Symbol Cell Color & Background Formatting — Research Findings and Decision

Date: 2026-07-20 Status: Accepted Type: ADR + research synthesis Research provenance: deep-research run, 2026-07-19 (synthesis phase ended at a session limit; verified/unverified claims below were preserved before that point and independently reconciled against the run’s source fetches for this document).

TL;DR

Oso Voice’s default cell design follows the research evidence, not just AAC convention: adversarially-verified findings show full-cell background color fill does not help — and may hurt — symbol search on the grid sizes the app actually uses, so the shipped default (border-only color accent, white cell background) is the evidence-backed choice, not an accident of convention. A desaturated-pastel “Color-filled buttons” toggle remains available, off by default, for the caregiver- and larger-grid use cases the evidence does support.

Background

Oso Voice color-codes symbol cells using the Modified Fitzgerald Key (MFK) “core 4” part-of-speech convention (decision docs/decisions/002-mfk-color-coding-as-accent.md, 2026-07-07), which already chose a border/corner-tag accent, never a saturated background fill, reasoning from general AAC clutter/abandonment literature available at the time. Adam’s own aesthetic instinct still leaned toward a fuller, filled-cell look, and that preference was never directly evidence-checked against fill-versus-border specifically — as opposed to the earlier decision’s broader “avoid visual clutter” reasoning. Before locking that preference in (or dismissing it), Adam asked for targeted evidence-based guidance on cell formatting, which prompted a dedicated deep-research pass on 2026-07-19.

That run’s adversarial-verification step (each claim checked by multiple independent reviewer passes) was cut short by a session usage limit before the synthesis report could be produced. The claims it had already fetched and, in several cases, fully verified were preserved in the Obsidian vault (~/Documents/Vaults/brightvoice/Oso Voice App.md, “Cell-formatting research findings” section). This document is the durable, in-repo synthesis of that material, written so it can be cited later — including in marketing copy — without re-deriving it from session logs.

Findings

Findings are grouped by how much verification they actually received. The distinction matters: only the first tier survived full adversarial cross-checking. The rest are real, independently-sourced leads that were never adversarially contested to conclusion — they should be treated as directional, not settled, until re-verified.

Tier 1 — Adversarially confirmed, 3-0

Full-cell background color does not aid symbol search on 12–24-symbol grids, and may disrupt search in younger children. The Fitzgerald-Key convention of coloring the entire cell background is a convention-only practice with a minimal underlying evidence base — i.e., its ubiquity in commercial AAC systems is not itself evidence that it improves the outcome it is often assumed to improve (locating a symbol quickly on the grid).

• Source: pmc.ncbi.nlm.nih.gov/articles/PMC6436972
• Verification: 3 independent adversarial review passes, unanimous (3-0).

Tier 2 — Adversarially confirmed, 2-0

Color placed inside the symbol artwork itself, combined with clustering same-colored (same-category) symbols together on the grid, does improve search. This is a materially different mechanism from cell-background fill — it locates the color signal in the icon and in spatial grouping rather than in a colored panel behind the icon.

• Verification: 2 independent adversarial review passes concurring (2-0).
• Note on sourcing: the surviving vault notes list this finding immediately after, and evidently drawn from, the same research pass as the Tier 1 finding above; the specific citation was not separately re-logged before the session limit hit. Before this claim is quoted on its own (e.g., in marketing copy, as a distinct citation from Tier 1), re-confirm its exact source against PMC6436972 or re-run the synthesis.

Tier 3 — Unverified but consistent (strong leads, verification lost to the session limit)

These are real, independently-sourced claims the research run surfaced and found no contradicting evidence for, but which never received a completed adversarial verification pass. Treat each as a lead worth acting on cautiously, not as settled fact, until independently re-checked.

• White backgrounds were often faster for young children in the studies reviewed. Sources: tandfonline.com/doi/full/10.1080/07434618.2017.1336571; pubmed.ncbi.nlm.nih.gov/28617614.
• Cell fills only helped on large (~50-symbol) grids, and initially acted as a distractor on small grids, per a pilot eye-tracking study. Source: researchgate.net/publication/385378705.
• That same pilot’s desaturated pastel per-category color tokens are directly usable as design tokens (full RGB values below). Source: researchgate.net, publication 385378705 (same pilot as above).
• Word-class ARRANGEMENT — the Fitzgerald layout principle of grouping cells by grammatical role positionally — is distinct from Fitzgerald’s colors, and DID speed message construction in the sources reviewed. This is a claim about spatial layout, not about color at all, and should not be conflated with the cell-fill question above.
• Thistle (2019): background color mainly helps communication PARTNERS (caregivers), not the AAC user themselves. If replicated, this reframes cell fill as a caregiver-facing affordance rather than a child-facing search aid — consistent with, and a plausible explanation for, the Tier 1 finding that fills don’t help user-side search.

