Mandates: reading what an institution may do against what it can do

In this method a mandate is an institution’s ability to take the next step. A text may authorise a step the institution cannot take, and an institution may take steps no text authorises. The method reads the authorising text, the ability and the practice against each other, and tests each reading at another level or by another agent.

Why mandates

Innovation units, labs, pilots and participation exercises live or die on whether someone can take the next step. Moore’s strategic triangle asks three questions of any public initiative: is it valued, is it authorised, and can it be done (Moore 1995). A practitioner working paper treats authority as something to be built and managed, since it is rarely concentrated in one office (Andrews et al. 2016). Two practitioner pieces state the failure from the side of the unit. Pilots judged yes or no lose their lessons, and a pilot labelled a failure is often a weak authorising environment and not a weak idea (Guillot et al. 2019). A 2026 provocation argues that visioning without execution is expensive dreaming: a scenario needs the groundwork of cost, policy change and who must be in the room (Collins 2026).

The literature has names for the gaps this method reads. Organisations adopt forms for legitimacy and keep them apart from activity (Meyer and Rowan 1977; Brunsson 1989). Governments can copy the look of working institutions without the function, which Pritchett, Woolcock and Andrews call isomorphic mimicry, and can ask a new tool to carry more than it can bear (Pritchett et al. 2013). Policy is made again at the counter by those who apply it (Lipsky 1980). Implementation is a chain of agreements, each of which can fail (Pressman and Wildavsky 1973), and layers of government do not reduce to one control relation (Hill and Hupe 2003). What the literature offers less of is a way to record a mandate that someone else can contest. That is what the gram adds.

Three things that are easily confused

Start from one next step and not from the institution in general: issue a licence, publish a rule, release data, admit a second cohort. Put Moore’s three questions to that step, then map the documents backwards from it. This keeps a study to a working definition and to three weeks.

Levels, and what each admits as a Trace

Authority nests and overlaps. A unit sits inside an authority, inside a jurisdiction, inside a polity, and some units answer to two. Hooghe and Marks distinguish governance built from nested, non-intersecting jurisdictions from governance built from overlapping, task-specific ones (Hooghe and Marks 2003). A part can also be exempt from the whole’s rules by the whole’s own instrument: the constitution of the United Arab Emirates leaves to federal regulation how far financial free zones are excluded from federal legislation (Art. 121) (UAE Constitution 2009). So the levels need a rule for carve-outs and overlap, and overlap is treated as normal and not as a defect. Constitutions also allocate competence differently, as lists of federal powers with the rest left to the parts (Arts 120 to 122 of the UAE text) or as lists of exclusive State and concurrent powers (Art. 117 of the Italian text) (UAE Constitution 2009; Italian Constitution 2012). An entry names which allocation applies and whether its claim concerns making law or carrying it out. These are a non-lawyer’s readings of constitutional text, and a legal reader should check each.

What a mandate entry records

For a claim to be contestable by someone who was not there, each entry on a mandate records:

1. The instrument: issuer, number or title, date, version, link, and its legal force (a binding act, a programme page, an announcement).
2. The passage, verbatim, as the Trace.
3. The statement, parsed: who is addressed, what is required, permitted or forbidden, the action, the conditions and the sanction (Crawford and Ostrom 1995). This is a Reading, with its hands note. Where the text can be coded as rules, as in the practice behind Rules as Code (Mohun and Roberts 2020), two readers code the same clause independently, and the divergence is logged and not resolved.
4. The levels of the issuer and of the addressee, and the competence regime of that level.
5. The upward link: the instrument it derives from, marked traced, unchecked or broke.
6. Who answers to whom: actor, forum, duty to explain and consequence (Bovens 2007).
7. The authorisers, each with a Trace of consent, and the means: money, staff, and the number of separate consents the step needs (Andrews et al. 2016; Pressman and Wildavsky 1973).
8. Dated acts and refusals (Lipsky 1980).
9. The coupling: whether the text is tied to a budget line, a named post, a deadline or an indicator. An uncoupled statement may be talk (Brunsson 1989).
10. The status of each test: who ran it, at which level, and when.

A claim that lacks the passage, the upward link or the test status cannot be contested by a stranger. Each reading also says who can say no, and by what means: complexity thinking has been criticised for lacking a theory of power (Kok et al. 2021), and a mandate method that cannot say who can refuse would repeat the gap.

The next-step test

The test at another level is simple to state. For any claim of mandate, name the next step, then follow the request until someone can say yes, or until it stops. Each hand-off is an entry. The chain of who answered, at which level, is the diagram.

