at a time. The negotiating leverage is qualitatively different.

This composes cleanly with the prior paper in this series. Paper #2³⁷ derived the take-or-reject rule for a single project from the perspective of whoever runs the model; this paper specifies that whoever should be the MoF, applying the country's published discount rate and standing relationship with the candidate guarantors. A MoF that has implemented Paper #2's framework against a single project has taken the first orchestrator step on Stage D — generalising it to every project in the pipeline against a single discount rate is the full implementation, and is what enables the programme-level negotiation above.

7. Stage E — FCCL and fiscal envelope

Why this matters: the FCCL ceiling is the only MoF-owned binding constraint that survives under all variants of the doctrine. Even a pure gatekeeper MoF must aggregate FCCL exposures and refuse projects that breach the ceiling. The orchestrator partition extends this ownership in two directions: the MoF sets the ex-ante envelope, and every project flows against the envelope at every gateway.

7.1 The ex-ante envelope

A standing FCCL envelope expressed in two numbers — firm commitments as a percentage of GDP and contingent liabilities as a percentage of GDP — is the country's binding fiscal constraint on the PPP programme as a whole. Uruguay's Article 62 of Law 18.786 is the cleanest example: a legally binding cap on firm and contingent PPP commitments, monitored annually by the MEF and published in the budget documentation¹⁰. Brazil's Lei 11.079 imposes a federal cap of 1% of Receita Corrente Líquida (current net revenue) on PPP commitments, monitored by the STN at the Tesouro Nacional¹². Honduras has a 5% cap which the MoF was unable to enforce — actual contingent liabilities reached 6.6% of GDP, against the legal cap, with the MoF as the formal gatekeeper — a clean illustration of why a gatekeeper MoF without analytical capacity cannot enforce its own limit¹³.

The envelope is an MoF-side variable by definition. Under the orchestrator partition the envelope is set transparently, published, and updated against the country's fiscal plan; under the gatekeeper partition it is often inferred ex-post from project-by-project decisions, with no clear connection to the country's debt-service capacity.

7.2 The per-project draw

Every project at every gateway draws against the envelope. The draw is computed by the MoF using a stress-tested PFRAM-style model on the project's cash flows, with a Monte Carlo distribution on the contingent risks (MRG, IMG, MAR, debt-service guarantee, termination payment). The expected draw is added to the firm-commitments line; the P95 of the contingent draws is tracked separately against the contingent-liabilities line. Both lines are visible to the contracting authority before the project enters structuring, so the CA can adapt the project to fit the envelope rather than discovering that it does not fit at the affordability gate.

7.3 The reporting

The orchestrator partition produces three standard outputs from the same engine: an FCCL note in the format used by the World Bank and IMF FCCL methodology; an IPSAS-32 reconciliation for the Sector Public Non-Financial balance sheet; a Government Finance Statistics 2014 (GFS 2014) fiscal-statistics breakdown for the IMF Article IV. All three are produced with one click from the same audit chain; the rating-agency sovereign-team conversation moves from defensive posture to validation exercise when the country can hand over the IMF Fiscal Risk Statement (FRS), the IPSAS-32 note, and the FCCL bulletin in one bundle, all referencing the same audit-chain identifier³⁶.

8. Evidence

Why this matters: the orchestrator hypothesis is contestable. It stands or falls on whether the LEAD designation predicts, in real data, the kind of programme outcomes the doctrine implicitly says it should. We test it against three independent bodies of evidence.

8.1 The literature scoring (n = 12 documents)

We scored twelve canonical doctrine documents on the four-level rubric (NONE, GATE, CO-LEAD, LEAD). The full per-document briefs and scoring rationales are in the library at references/01_canonical_frameworks/. The summary distribution:

Two findings stand out from this scoring. First, the LEAD cell is empty — no canonical doctrine document prescribes that the MoF should run the analytical chain end-to-end. Countries that reach LEAD have done so without doctrinal cover. Second, the CO-LEAD documents (PFRAM, OECD, Irwin 2007, IADB LAC) make a methodological case that the GATE documents (WB v3, APMG, EPEC × 2, EBRD) have not operationalised. The argument the MoF should run the analytical chain is available in the literature; the doctrine has chosen not to follow it.

8.2 The country scoring (n = 50 countries)

We scored fifty countries on the same rubric. Country-by-country briefs and source files are in the library at references/02_country_legal/. The summary distribution:

Three structural findings stand out. Programme scale is uncorrelated with MoF analytical leadership. Japan (a multi-trillion-yen cumulative programme, scored GATE¹⁴) and Türkiye (approximately USD 70-plus billion across roughly 200 projects since the early 2000s, scored GATE¹⁵) are the largest GATE programmes in their regions; Uruguay (small but LEAD) and New Zealand (small but LEAD) show that orchestration capacity is design-driven. Three regression episodes are documented, with different durabilities. Canada dissolved PPP Canada in 2018 (federal PPP function downgraded from CO-LEAD to GATE; persistent)¹⁸; Pakistan moved its PPP Authority out of the Finance Division by statute in 2021 (dropping to GATE) but the Finance Division partially restored the analytical function in 2024 via a Risk Management Unit (returning the country to CO-LEAD)¹⁹; the World Bank doctrine itself regressed from CO-LEAD to GATE between Irwin 2007 and the WB Reference Guide v3 (2017) and the regression persists. The doctrine and the practice can both retreat, and §13.2 takes up the durability distinction between the three. De jure design and de facto practice diverge. Several CO-LEAD countries (Morocco, Pakistan-RMU, UAE federal, NZ NIFFCo) have legal designs ahead of demonstrated practice; the IMF Morocco Article IV 2024 explicitly recommended MEF add a PPP budget annex and set a global annual envelope, both of which the law contemplates and the practice does not yet produce¹⁶.

8.3 World Bank PPI Database, 1990 to 2024 (n = 31 in-scope countries, 7,135 projects)

The World Bank Private Participation in Infrastructure database³⁵ covers low- and middle-income economies; nineteen of our fifty scored countries are out of scope by design (all of Western Europe; Australia, Canada, New Zealand; Japan and Korea; Chile; Saudi Arabia and UAE). Thirty-one countries are in scope, with at least one project each: four LEAD (Egypt, India, South Africa, Uruguay), seventeen CO-LEAD, and ten GATE. Replication notebook at references/03_empirical/ppi/.

¹ Means are unweighted across countries within each score group, not weighted by project count. This treats each country as an institutional-design unit of observation, which is the variable the orchestrator hypothesis predicts on. Project-weighted means are reported in the replication notebook (Appendix B) for readers who prefer that aggregation.

Statistical tests (Mann-Whitney U, two-sided): LEAD vs CO-LEAD on cancellation rate, p = 0.065 (effect size r = 0.41, marginally significant at p < 0.10); LEAD vs GATE on cancellation rate, p = 0.18 (the difference is large in magnitude but the test is underpowered and outlier-driven — Bolivia's 57% cancellation rate is a 1990s capitalisation reversal not an institutional-design failure in the current sense, and Argentina's 36% is the post-2018 dormancy of the Ley 27.328 regime). Excluding Bolivia and Argentina the GATE mean drops from 13.5% to 5.3% and the gap to LEAD remains directionally large but ceases to be testable on this sample.

Two caveats are worth stating explicitly. The PPI Database excludes high-income economies, so the LEAD signal is generated by four MIC LEADs (Egypt, India, South Africa, Uruguay); the high-income LEAD cases (Korea, Portugal, NZ) are not in the test. This is informative rather than disabling: it shows that MoF orchestration is associated with better project outcomes at the income level where development gaps are largest. The PPI Database also does not carry renegotiation data, which the literature on Latin American PPPs identifies as the most theoretically important PPP-outcome variable³³; the IADB renegotiation dataset would be the right complement and is flagged as a follow-up.

One outlier deserves named attention: Senegal, scored GATE on our rubric, has a 2.4% cancellation rate and a 70.7% MDB-involvement rate — the best programme performance in the GATE cell. The reading is that intensive MDB transaction support can substitute, in the short run, for MoF orchestration in programmes that are small enough to fit inside the MDB pipeline. It is a substitute, not an equivalent: the substitute does not scale beyond what the MDB can carry, and it externalises the orchestration function rather than building it.

8.3.1 Robustness — the LEAD signal is India-dominated

A peer-reviewed version of this paper would lead the empirical section with the robustness check this subsection makes. India contributes 1,135 of the 1,547 projects in the LEAD cell (about 73%). The LEAD-group mean cancellation rate is therefore heavily a function of India's number. The remaining three LEAD countries (Egypt, South Africa, Uruguay) carry small project counts, and a single high-profile cancellation in any of them would visibly move the LEAD mean.

The honest reading of the test is therefore: the LEAD signal in the PPI sample is reliable as a statement about India and indicative as a statement about the LEAD cell. Excluding India from the test reduces the LEAD cell to N=3 countries with about 400 projects, drops the unweighted-mean cancellation rate from 1.0% to a number that varies materially with how a small handful of South African / Uruguayan projects are classified, and removes the statistical power needed to test against CO-LEAD or GATE. The directional finding (LEAD cancellation lower than CO-LEAD lower than GATE) survives the India exclusion; the strength of the result does not.

