open-standards/aac/aac-rs/tools/pagesets/DESIGN-motor.md

Fixed Paths — design

A motor-plan vocabulary: every word has one place to be reached from, by the same 1, 2 or 3 movements every time, starting from Home. It is designed from first principles — frequency times motor cost — and is ours: it does not reproduce LAMP Words for Life's, Speak for Yourself's or Unity's layout, category scheme or names, and it uses no multi-meaning icon sequences.

• Source: data/fixed-paths.yaml (the words, the page each belongs to, the groups pages fall back on), data/fixed-paths.layout.yaml (where everything sits, solved once and reviewed), data/lexicon.en.yaml.
• Output: fixed-paths-{30,60,84}.obz, and each with the starter filter; built by catalog.py wherever they are served, never committed (see DESIGN-core.md for the catalogue, aac-rs/data/pagesets/index.json).
• Check: python3 tools/pagesets/build.py --check, python3 tools/pagesets/catalog.py --check and cargo test -p aac-core --test pagesets.
• Rebuild: python3 tools/pagesets/build.py and catalog.py --write (Python 3.11+ and PyYAML, no network). Re-solve the layout: python3 tools/pagesets/solve.py motor, which reproduces the committed one byte for byte; any change to the design means re-solving, reviewing the grids, and committing the layout and the catalogue with it.

Tables between <!-- table:… --> markers are written by build.py.

1. Rules

1. One meaning per location. A location says one word, or opens one page. Every button shows the word it says or the name of the page it opens.
2. One path per word, at each size: Home → word (1 hit), Home → page → word (2), or Home → page → group → word (3). The test walks every package and fails on a word reachable twice or in more than three hits.
3. Every selection returns Home. The app has this as the person's setting ("go back to Home after a word", access.return_home); SPEC §5.6 item 6 makes it the person's choice, never the board's. So the set does not turn it on: its manifest says it is made for it (ext_aurasense_pageset.recommends.return_home_after_word), and the chooser asks when the set is added.
4. The place that opened a page closes it. On a page, the cell in the place of the key that opened it is Back, to the board that key is on (from a group, its page; from a page, Home). A wrong page is undone by pressing the same place again — the most automatic movement there is — at no cost of a separate location, and it is the same rule on every page and group at every size. It is labelled Back, not Home, with the app's arrow and the control's word in each language: the Places page has the word home, and two buttons reading alike must never mean two things on one page (a build check; it is also why the body part back is not in the set).
5. The same 472 words at every size: 30, 60 and 84 locations differ only in how many words are one hit away.

The patent constraint

Three PRC patents on multi-meaning symbol sequences are listed as active to about 2033–34 (US 9,202,298; 9,183,655; 9,239,824). Fixed Paths has no multi-meaning icons and no icon sequences:

• No picture stands for different words depending on what was pressed before it. A word's button shows that word; a page key shows the page's name.
• A page key's picture is an exemplar of the words behind it (Family shows "mom"), as folders do in every grid app; no two keys share a picture, and a key's label is never a word on the page it sits on (build check).
• A location's meaning never depends on a previous selection: on Home it means one thing; on a page, that page's cell means one thing.

This is the ordinary navigation of every dynamic-display AAC app (pages of words), not semantic compaction. It is a design statement, not legal advice; a freedom-to-operate review is still the founder's call before anything like icon sequences is ever added.

Not LAMP, not Speak for Yourself

• LAMP and Unity reach words by sequences of multi-meaning icons on one fixed keyboard. Fixed Paths has none: two-hit words sit on pages of explicit words.
• Speak for Yourself gives each core word its own secondary page of related words. Fixed Paths groups by meaning (twelve pages: Social, People, Actions, Describe, Little words, Feelings, Food, Places, Things, Body, Time, Word forms), and which words are on Home is computed from frequency, not chosen as a set of keys.
• TD Snap Motor Plan is taxonomic too; its layouts and words are not used. Its published averages (2.3 hits on 30 locations, 2.0 on 66) are a reference point: Fixed Paths averages 2.36 on 30, 2.00 on 60, 1.86 on 84 (unweighted, §5).
• Every position comes from the optimiser below and the committed layout; the page and group names are plain English.

