The Design

2 · Experimental structure — two arms, within participant

Each participant completes two arms of a forced binary choice task. Presentation is binocular (both eyes, no dichoptic rivalry) — despite “CFS” in the project name, Part 1 is deliberately the pre-CFS validation baseline: the flashing Mondrians are a graded masking/degradation of a binocularly-viewed face, not interocular suppression. Both arms yield choice + RT on every trial → DDM.

The scientific tests

• T1 — graded within-subject drop in v across obscuring (does obscuring lower drift?).
• T2 — parameter selectivity (is the effect drift-specific — v moves, a/t₀ don’t?).
• T3 — cross-arm equivalence (TOST) — is the coherence→v slope quantitatively equivalent to the obscuring→v slope in the same person? This is the claim that certifies the pipeline as a reusable tool.
• T4 — individual-differences convergent validity — do motion-drift and face-drift covary across people?

3 · The three design versions

Stimulus set for all versions: RaFD (Radboud Faces Database) — on disk, 8,040 images, 67 IDs × 8 emotions × 5 camera angles × 3 gaze directions (left / frontal / right). Frontal camera angle used throughout.

Example stimuli — the manipulated dimensions

Gaze LEFT

Gaze RIGHT

ANGRY

HAPPY

Radboud Faces Database (RaFD), Langner et al. 2010^PDF — academic use, kept local. Frontal camera (Rafd090). These are the real manipulated dimensions the task discriminates.

A separate in-house synthetic face set (assets/faces/ex_*.png) exists only as an illustrative pipeline-test set — the study stimuli are RaFD.

4 · Stimuli, apparatus, task — and where each choice comes from

• RaFD faces — used by our anchor paper (Williams 2023, Study 1), so it is not novel to the corpus; using it is justified on RaFD’s own validation (Langner et al. 2010^PDF) and norming and its 3 gaze directions (needed for V1/V3), not on face-DDM precedent. Low-level confounds (luminance, contrast, spatial frequency) controlled per Purcell 1996, Becker 2011, Calvo 2016.
• Graded binocular Mondrian masking 0/19 precedent — only 1 distinct precedent masks at all (Williams 2023; Hartmann 2021 is its preprint — not an independent second study, so the corpus is effectively ≤18 distinct studies), and it is binary. A continuously graded mask is novel — justified from the RDM/CFS graded-noise tradition (Newsome & Paré 1988 for per-subject coherence/noise-dot calibration; Tsuchiya & Koch 2005 for 10 Hz Mondrian dynamics), not from a face-DDM precedent. Mask levels titrate discriminability, mirroring coherence in Arm A.
• Adaptive per-subject calibration — QUEST/Ψ 0/19 precedent — not one corpus paper uses adaptive staircase calibration of face-task difficulty. Imported wholesale from psychophysics (Watson & Pelli 1983^PDF; Kontsevich & Tyler 1999) to equate difficulty across participants and across the two arms. A genuine methodological contribution for face-DDM.
• RDM task: built on the PI’s own open PsychoPy RDK (abcsds/RDM, 2024). Coherence levels per Palmer, Huk & Shadlen 2005^PDF (PHS05).
• Gaze angle calibration (V1): seed left/right separation from Palmer, Caruana, Clifford & Seymour 2018 (cone half-difference ≈ 10°); confirm in pilot.
• Trials: target ≥100 correct trials/cell for the EZ baseline; ~20–40/cell adequate under HDDM shrinkage (Lerche, Voss & Nagler 2017^PDF; Wiecki 2013^PDF). Williams ran ~18/emotion×mask cell (sparse) — our design exceeds this.

The obscuring mechanism — a novel contribution

The single largest methodological gap the golden-standard comparison surfaces (see Comparison) is that no paper in the 19-paper corpus grades face-masking severity continuously — occlusion/masking has only 1 distinct precedent (Williams 2023; Hartmann 2021 is its preprint, not an independent study), and it is binary (masked vs. unmasked). A continuously graded occlusion/masking manipulation (percentage of face area obscured, or a stepped noise-mask series) has zero precedent in the reviewed literature. This design's graded binocular flashing-Mondrian mask — mirroring the coherence ladder in the RDM arm — must be justified from the RDM/b-CFS graded-noise literature (with Newsome & Paré 1988 cited specifically for per-subject coherence/noise-dot calibration) rather than from a face-specific precedent, since none exists.

5 · Analysis pipeline (open-source)

• Per-subject baseline: EZ-diffusion (Wagenmakers 2007^PDF) — fast, closed-form 0/19 precedent; chosen from best-practices, not corpus.
• Confirmatory: HDDM (Wiecki, Sofer & Frank 2013^PDF) — hierarchical Bayesian; the plurality choice among corpus papers that name a method (4/6). Required because EZ recovers a/t₀ poorly (see Power §1 caveat).
• Parameter recovery validated by simulation before fitting real data (r_v=0.99; r_a, r_t₀≈0.55–0.59 under EZ — motivates HDDM for the selectivity claim).
• Equivalence: TOST on standardized difficulty→v slopes across arms, pre-specified margin (see Power §3.3).
• Everything pre-registered; task code, analysis, and (permitted) data open.

7 · Best-practice spine (golden standards)

1. Pre-registration of hypotheses, N, exclusion rules, and the equivalence margin (before data).
2. Adaptive per-subject calibration (QUEST/Ψ) to equate difficulty across arms — novel for face-DDM.
3. Graded difficulty in both arms (coherence ↔ mask), matched in number of levels.
4. EZ (screen) + HDDM (confirm); report parameter recovery.
5. Low-level confound controls on RaFD stimuli; counterbalancing of arm order, response mapping.
6. Distributional RT analysis — never mean-RT (Tipples 2023^PDF: conclusions flip with outlier/model choice).
7. Open task, analysis, and data; local-first tooling.

8 · Roles, milestones, deliverables

Roles

Milestones (relative to 2026-07-23)

Deliverables of this KB effort: the contract; this interactive KB website; the illustrated slide deck (main acts + annex); the golden-standard comparison; the grounded power analysis.