This is a known limitation, not a hidden one. India's "surprise LEAD" classification is itself contested in the Indian PPP literature, where the DEA-PPPAC architecture is sometimes characterised as catalytic-coordinative rather than analytical-end-to-end. A reviewer who re-scores India as CO-LEAD (defensible on either reading) would observe the LEAD signal in this sample evaporate. The paper's empirical case rests, in the end, on a small N of LEAD countries operating without doctrinal cover and producing outcomes that — at the unweighted-country level of observation — sit below the GATE distribution. The argument should not be made stronger than that.

The qualitative case for the orchestrator partition, which does not rest on the empirical signal, is in §13.6.

8.4 Procuring Infrastructure PPP 2018 and Benchmarking Infrastructure Development 2023 (n = 49 in-scope countries)

The World Bank's Procuring Infrastructure PPP 2018 product and its 2023 successor Benchmarking Infrastructure Development³⁴ score countries on four sub-dimensions of PPP procedural design: Preparation, Procurement, Contract Management, Unsolicited Proposals. Forty-nine of our fifty scored countries are in scope (Bolivia is not). Replication notebook at references/03_empirical/infra_benchmark/.

The headline result is that LEAD countries score significantly higher than GATE countries on the sub-score that, in each edition, carries the fiscal-orchestration weight:

LEAD-vs-CO-LEAD comparisons are uniformly small and statistically insignificant. The meaningful divide is between the upper two cells (LEAD + CO-LEAD) and GATE — not between LEAD and CO-LEAD. The orchestrator-vs-gatekeeper boundary is the analytically useful boundary; the LEAD/CO-LEAD subdivision is real on the institutional-design rubric but does not yet show up in outcome-relevant data.

The 2018 → 2023 shift in which sub-score carries the signal is itself a finding. The 2018 LEAD advantage on Preparation (+26 points, p < 0.01) evaporated in 2023 (−3 points, n.s.), but a new LEAD advantage on Procurement (+15 points, p ≈ 0.01) appeared. The most plausible interpretation is that the 2020 redesign of the World Bank product moved the fiscal-rigour content (PFRAM-style affordability checks, FCCL-style commitment ceilings, value-for-money methodology) out of the Preparation dimension and reweighted it under Procurement. The orchestrator advantage tracks the dimension that asks the right question, not a fixed-name dimension. The paper therefore treats the two editions as a methodological evolution rather than a longitudinal series. The defensible claim is the cross-edition one: LEAD countries score 15 to 26 points higher on whichever sub-score, in each edition, captures fiscal orchestration.

8.5 The signal is detected through an instrument not designed to detect it

The same World Bank teams that produced the BID/PI-PPP scoring methodology produced the canonical doctrine documents we score in §8.1 — and none of those documents prescribes LEAD. The methodology itself carries the gatekeeper frame. That the orchestrator signal is nonetheless detectable through the methodology is informative: it suggests that a purpose-built measurement instrument that included an explicit "MoF analytical leadership" dimension would likely show a stronger effect. We treat this as a lessons-learned recommendation in §13.

8.6 What the evidence does not show

The PPI sample sizes per cell are small (LEAD = 4) and confounders are not controlled for (income level, legal tradition, World Bank technical-assistance history, sector composition, political stability, IMF programme presence). The BID/PI-PPP scoring measures de-jure design, not implementation. The PPI Database does not carry renegotiation data. We cannot make a causal claim. The honest reading is: there is correlational evidence consistent with the orchestrator hypothesis, with effect sizes that are large but sample-size-constrained. The hypothesis is supported, with caveats; it is not proven.

9. Comparator cases

Why this matters: the seven LEAD countries, four GATE-failure cases, and three design-space cases are the empirical anchor of the paper's argument. Each is scored and source-referenced in the library; what follows is the in-paper digest.

9.1 Seven LEAD orchestrators

Uruguay (LAC, LEAD). Law 18.786 of 2011 places the Comisión Interministerial APP under the chairmanship of the Minister of Economy and Finance (Article 8) and the Unidad de Proyectos PPP under the MEF directly (Article 13). Article 62 imposes a binding legal cap on firm and contingent PPP commitments. The MEF publishes an annual Compromisos PPP report; the Tribunal de Cuentas audits it; rating agencies cite it as a positive factor in Uruguay's BBB+/Baa1/BBB sovereign ratings. The Ferrocarril Central PPP (USD 1 billion total project cost, the largest PPP in Uruguayan history) closed under this design, with BID Invest leading a USD 855 million senior financing tranche. Uruguay is the cleanest LAC LEAD case¹⁰^,^¹¹.

Portugal (Europe, LEAD). Decreto-Lei 111/2012 places the Unidade Técnica de Acompanhamento de Projetos (UTAP) inside the Ministry of Finance by statute. UTAP holds mandatory prior-opinion authority on every PPP, monitors execution, and publishes the Boletim Informativo das PPP annually. The 2025 reorganisation into the ETF (Entidade de Tesouraria e Finanças) reinforced rather than diluted the MoF placement. Portugal is the cleanest European LEAD case²⁴.

South Africa (Africa, LEAD). The Public Finance Management Act of 1999 and Treasury Regulation 16 (amended February 2025) place the GTAC PPP Advisory Unit inside the National Treasury. Treasury approval is the binding step at four gateways: Treasury Approval I (project registration and inception), TA:IIA (feasibility study), TA:IIB (procurement documentation), and TA:III (final award). The PPP Manual (refreshed 2024) is the published methodology. South Africa is the cleanest African LEAD case²².

Egypt (Africa, LEAD). PPP Law 67 of 2010 established the PPP Central Unit (PPPCU) inside the Ministry of Finance by ministerial decree. The PPPCU selects transaction advisers, maintains the FCCL database, and is the IMF and PPIAF counterpart for PPP fiscal-surveillance tooling. The IMF explicitly engaged with the PPPCU on the Fiscal Risk Statement under the EFF arrangement. Egypt is co-equal with South Africa as Africa's clearest LEAD case, frequently understated in cross-region comparisons²³.

Korea (Asia, LEAD). The Act on Private Participation in Infrastructure (1994, reformed 2005) places the Public and Private Infrastructure Investment Management Center (PIMAC) at KDI under the supervisory authority of the Ministry of Economy and Finance. PIMAC runs the Preliminary Feasibility Study, the Value-for-Money Test, and the RFP review for every PPP. The annual PPP Basic Plan is issued by MOEF. Korea is the textbook LEAD case²⁰.

India (Asia, LEAD). India has no single PPP Act, but the Department of Economic Affairs (DEA) at the Ministry of Finance hosts the Private Investment Unit, secretariates both the Public-Private Partnership Appraisal Committee (PPPAC) and the VGF Empowered Committee — both chaired by the Secretary (DEA) — and publishes the Model Concession Agreements that bind the sectoral programmes. The Kelkar Committee Review (2015), commissioned by the MoF, set the post-2015 reform agenda. India is the surprise LEAD case: a de facto leadership architecture built through executive instruments rather than statute. It demonstrates that LEAD is reachable through accumulated convention, not only through law²¹.

New Zealand (Anglosphere, LEAD). Treasury Circular 2024/18 and the December 2024 NZ PPP Blueprint place the PPP function inside the Treasury (Te Tai Ōhanga) for policy, methodology, and fiscal disclosure; the National Infrastructure Funding and Financing Company (NIFFCo, a Crown company under Ministry of Finance mandate from December 2024) holds execution. The Long-Term Fiscal Statement carries the contingent-liability disclosure. New Zealand is the most recent LEAD designation in the sample. The Transmission Gully PPP is a credible counter-example to a strong LEAD reading: institutional design did not prevent project-level operational failure, and the post-Transmission Gully Blueprint reforms are precisely the Treasury's self-correction response²⁵.

9.2 Four GATE-failure cases

These four cases are the paper's evidence on what happens when the orchestrator role is absent. They are not presented as causal proofs; they are presented as illustrations consistent with the orchestrator hypothesis.

Türkiye (MENA, GATE). A programme of approximately USD 70-plus billion across roughly 200 projects since the early 2000s, with MoF contingent exposure on the order of 3% of GDP (debt assumptions, repayment guarantees, minimum-revenue payments combined). The IMF Article IV 2024 staff report states that no single agency is responsible for monitoring and managing the fiscal costs and risks of the overall portfolio. No central PPP unit; sector ministries run the analytical chain. Treasury and Finance Ministry instruments are reactive financial backstops triggered at financial close or early termination, not analytical inputs at project entry. Largest GATE programme in the Middle East, defensively governed¹⁵.

Japan (Asia, GATE). The PFI Act of 1999 (amended 2011 and 2018) places the analytical lead with the Cabinet Office's PPP/PFI Promotion Office, not with the Ministry of Finance. The country runs a multi-trillion-yen cumulative PFI programme — the largest mature programme in the sample by deal count — and the MoF has no project-level gateway role at all. A clean test case: programme scale does not require MoF analytical leadership, but it also does not produce it¹⁴.

Spain (Europe, GATE). Ley 9/2017 governs concessions but places no central PPP unit inside the Ministerio de Hacienda. The radial-motorway portfolio (the radiales around Madrid, conceded in the early 2000s) collapsed in 2010 to 2018; the Spanish state absorbed approximately EUR 5 billion in costs it had not modelled ex-ante via the responsabilidad patrimonial de la Administración mechanism. No MoF analytical gate, no mandatory ex-ante modelling, no published methodology. The textbook case of GATE-failure at scale²⁸^,^²⁹.