Illustration, not data. A request starts at the act level, is forwarded by the unit, needs sign-off at the authority level, and stalls at the jurisdiction level where no one can say yes. A counterpart at the authority level declines. The chain is the record of the mandate.

Some legal orders already provide a procedure for one level to contest another’s competence. In the UAE the Federal Supreme Court hears a challenge by an Emirate to a federal law and by a federal authority to an Emirate’s legislation (Art. 99), and disputes between federal and Emirate provisions go to the same court (Art. 151). In Italy the Government can challenge a regional law, and a Region a State or regional law, on grounds of competence within sixty days of publication (Art. 127), and the Constitutional Court rules on conflicts of attribution (Art. 134) (UAE Constitution 2009; Italian Constitution 2012). The common features are a claimant at the other level, a named forum, a ground (competence) and sometimes a window. A Holagram test borrows the shape: a unit-level claim is tested by a reader at authority level, and a jurisdiction-level claim by a reader from the other jurisdiction. The result stays unchecked, and is labelled unchecked by the competent forum, until that forum has answered.

The text and the ground

The text is the authorising paper read as invisible literature: decrees, statutes, charters, terms of reference, delegation letters, memoranda, sandbox conditions, performance sheets. The ground has three columns. Text: what is authorised. Means: money, staff, and the consents the step needs. Act: dated instances of the unit acting or being blocked, including refusals. A Reading about ability that rests on one column is labelled text-only or act-only.

Fiction

Alter writes the institution as its text reads literally: a year in the life of the unit if it exercised every power the instrument names. A second fiction writes the unit as it is funded and staffed. The gap between the two is a list of tests. Neither is evidence. In politically sensitive settings a fiction about a named authority can be read as an allegation however it is labelled, so keep it off real decisions or label it so plainly that it cannot be quoted as a finding.

Agents, including the ones we delegate to

The unit, its counterparts, those governed, regulators, and models that hold delegated tasks. When a model drafts a clause, screens an application or recommends a decision, someone holds the mandate for it. Existing formats record part of this. The UK Algorithmic Transparency Recording Standard asks for an owner by role title, the human involvement and a review route (GDS 2021). French law requires a notice when a decision rests on an algorithm and lets the person ask for the degree and mode of its contribution (French and Italian law). The EU AI Act asks that the people assigned to oversight have competence, training and authority (EU AI Act 2024). None records what the reviewer changed, the time available or the authority to override, which is where the hands card adds something.

Announcements that create roles are traces at one level and need a document at another. In 2024 Dubai appointed Chief AI Officers in 22 government entities (Government of Dubai 2024). The delegation or terms of reference that give such a role authority to act is the text to read against the announcement. In April 2025 the UAE Cabinet approved a Regulatory Intelligence Office to lead an AI-based legislative system, which the announcement describes as combining human insight with AI efficiency (UAE Cabinet 2025–2026). A mandate entry asks the next question: at which stage, by whom, and with what authority to stop.

A worked entry

Invented, to show the shape.

Where it goes wrong

• Authority taken for ability. The decree says so. The next-step test asks whether anyone can say yes.
• The organisation chart taken for the mandate. Charts show lines of reporting, not who can take the step.
• A “broke” read as bad faith. A structure kept apart from activity may be doing a different job (Meyer and Rowan 1977; Brunsson 1989). Say what job before judging.
• A pilot read as a mandate. A pilot is a whole at one level. That it worked at the unit level says nothing about who can take it up at the authority level (Guillot et al. 2019).
• Overlap read as conflict. Overlapping authority can be designed redundancy. Test before calling it a clash.
• Mimicry and premature load. A tidy map can look like analysis, and an entry can be used to settle an authority dispute before it has been tested (Pritchett et al. 2013).
• Capture. An entry written with the unit it describes.
• Misreading statute. A non-lawyer’s misreading, which a synthetic reader may repeat. Use a legal outside reader.
• Silence read as concealment. Test the paranoid reading against incapacity, delay and conflicting demands.
• Readings about people. Readings are about offices, instruments and units, not the motives of named individuals.

