Practice-based research in audiovisual art: a worked example

The canonical distinction belongs to Linda Candy’s 2006 guide. Practice-based research is an original investigation undertaken to gain new knowledge partly by means of practice and the outcomes of that practice: an artwork or prototype is submitted together with documentation that makes the knowledge claims explicit. Practice-led research is the mirror case: it produces knowledge about practice — the artefact is not the primary outcome, and the findings feed back into how practitioners work.

The older triad of research into, through and for art and design (Christopher Frayling, 1993) is the common ancestor of both terms: research through practice sits in the middle, and the practice-based label sharpens it by requiring an outcome — the work itself — plus an explicit written account. Universities that run such degrees formalise the same balance. Otago’s 2024 guidelines pair the creative practice with a written exegesis of 30,000–40,000 words that situates the work in a conceptual framework; York’s doctorate by creative practice combines a portfolio with a 20,000–50,000-word dissertation and a critical reflection.

The case unpacked below is practice-based in Candy’s sense: the prototype is the instrument through which the questions are investigated, and the knowledge claims travel in the theses, the measurement methodology and the papers — all of it public. If you need the vocabulary in one place, the bilingual glossary entry on practice-based research sums it up in two sentences.

The KVANT 1993 Archive is a research prototype on the border of film and play: an abandoned late-Soviet research lab from an imagined 1993, explored in first person. The project is framed as a doctoral study in audiovisual art (field 021). Everything is open: the prototype, session journals, methodologies and paper drafts live in a public repository, and a browser portal mirrors the artifact — world inspector, paper reader, media archive included.

The prototype as a measuring instrument

The point of the artifact is not to be a game but to be an instrument. Technically it is a custom raycaster in the spirit of Wolfenstein 3D — 320×200 internal frame, software rendering, C++/SDL2 — with a narrative system on top: text scenes, choices with consequences, film inserts, procedural audio. At the current v0.4+ stage it offers a 16×16 zoned space, boot and title screens, three terminal endings (TECH / CRITICAL / META), FMV-lite with voice, three BGM loops, and all 7 automated tests pass. Because the whole world is generated from code, nothing about an experiment depends on shooting logistics: a parameter changes, the world changes, the frame dumps record it.

The variables: directorial parameters made switchable

Five directorial parameters of interactive audiovisual storytelling became controllable variables: image density, fog, the pause of participation, montage, and sound. Each can be switched on and off within the same space — and that is the whole trick. What directors normally tune between films, this instrument tunes inside one world, which makes the direction of participation comparable: the same corridor exists with and without fog, the same scene exists with and without a consequential pause. How such variables sit inside a larger analytical frame — story implementation, interaction, viewer experience — is the subject of my analysis model for interactive cinema; the montage variable in particular continues the argument of the article on montage as a tool for variable narrative.

The measurement methodology MV-1…MV-4

Measurement needs rules fixed before the data arrives. The MV-1…MV-4 methodology combines scales, a behavioural channel, and pre-registered rules for reading them; its full specification is documented in the repository and on the case page. In the synthetic phase it passed saturation with 24 personas — and the key property held: the distributions for the base world and the “flat” world (participation stripped of consequences) do not overlap. That is what makes the instrument credible enough to carry hypotheses forward, and it is also the level at which claims stop: no audience generalisations are made before the human phase.

Theses T1–T7 and ten hypotheses, honestly yellow

The theory layer consists of seven theses: registers of presentation, the consequential pause, the motor episteme, retroactive montage, distributed interpretation, presentation validity, and prior inversion. Each thesis connects the measured differences to a claim about the form — for instance, memory beats the frame: the authentic 1993 textures read as a “modern filter” (thesis T7). Ten working hypotheses hang on these theses, and every one of them is marked 🟡 — formulated, instrumented, not verified. The rule is strict: synthetic observers can refute, sharpen or expose an instrument; only human playtests can verify. Until the human phase runs, no hypothesis advances.

Contrast builds and synthetic observers

Every research question gets its own contrast build. Technological ascesis (HA) is tested with builds HA-2 and HA-4: frame dumps capture the objective difference of the modes, and walkthrough sessions surfaced the “a step teaches” mechanism — fog works as a depth cue that reveals itself through movement. Hybrid agency (HB) uses participation cycles, a “flat” agency contrast and a type matrix (HB-3); the flat contrast produced the same verdict from every observer so far: a world without consequential pauses reads as broken, not merely calmer. Montage (HC) runs through film routes and montage inserts with dumps HC-1…HC-3. Observation is two-layered: blind, counterbalanced synthetic sessions for fast coverage; human playtests with a ready package (HB-1) for verification — currently deferred.

The pattern is transferable to any artwork that can host a switched parameter — a game, an interactive film, an installation. The order matters: measurement rules come before data, and honest status comes before both.

• Formulate one research question per build. If a question cannot name its contrast, it is not ready.
• Turn directorial decisions into switchable variables inside one space, not across different works.
• Build the minimal instrument that can hold those variables — nothing more than the questions need.
• Pre-register the measurement rules: scales, behavioural channel, thresholds — fixed before any data.
• Automate the evidence: frame dumps, automated tests, session journals. The dumps in the case cost nothing to re-run.
• Run synthetic saturation first (blind, counterbalanced, cheap) and watch for non-overlapping distributions as the credibility bar.
• Attach theses to the measurements: say in advance what a difference will mean if it holds.
• Keep a public status board for hypotheses and never promote a status by hope — only by the human phase.
• When the artifact stage pauses, freeze the interim results with a date and switch to writing.

By the author’s decision of 14 September 2026 the artifact stage is paused: the collected data was sufficient to move to analysis and the research paper, which consolidates the SYN sessions, MV saturation, think-alouds and walk logs on the basis of three working texts. The human playtests — the HB-1 package is ready — return after publication. Until then, the project reports exactly what it has: a working instrument, a saturated methodology, seven theses, and ten hypotheses that remain unverified. Interim results describe the instrument and its behaviour, never audience effects.

That restraint is not modesty but method. A practice-based PhD stands or falls on whether its written account distinguishes what the artefact demonstrates from what it merely suggests. The case shows the discipline: statuses are public, synthetic data is never dressed as verification, and the “what comes next” is dated and checkable. An examiner — or a reader — can audit every step without trusting anyone’s enthusiasm.

Otago’s guidelines are the clearest single description of what a PhD with practice research consists of — practice plus exegesis, with explicit rules for attribution and examination. York’s programme page shows the same structure as a live degree with its word counts. The UEA thesis by Daryl Scott (2024) is a completed example in adjacent territory: a critical video essay submitted with a written reflection on the researcher-practitioner-spectator position. Candy’s 2006 guide remains the standard reference for the terminology.

For the interactive-cinema side of the practice — why participation can be directed at all, and where the measure sits in a typology of the form — start with What Is Interactive Cinema and the KVANT case page itself.