Honduras (LAC, GATE). Decreto Legislativo 143-2010 created COALIANZA, the PPP Authority, under the Presidency rather than the Ministry of Finance. The MoF's Fiscal Capacity Unit (FCU) was established in 2014 but has been documented by the IADB as "unable to function in its oversight/approval capacity". Actual PPP contingent liabilities reached 6.6% of GDP against a 5% legal cap — a binding statutory MoF gate that the country's institutional reality could not enforce. The case where the gatekeeper design failed for the precise reason the orchestrator partition argues against: the MoF without analytical capacity cannot defend a refusal¹³^,^²⁷.

9.2.1 What an orchestrator MoF would have caught — Spain and Türkiye walked through the five stages

The four GATE-failure cases are not natural experiments — there is no parallel-universe orchestrator-Spain to test against actual-Spain. But the failures have a structure, and the structure maps to the five MoF-led decisions in §3 to §7. Walking Spain's radial-motorway collapse and Türkiye's portfolio accumulation through the five stages produces a concrete, falsifiable account of where the analytical chain would have intervened. The reader is invited to challenge each step.

Spain — the radiales around Madrid (2003–2018). Six toll-road concessions (R-2, R-3, R-4, R-5, M-12, AP-36, AP-41) were tendered in the early 2000s on traffic forecasts that priced in continued economic expansion and continued price-inelastic willingness-to-pay. The 2008 recession cut both volumes and prices simultaneously; six concessions filed for bankruptcy protection between 2010 and 2014; the Spanish state absorbed approximately EUR 5 billion in responsabilidad patrimonial de la Administración payments. Walking this through the orchestrator chain:

• Bankability (§3). The MoF's canonical bankability model, run at the central discount rate against the bidders' traffic forecasts, would have produced ADSCRs that depended sensitively on the demand assumption. A Monte Carlo on the traffic input — standard practice in a canonical MoF chain — would have generated a P5 / P95 range of equity IRRs that flagged the unhedged demand exposure visibly. The arrangers' own models did this analysis privately; the MoF saw only the headline ADSCR.
• Risk allocation (§4). A standing per-country policy that placed catastrophic-demand-loss risk on the sponsor (with a defined cap) rather than ambiguously between sponsor and state would have priced into the tariff up-front. The actual contracts were silent on the catastrophic-loss case, which is why the responsabilidad patrimonial de la Administración mechanism was triggered.
• VfM (§5). A PSC computation at the MoF's discount rate that retained demand risk on the public-sector counterfactual at the same level as the PPP would have produced a smaller VfM differential than the bidder-side computation showed. The PPP-vs-PSC verdict for at least two of the six concessions is unlikely to have cleared a 5% threshold under the corrected risk-retention assumption.
• Guarantees (§6). The portfolio did not use multilateral guarantees, but a standing MoF process for guarantee selection would have surfaced the absence — and asked why a major motorway portfolio with concentrated demand exposure was not using a partial-credit or partial-risk facility from EBRD or IFC.
• FCCL (§7). Aggregating six toll-road concessions into a single contingent-liability line, with the demand-stress correlated across roads, would have produced a programme-level VaR that the affordability gate would have refused at the bid date. The country's debt-sustainability analysis at the time did not carry this line.

The orchestrator chain would have intervened at every one of the five steps. Whether it would have changed the political decision is a separate question — the radial motorways were strategic infrastructure for Madrid's growth, and the political imperative may have absorbed even a clear analytical refusal. But the analytical artefacts the MoF would have produced would have created an audit trail at the IGAE, the Tribunal de Cuentas, and the IMF Article IV mission that did not exist at the time and that would have made the post-2010 absorptions politically more costly to defend. The mechanism through which orchestration improves outcomes is not always "refusal at the gate"; it is often "audit-defensible warning that constrains the political decision over time."

Türkiye — the portfolio accumulation (2003–present). Türkiye built a ~USD 70-plus billion programme across roughly 200 projects (airports, hospitals, motorways, bridges) using a sector-by-sector procurement model with Treasury and Finance Ministry providing reactive financial backstops triggered at financial close or early termination. The IMF Article IV 2024 estimates MoF contingent exposure on the order of 3% of GDP from guarantees, debt assumptions, and minimum-revenue payments combined. Walking this through:

• Bankability (§3). Project-by-project bankability decisions sit with the sector ministries' arrangers; the MoF receives summaries. A canonical MoF bankability run on every project would have surfaced the demand-risk concentration in the airport portfolio (where minimum-revenue payments are the default counterfactual instrument) and the FX exposure on the energy portfolio.
• Risk allocation (§4). No standing per-country risk-allocation policy exists; each sector ministry instantiates its own. The Treasury's 2018 Guidelines on Fiscal Commitments names the categories of commitment but does not bind sector ministries to a standing position.
• VfM (§5). PSC computations are run at the sector level with sector-specific discount rates; there is no central VfM methodology. The 2024 Article IV recommends one. A canonical MoF VfM run would have flagged at least the hospital portfolio — where the procurement model has been studied and the PPP-PSC differential is contested — for explicit verdict.
• Guarantees (§6). Türkiye uses sovereign instruments (debt assumption, payment guarantee, MRG) routinely. A canonical MoF guarantee-selection process against an external menu (MIGA, IBRD, IFC) would have asked, for several large projects, why the sovereign was the guarantor rather than a multilateral — moving exposure off the sovereign balance sheet at the cost of a multilateral fee.
• FCCL (§7). This is the gap the Article IV most directly names: no single agency aggregates the portfolio. A standing FCCL aggregation would have produced a cumulative number that the MoF would have had to defend against the country's debt-sustainability analysis. The MoF would not have stopped the programme but would have constrained the pace and the instrument mix.

In both cases the orchestrator chain would have intervened, would have produced audit-defensible warnings, and would have shifted the political conversation toward instruments that reduce sovereign exposure. Neither case shows that orchestration would have stopped a bad project from being signed — political pressure, in both Spain and Türkiye, would likely have absorbed a clear analytical refusal. But both cases show that orchestration raises the cost of ignoring the warning, which is the mechanism through which institutional design improves outcomes over time.

9.3 Three design-space cases

These three cases show where the partition is harder than the LEAD-or-GATE binary suggests. The third case treats Australia and Canada jointly because their pattern is the same: federal-level gatekeeper, sub-national orchestrator.

United Kingdom (Europe, CO-LEAD). HM Treasury designed the Standard Form Contract (SoPC) and killed PF2 with a single Budget 2018 announcement — strong policy primacy. But HMT never ran the project-level VfM models; the sector departments and their advisors did. The 2018 National Audit Office report on PFI/PF2 is devastating precisely because it shows MoF policy leadership without MoF analytical control: the policy frame was strong, the project-level analysis was not the Treasury's, and the country paid for the gap. The IPA — jointly sponsored by HMT and the Cabinet Office — is the institutional residue. The UK is the clearest evidence that policy leadership does not equal analytical orchestration, and the gap is where the cost shows up³¹.

Chile (LAC, CO-LEAD, near-LEAD). DFL 164/1991 established the concessions regime; DS 900/1996 and Ley 20.410/2010 refined it. Hacienda's Subdepartamento de Pasivos Contingentes y Concesiones at DIPRES has co-signature on every concession decree alongside the Ministry of Public Works (MOP). The annual Informe de Pasivos Contingentes is regional best practice. Chile remains CO-LEAD because the transactional unit (MOP's Dirección General de Concesiones) sits at MOP and DIPRES retains only the fiscal-monitoring co-signature — the split keeps the analytical chain divided across two ministries. The most-cited LAC reference and the model many subsequent designs were built against²⁶.

Australia and Canada (Anglosphere, CO-LEAD sub-national, GATE federal). The two federal Anglosphere cases reveal that "MoF as orchestrator" in federated systems means the sub-national finance ministry. State Treasuries in Australia (NSW Treasury, Victoria's Department of Treasury and Finance, Queensland Treasury) and provincial governments in Canada (Ministry of Infrastructure / Infrastructure Ontario in Ontario, Infrastructure BC under the Ministry of Finance in British Columbia) hold the analytical chain. The federal MoF in each country holds at most a GATE function — Infrastructure Australia is advisory only at the Commonwealth; in Canada, the federal PPP function contracted to a financial-instrument veto via the Canada Infrastructure Bank after PPP Canada's 2018 dissolution, with the Department of Finance providing budget-level concurrence rather than analytical leadership. The orchestrator partition is therefore tier-specific in federated systems, and the fiscal-risk visibility gap that this creates (the federal MoF carries the consolidated balance-sheet exposure but does not exercise the orchestrator function) is itself a governance issue, taken up explicitly in §14.1³⁰^,^¹⁷.

10. Architecture — how the Austral platform closes the asymmetry

Why this matters: the orchestrator partition is implementable on existing tooling. The Austral platform implements all five MoF-led decisions in §3 to §7, against the same financial-model engine, with the same audit chain, in the same workspace. The platform does not displace the MoF's judgement; it supplies the MoF's tooling.