2. Frequency × motor cost

Weight of a word

A word's weight is how often it is likely to be chosen: its spoken frequency (FrequencyWords 2018, OpenSubtitles dialogue, CC BY-SA 4.0), times 4 (aac_boost 3) if it is in Universal Core or Banajee et al. (2003). Pure adult spoken frequency would put "the", "of" and "to" ahead of "want", "more" and "help"; the boost is the evidence that early communicators need the latter, from the two open lists made for them. A message (a phrase) weighs a quarter of its proxy word ("I'm not finished" → "finished"). The metrics show both weights.

Home at 30 locations is for a person starting out (decision of 2026-10-05): its 18 words are the ones that carry a message on their own, from Universal Core and Banajee et al. (2003) — I, you, it, that, what, want, more, go, help, stop, not, like, all done, yes, no, my, look, do — and the most frequent function words ("the", "a", "to", "is", "in", "on") are one page away (Little words). At 60 and 84 Home keeps those 18 where they are and adds the heaviest of the rest by weight, to 48 and 72. The sets nest: a word on Home at 30 is on Home at 60 and 84.

What the beginners' Home costs, measured on the 30-location set (by spoken frequency; § 5): a spoken word takes 1.84 hits on average instead of 1.67 with the heaviest words on Home, and 9 of the 25 most-spoken words are one hit instead of 16 — the words a fluent speaker uses most, for the words a person starting out needs most. At 60 and 84 the difference is under 2% (the beginners' words are among the heaviest there anyway).

What "easy" means

Each model is scaled 0–1 over the grid and blended per size (pslib/costs.py); none is a measurement of a person.

• Touch: Fitts' index of difficulty, log2(1 + D/W), from a hand resting at the bottom centre below the board to the cell, W the cell's smaller side, on a 10.2-inch tablet in landscape. Easy: low and central.
• Scanning: row-column steps from the top left (rows highlighted until the row, then cells until the cell, both presses counted). Easy: the top left. The app's opt-in adaptive scan order (SPEC §5.4) can put heavy rows first without moving a cell, so position matters less here than for touch.
• Gaze: distance from the centre of the board, 0 at the centre, 1 at a corner. A webcam's gaze estimate is least accurate towards the edges, and a target at the edge has nowhere to put an overshoot. Easy: central, large.

They disagree: touch wants the bottom centre, scanning the top left, gaze the middle. Blends: 30 locations (the gaze size) gaze 0.5, touch 0.25, scan 0.25; 60 and 84 touch 0.5, scan 0.25, gaze 0.25.

Placement

Home items live at points, like the Growing Core's words (DESIGN-core.md §1), on the ladder 5×6 → 6×10 → 7×12, so an item on Home at 30 stays in the same region at 60 and 84 and a person moving up keeps their Home motor plan. The solver (pslib/solve.py) minimises Σ weight × blended cost over the three sizes; a page key weighs the sum of the words behind it at that size. Pages are then laid out per size: heaviest words in the cheapest cells, the Back cell where the key was.

Groups. When a page has more words than places (always at 30), it frees places one at a time at the least weight: opening a group moves that group's two lightest words behind its key (two words out, one key in), and an open group takes its next lightest. So the heaviest words stay on the page whatever their group, the lightest of a meaning (Family, Moving, Colors, …) go behind its key, and no key ever stands for a single word. The Social page keeps its repairs on the page at every size, however full: wait, I'm not finished, that's not what I meant, say it again, I don't know.

Page	Words	Groups (by meaning)	Groups in use, by size
Social	41	Answers, Manners, Greetings, Turn taking, Likes, Pauses, Exclamations	30: Exclamations, Greetings, Pauses, Turn taking; 60: none; 84: none
People	43	Pronouns, Owners, Family, Helpers, Others	30: Family, Helpers; 60: none; 84: none
Actions	101	Wanting, Doing, Trying, Moving, Thinking, Talking, Daily	30: Daily, Doing, Moving, Talking, Thinking, Trying, Wanting; 60: Daily, Doing, Moving, Talking, Thinking, Trying; 84: Daily, Moving
Describe	72	Amount, Opinions, Size and feel, Colors, Numbers	30: Colors, Numbers, Opinions, Size and feel; 60: Colors, Size and feel; 84: none
Little words	51	Questions, Joining, Linking, Position, Negatives	30: Joining, Linking, Negatives, Position; 60: none; 84: none
Feelings	24	–	30: none; 60: none; 84: none
Food	26	Flavors	30: none; 60: none; 84: none
Places	17	–	30: none; 60: none; 84: none
Things	45	Play things, Home things, Clothes, Animals, Outside things	30: Animals, Clothes, Home things, Outside things, Play things; 60: none; 84: none
Body	14	–	30: none; 60: none; 84: none
Time	25	–	30: none; 60: none; 84: none
Word forms	21	–	30: none; 60: none; 84: none