Objections the literature raises

• “We already do this.” The strategic triangle and the authorising-environment exercises are in use (Moore 1995; Andrews et al. 2016), and so are portfolio sensemaking, peer review between units and failure reports (OECD 2022; UNDP Strategic Innovation Unit; Sabel and Zeitlin 2008; Hartley and Knell 2021; UN System Staff College 2019). Holagram has to show a case in which a test changed a decision. Until then its added value is a claim. What it offers is a Trace for each claimed authoriser, a separation of text, means and act, a test by a reader at another level, and a log.
• “A portfolio is a hedge, not a set of claims.” The OECD portfolio approach expects some items to fail and manages the spread (OECD 2022), and UNDP’s portfolio page calls the approach a start in changing ways of thinking and not a management tool (UNDP Strategic Innovation Unit). Saying that no level sums up another does not forbid managing a portfolio. Adding is legitimate for managing. A claim about the portfolio is its own claim, with its own test by someone outside the portfolio team.
• “The map is the mimicry.” A clean diagram of mandates can become what a unit shows in order to look diligent (Pritchett et al. 2013; Meyer and Rowan 1977), and a map that makes a unit legible to outsiders can be used to control it (Scott 1998).
• “It is not a manager’s task.” Rhodes and Wanna argue that where ministers are strong the authorising environment belongs to elected office, and that tools helping officials manage it risk displacing it (Rhodes and Wanna 2007). So entries describe and do not advise on extending a mandate, and the body that authorises a unit reads that unit’s entry first.
• “Ambiguity is for courts.” Competence disputes are settled by a competent forum (UAE Constitution 2009; Italian Constitution 2012). The labels held, broke and unchecked must not read as legal findings, which is why the result says unchecked by the competent forum.
• “The act is the mandate.” Practice at the street makes policy (Lipsky 1980), so a text-first map can misdirect effort, and act-level evidence is anecdotal. That is why the ground has three columns.
• “Gaps are not defects.” Divergence between talk, decision and action can be how an organisation copes with conflicting demands (Brunsson 1989). A map that ranks gaps as failures will be read as an audit, and units will not cooperate.
• “The tests are too weak.” Accountability, as Bovens defines it, requires consequences (Bovens 2007). Holagram’s tests have none: they are commentary, and should say so.
• “A public failure log is unsafe.” One lab’s failure report depended on trust and on none of the failures being critical (UN System Staff College 2019). Log readings and not people, allow a delay before publication, keep an outside reader with a stop right, and treat unchecked as an honest published result.
• “Practitioners need something to do on Monday.” The method makes no recommendation. It improves what is read before action and shows which reading broke.

Uses in governance innovation

Start with what practitioners already do, and add two things: a test by someone else at another level, and a log that records the level of each broken reading.

• The unit’s own mandate as the first entry, written by someone outside it, and read first by the body that authorises the unit.
• An authorising-environment map for a lab or innovation unit: the nested mandates it sits inside, the next-step chains for its last requests, and a log of what stalled and where.
• A mandate test before an exercise: name the next step after a workshop, a foresight exercise or a pilot, and who can take it. Classify what a participation or visioning clause is linked to: who takes part, how they communicate and decide, and how the discussion connects to policy or action (Fung 2006). If no one can take the next step, say so, and call the exercise education and not mandate.
• A hands card per delegated decision, so that a model’s part in a decision is recorded with the person who holds the mandate for it.
• A portfolio read as a set of mandates in use: for each project, what could be done next and who can do it, to show where results can travel and where they stop.

First study: one instrument, one unit, three weeks

Choose one public instrument that creates or empowers a unit, such as a decree, charter, sandbox regime or terms of reference, together with the unit’s public outputs and one counterpart’s public records. Public material only; no client material. Candidates, each checked on a public page:

• A sandbox regime. The UAE government portal lists regulatory sandboxes at federal, financial-free-zone and sectoral level, and says RegLab began in January 2019 under a 2018 federal law that lets the Cabinet grant temporary licences. The portal gives no law number (UAE Government 2024). Finding the law is the first task. Trace one licence from rule to announcement (the portal lists dated licences, among them an electric cargo aircraft in March 2023) and set it beside the constitutional allocation of aircraft licensing to the federation (Art. 120) (UAE Constitution 2009). Expected first result: unchecked, pending a legal reader.
• A programme. The OECD record of Accelerate Estonia, a programme self-report, gives about 25 experiments, about a fifth of which failed, with the government changing two months in (OECD 2022 (Accelerate Estonia)). Take three experiments and look for each in ministry documents. The record gives a failure share and no log, so the individual cases are the trace.
• A delegated tool. Take one government language-model tool with officer review, such as the UK planning-consultation tools, collect its public evaluation and its transparency record, and ask what the public record shows of machine origin and of officer edits (MHCLG et al. 2025–2026; GDS 2021).

What would show it works. The counts on the Research page, and at least one next-step chain that ends somewhere the instrument did not predict.

Read next

• Mandates and the authorising environment, governance innovation and synthetic agents in government in the literature review.
• The method, synthetic hands, and the neighbouring proposals: Cities and Strategy making.