Disclosure. The Austral platform is built by the author and is the analytical tooling that this paper's three working papers in the Austral Working Papers in Infrastructure Fiscal Intelligence series implement. The orchestrator partition is the paper's thesis; the platform is one implementation. Alternative implementations exist and are not displaced by this paper's argument — the IMF / World Bank PFRAM as a standalone, the World Bank PPP Reference Guide companion Excel templates, country-specific tooling stacks such as Korea's PIMAC suite or South Africa's National Treasury PPP Manual workbooks, and commercial offerings from the major Big-4 advisory firms all reduce the MoF-side information cost on different cuts of the analytical chain. The case for orchestration in §3 to §7 stands on the partition rule (§2.4), not on the choice of tool. This paper foregrounds the Austral platform because (a) it implements all five MoF-led decisions against a single canonical engine with a shared audit chain, which is a stronger condition than any alternative meets at the time of writing; and (b) it is the implementation the author can speak to with full evidence. A reader who prefers a different implementation should still find the partition argument intact.

10.1 The five MoF-led modules

All five modules read and write the same canonical ModelInputs schema. The third-party guarantee decision uses the framework from the prior paper in this series³⁷; the VfM and bankability decisions use the engine from the framework before it³⁶.

10.2 The audit chain

Every model run, every guarantee evaluation, every FCCL aggregation writes an immutable entry to the workspace activity log via the platform's append-only, SHA-256 hash-chained audit trail. The log entry records the inputs, the outputs, the user, the timestamp, and a content hash. Every PDF, DOCX, and XLSX export references the audit-chain ID at the bottom of every page. An auditor — internal compliance, supreme audit institution, IMF Article IV mission, MDB compliance team, parliamentary committee — can re-run the same call against the saved inputs and verify the result matches the audit-chain entry to the cent.

This is the institutional answer to a question the orchestrator partition raises and the gatekeeper doctrine ducked: how does the MoF defend a verdict that overrides a contracting authority's advisor? The answer is with the audit chain in hand. The MoF's verdict is reproducible by anyone with the inputs, and the inputs are signed and time-stamped. The contracting authority's advisor can challenge the inputs; it cannot challenge the computation.

10.3 The guest-mode walkthrough

Anonymous "guest mode" access is sufficient to follow the walkthrough in §11 end-to-end without creating an account. The platform is at https://platform.austral-intelligence.com.

11. Walkthrough — reproduce in 10 minutes

Why this matters: a paper that proposes an institutional re-partition has to show that the partition is operationally light. Ten minutes of guest-mode interaction takes a single project through all five MoF-led decisions and produces a defensible audit-chained bundle.

11.1 The synthetic project

Open the platform at https://platform.austral-intelligence.com and load the Country X · Solar Availability-Payment sample workspace. The project is a USD 500 million utility-scale solar facility in a frontier-market jurisdiction with an OECD Country Risk Classification (CRC) of 6 on the 0-to-7 scale, procured under a 25-year availability-payment contract with a sovereign offtaker. Capex is USD 500 million spread over 36 months of construction; opex is USD 12 million per year; the indexed availability payment produces approximately USD 65 million per year of revenue. The financing is structured 70/30 debt-to-equity with a single senior tranche and a 1.40× minimum DSCR covenant under market terms. The walkthrough exercises one project end-to-end; §12 then exercises the same five modules across a four-project programme.

11.2 The five-step sequence

(The numbers below are illustrative — the platform produces the actual figures from the saved inputs.)

1. Bankability (§3). Open the Financial Model module. Load the saved ModelInputs, run the model, observe the RunSnapshot: Min ADSCR, equity IRR, equity NPV, project IRR, feasibility flag. Note the project is unbankable at market terms (Min ADSCR below 1.40×, equity IRR below the 18% hurdle).
2. Risk allocation (§4). Open the Risk Allocation module. Compare the project's allocation to the country's standing policy. The default allocation has the SPV bearing FX risk in full on the indexed availability payment; switch to the country's standing partial-pass-through policy (sponsor bears the first 10% of devaluation against the USD-bond curve, sovereign bears beyond). Observe the bankability snapshot update — the equity IRR rises by approximately 1.2 percentage points and the implied availability-payment escalation falls by approximately 6%.
3. Value for Money (§5). Open the VfM module. Run the PSC; the PPP delivers a 12% NPV saving against the public delivery comparator at the country's published 8% public discount rate. The project clears the country's 5% VfM threshold.
4. Third-party guarantees (§6). Open the Credit Enhancement Decision sub-module under Financial Model. Sweep MIGA Political Risk Insurance, IBRD Project-Based Guarantee, and IFC Partial Credit Guarantee at 50%, 75%, 90% coverage. The IBRD PBG at 75% coverage produces the highest society-saving NPV; the project becomes bankable under the guaranteed terms with material headroom on both DSCR and equity IRR.
5. FCCL & fiscal envelope (§7). Open the FCCL aggregation view (under FARO for project-level draws, under the Infrastructure Fund module for portfolio-level capacity, and under the FCCL Alerts cross-cutting view for ceiling rules). Add the project to the country's pipeline. Observe the firm-commitments draw (the discounted availability-payment stream against the firm-commitments line) and the contingent-liabilities draw (driven by the IBRD PBG counter-guarantee and the FX pass-through above the 10% sponsor band). Confirm the country's envelope still has headroom on both lines.

Total elapsed time on the platform: approximately ten minutes for a practitioner familiar with the modules. Each step writes to the audit chain; the final exported bundle (PDF + DOCX + XLSX) carries the audit-chain identifier at the bottom of every page.

11.3 The exported bundle

The exported bundle contains a methodology note (DOCX), a project model spreadsheet (XLSX), a dashboard PDF (with the bankability snapshot, the VfM differential, the guarantee decision, and the FCCL draw), and the per-step audit-chain identifiers. Any auditor can re-run the same five steps against the same inputs and verify the bundle reproduces bit-for-bit.

12. Worked example — a synthetic frontier-market pipeline

Why this matters: a single project illustrates the per-project mechanics; a small pipeline illustrates the programme-level orchestration that is the point of the partition.

12.1 Country X — context

Country X is a synthetic LAC frontier-market sovereign, OECD CRC 6 (see §11.1 for the scale), with a recently reconstructed fiscal anchor (a multi-year IMF Extended Fund Facility, sovereign spread approximately 400 basis points over US Treasuries, sovereign rating B+/B1/B+). The country has a USD 6.6 billion candidate PPP pipeline of four projects: a 1,000 MW hydroelectric IPP (USD 4.4 billion capex), a 400 MW hydroelectric IPP (USD 1.7 billion capex), a 90 km motorway corridor (USD 350 million capex), and a multi-purpose grain terminal (USD 230 million capex). The country has reaffirmed an EFF programme that endorses the PPP framework and explicitly requires a Fiscal Risk Unit able to monitor PPPs and state-owned enterprises.

The CA's portfolio is sectorally mixed: two energy projects, one transport project, one port project. Three line ministries are the contracting authorities. The MoF holds the orchestrator role under the partition rule.

12.2 The orchestrator's programme-level decisions

Under the partition rule the MoF makes five programme-level decisions before the projects enter detailed structuring.

1. Discount rate. The MoF publishes an 8% public discount rate for VfM and PSC computations across the pipeline.
2. Risk-allocation policy. The MoF's standing policy places construction-overrun risk on the SPV (with a 12% capex-cap force-majeure carve-out above which a sovereign back-stop applies) and offtake risk on the public offtaker via take-or-pay PPAs for energy projects. Transport projects use an availability-payment regime with FX risk shared on a partial-pass-through basis. The grain terminal is a fully user-pays concession with the country retaining termination-payment risk only.
3. VfM threshold. The MoF sets the threshold at 5% PSC saving — the Latin American median — with a published methodology.
4. FCCL envelope. The MoF sets an envelope of 4% of GDP firm + 3% of GDP contingent for the PPP programme as a whole, leaving the Article-equivalent ceiling at 7% combined. The envelope is published in the budget annex and reviewed annually against the country's debt-sustainability analysis.
5. Guarantee policy. The MoF runs the take/reject decision on the candidate guarantees for each project against the published 8% discount rate and selects the cheapest feasible guarantee per project on a society-saving basis.

12.3 Per-project verdicts

Running the five modules against the four projects produces the following table (illustrative — exact values depend on financing-market assumptions and sectoral cost stacks):

The programme-level FCCL draw fits inside the country's published envelope with material headroom on both lines. Two of the four projects take a multilateral guarantee; two do not. The MoF's verdict on each project is reproducible from the saved inputs and reviewable by the IMF Article IV mission, the supreme audit institution, the MDB credit teams, and the parliamentary fiscal committee.

12.4 The contracting authorities are not displaced

Each contracting authority continues to own its project's technical design, procurement, contract structure, performance regime, and operational management. The MoF's verdicts cover bankability, risk-allocation fit, VfM, guarantee selection, and envelope compliance. The two halves of the chain communicate through the published methodology and the per-project audit-chain identifier.

13. Lessons learned — what makes orchestration stick

Why this matters: the seven LEAD cases reach the cell through six recurring institutional design choices. Three documented regression episodes — one persistent country-level (Canada federal), one recovered country-level (Pakistan), one persistent doctrine-level (WB v3) — show the orchestration can be lost and, in one case, rebuilt. A MoF planning to move toward the orchestrator role should know what the prerequisites are, what the regression risks look like, and which failure modes are reachable on the analytical side and which need institutional rebuilding.

13.1 Prerequisites for sustainable orchestration

Across the seven LEAD cases (Uruguay, Portugal, South Africa, Egypt, Korea, India, New Zealand) and the strongest CO-LEAD cases (Chile, Jordan pre-2020, Philippines, Mexico-pre-2019, the Australian and Canadian sub-national leaders), six prerequisites recur.

