Using textweaver with a screen reader

This guide is for people who use a screen reader, such as JAWS, NVDA, VoiceOver, or Orca, and want to use textweaver's terminal reader alongside it. It explains the three accessibility modes, how to stop hearing things twice, what your screen reader can read, the settings to try in NVDA and JAWS, which terminals to use, and which keys may clash. It also covers braille displays, with a checklist for a 40-cell display, and the app, and ends with a checklist for trying each mode.

textweaver's keys are the same in every mode. See Reading and moving around and the keyboard reference.

What this guide says about textweaver comes from its code. Screen reader and terminal settings are marked not yet verified until they have been tried by ear. Where something is untested, please tell us how it goes (see Troubleshooting, "Report a bug").

Three modes

textweaver can speak with its own voice, write on its status line, or both. A screen reader that reads new text in a terminal speaks the status line too. The accessibility mode decides which of the two each kind of output uses, so you hear it once.

Self-voicing

textweaver speaks everything with its own voice: reading, messages such as "Paused." or "Heading level 2: Methods", typing echo, and the word or line a cursor key reaches. Each message is also written on the status line.

This is the default, and it is what textweaver always did. Choose it when you use textweaver without a screen reader, or when you quiet your screen reader in the terminal (see Avoid hearing things twice).

Hybrid

textweaver reads documents aloud: continuous reading, and reading a character, word, sentence, line, paragraph, or selection. That reading includes the narration only textweaver can do: math in words, table rows with their headers, citations, and structure such as "heading level 2" and "list with 3 items".

Your screen reader does the rest from the status line and the cursor: messages, typing echo, and cursor moves. textweaver does not speak those, and it does not copy what it reads aloud onto the status line.

Hybrid is recommended when a screen reader is running. You keep textweaver's voice and rate for long reading, and your screen reader's voice for everything else.

textweaver never speaks on its own. Everything goes to the status line, written for your screen reader:

Messages, moves, and questions, as in the other modes.
The word, line, or character a cursor key or a "say" key reaches.
Text you ask to hear, such as the sentence (.) or the line (Alt+Shift+L), narrated as textweaver would say it: math in words, and tables with their headers.

Continuous reading still works. By default, Space moves through the text a sentence at a time. Each sentence goes to the status line for your screen reader, the cursor moves with it, and the next one follows after the time textweaver's rate allows. Press + or - to match the pace to your screen reader. A message that comes up while it reads, such as "Saved.", goes after the sentence on the status line, so the sentence stays first on your Braille display. Space pauses and resumes, and Escape stops. To have textweaver's own voice read instead, set say_all = "voice"; that is the only thing it then says.

You can also move through the text yourself, by sentence (Alt+Down, or Alt+.), line (Down), or word (Right), and your screen reader reads each place from the status line.

Choosing a mode

Alt+Shift+A cycles self-voicing, hybrid, and screen reader, and saves your choice. The new mode is announced the way the old one spoke, so you hear it.
--mode hybrid, --mode screen-reader, or --mode self-voicing on the command line sets the mode for one run without saving it. tw open --mode hybrid essay.md works too.
--no-speech starts textweaver with no voice at all, in screen-reader mode.
In settings.toml:

[accessibility]
mode = "hybrid"

The title line shows hybrid or screen reader mode when one of those is in effect.

The very first run starts with the list of interface languages instead (see Language); the question below comes on the next start.

When textweaver starts, the mode has never been chosen, and a screen reader is running, it asks once: "NVDA is running. Use hybrid mode, where textweaver reads documents aloud and your screen reader speaks messages and typing? y or n". The question stays on the status line until you answer. y switches to hybrid and saves it. n or Escape keeps self-voicing. Either way it is not asked again until you choose Tools, Ask again about first-run choices (or set hybrid_offered = false under [accessibility]); it then asks at the next start.

How textweaver notices a screen reader:

Windows: the system's screen reader flag, which NVDA, JAWS, and Narrator set, and the running programs nvda.exe, jfw.exe (JAWS), and narrator.exe, which also give its name.
macOS: VoiceOver's setting (defaults read com.apple.universalaccess voiceOverOnOffKey).
Linux: GNOME's screen reader setting, the accessibility bus's "screen reader enabled" flag, or a running orca.

If textweaver gets it wrong, set TEXTWEAVER_SCREEN_READER=0 to say there is none, or TEXTWEAVER_SCREEN_READER=NVDA (any name) to say there is one.