Home at each size (▸ opens a page):

30 (5x6)

| help        | Time▸       | yes         | Things▸     | stop        | Body▸ |
| more        | my          | People▸     | Word forms▸ | like        | Feelings▸ |
| not         | Describe▸   | Little wor~ | Actions▸    | Social▸     | Places▸ |
| go          | what        | you         | I           | no          | all done |
| look        | do          | that        | it          | want        | Food▸ |

60 (6x10)

| help        | have        | this        | we          | yes         | up          | Things▸     | stop        | was         | Body▸ |
| more        | good        | Time▸       | my          | here        | Word forms▸ | like        | when        | your        | Feelings▸ |
| why         | she         | can         | People▸     | in          | Actions▸    | Social▸     | and         | for         | Places▸ |
| out         | not         | Describe▸   | what        | Little wor~ | I           | on          | get         | who         | be |
| of          | go          | all         | is          | you         | a           | he          | no          | where       | all done |
| me          | look        | do          | that        | the         | it          | to          | want        | some        | Food▸ |

84 (7x12)

| help        | have        | now         | this        | we          | yes         | him         | one         | stop        | got         | her         | Body▸ |
| more        | good        | Time▸       | there       | off         | just        | up          | Things▸     | when        | was         | your        | Feelings▸ |
| why         | they        | she         | my          | People▸     | here        | with        | Word forms▸ | like        | at          | come        | Places▸ |
| out         | not         | Describe▸   | can         | don't       | in          | Actions▸    | know        | Social▸     | and         | for         | well |
| of          | go          | all         | what        | are         | Little wor~ | I           | on          | get         | who         | where       | be |
| right       | look        | but         | do          | is          | you         | a           | he          | no          | if          | put         | all done |
| oh          | me          | so          | make        | that        | the         | it          | to          | want        | about       | some        | Food▸ |

Stability of Home across sizes:

Home items present at two sizes: 120 pairs; mean shift 0.072 of the board; largest 0.79 cells of the smaller board (Things, 6x10 → 7x12); every item's buttons share a point: yes.

The trade-offs, measured

Each size, laid out four more ways — the blend alone (this size only, no tie to the other sizes), and each access method alone — then every layout measured under every method (design-weighted mean per word; hits are the same for all, since which words are on Home does not depend on where):

Locations	Layout	Hits	Scan steps	Touch (bits)	Gaze
30	chosen	1.614	10.488	3.208	0.502
30	blend alone	1.614	10.272	3.292	0.483
30	touch only	1.614	12.351	2.646	0.798
30	scan only	1.614	6.314	4.152	0.997
30	gaze only	1.614	10.498	3.324	0.478
60	chosen	1.178	11.149	2.371	0.486
60	blend alone	1.178	11.203	2.312	0.512
60	touch only	1.178	12.314	2.225	0.597
60	scan only	1.178	6.555	3.228	0.717
60	gaze only	1.178	10.543	2.739	0.383
84	chosen	1.113	12.455	2.383	0.486
84	blend alone	1.113	12.604	2.298	0.49
84	touch only	1.113	13.699	2.21	0.576
84	scan only	1.113	6.692	3.33	0.704
84	gaze only	1.113	11.687	2.796	0.331

What it shows:

• A scan-only layout needs 40–46% fewer scan steps than the chosen one (6.3–6.7 against 10.5–12.5) and is the worst for touch and gaze.
• A touch-only layout is 6–18% cheaper to touch and the worst to scan.
• Gaze-only is within 5% of the chosen layout at 30 locations, where gaze weighs most; at 60 and 84 it would be 21–32% better for gaze and 16–17% worse for touch.
• Tying the three sizes together (chosen against the blend alone) costs at most 4% on one measure (gaze at 30, touch at 84) and is better on others.

One layout for every access method (decision of 2026-10-05): the layout is the motor plan, and stays; a person who scans can opt into the app's adaptive scan order (SPEC §5.4), which puts the rows they use first without moving a cell.