1. Statutory or convention-level anchor. Four of the seven LEAD cases place the PPP unit physically inside the MoF by law (Uruguay, Portugal, South Africa via the PFMA, Egypt). Korea reaches LEAD through statutory delegation rather than physical placement: PIMAC sits at KDI but is bound by the PPP Act to operate under MOEF supervisory authority. India is the surprise — LEAD is reached through the executive instruments DEA controls (PPPAC, VGF, Model Concession Agreements) without a single PPP statute. New Zealand's anchor is the most recent (Treasury Circular 2024/18 + the December 2024 Blueprint). The lesson: institutional placement that is hard to undo without an explicit political act is necessary; the form (law, regulation, decree, statutory delegation, accumulated convention) matters less than the durability. Korea's three-decade statutory delegation and India's two-decade DEA convention are more orchestrator-anchoring than New Zealand's barely-18-month-old regulatory placement (the youngest LEAD anchor in the set).

2. Technical unit physically inside the MoF. Uruguay's Unidad de Proyectos PPP, Portugal's UTAP/ETF, South Africa's GTAC, Egypt's PPPCU, India's DEA Private Investment Unit, New Zealand's Treasury PPP team — the analytical unit reports up the MoF chain. (Korea is the partial exception treated in #1: PIMAC sits outside MOEF physically but inside it functionally via statutory delegation.) The lesson: a PPP unit located outside the MoF without that statutory delegation (Bangladesh's PPP Authority at PMO, Pakistan's P3A at the Ministry of Planning, Philippines' PPP Center at NEDA, Brazil's PPI Secretariat at Casa Civil) caps the country at CO-LEAD even when the design otherwise looks strong.

3. Published methodology. Every LEAD case publishes the methodology it uses: Portugal's Boletim, South Africa's PPP Manual, Korea's PPP Basic Plan, India's Model Concession Agreements and Kelkar Committee Review, NZ's PPP Blueprint, Uruguay's Compromisos PPP report. The methodology is what allows the MoF's verdict to be defended to a contracting authority's advisor without re-running the project-level analysis adversarially.

4. Audit-defensible outputs. Every LEAD case produces outputs that the supreme audit institution and the IMF Article IV mission accept as primary evidence: signed reports, traceable inputs, time-stamped methodology versions. This is also the gap at which Honduras failed — the FCU produced no defensible outputs and its veto was inoperable.

5. Affirmative — not merely negative — decision power. Ghana's Seal of Quality (the Minister of Finance must personally issue an affirmative seal before procurement) and South Africa's four Treasury Approvals are affirmative gates, not passive vetoes. Affirmative gates raise the cost of bypass and give the MoF a constructive role. The lesson: a MoF role designed around refusing on bad grounds tends to atrophy; a MoF role designed around affirming on good grounds tends to consolidate.

6. An external interlocutor that quotes the MoF. Chile, Uruguay, Portugal, South Africa, Egypt, Korea, and India are all cited by the IMF Article IV missions and / or by the major rating agencies as the institutional anchor for PPP fiscal management in their countries. Egypt's PPPCU is the IMF's direct counterpart on the Fiscal Risk Statement under the country's EFF arrangement — the most operationally tight external citation in the LEAD set. The external citation reinforces the internal authority: a MoF that publishes and is cited by the IMF is harder to override politically than a MoF that is silent. The lesson: external acknowledgement is itself an institutional asset.

13.2 Failure modes — three documented regression episodes, with different durabilities

Canada 2018 — dissolution of the federal PPP unit (persistent regression). PPP Canada was dissolved in 2018; the federal PPP function contracted to an instrument-level veto via the Canada Infrastructure Bank, with the Department of Finance providing budget-level concurrence rather than analytical leadership. Pre-2018 Canada would have scored CO-LEAD; post-2018 it scores GATE at the federal level. The provincial agencies (Infrastructure Ontario, Infrastructure BC) score CO-LEAD because Treasury Board / Management Board concurrence is required at every gateway, even though Infrastructure Ontario itself sits at the Ministry of Infrastructure rather than at Finance. They continue to deliver large programmes. The federal regression is real and persists at the time of writing — the orchestration moved down a tier, not away¹⁷^,^¹⁸.

Pakistan 2021 → 2024 — statutory removal of P3A from the Finance Division, followed by partial recovery. The Public-Private Partnership Authority Act 2017 placed P3A under the Finance Division. The 2021 amendment moved it to the Ministry of Planning. The Finance Division walked the change back in 2024 by issuing FCCL Guidelines and establishing a Risk Management Unit with a binding FCCL review function — restoring much of the MoF analytical role even though the P3A itself remains housed outside the Finance Division. Pakistan is therefore not a persistent regression: the country dropped to GATE between 2021 and 2024 and is back at CO-LEAD via the RMU restoration. The episode is informative anyway: it shows that the analytical function can survive an institutional displacement of the PPP unit if the MoF has the political space and tooling to rebuild the function on its own. The institutional placement matters; the analytical function matters more¹⁹.

World Bank doctrine 2007 → 2017 — the slow internal regression (still in force). Irwin's 2007 Government Guarantees prescribed CO-LEAD; the WB Reference Guide v3 (2017) prescribed GATE. The doctrine itself walked back, in a decade, an analytical responsibility it had previously assigned to the MoF. The PFRAM 2.0 user guide (2019) partially restored the framing on the contingent-liability side, but the headline reference document remains GATE. The lesson: doctrine is a slow lever; it can regress as well as progress.

The honest characterisation across the three episodes is: one persistent country-level regression (Canada federal), one recovered country-level regression (Pakistan), one persistent doctrine-level regression (WB v3). The recovery in Pakistan is itself a finding — it suggests that the orchestrator function rebuilds more easily than the orchestrator institution, when the political space exists.

13.3 What the analytical tooling adds — and what it does not

The argument of this paper is that analytical tooling collapses the information asymmetry that justified the gatekeeper frame. It does not argue that tooling is sufficient. The seven LEAD cases reached LEAD largely without modern tooling — they built durable institutional designs first and added tooling later. The contribution of modern tooling is to make orchestration cheaper and defensible for countries whose institutional design is not yet at LEAD: a CO-LEAD country can produce LEAD-quality analytical artefacts (a reproducible bankability snapshot, a defensible PSC, an FCCL aggregation that the IMF Article IV will quote) before completing the political work to formalise the institutional change. Higher-quality artefacts do not automatically translate into LEAD-quality programme outcomes — outcomes also depend on the surrounding political authority and contracting-authority compliance — but they are a necessary precondition. The artefact upgrade is what most CO-LEAD MoFs will get from the tooling; the outcome upgrade requires what comes next.

Tooling cannot manufacture political capital. A MoF that lacks the political space to override a sectoral ministry's advisor will not gain that space because it now has a better model. What tooling does is change the cost of using the political capital the MoF already has. A MoF that had to choose between spending political capital on every project or accepting every project can now spend it on the projects where the model produces a clearly defensible refusal — and accept the rest under the same defensible framework.

13.3.1 Five failure modes where tooling does not help

The paper has named several places where the orchestrator partition is hard to enforce. Pulled together, they form a catalogue of five failure modes that tooling cannot address on its own.

(a) Captured MoF. The MoF is politically captured — by the ruling party, by a sector incumbent, by a debt-service constituency — and uses the analytical chain to justify pre-decided verdicts rather than to test them. No tool prevents this. The audit-chain output remains the form of analytical rigour without the substance of independent verdict. The mitigation is institutional: the same external interlocutors (IMF Article IV missions, supreme audit institutions, rating-agency sovereign teams) that quote LEAD MoFs as institutional anchors also expose captured MoFs as performative — eventually, but with lag. The tool helps the interlocutor; it does not protect the MoF from capture in the first place.

(b) Under-staffed MoF. The orchestrator function for a mid-sized country requires somewhere between four and twelve dedicated analytical staff (financial-model analysts, a sector specialist or two, a standing legal/contracts capacity, an audit-systems analyst, a multilateral-relationships officer). For an MoF that does not have this capacity, the tooling provides leverage but not staff. Honduras's FCU was a single-digit-headcount unit asked to oversee a programme exceeding 5% of GDP in contingent liabilities; the legal mandate was real, the staff was not. Tooling helps a four-person team do the work of a six-person team; it does not help a one-person team do the work of a six-person team.

(c) Wrong-tier orchestration in federated systems. Australia and Canada (§9.3) show that in federated systems the orchestrator function in practice lives at the sub-national level (state Treasuries in Australia; provincial Infrastructure agencies in Canada). The federal MoF retains the consolidated fiscal-risk visibility (it carries the consolidated balance-sheet exposure) but does not exercise the analytical function. Tooling at the federal level produces the right aggregations on the wrong analytical inputs. Tooling at the sub-national level produces the right analytical inputs but cannot consolidate across tiers. The federation problem requires institutional design at the federal-sub-national interface that no tool can supply.

(d) Doctrine outpacing capacity. A country adopts a strong de jure design — Morocco's PPP framework, the UAE federal framework — without the de facto analytical depth to operate it. Tooling can populate the de facto layer faster than capacity-building can, but only after the country decides to use the tool. The IMF Morocco Article IV 2024 explicitly named the de jure / de facto gap and recommended the analytical layer be populated; the recommendation has not been operationalised at the time of writing. Tooling adoption is itself a political act that the doctrine-leading countries have not yet made.