Avoid hearing things twice

Use hybrid mode (Alt+Shift+A). textweaver reads; your screen reader speaks the rest.
Use screen-reader mode for one voice for everything, or with a braille display.
Stay in self-voicing and quiet your screen reader while you are in textweaver. NVDA's speech modes (NVDA+S) include "on demand" in recent versions, which speaks only when you ask, for example to read the current line. JAWS has a way to turn speech off (Insert+Space, then S) and, in recent versions, a Speech on Demand setting. These belong to your screen reader; check its own documentation. Not yet verified.

While textweaver reads aloud, the terminal's cursor moves to each word. If your screen reader speaks as the cursor moves, it talks over the reading. Two settings help:

quiet_screen = true keeps the screen still while textweaver reads continuously: the title line's position stops updating, and the text being read is not copied to the status line. It is already on in hybrid mode unless you set it to false.
cursor = "status" parks the cursor on the status line instead of on the spoken word.

[accessibility]
quiet_screen = true
cursor = "status"

With cursor = "status", the cursor waits at the start of the status line, as it did in star, so your screen reader's "read current line" (NVDA+Up, JAWS Insert+Up) repeats the last message. A prompt, such as Find, still puts the cursor where you type.

textweaver is built so that everything it would say is also on the screen, in screen-reader mode, and for the parts left to your screen reader in hybrid mode.

The status line is the line above the bottom line. Messages go there: moves, modes, settings, errors, and questions such as "Quit textweaver? y or n". Screen readers that read new text in a terminal read it as it changes. While a yes-or-no question waits, it stays on the status line in front of any other message.
Math on the status line is written in words for a screen reader, as textweaver would say it: $x^2$ becomes "x squared".
Repeated messages. A terminal screen reader speaks the status line only when it changes. When the same message comes twice in a row, such as "No next heading." after pressing h twice at the end, textweaver blanks the status line for 150 milliseconds first, so the message changes and is read again.
The cursor. With cursor = "follow", the default, textweaver parks the terminal's cursor where your attention is: on the chosen item of a list, where you type in a prompt, on the Speech Cursor line, on the word being read, or else on the reading cursor. Screen readers, braille displays, and magnifiers that follow the cursor follow it there.
Lists. The help, the keyboard shortcuts, bookmarks, notes, the library, and the voices appear in a box over the document. When a list opens, the status line says what it is and how to use it, followed by the first item. As you press Up and Down, the status line shows the chosen item.
The prompt line. The bottom line shows the prompt and what you type, with the cursor where you type. Find, Go to, Open file, the command palette, and the note prompt all use it. In hybrid and screen-reader modes textweaver does not echo typing; your screen reader does.
Moves. After a sentence, paragraph, heading, table, link, find, go to, or bookmark jump, the status line shows where you arrived, with a preview of the text.
Cursor keys. Right and Left put the word on the status line; Down and Up put the whole line there. c puts the character's name there.
Where am I. Shift+W puts the line, the percentage, and the heading on the status line.

How much textweaver says is set by [speech] verbosity. See Reading and moving around.

Keys like a screen reader's browse mode

textweaver's default keys are the quick navigation keys of NVDA's and JAWS's browse mode, so they are already familiar:

h and Shift+H: next and previous heading.
1 to 6: next heading at that level; Shift with the digit: the previous one. textweaver matches the digit key itself, on any keyboard layout (on Windows it reads the key from the console; elsewhere it knows the US, UK, German, Spanish, Nordic, and Italian layouts, and French with digit_row = "azerty").
l and Shift+L: list. i and Shift+I: list item.
t and Shift+T: table.
k and Shift+K: link.
q and Shift+Q: block quote.
s and Shift+S: separator (a horizontal rule).
g and Shift+G: graphic (an image).
d and Shift+D: section or chapter, the nearest thing to a landmark.
Backspace: go back after a jump.
Alt+Down and Alt+Up: next and previous sentence, as in JAWS. Ctrl+Down and Ctrl+Up: next and previous paragraph.

These follow NVDA's letters; where JAWS differs (its q, d, r, and link keys), the keyboard reference says how. Quitting is Ctrl+Q, and it still asks first.

The earlier keys are the classic preset. The screen-reader preset is now the default, and naming it still works:

[keyboard]
preset = "classic"    # the earlier keys; "default" (or "screen-reader") for these

Your keymap.toml overrides apply on top. The keyboard reference lists every change and the classic preset.

These are the settings to look at in NVDA and JAWS for each mode. Names and keys differ between versions. Every item here is not yet verified by ear.

