The economic theory of state enterprise locates the problem in the objectives of those who control the firm. Shleifer and Vishny (1994) model politicians who use public firms to pursue political ends, such as employment or favours to supporters, at the expense of efficiency. Kornai (1986) described the soft budget constraint: a firm that expects to be rescued has weak reasons to control costs. A state-owned utility has several principals with different aims, which blurs accountability. The guidelines of the OECD respond to these problems by asking the state to separate its role as owner from its roles as policy maker and regulator, to act as an informed and active owner, and to leave boards with real authority (OECD, 2015).

A design must start from the rules that were meant to prevent failure. In South Africa these include the Public Finance Management Act, which makes the board of a public entity its accounting authority with fiduciary duties and responsibility for preventing irregular expenditure (Republic of South Africa, 1999); the King IV code on corporate governance, which applies to state-owned entities through a sector supplement (Institute of Directors in Southern Africa, 2016); and the statutes that establish the energy regulator and give it power to license and to set tariffs. Other countries in the region have their own equivalents. The first analytical task in any case is to establish which rule assigned which control to which body.

Studies of organisational failure suggest that breakdowns are seldom sudden. Turner (1976) described an incubation period in which warning signs accumulate and are discounted because they do not fit accepted beliefs. That idea directs attention to the years before a crisis becomes visible, and to the treatment of early warnings by auditors, board committees and the regulator. On the regulatory side, Ayres and Braithwaite (1992) proposed that regulators escalate from persuasion to sanction as cooperation fails. Their enforcement pyramid offers a yardstick against which the actual sequence of regulatory responses can be compared.

The researcher fixes the period, the entity and the decisions of interest before reading the record in detail. Procurement of fuel, the appointment and removal of directors and executives, and the financing of new generating plant are natural candidates, because each is governed by identifiable rules. Propositions are then written down from theory. One might hold that failures followed changes in board composition initiated by the shareholder; another, that the soft budget constraint weakened cost control regardless of who sat on the board. Stating propositions in advance is what separates a case study from a chronicle (Eisenhardt, 1989).

Documents are the core of the evidence, and they should be collected systematically and logged with their origin, date and purpose. Table 1 lists the main types and what each can and cannot show. Document analysis has known weaknesses: documents are written for a purpose, survive selectively and may be incomplete (Bowen, 2009). An inquiry transcript records what witnesses chose to say under legal advice. An annual report is drafted by the entity under study. The log should therefore record who produced each document and why.

Every dated event relevant to the bounded decisions is entered in a single chronology with its source: appointments, resolutions, contract awards, audit findings, regulatory applications and decisions, ministerial directives. The chronology is the main analytical instrument for process data (Langley, 1999). It exposes sequence, which narrative sources often obscure, and it shows where the record is silent. Gaps are themselves recorded, since an absence of board minutes on a large decision is evidence of a kind.

With the chronology in place, the researcher asks, for each decision, which control should have operated and what happened at that point. Process tracing provides the logic. Each proposition implies observable traces: if board changes drove the failure, the chronology should show contested decisions clustering after those changes, and committee records should show dissent ending. The researcher states in advance what evidence would count for and against each proposition and how strongly (George and Bennett, 2005; Beach and Pedersen, 2013). A trace that would be expected under every explanation proves little, while one that only a single explanation predicts carries weight.

The response is treated as a second sequence running alongside the first. For each warning that reached a body with power to act, the researcher records what that body knew, what it could lawfully do, and what it did. The bodies include the energy regulator, the shareholder ministry, the treasury, the auditor and the legislature. Comparing the observed sequence with the escalation that responsive regulation would predict shows where responses stalled, and the formal framework from the background section shows whether the cause was a lack of power or a failure to use it.

Before conclusions are drawn, at least three rival accounts should be tested against the chronology: that the failure was mainly one of technical and financial circumstance, such as ageing plant or tariffs held below cost; that it was a failure of individuals; and that it was a failure of institutional design that would have occurred with different people. The report should present the chronology, the evidence for and against each account, and the points on which the record does not allow a conclusion.