(e) Sector-specific complexity the canonical chain does not absorb. The five MoF-led decisions are sector-agnostic by design. Some sectors carry analytical features the canonical chain has to extend to handle well — energy storage portfolios with dispatchable / non-dispatchable economics, social-infrastructure availability-based concessions with KPI-linked deductions, sub-national tariff politics in water concessions, traffic models with induced-demand feedback. The orchestrator MoF that runs the canonical chain on every project also has to recognise where the canonical chain under-models the project; the sector ministry's advisor can usually see this faster. The mitigation is a methodology-revision discipline at the MoF (sector extensions are added to the canonical chain when needed, not bolted on) and a published methodology that names the extensions. Tooling supports the discipline; it does not substitute for the sector-extension judgement.

The honest reading of these five modes is that the orchestrator partition is reachable by countries that are at the upper half of the state-capacity distribution and is harder for countries below it. The platform makes the upper-half countries faster to reach LEAD; it does not lift lower-capacity countries above the threshold at which the orchestrator role is operable.

13.4 Sequencing — what an MoF should build first, second, third

For an MoF moving toward the orchestrator role, the sequencing across the five MoF-led modules is largely empirical. The seven LEAD cases suggest an order:

1. FCCL and fiscal envelope first. This is the role no doctrine disputes. Building the FCCL aggregator, publishing the envelope, and running the per-project draw establishes the MoF's standing as the analytical home of PPP fiscal information without claiming any role the gatekeeper doctrine refuses.
2. Bankability second. The bankability model is the foundation for all the other modules. Once it is in place and run on every project, the MoF holds the de facto analytical chain even before the partition is formalised.
3. Value for Money third. With bankability in place, the PSC computation is incremental and produces an output (the PPP-vs-PSC differential) that is hard for any contracting authority's advisor to dispute without showing its workings.
4. Risk allocation fourth. The standing policy is the most consequential of the five but also the most political. It is best built once the methodology has been demonstrated on the prior three modules.
5. Third-party guarantees fifth. The guarantee decision is technically the most contained and politically the lightest — it is straightforwardly an MoF computation against an external menu. It is therefore a low-risk capstone for the analytical chain, and the place where the MoF can demonstrate the orchestrator role to the multilateral guarantors with whom it deals at the programme level.

This sequence is empirical, not theoretical. The five-stage chain in §3 to §7 is logically simultaneous; the ordering above is what the LEAD cases show as the path of least resistance.

13.5 The doctrine should catch up

The strongest implication of the evidence in §8 is that the canonical doctrine has not kept pace with the LEAD cases. The doctrine assigns the MoF a role that the LEAD countries have outgrown, and the empirical signal — read with the §8.3.1 caveats on India-dominance and small N — is directionally consistent with the LEAD countries outperforming. The doctrine's reform is not the lever that will move the next country to LEAD — institutional and analytical work inside the MoF is — but the doctrine's reform would lower the cost of every subsequent move. A revised World Bank Reference Guide v4 that retired the word gatekeeper, an updated APMG Certification Program, an EPEC paper that treated the PPP unit as the analytical arm of a programme the ministry owns, an OECD Principle 3 that named the MoF as the analytical owner of the PPP programme — any of these would shift the equilibrium for the next decade of programmes.

13.6 The case for orchestration when the empirical evidence is incomplete

The empirical case in §8 has known limits. The PPI sample is small at the LEAD cell (N=4); the LEAD signal is India-dominated (§8.3.1); the BID / PI-PPP signal is detected through an instrument not designed to detect it (§8.5); the orchestrator vs gatekeeper boundary is sharper than the LEAD vs CO-LEAD subdivision; and the regression-with-controls that would test the hypothesis at a higher inferential bar is not in this version of the paper (§14.2). A skeptical reviewer who discounts the empirical case heavily — or who excludes India from the test on the grounds that the country's LEAD classification is contested — is left with a small-N correlational signal that is consistent with the orchestrator hypothesis but does not establish it.

The case for the orchestrator partition does not depend on the empirical signal. It rests on five qualitative arguments that follow from the partition rule (§2.4) and the institutional structure of a PPP programme. A reader who weights the empirical evidence at zero should still find the qualitative case sufficient.

1. Information-cost first principles. The partition rule assigns each analytical step to the actor with the lower marginal information cost on the variables that step requires. The two-dimensional asymmetry in §2.2 is the empirical claim about where the marginal costs sit. Once granted, the partition rule is just price theory applied to the bureaucracy: assigning a step to the higher-cost actor wastes resources. The MoF's comparative advantage on cost of capital, country spread, FCCL envelope, multilateral-guarantee menu, and contingent-liability quantification is structural — these are country variables held by country-level institutions. The contracting authority's comparative advantage on engineering, demand modelling, sector regulation, and operations-and-maintenance design is equally structural — these are project variables held by sector-level institutions. The partition follows from the structure, not from a correlation across thirty-one countries.

2. The portfolio view is only available at the MoF. Risk allocation, fiscal envelope, and guarantee-decision economics are portfolio problems, not project problems. A country that allocates demand risk to the SPV on every road project and to the public on every airport project sees a different aggregate FCCL profile than a country that allocates project-by-project. A country that pools its multilateral-guarantee relationships at the programme level negotiates different terms than a country that does it one project at a time. A country that publishes one VfM methodology and one discount rate produces comparable verdicts across sectors that a sector-by-sector approach cannot produce. The portfolio view is institutionally only available to whichever actor sits above the sector level — and that actor is the MoF, by definition. Even if no measurable outcome difference were detectable, the analytical view itself only exists at the MoF.

3. The audit institution can only effectively audit one methodology. A supreme audit institution evaluating a PPP programme can audit the methodology the country uses (was it applied consistently? were the inputs traceable? were the verdicts defended?) more cheaply than it can audit the project-by-project judgement of twenty sector-ministry advisors. Methodology consolidation is therefore cheaper in audit terms — it reduces the supreme audit institution's marginal cost per project and increases the rate at which audit findings can be acted on. The only actor with cross-sector standing to consolidate the methodology is the MoF. This argument holds regardless of whether the methodology-consolidated programme produces lower cancellation rates; it holds because audit cost is a real social cost and methodology fragmentation is a real source of it.

4. Stress-moment defensibility. Fiscal envelopes are most tested at the moment they bite — a macroeconomic shock, a contingent-liability call, a political-economy disruption that exposes a portfolio of underpriced commitments. At that moment, a MoF that has the analytical chain ready (PFRAM running every quarter on the portfolio; FCCL aggregations current; risk-allocation policy published and defended) can refuse projects, restructure commitments, or negotiate amendments from a position of analytical credibility. A gatekeeper MoF without the chain folds — Honduras is the documented example, and Spain and Türkiye are the documented examples of the same dynamic at scale. The stress moment is not in the PPI cancellation-rate dataset because the dataset does not separate "cancelled at financial close" from "cancelled in operations after a fiscal stress." But the qualitative pattern is clear: institutions test in stress, not in calm.

5. The MDB-relationship leverage argument. A country that negotiates one MIGA wrap at a time deals with MIGA as a transactional counterpart. A country that maintains a programme-level relationship with MIGA, IBRD, IDA, the regional MDBs, and the home-country ECAs deals with them as institutional counterparts. The leverage is qualitatively different: a programme-level relationship buys cheaper fee rates, faster turnaround, and (in the orchestrator-MoF version) a master arrangement that prices the country's pipeline rather than each project. Only the MoF can hold a programme-level relationship with a multilateral guarantor; a sector ministry cannot. This argument depends on the multilateral-guarantor side being willing to engage at the programme level — which the LEAD-case evidence shows they will (MIGA's engagement with Egypt's PPPCU on the Fiscal Risk Statement is the most operationally tight example in our sample), and which the institutional architecture of the major multilaterals (the IBRD Country Partnership Framework, the MIGA Country Strategy, the IDB Country Strategy) is built to support. Whether CO-LEAD MoFs that have not yet asked for a programme-level arrangement would obtain one is a question the evidence cannot settle in advance, but the multilateral side's revealed preference is for programme-level engagement where the country side is capable of it.

13.7 What the LEAD cases embody, beyond the statistical signal

The pattern across the seven LEAD countries is not that they implemented a doctrinal recommendation; the pattern is that they each, through different national paths (Uruguayan statute, Korean statutory delegation, Indian executive convention, New Zealand 2024 regulatory placement, Portuguese 2012 placement, South African 1999 PFMA, Egyptian 2010 ministerial decree), arrived at the same partition. Convergent institutional evolution across regions and legal traditions to the same MoF placement is itself evidence, even when the outcome-side measurement instruments do not show the signal cleanly. The convergence cannot easily be explained by income level (Uruguay and India are not high-income), legal tradition (the seven span common law, civil law, Roman-Dutch hybrid, Islamic-law-influenced), or IMF-programme presence (Uruguay had no programme; Egypt did). The convergence is consistent with the qualitative argument above: the partition rule is right, and capable countries arrive at it from different starting points.

The case for the orchestrator partition, finally, is that it is what capable countries do. The empirical literature has not yet built the measurement instrument to detect it cleanly. The qualitative argument is sufficient until the instrument is built.

14. Limitations and roadmap

Why this matters: the orchestrator hypothesis is testable and partially supported. It is not proven, and its operational implementation has limits that deserve to be stated openly.

