Research
Holagram is a research programme before it is a result. The way of thinking is stated, the method is written down, and each of three method proposals has a first study. None has been run. This page says what exists, what would count as it working, what is open, and how the literature behind it was built.
Where the work stands
| Part | State |
|---|---|
| Hola, the way of thinking | Stated as seven principles. Each rests on sources in the literature review. Not tested as a whole. |
| Gram, the method | Written: the entry, the levels, seven rules, care rules. |
| Synthetic hands | Protocol drafted (v0.1) from a second round of the review. Untested; calibrated inside the first studies. |
| Cities | First study proposed: one plan, one place, three weeks. |
| Strategy making | First study proposed. |
| Mandates | First study proposed. |
| Literature | 15 threads, 387 sources, 285 re-checked by an auditor. |
What would count as it working
Each first study reports the same counts, with the criteria used. The values are starting points to adjust before the study begins.
- At least a third of the readings are tested at another level or by another agent, and some of them break. If none break, the tests are too soft.
- A conventional write-up can be produced from the log alone.
- The Fiction register raised at least one question that Trace and Reading had not.
- For a random half of the readings a person also wrote a Test, so the synthetic Adversary can be compared with a human one: how many of its objections were checkable, how many generic, how many invented. If the Adversary adds nothing, it is dropped.
The counts calibrate the method. They are never a score and are never used to rank people, units or organisations: any count becomes a target once someone funds or ranks on it (Muller 2018; Hood and Piotrowska 2020).
What is inherited, and what is added
Each of the seven lines has an older statement, and each has a known limit. Setting them side by side is the honest way to say what Holagram repeats.
| Principle | Older statements | Main limit |
|---|---|---|
| 1 · Say where you stand | Bateson’s logical types and levels of learning (Bateson 1972); Beer’s recursion (Beer 2002); Meadows on places to intervene (Meadows 1999). | Boundaries are not given. Cilliers argues that the boundaries of complex systems are unusual in kind and the hierarchies less nested than assumed (Cilliers 2001). If the levels are read as a claim about reality, the objection applies. |
| 2 · No view adds up to the whole | The ecological and the individualistic fallacy (Subramanian et al. 2009); Cilliers (Cilliers 2001). | Morin’s hologrammatic principle puts the whole inside the part (Morin 2008), which points the other way. |
| 3 · Many perceivers, many holas | Soft systems methodology, in which a model is a device for debate tied to a declared worldview (Checkland 1981); second-order cybernetics, in which what is said is said by an observer (von Foerster 2003); situated knowledge (Haraway 1988). | Complexity thinking lacks a theory of power (Kok et al. 2021). Treating positions as comparable can hide who can say no. |
| 4 · Keep the ambiguity | Ambiguity as a resource in design (Gaver et al. 2003); wicked problems (Rittel and Webber 1973). | Wicked problems have no stopping rule and no right to be wrong (Rittel and Webber 1973). A three-week study that ends in held, broke or unchecked can look like closure on a question that has none. |
| 5 · Let it be answered | Second-order cybernetics (von Foerster 2003); Luhmann on observing observers (Luhmann 2018); boundary critique (Ulrich 2022). | The regress: the other observer sits in the same frame. See the stopping rule above. |
| 6 · Imagination, labelled | Scenario planning (Healey and Hodgkinson 2024); design fiction (Bleecker 2009). | A vivid fiction can move people whether or not it is sound (Healey and Hodgkinson 2024). |
| 7 · Draw it so that others can meet it | Inscriptions (Latour 1986); boundary objects (Star and Griesemer 1989). | Making a society legible is itself a way of seeing from above (Scott 1998). A diagram of linked entries is a legibility device. |
What Holagram adds is narrower than the way of thinking. It is a trace that anyone can open, a record of the hands behind each register, a labelled fiction, a failure log, and a rule against rolling readings up, together with the definition above, which treats every agent in the same way, synthetic ones included, and makes the gram the unit of work. The rest is inherited. The nearest existing methods are soft systems methodology (Checkland 1981) and boundary critique (Ulrich 2022), and a reader who knows them will ask what is left once they are removed. That list is the answer.
What the entry is for. It improves what is read before action, it shows which reading broke, and it makes no recommendation. The entry is a writing form and a log and not a process model: the four registers are not four steps of a design method.
What the literature would say against it
The review sets out the objections as well as the support. The strongest are below, with where each is answered or left open.
- “This is soft systems methodology and boundary critique renamed.” The nearest existing methods are Checkland’s (Checkland 1981) and Ulrich’s (Ulrich 2022). What Holagram adds is the narrower list above: a trace anyone can open, a record of hands, a labelled fiction, a failure log and a rule against rolling readings up.
- “The levels are not given.” Boundaries of complex systems are unusual in kind and hierarchies less nested than assumed (Cilliers 2001). Holagram treats levels as a device chosen for a question, and each method proposal gives a rule for assigning one.
- “The regress.” An observer of an observer is still an observer (von Foerster 2003; Luhmann 2018). Holagram stops at an outside reader and the log, as a decision, and claims contestability and not validity.
- “Complexity thinking lacks a theory of power.” (Kok et al. 2021) The mandate method asks of each reading who can say no and by what means.
- “Wicked problems have no stopping rule and no right to be wrong.” (Rittel and Webber 1973) Each test says who could act on the result and what being wrong would cost them.
- “We already do this.” In each domain practitioners already run something close: plan evaluation, strategy-as-practice, authorising-environment exercises, portfolio sensemaking and failure reports. Each of these pages says what is added and asks for one case where a test changed a decision.
- “A model Adversary may add nothing.” No study tests a model critic on a reading of an ecology. See synthetic hands.
- “Public documents are the front stage.” Talk aimed at legitimacy is most likely to sit there (Brunsson 1989), so a gap proves little and a three-week study may find decoupling by construction. Each study requires at least one lower-level trace per entry.
Open questions
- Does a model argue a useful Test against a person’s interpretation of an ecology? No study found tests this. The first studies are its calibration.
- Do levels chosen for one domain travel to another, or does each method proposal need its own?
- Do the care rules prevent self-confirmation in practice, or only in principle?
- How short can the three weeks be made without losing the tests at another level?
How the literature was built
The concept was developed from a 2025 statement about cities. The literature review was built in three rounds by AI scouts working in parallel, one per thread, each checked by a separate AI auditor that tried to find which sources were wrong. Every source is marked on the literature pages: checked, corrected, not re-checked or recalled, unverified. The auditors found no invented work in the samples they checked, a number of wrong links and labels, and a few entries they could not confirm; each is marked.
This follows the method’s own rule: models draft and object, and say so; a person decides.