Museum feasibility pilot — draft 0.1.0

Primary: can the procedure retain a technically complete encounter and first account for at least 22 of 24 enrolled participants? Secondary: can at least 20 of 24 complete the encounter and account without unplanned facilitator assistance? Exploratory: how do participant-level selection shares, regional pointer-centre dwell and account descriptions differ between introductory framings? Interaction, self-report and researcher inference remain separate. No claim about eye movement, aesthetic quality, preference or physical-gallery transfer is tested.

Proposed setting: one museum study room, one fixed desktop and mouse, the same browser/app build, viewport, display scale and three licensed images for everyone. Institution, principal investigator, facilitator, data custodian, independent coders, ethics determination, accessibility procedure and session dates are unassigned and must be recorded before registration. Confirm image and text permissions, storage policy, consent and export review with the host. No institutional participation is implied by this draft.

Use the same three works, source-image hashes, simultaneous arrangement, positions, sizes, neutral work labels and instructions in both arms. Select and freeze them with a curator before registration. Condition A candidate introduction: ‘These three works are presented together. Take time to look at them.’ Condition B candidate introduction: ‘These three works are presented together. Consider how repeated forms and contrasts connect them.’ These candidate texts intentionally differ in semantic framing and must be reviewed for relevance and accessibility, then retained verbatim; the estimand concerns this whole framing change, not an isolated word or text-length effect. Read the introduction once before beginning. Keep a clear glass at 1× in both arms to avoid changing magnification. Define one named non-overlapping polygon per work in normalised source coordinates before collection; freeze vertices, revision and hash. Do not alter boundaries after inspecting outcomes.

Propose 24 consenting adults, 12 per arm, recruited from a convenience sample of adult museum visitors able to use this supported setup and study language. This sample is a pragmatic feasibility target, not a power calculation or a representative visitor sample. Offer an equivalent non-research museum activity to people who cannot or do not wish to take part. Record access needs and recruitment limitations without treating disability as an outcome. Use a computer-generated sequence of six randomly ordered blocks of four (two A and two B), generated and retained by someone outside recruitment; conceal upcoming assignments until consent and baseline information are recorded. Stop after 24 enrolled people have begun a trial; do not replace early exits or incomplete records. Before launch, set a calendar recruitment deadline; if reached early, report the smaller sample and denominators. Stop immediately for a critical data exposure, consent failure or a bug that corrupts acquisition; document an amendment before resuming.

Allow approximately 10–15 minutes total. Obtain consent covering pointer/lens records, first accounts and coded research outputs; participation and account writing are voluntary. Assign a pseudonymous code and keep any contact/withdrawal key separately. Record optional artwork familiarity and required device/display metadata before allocation is revealed. Practise on an unrelated image without research recording. Present the assigned introduction, then the three-work field for 60 active seconds; permit early exit at any time. Standard instruction in both arms: ‘Move the looking glass as you explore. There is no correct route. You may pause or stop at any time.’ This instruction is an intervention and its influence is a limitation. Collect the first account before showing any trace: ‘Describe what appeared to you. Include anything that did not fit or remained unresolved.’ Allow skipping; do not require an interpretation. Collect later reflection and facilitator notes separately. Record interruptions, assistance, early exit, missing account and reasons offered without prompting for sensitive details.

Primary feasibility numerator: enrolled records with a retained version/hash, source hashes, timing and event flags, completed 60 active seconds, and a saved first account. Denominator: all enrolled people who began a trial (target 24), including early exits and missing accounts. Report consent withdrawals and limits on retained data in the flow record; never retain withdrawn content merely to keep a denominator. Report analysis use separately from consent-authorised anonymous recruitment totals. Secondary numerator: people completing the encounter and first account without unplanned assistance, using the same enrolled denominator. Report every component separately so a missing account is not mistaken for an acquisition failure. Exploratory units are people, with one trial each. Report selected dwell/active time ×100 for each work; unassigned time remains in the denominator and is separately listed. Report ROI pointer-centre dwell in milliseconds, entries and returns with the frozen region definition; selected dwell and regional dwell are different measures.

Exclude pre-labelled facilitator rehearsals from participant analysis. Retain consent-authorised early exits, interruption flags and incomplete records in the feasibility flow; do not silently drop them. Mark a trial unsuitable for the exploratory completed-trial summary if it ended early, lost/truncated required events, used a different frozen design or had an unrecorded display change. State every reason before inspecting condition differences. Do not impute missing accounts or interaction intervals and do not treat missingness as zero. Describe missingness by arm and repeat the exploratory summary including consent-authorised early exits, clearly identified with their own active-time denominator. A participant can decline an account without losing compensation or museum access.

Draft coding unit: the whole first account. Two coders, blind to arm, trace and each other’s decisions, code every retained account before discussion. Proposed non-exclusive binary codes: specific visible detail; relation between works; interpretive explanation; explicit uncertainty or resistance. Freeze definitions, positive and negative examples and rules for ambiguous/empty accounts after training on separate non-study examples. Preserve the original text, each independent code, uncertainty notes and later adjudication separately. Report each 2×2 agreement table, raw agreement and Cohen’s kappa per code with denominators. If a marginal is constant and kappa is undefined, report that with the counts; do not substitute perfect reliability. Describe semantic disagreements and resistant accounts in their own right. Do not use automated classification as the ground truth.

Report recruitment flow, each feasibility proportion and a two-sided 95% Wilson interval, assistance, time burden, missingness and acquisition problems. Proposed progression: proceed to redesign/confirmatory-study planning only if at least 22/24 meet technical completeness, at least 20/24 finish without unplanned assistance, all retained exports pass integrity/recomputation checks, and there is no unresolved critical consent or data incident. At a smaller achieved sample, apply the prespecified proportion targets (90% and 80%) and report the intervals; treat the decision as provisional. These are proposed practical thresholds, not established field standards. Any failed criterion leads to a documented revision and a fresh feasibility test, not post-hoc relaxation. For exploratory arm differences, show participant values, means/medians, ranges and A/B differences with uncertainty; no confirmatory p-value or claim of efficacy. Counts of pointer samples are not the sample size. Record order, familiarity and facilitator issues as limitations, not opportunistic adjustment choices.

Collect no video, audio, face image, gaze stream, name or email in the research notebook. Warn participants not to include identifying information in free text; review accounts and notes before sharing. Propose restricted encrypted institutional storage and named access roles; the host must set actual retention, lawful/ethical basis, withdrawal cutoff, compensation, accessible consent and contact arrangements before recruitment. A public package may contain only consent-compatible de-identified or constructed material after a disclosure review; never publish the participant-code key. Document access barriers and offer supported alternatives; analyse any alternative-device cohort separately under a declared amendment rather than assuming measurement equivalence.

Resolve every prerequisite, freeze the codebook, stimulus set, regions, allocation, software version and analysis rules, then externally register the final plan before recruitment/collection. An app lock is not external preregistration. Record the registry URL and timestamp only after registration occurs. Retain dated amendments without overwriting the original and identify whether each occurred before or after outcome inspection. Planned outputs: recruitment flow and feasibility report, independently coded account results with disagreement, participant-level descriptive plots and accessible tables, deviation log, versioned methods and consent-compatible evidence package. A separately designed physical-gallery comparison is required before claiming museum-floor transfer.

OSF registrations guidance: https://help.osf.io/article/330-welcome-to-registrations . W3C guidance for complex images and data alternatives: https://www.w3.org/WAI/tutorials/images/complex/ . These references guide record transparency and presentation; they do not validate the instrument or this proposed sample size.