3. Vocabulary filter, and "show everything for now"

A filter shows only its words, each in its place; every other word is hidden in place (OBF hidden: true, the app's Cell.hidden). A page or group key shows while any word behind it shows; Back cells always show. Positions never change (test). The starter filter is Universal Core ∪ Banajee et al. (2003): 45 words for a person starting out. Filters are lists in fixed-paths.yaml, so a team can write its own; the chooser could also build one from the words a person used (the app's usage counts), never by moving anything.

"Show everything for now" (Babble, in other apps' words) is Track B's "Show hidden words for now", in the partner menu and the edit bar, and scanned (decision of 2026-10-05): it shows every hidden word in its place until it is turned off, and changes nothing on the board. A word hidden because it has no word in the person's language stays hidden (DESIGN-core.md §4).

4. How the app offers it

As for the Growing Core (DESIGN-core.md §5): at setup and in Settings → Words and boards, downloaded once, checked, kept on the device, and added as one undoable change. Then:

1. Choose 30 places (the largest targets, for eye gaze and head pointers), 60 or 84; when a size's targets come under 10 mm on this screen, the chooser says so with the measured size.
2. "Start with the 45 first words" picks the starter package (the filter's hidden flags are in it: fixed-paths-<n>-starter.obz).
3. Its Home becomes the person's home board (going back to Home goes there), and the app asks once, in words: "Fixed Paths is made to go back to Home after each word … Turn that on?" — at setup as part of the choice, in Settings after the set is added; access.return_home is set only on yes, and the question is not asked again (SPEC §5.6 item 6).

5. Metrics

Locations	Grid	Words on Home	Pages	Groups	Words	1 hit	2 hits	3 hits	Mean hits	Hits per word, spoken weight	Hits, design weight	Top 25 in 1 hit	Top 100 in 1 hit	Top 100 in reach	Top 200 in reach	Spoken tokens covered
30	5x6	18	12	26	473	18	268	187	2.357	1.837	1.614	9/25	14/100	99/100	156/200	73%
60	6x10	48	12	8	473	48	377	48	2.0	1.397	1.178	25/25	44/100	99/100	156/200	73%
84	7x12	72	12	2	473	72	395	6	1.86	1.259	1.113	25/25	65/100	99/100	156/200	73%

"Top 100 in reach" counts words the set holds as words: 99 of the 100. "Going" is "go" and the -ing ending. Beyond the top 100, metrics.json lists every size's coverage of the top 25 to 400.

Spoken-weighted cost per word, by method:

Locations	Hits per word	Row-column scan steps	Touch (Fitts ID, bits)	Gaze (eccentricity)
30	1.837	11.636	3.78	0.605
60	1.397	12.961	2.936	0.566
84	1.259	13.826	2.862	0.548

Target sizes (mm, width × height), against the app's minimums (touch 10, switch 15, head 18, gaze 20):

Locations	Grid	10.2" tablet, landscape (mm)	tablet, portrait	6.1" phone, portrait	phone, landscape	Meets on the tablet (landscape)	Meets on the phone (portrait)
30	5x6	34×24	25×34	10×19	20×6	touch, switch, head, gaze	none
60	6x10	20×20	15×28	6×16	12×5	touch, switch, head	none
84	7x12	16×17	12×24	5×13	10×4	touch, switch	none

30 locations meets the app's 20 mm gaze minimum on a 10.2-inch tablet in landscape; 60 and 84 do not. No size meets touch's 10 mm on a 6.1-inch phone held upright (30 locations misses by a fraction of a millimetre): Fixed Paths is a tablet set.

6. Decisions (2026-10-05)

Taken by the coordinator for the founder, and reversible:

1. Home at 30 is for beginners (§2), with the frequency × 4 weighting kept for 60 and 84.
2. One layout for every access method, with the adaptive scan order for scanners (§2).
3. "Show everything for now" is Track B's "Show hidden words for now" (§3).
4. "Back", the word, has its place (Little words → Position at 30, Little words at 60 and 84, by its weight): the Back control is drawn as its arrow, so the two never read alike. Its glosses are the body part, so it is English only and hidden in place in other languages (DESIGN-core.md §2).

Still open: irregular forms sit on the Word forms page as words until the UniMorph plan (DESIGN-core.md §3) gives a Word Forms page that re-inflects; until then "went" is two hits and "goed" is what the ending says.