RDMxbCFS · Part 1 · §9

Research Gaps & Open Questions

Five gaps surfaced by the 19-paper golden-standard comparison, framed as candidate research questions — each tied to the power tier and test that would actually answer it.

From ddm-face-comparison.md §3.5 and effect-sizes-retrieved.md

These are not softened as afterthoughts: the gaps and the zero-precedent flags (see Comparison §4.5) are the highlight of this project, not a caveat to it. A design that scores PARTIAL on every published precedent, in combination, is exactly where the publishable contribution sits.

Where our design has zero precedent tallied across the 19-paper face-2AFC-DDM corpus Emotion as discrimination axis 12/19 Gaze as discrimination axis 2/19 Full combination (this design) no precedent 0/19 Graded masking (any axis) no precedent 0/19 Adaptive staircase (QUEST/Ψ) no precedent 0/19 EZ-diffusion estimator used no precedent 0/19 Emotion outnumbers gaze 6 : 1 among papers judging the face’s own content 0 / 19 strict PASS
The tally at a glance. Across the 19-paper face-2AFC-DDM corpus, emotion outnumbers gaze 6:1 as the discrimination axis, and four of this design’s choices (the full combination, graded masking, adaptive staircase, EZ-diffusion) have zero precedent. No paper is a strict PASS on all eight criteria.
Gap 1 · addressed by V1 pilot · Tier ① Validation (N≈12–16, pilot ~10)

No drift-rate estimate exists for gaze-direction discrimination

Alister 2023 shows gaze cueing is a non-decision-time effect, not a drift-rate effect (v inclusion probability only 3–13% across three datasets); Palmer & Clifford 2018 fits no DDM at all. Open question: does discriminating left/right gaze direction (as opposed to being cued by it) accumulate perceptual evidence the way motion coherence does?

V1 answers this — a ~10-person gaze pilot run in parallel with the V2 flagship, per Design §3.

Gap 2 · addressed by T3 · Tier ③ Strong equivalence (N=130–160 at Δ=0.3)

Graded masking of faces + full DDM has never been run

All face-masking precedent in the 19-paper corpus is binary (masked/unmasked; Williams 2023PDF, Hartmann 2021PDF) — a continuously graded severity manipulation has 0/19 precedent. Open question: is the mask→drift function shaped like the coherence→drift function (a proportional-rate model, μ′=k·x, as in Palmer/Huk/Shadlen 2005PDF)?

T3 (TOST equivalence) answers this — see Power §Equivalence for the margin trade-off.

Gap 3 · addressed by T3 · Tier ③ Strong equivalence

Within-person equivalence of a "high-level" and "low-level" decision is untested

No published study has compared face-judgment drift and motion-judgment drift within the same person under a common DDM framework. If the coherence→v slope and the obscuring→v slope are quantitatively equivalent, the DDM is a domain-general accumulation readout — a strong, citable claim, and the one that certifies this pipeline as a reusable tool (see Design §2, test T3).

Cost: this is the expensive claim — N=50 (Δ=0.5, weak) to N=260 (Δ=0.2, strict). The PI's margin choice determines how strong a claim Part 1 can defend.

Gap 4 · addressed by T4 · Tier ② Convergent validity (N≈50–65)

Reliability/individual-differences of face-DDM parameters is essentially unstudied

Whether a person's motion-drift and face-drift covary across individuals (T4) has not been tested in this literature. A pipeline that yields a reliable per-person v across domains is a tool for clinical/developmental work (anxiety, psychopathy, aging — all present in the 19-paper corpus but methodologically fragmented, see Comparison §3).

T4 answers this at the recommended default tier, N≈50–65 for a true cross-arm r=0.5.

Gap 5 · a literature-completion task, not a design decision

The Sawada 2022 magnitude gap

Sawada, Sato, Nakashima & Kumada (2022, Cognition) is the cleanest mechanistic anchor for the anger/happy→drift effect (energy-matched anti-expression controls) — but the paper is paywalled and no numeric drift-rate, boundary, or non-decision-time value could be retrieved (PubMed abstract gives direction only: larger v, shorter t₀, larger a for normal vs. anti-expressions). A concrete task for a student: obtain institutional access to Cognition 229:105235, or the corresponding author's data, and extract the numeric effect sizes needed to properly anchor the V2 emotion-arm power analysis.

Where the impactful science is

Per the 19-paper tally (Comparison §3.5) and the retrieval work (Power §Assumptions), the impactful science in this project is not any single novel stimulus or manipulation in isolation — RaFD is already used by our anchor paper (Williams 2023, Study 1) and is not novel to the corpus, while graded masking and adaptive calibration are each individually defensible imports from adjacent literatures. It is the combination, run within the same participants under a common DDM framework, that no paper in the reviewed corpus has attempted:

These are seeds; the claim-tier choice determines which of them Part 1 can claim now versus defer to a committed follow-up. The team decides in the room — this contract lays out the full decision space so that choice is explicit rather than accidental.