Evidence at scale

    Most organisations are not short of evidence. They are short of a way through it. We do both halves of that problem, the synthesis and the thing people actually use.

    Synthesis

    From a folder of documents to a picture you can act on.

    A large evidence base is not hard because it is large. It is hard because it was written for other purposes, by different people, at different times, to answer different questions. This is the route through it.

    1. Stage 01, Catalogue. Every document logged in a structured, maintainable format. Type, date, origin, programme, population, method, status. The catalogue is a deliverable in its own right, not scaffolding we throw away.
    2. Stage 02, Appraise. Each source assessed for what it can and cannot support. A well written report built on twelve interviews does not carry the same weight as a weak one built on four hundred, and the synthesis has to know the difference.
    3. Stage 03, Code. Machine assistance does a first pass across the full corpus so nothing is skipped for reasons of volume. Human researchers code a stratified sample independently. Agreement between coders is measured and reported rather than assumed. Control point after this stage: Coding frame signed off by the community before analysis proceeds.
    4. Stage 04, Validate. The coding frame goes to the community before analysis proceeds. If the categories do not match how people describe their own experience, the categories change. This is the stage most synthesis work does not have.
    5. Stage 05, Map. Themes plotted against programmes, populations and time, so the gaps are as visible as the findings. What is missing from an evidence base is usually the most useful thing in it.
    6. Stage 06, Publish. Synthesis report, executive summary, evidence map and the interactive output, plus the training and documentation that lets the organisation maintain all of it without us.
    Methods note

    Where the machine is used, and where it is not.

    Anyone synthesising several hundred documents is using computational methods. The question is whether they will tell you where.

    What the machine does

    • A first pass across the entire corpus, so no source is skipped because of volume.
    • Candidate themes and clusters, proposed rather than adopted.
    • Retrieval across the full evidence base, so a question can be asked of everything at once.
    • Consistency checks that flag where a code has been applied unevenly.

    What people decide

    • Whether a proposed theme is real or an artefact of how the documents were written.
    • What the finding means, which is a judgement about people and never a computation.
    • Which sources carry weight and which do not.
    • Whether the coding frame reflects how the community describes its own experience.

    Every use of machine assistance is documented in the methods section of the output: what was used, on what, at what stage, and what a human checked afterwards. A reader who wants to know whether a finding came from a person or a process can always find out.

    Platforms

    The output people actually use.

    A synthesis nobody can navigate is a report nobody opens. We design and build the thing the evidence lives in, and we hand it over so it keeps working.

    01

    Blueprint before build

    User journeys mapped with the people who will actually use it, internal teams and external audiences separately, because they arrive with different questions. Structure agreed before a line of code.

    02

    Prototype and test

    A working prototype in front of real users, not a slide deck. Tested, changed, tested again. What survives that is what gets built.

    03

    Built to be explored

    Filterable by theme, programme, population, method and time. Every view addressable by its own link, so a finding can be sent to a colleague. Accessible to WCAG AA as a floor, and tested with assistive technology rather than assumed.

    04

    Handed over properly

    Code in your repository, not ours. Documentation, maintenance guidance and training for the team who will run it. Intellectual property sits with the commissioner. We build things organisations keep.

    Working example

    This site is one.

    The explorer below runs on our own body of work. Same mechanism, at a scale you can see the whole of.

    Filter by:

    11 results

    Filter this view and the address bar changes with it. Send someone the link and they land on what you were looking at.

    Have an evidence base nobody can navigate?

    Tell us what you are sitting on, and we will tell you what a route through it would look like.

    Start a conversation