14.1 What the paper does not establish

The empirical evidence in §8 is correlational, not causal. The country scoring rests on observable institutional design and observable published outputs; it does not claim to measure the unobservable variable MoF analytical leadership in practice. The PPI cancellation-rate result is statistically marginal at p = 0.065 between LEAD and CO-LEAD, outlier-driven between LEAD and GATE, and India-dominated within the LEAD cell (§8.3.1). The BID/PI-PPP results are de jure design scores, not implementation scores. We cannot rule out that the LEAD-score countries are higher-performing on the relevant dimensions for reasons unrelated to the MoF's role — income level, legal tradition, World Bank technical-assistance history, sector composition, IMF programme presence — and §14.2 names the regression-with-controls that would test this directly but that we have not run on this sample.

The framework's operability assumes a unified MoF voice. Federated systems break that assumption — Australia and Canada show that the orchestrator function in practice lives at the sub-national level, where the consolidated fiscal-risk visibility may be lost. A federal-level orchestrator framework is a separate question from the one this paper addresses.

The framework also assumes a MoF with the political space to use the analytical role. A captured MoF, an under-resourced MoF, or an MoF without standing in cabinet cannot be made into an orchestrator by tooling alone. The framework is necessary but not sufficient.

14.2 What a stronger version would add — and what this version deliberately stops short of

The most consequential gap in the empirical work, and the one a peer-reviewed version of the paper would have to close, is a regression of project-level outcomes on score group with the standard confounder controls. The regression that the data supports is:

cancellation_rate(country) ~ score_group + log(GDP per capita) + legal_origin + IMF_programme_years + sector_concentration + WGI_control_of_corruption + ε

with country as the unit of observation, weighted by project count. The controls absorb the most-likely-confounding variation in state capacity, legal tradition, IMF discipline, sector mix, and corruption baseline. The expected reading of the result is one of three:

1. The score-group coefficient survives controls and remains directionally as the unweighted means suggest. The orchestrator hypothesis is supported.
2. The coefficient becomes statistically insignificant once controls are added. The unweighted-mean finding is then an artefact of state-capacity covariation. The qualitative case in §13.6 still stands; the empirical case becomes correlational-only.
3. The coefficient flips sign on controls. The unweighted-mean finding is then misleading and the paper's empirical claim would need to be retracted in favour of the qualitative case.

We do not run this regression in this version of the paper. The reason is honest rather than methodological: the country sample of in-scope LMICs in the PPI dataset is N=31, with N=4 in the LEAD cell (Egypt, India, South Africa, Uruguay) and India dominating the LEAD project count. A regression with five controls and N=31 is severely underpowered, and the inferential gain over the unweighted-mean comparison would be modest. The honest move is to wait for the IADB renegotiation dataset (which would add a second outcome variable and roughly double the effective sample) and the BID 2026 refresh (which would add a longitudinal dimension). A v2 of the paper, prepared after those releases, will run the regression and report the result against this version's prediction.

The other items a stronger version would add:

• Replace the binary score (LEAD / CO-LEAD / GATE / NONE) with a continuous index built from a checklist of observable practices, scored by multiple raters, with inter-rater reliability reported.
• Add the IADB renegotiation dataset for the LAC subset, allowing the renegotiation hypothesis to be tested against the score group.
• Add a longitudinal dimension by re-scoring the same countries at three points — say 2010, 2017, 2024 — to test whether transitions in score (e.g. Pakistan 2017 → 2021 → 2024, Canada pre/post-2018) correspond to transitions in outcomes.
• Run a purpose-built measurement instrument: an explicit "MoF analytical leadership" index, with sub-dimensions for the five MoF-led decisions in §3 to §7. This would address the §8.5 finding that the existing instruments were not designed to detect the signal.

14.3 The roadmap

This is the third paper in the Austral Intelligence Working Papers in Infrastructure Fiscal Intelligence series. The first paper sized the Viability Gap Fund as a constrained optimisation³⁶; the second framed the third-party guarantee decision³⁷; the present paper re-partitions the analytical responsibility between the contracting authority and the MoF. The next paper in the series will address the closely related question of how the MoF orchestrator interacts with the multilateral guarantor at the programme level — a master arrangement that prices a country's pipeline rather than each project, with audit-evidence carried forward.

The platform's roadmap incorporates the empirical instruments described in §8: the PPI replication notebook, the BID/PI-PPP cross-tab, and the literature scoring rubric are exposed as standalone tools so any reader can re-score a new doctrine document, re-classify a new country, or replicate the cross-tabs against an updated PPI release. Replication notebooks are at references/03_empirical/.

Appendix A — Scoring rubric and per-document scores

The two rubrics below — one for canonical doctrine documents, one for country institutional design — use the same vocabulary so the evidence in the paper lines up cleanly. Both are applied on observable evidence (what the document prescribes; what the country's framework and published outputs show) rather than on what the law intends. The rubrics and the per-document / per-country briefs are mirrored in the paper's reference library at references/.

A.1 Canonical doctrine — how the document frames the MoF role

A.2 Country institutional design — what the legal framework + actual practice show

A.3 Scoring rules

• Score on observable evidence, not on what the law intends. A statute that says "MoF approves" but is in practice routed through a sector-ministry advisor scores GATE, not LEAD.
• The four cells are mutually exclusive. Pick one. Use the brief's narrative to nuance borderline calls.
• Down-score on weak observation: if you cannot find a published methodology or a public output bearing the MoF imprint, the country is at most CO-LEAD, not LEAD.
• Document the rationale in the brief. The score is checkable; the narrative explains why.

A reader who disagrees with a specific score can re-classify the document or country against the rubric above and re-run the per-cell statistics. The paper's per-document briefs are in the library at references/01_canonical_frameworks/; the per-country briefs are at references/02_country_legal/.

A.4 The twelve canonical documents — per-document scores

Appendix B — Empirical replication notes

PI-PPP 2018 + BID 2023. The replication notebook is at 03_empirical/infra_benchmark/infra_benchmark_pull.py. Inputs are pulled from the World Bank Indicators API (Source 61, indicators IQ.PPN.REGQ.S0..S3) and from the Data360 portal (database WB_BID, indicators WB_BID_PPI_PRS / PPS / CMS / USP). The slug-to-ISO3 mapping is hard-coded for the 50 scored countries. Mann-Whitney U tests via scipy.stats.mannwhitneyu. Outputs: derived/by_country.csv, derived/by_score_group.csv, derived/tests.csv. Findings document at findings.md.

PPI Database 1990–2024. The replication notebook is at 03_empirical/ppi/ppi_pull.py. Input is the World Bank PPI Full DTA file (raw/2024-PPI-Full-DTA.dta), 11,640 projects across 133 LMIC countries. Cancellation-rate metric defined as (Cancelled + Distressed) / Total. MDB-involvement rate from the multilateral-support flag. Outputs: derived/by_country.csv, derived/by_score_group.csv, derived/tests.csv. Three figures (cancellation by score, box plot, MDB rate) at figures/. Findings document at findings.md.

