When a Theory of Change has to hold up

The direction of travel

The centre of government is steadily raising the evidential bar behind spending. The Evaluation Task Force - a joint Cabinet Office and HM Treasury unit set up after the 2020 Spending Review, once it emerged that only 8% of the government’s £432 billion of major-project spend had a robust impact-evaluation plan - has since advised on over 380 programmes worth £202 billion. Its 2026–2029 strategy marks a deliberate shift: from checking that evaluation happens, to ensuring the evidence actually shapes what gets continued, expanded or stopped.

For services that support victims and witnesses the pull is sharper still. The Victims and Prisoners Act 2024 puts the Victims’ Code on a statutory footing, adds a duty to collaborate on commissioning victim support, and gives the Victims’ Commissioner powers to request compliance data; the Victims’ Funding Strategy already commissions against four national outcomes and asks for evaluation plans in contracts. The direction is unmistakable: these services will increasingly have to show, proportionately and on shared outcomes, that they work - and to do it with the Magenta Book as the standard and, per the Ministry of Justice’s own Areas of Research Interest 2025, an openness to AI-assisted evidence synthesis.

A method problem, long known

Most of that evaluation will rest on a Theory of Change, because randomised designs are rarely feasible for support services. Yet theory-based evaluation has carried the same flaw since Carol Weiss set it out in the 1990s: teams draw the programme theory and then do not use it to steer the evaluation. Pawson and Tilley’s realist evaluation sharpened the question a Theory of Change exists to answer - what works, for whom, in what circumstances - but the artefact itself stays a diagram with a narrative, and on the page a link with no evidence behind it looks exactly like a link with strong evidence. (Blamey and Mackenzie, in Evaluation, 2007, mapped how closely the two traditions actually sit.)

For services whose outcomes are diffuse and whose evidence base is thin - much of the victim and witness landscape - that is exactly where an evaluation goes wrong before it begins: effort pours into the links that are already well understood, while the load-bearing assumption nobody has tested sits unexamined in the middle of the diagram.

The method: treat the Theory of Change as data

The fix is not a better drawing. It is to make the Theory of Change computable: type every element, anchor each outcome to a published national outcome rather than a bespoke one, attach an evidence grade to every causal claim, declare the assumptions each pathway rests on, and let a validator check the whole graph for completeness. What was a persuasive picture becomes a structure you can query, test and reuse.

A validator that finds the gaps for you

Because the pathways are typed, SHACL shapes check the graph the way a spell-checker checks prose: no outcome without a published anchor, no causal claim without an evidence grade, no pathway without its declared assumptions. Running the validator on this graph returns SHACL conforms: true, then surfaces the evidence-gap map by query - two genuine gaps across seven pathways. Structure is guaranteed; the gaps are recorded, not hidden. That is what turns “name the weak links” from a hope into a mechanical result - and it is precisely what a feasibility assessment or an Evaluation Task Force reviewer is looking for.

Where this goes

An open, machine-readable Theory of Change and evidence schema released under CC-BY-4.0, built on Tesseract’s open-ontologies approach and validated with SHACL. As evaluation moves to the centre of spending decisions, a Theory of Change stops being a document produced once and filed. It becomes standing infrastructure - auditable, internally consistent, and carried forward into the full evaluation rather than redrawn from scratch. The products a commissioner sees are still the familiar ones - the diagram, the narrative, the monitoring-and-evaluation framework; the structured graph underneath is what makes them hold up.

This is an independent, self-initiated reflection on method. It is built entirely from public information and open standards, uses no confidential or commissioned data, and represents no organisation’s findings but our own.