Design Tokens

The pilot eye-tracking study (researchgate.net/publication/385378705, Tier 3) used a desaturated pastel palette, one tint per grammatical category, distinct from Oso Voice’s existing saturated MFK border-accent hues (Yellow/Green/Orange/Blue — see decision 002). These are the tokens the “Color-filled buttons” toggle uses:

Grammatical category	Oso Voice MFK category	Pastel token (source)	RGB	Hex
Persons / pronouns	Yellow (people)	Purple pastel	rgb(246, 183, 231)	#F6B7E7
Verbs / actions	Green (verbs)	Green pastel	rgb(216, 253, 185)	#D8FDB9
Objects / nouns	Orange (nouns)	Red/salmon pastel	rgb(255, 216, 193)	#FFD8C1
Adjectives / descriptors	Blue (descriptors)	Blue/teal pastel	rgb(205, 238, 234)	#CDEEEA

Note the hue names in the pilot source do not line up 1:1 with Oso Voice’s MFK border-accent hues (e.g. the “persons” category is a purple pastel in the pilot but maps to Oso Voice’s Yellow MFK category; “objects” is a red/salmon pastel mapping to Oso Voice’s Orange category). The toggle should map by grammatical category, not by matching hue name, so a filled cell’s pastel tint and its MFK border accent will legitimately be different colors on the same cell. This is expected, not a bug.

Decision (ADR)

Context: Oso Voice already shipped border-only MFK color accents (decision 002, 2026-07-07) reasoning from general clutter/abandonment literature. Adam’s own aesthetic preference for a filled-cell look remained an open, evidence-unchecked question. This research run was commissioned to check fill-vs-border specifically before deciding whether to change the shipped default or add filled cells as an option.

Decision: Oso Voice ships border-only white cells as the child-facing default. MFK color continues to appear as a border accent on every cell, and the same color signal will additionally live inside future symbol artwork (Tier 2 lever, not yet implemented — see Implications below). A “Color-filled buttons” Settings toggle, default OFF, is added, offering desaturated-pastel fills using the four tokens above. This toggle exists for the use cases the evidence does support: communication partners/caregivers (Tier 3, Thistle 2019) and larger future grid sizes (Tier 3, pilot eye-tracking study) — not as the child-facing default.

Status: Accepted 2026-07-20. The toggle is being implemented in the same working session that produced this document.

Consequences:

• The shipped default is now defensibly evidence-based rather than convention-based: it matches the only Tier 1 (3-0 adversarially confirmed) finding in this research — full-cell fill does not help, and may hurt, search on grids in Oso Voice’s actual size range (12–24 symbols; the current starter vocabulary and category views are in this range). This is the basis for the “evidence-based design” framing this document is meant to support.
• The higher-value lever for search performance — color inside the symbol artwork plus clustering same-category symbols (Tier 2) — is not yet implemented by this decision. Shipping the border-accent-only default plus an optional fill toggle does not capture that gain; it is future work (see below).
• Several supporting claims (word-class arrangement speeding message construction; Thistle’s partner-vs-user framing) are Tier 3 (unverified-but-consistent). They should not be cited as confirmed in marketing copy without a re-verification pass; they may currently be cited only as “early/consistent evidence” or “a pilot study suggests.”
• The pastel token palette is sourced from a single pilot eye-tracking study with an unreported/likely-small sample size and no known independent replication. It is reasonable as a default token set for an opt-in toggle; it is not strong enough evidence on its own to justify making fills the default.

Implications for Future Work

• Larger, paged grids may justify fills. The Tier 3 pilot finding — fills helped on ~50-symbol grids but acted as a distractor on small grids — is directly relevant to Oso Voice’s planned paged fixed-position grid work (Adam’s 2026-07-19 simulator feedback: fixed word positions, no scrolling, page dots/next-page). If future page sizes grow well past the 12–24-symbol range this research covers, revisit whether fills should default ON for those larger layouts specifically.
• Color inside symbol artwork + clustering is the higher-value lever. This is the one search-improvement finding with adversarial confirmation (Tier 2), and it is not yet built. It depends on the artwork itself (VoicePack Studio’s generation pipeline) and on grid-arrangement logic to cluster same-category symbols — a cross-repo dependency, not a Settings toggle. Consider prioritizing this over further cell-formatting options once artwork categorization data supports it.
• Partner-facing modes could default fills ON. If Oso Voice ever ships a caregiver/partner-facing view (as distinct from the child-facing grid), the Thistle 2019 lead (Tier 3) — that background color mainly benefits partners, not users — would argue for defaulting the fill toggle ON in that context specifically, inverting today’s default. Re-verify before committing to this as a stated design rationale.
• Word-class arrangement is a layout question, not a color question, and is still open. The Tier 3 finding that Fitzgerald-style grammatical grouping (position, not color) speeds message construction is independent of this decision and worth folding into the in-progress paged-grid layout work regardless of how the color question resolves.
• Re-run verification before further marketing claims. The vault notes reference a resumable research session (resumeFromRunId wf_a81c19c9-c0d) that can replay cached agents. If Tier 3 claims need to move to Tier 1/2 confidence for stronger marketing language, resume that run rather than re-researching from scratch.