Appendix C — Glossary

• ADB — Asian Development Bank.
• APMG — Association for Project Management Group, the body that administers the PPP Certification Program (CP3P).
• BID — Benchmarking Infrastructure Development, the 2023 successor to PI-PPP 2018; published on Data360.
• CA — Contracting Authority. The line ministry or sectoral agency that procures and signs the PPP contract.
• CO-LEAD — In this paper's rubric, the cell where the MoF shares analytical leadership with another body (PPP unit, planning ministry, sectoral department).
• CRC — OECD Country Risk Classification, an eight-class scale (0 lowest risk to 7 highest) used as the index for export-credit fee schedules and counterfactual-term tables.
• DGCPTN — Dirección General de Crédito Público y del Tesoro Nacional, the public-credit and Treasury arm of Colombia's Ministerio de Hacienda y Crédito Público.
• DIPRES — Dirección de Presupuestos, Chile's budget directorate within Hacienda.
• DSCR — Debt-Service Coverage Ratio. Cash from operations divided by debt service in a given year.
• EBRD — European Bank for Reconstruction and Development.
• ECA — Export-Credit Agency. Government agency in OECD countries that provides insurance and guarantees on export-related project finance.
• EFF — Extended Fund Facility. IMF lending instrument for medium-term balance-of-payments support, conditional on a structural-reform programme.
• EPEC — European PPP Expertise Centre, hosted at the European Investment Bank.
• FCCL — Fiscal Commitments and Contingent Liabilities. The standard framework for analysing PPP fiscal risk.
• FCU — Fiscal Capacity Unit. Honduras's MoF unit established by Decreto 35-2014 to oversee FCCL on PPPs; documented by the IADB as non-functional.
• GATE — In this paper's rubric, the cell where the MoF holds a fiscal-affordability or contingent-liability veto but does not run the analysis.
• GFS 2014 — Government Finance Statistics Manual 2014 (IMF). The international standard for the classification and presentation of public-sector fiscal statistics.
• GIH — Global Infrastructure Hub, the G20 infrastructure platform.
• GTAC — Government Technical Advisory Centre, the unit at South Africa's National Treasury that includes the PPP Advisory Unit.
• IADB — Inter-American Development Bank.
• IBRD — International Bank for Reconstruction and Development. The original lending arm of the World Bank Group.
• IDA — International Development Association. The concessional-lending arm of the World Bank Group, for low-income countries.
• IFC — International Finance Corporation. The private-sector arm of the World Bank Group.
• IMF FRS — International Monetary Fund Fiscal Risk Statement. Country annex disclosing fiscal risks (including contingent liabilities, public-sector exposures, and natural-disaster shocks) under IMF transparency standards.
• IMG — Ingreso Mínimo Garantizado. Chile's minimum-revenue guarantee on toll-road concessions.
• IPA — Infrastructure and Projects Authority (United Kingdom). The centre-of-government infrastructure delivery body jointly sponsored by HM Treasury and the Cabinet Office.
• IPP — Independent Power Producer. A private power-generation project that sells electricity to a utility under a long-term Power Purchase Agreement.
• IPSAS-32 — International Public Sector Accounting Standard 32, governing the accounting treatment of service-concession arrangements.
• KDI — Korea Development Institute, the Korean national policy-research institute that hosts PIMAC under MOEF supervisory authority.
• LEAD — In this paper's rubric, the cell where the MoF (or a unit reporting to it) runs the analytical chain end-to-end.
• MAR — Mecanismo de Ajuste de los Ingresos. Income-adjustment mechanism on certain LAC concession contracts; analogous to a minimum-revenue guarantee with a true-up.
• MEF — Ministerio de Economía y Finanzas (the Spanish-language name for several Latin American MoFs).
• MIGA — Multilateral Investment Guarantee Agency, the political-risk-insurance arm of the World Bank Group.
• MoF — Ministry of Finance.
• MOEF — Ministry of Economy and Finance, Republic of Korea.
• MOP — Ministerio de Obras Públicas. Chile's Ministry of Public Works, contracting authority for concession contracts under DFL 164/1991.
• MRG — Minimum Revenue Guarantee. Common contractual mechanism by which a contracting authority compensates the SPV when project revenue falls below a defined floor.
• NEDA — National Economic and Development Authority, the Philippines' central planning agency, which houses the Philippines PPP Center.
• NHSFO — Non-Honoring of Sovereign Financial Obligations. MIGA cover for sovereign payment risk on debt-service obligations.
• NIFFCo — National Infrastructure Funding and Financing Company, the New Zealand Crown company under Ministry of Finance mandate from December 2024.
• NONE — In this paper's rubric, the cell where the MoF has no specific PPP role beyond ordinary budget approval.
• P3A — Public-Private Partnership Authority, Pakistan. Established under the Finance Division by the PPP Authority Act 2017; statutorily relocated to the Ministry of Planning, Development and Special Initiatives in 2021.
• PBG — Project-Based Guarantee. IBRD or IDA guarantee on a specific project's debt service or government-payment obligation.
• PCG — Partial Credit Guarantee. IFC, ADB, or comparable MDB guarantee on a portion of a commercial loan's principal and interest.
• PF2 — Private Finance 2. The UK successor regime to PFI introduced in 2012 and ended by HM Treasury in Budget 2018.
• PFI — Private Finance Initiative. The UK's flagship PPP regime from 1992 to 2012; replaced by PF2, which itself was terminated in 2018.
• PFRAM — PPP Fiscal Risk Assessment Model. IMF / World Bank tool for quantifying fiscal commitments under PPP contracts.
• PIMAC — Public and Private Infrastructure Investment Management Center, Korea's PPP technical arm at KDI under MOEF.
• PI-PPP 2018 — Procuring Infrastructure Public-Private Partnerships 2018, the World Bank product on procedural design.
• PMO — Prime Minister's Office. Used in this paper for the central executive secretariat that houses some non-Finance PPP units (e.g. Bangladesh).
• PPA — Power Purchase Agreement. Long-term contract between a power producer and an offtaker (utility or government) defining price, volume, and terms of electricity sales.
• PPI — Private Participation in Infrastructure, the World Bank's project-level PPP database.
• PPP — Public-Private Partnership.
• PRI — Political Risk Insurance. MIGA or private-insurer cover against transfer restriction, expropriation, war and civil disturbance, and breach of contract.
• PSC — Public Sector Comparator. The benchmark cost the country would bear if it built and operated the project with public capital.
• RMU — Risk Management Unit. Pakistan's Finance Division unit established in 2024 with a binding FCCL-review function, partially restoring MoF oversight after the 2021 statutory removal of the P3A.
• SPV — Special Purpose Vehicle. The project company that holds the PPP contract and the underlying financing.
• STN — Secretaria do Tesouro Nacional, the Treasury arm of Brazil's Ministério da Fazenda.
• UTAP — Unidade Técnica de Acompanhamento de Projetos, Portugal's PPP unit inside the Ministério das Finanças (reorganised into ETF in 2025).
• VfM — Value for Money. The PPP-vs-PSC differential analysis.

Appendix D — API and pseudocode reference

The five MoF-led modules are exposed as REST endpoints under /api/v1/. The end-to-end orchestrator workflow is exposed as a single endpoint that runs the five-step chain:

POST /workspaces/{wid}/orchestrator/evaluate

Pseudocode for the orchestrator workflow:

def evaluate_orchestrator(country, project_inputs):
    # Country-level preconditions (set once, applied to every project)
    cfg = load_country_config(country)        # discount rate, DSCR floor,
                                              # risk-allocation policy, VfM
                                              # threshold, FCCL envelope,
                                              # guarantee menu

    # Stage A — Bankability
    snap = run_model(project_inputs, cfg.discount_rate, cfg.dscr_floor)
    if not snap.feasible:
        return Verdict(stage="bankability", decision="restructure", snap=snap)

    # Stage B — Risk allocation fit check
    fit = check_allocation_fit(project_inputs.allocation, cfg.allocation_policy)
    if not fit.ok:
        return Verdict(stage="risk_allocation", decision="restructure", fit=fit)

    # Stage C — Value for Money
    vfm = compute_vfm(project_inputs, cfg.discount_rate, cfg.vfm_threshold)
    if vfm.differential < cfg.vfm_threshold:
        return Verdict(stage="vfm", decision="reject", vfm=vfm)

    # Stage D — Third-party guarantee selection
    ce = evaluate_guarantee_menu(project_inputs, cfg.guarantee_menu)
    chosen = ce.cheapest_feasible_or_none()

    # Stage E — FCCL draw and envelope check
    draw = compute_fccl_draw(project_inputs, chosen)
    envelope = load_envelope_state(country)
    if envelope.firm + draw.firm > cfg.envelope.firm_cap:
        return Verdict(stage="fccl", decision="reject", draw=draw)
    if envelope.contingent + draw.contingent_p95 > cfg.envelope.contingent_cap:
        return Verdict(stage="fccl", decision="reject", draw=draw)

    # All five stages cleared
    return Verdict(
        stage="all_cleared", decision="proceed",
        snap=snap, vfm=vfm, ce=ce, draw=draw,
        audit_chain_id=record_to_audit_chain(...),
    )

The full per-stage endpoints, their inputs, and their outputs are documented in the prior two papers in this series for stages A, C, D³⁶^,^³⁷; stages B and E are documented in the platform's API reference at https://platform.austral-intelligence.com/api.

Appendix E — Source list and library

The full reference library is at references/. The library contains:

• 12 canonical doctrine briefs at 01_canonical_frameworks/ with per-document scores.
• 50 country institutional briefs at 02_country_legal/ with per-country scores, source files, and a compiled comparator table at 02_country_legal/_scoring.csv.
• Two empirical pulls at 03_empirical/ (PPI Database 1990–2024 and Benchmarking Infrastructure Development 2023 + PI-PPP 2018) with replication scripts and findings notes.
• A consolidated Web_References.docx listing every web-only and downloaded source with URL and access date — 330 entries as of 2026-05-05, generated by build_web_sources_docx.py from the _sources.jsonl files distributed across the references tree.

E.1 Errata

28 September 2026 — misattribution of the "gatekeeper" quotation. Versions of this paper issued before this date attributed to EPEC's 2014 Establishing and Reforming PPP Units the statement that MoF-based PPP units exist to "fulfil [the MoF's] role as gatekeeper", and claimed on that basis that the EPEC report was the only document in our canonical corpus to use the word gatekeeper literally. Both claims were wrong, and in the same way.

The sentence belongs to the World Bank's PPP Reference Guide, Version 3, Module 2, §2.3.4, p. 78, where it reads: "Many countries do establish their central PPP Unit in the Ministry of Finance, to better fulfil its role of gatekeeper." The EPEC report does not contain it — the words fulfil and fulfill do not appear in that document at all — and its own two uses of gatekeeper describe the PPP unit, not the ministry: a unit required to review proposals and recommend to a separate approval committee, and units that act as gatekeepers to project funding. The error was one of attribution, not of invention: the quotation is real and the content of the claim held; the source named for it did not.

The correction was found in the course of Austral's own source verification for the institutional-design manual (M13), on a second pass that read both documents directly rather than through indexed search — which is how the earlier footnote had obtained the quotation, and which the earlier footnote disclosed as an unconfirmed page reference.

The substance of the paper is unaffected. The doctrinal scoring of Appendix A.4 does not depend on the quotation: EPEC 2014 scores GATE because it prescribes a fiscal-affordability gate, and the World Bank Reference Guide v3 scored GATE already. If anything the corrected reading strengthens §2: what the EPEC report actually documents — a PPP unit that carries the technical review without holding the authority to approve — is the very separation of analytical function from decision right that the partition rule assumes. Affected passages: the abstract, §1.1, §13.5 and the footnote epec2014; the World Bank quotation now carries its own footnote, wb_gatekeeper.