NVDA

Report dynamic content changes (NVDA+5, in Object Presentation settings). On for hybrid and screen-reader modes: that is how NVDA reads the status line as it changes. Off, or speech on demand, for self-voicing.
Speak typed characters (NVDA+2) and Speak typed words (NVDA+3), in Keyboard settings. On for hybrid and screen-reader modes, where textweaver leaves typing echo to NVDA. Off for self-voicing, where textweaver echoes.
Speech mode (NVDA+S): talk, beeps, off, and, in recent versions, on demand. "On demand" suits self-voicing: NVDA stays quiet until you ask it to read.
Windows Terminal: in Advanced settings, NVDA can read new text in Windows Terminal by "diffing" or by UIA notifications. Try both if new text is missed or read late.
Windows Console support (Advanced settings): UIA or legacy, for the classic console.
Read current line: NVDA+Up (laptop layout NVDA+L). With cursor = "status" it repeats the last message.

JAWS

Screen echo (Insert+S cycles none, highlighted, and all). "All" for hybrid and screen-reader modes, so JAWS reads the status line as it changes. "None" for self-voicing.
Typing echo (Insert+2 cycles characters, words, both, and none). On for hybrid and screen-reader modes; none for self-voicing.
Speech on and off: Insert+Space, then S. Recent versions also have Speech on Demand in Settings Center, which suits self-voicing.
Read current line: Insert+Up. With cursor = "status" it repeats the last message.
JAWS's PC cursor follows the terminal's cursor; with cursor = "follow" it moves with the spoken word while textweaver reads.

VoiceOver (macOS)

textweaver runs in Terminal.app. VoiceOver reads new text in Terminal as it appears. Basic functionality has had some testing; more extensive testing is planned.
textweaver's Alt chords, such as Alt+P and Alt+Shift+A, need the Option key to act as Alt (Meta). In Terminal, open Settings, Profiles, Keyboard, and turn on "Use Option as Meta key". Untested.
VoiceOver's own keys use Control+Option; textweaver's terminal keys use no Ctrl+Alt chords, so there is no clash by default.

Orca (Linux)

textweaver runs in GNOME Terminal and other terminals. Orca reads new text in terminals as it appears. Basic functionality has had some testing; more extensive testing is planned.
Orca's typing echo is in Orca Preferences, Echo: on for hybrid and screen-reader modes, off for self-voicing.
Orca also says each key you press by default, including modifiers ("left control", "space"), so every textweaver command is announced before textweaver answers. To hear only textweaver's answer, turn off key echo in Orca Preferences, Echo ("Enable key echo"), or leave only the kinds of keys you want spoken ticked. This is Orca's own setting; textweaver does not change it.

Terminals

textweaver runs in any terminal that sends key presses in the usual way.

Windows Terminal or the classic console

textweaver runs in Windows Terminal and in the classic console (conhost, the window cmd opens by default).

Windows Terminal exposes its text through UI Automation, which recent NVDA and JAWS versions use. The classic console is older and well known to screen readers. Try both; which works better with JAWS and NVDA for each mode has not been recorded yet. Not yet verified.
Paste with Ctrl+V in Windows Terminal, or with right-click in either.

Windows Terminal keys that clash

Windows Terminal keeps some keys for itself, so textweaver never sees them. These come from Windows Terminal's defaults and the settings file a new installation writes (checked against Windows Terminal 1.24):

Alt+Left and Alt+Right move between panes. In textweaver they are history back and forward. Use Backspace and \ instead (Shift+H and Shift+L with the classic preset), or unbind them in Windows Terminal. With only one pane open, Windows Terminal may pass them on; not yet verified.
Alt+Up and Alt+Down move between panes. In textweaver they move by sentence (JAWS's keys). Alt+. and Alt+, do the same and never clash; or unbind them in Windows Terminal. (With the classic preset they step through notes; e and Shift+E, or F12 and Shift+F12, do that too.)
Alt+Shift+Up and Alt+Shift+Down resize panes. In textweaver they make RSVP faster and slower; Alt+Shift+PageUp and Alt+Shift+PageDown do the same and reach textweaver.
F11 and Alt+Enter switch full screen. In the terminal reader F11 is the next chapter (in the textweaver app it is full screen). Use Alt+PageDown and Alt+PageUp for chapters.
Ctrl+C copies when text is selected in Windows Terminal; otherwise textweaver gets it and copies. Ctrl+V pastes, which textweaver takes as pasted text.
Alt+Shift+D, Alt+Shift+minus, and Alt+Shift+plus split the window into panes. So the terminal adds a reference by DOI or ISBN with Alt+B (the app keeps Alt+Shift+D), and "add reference" is in the command palette (F2). Pitch is Alt+= and Alt+-: pressing Shift by mistake splits the window.
Ctrl+Alt+Left moves to the previous pane in Windows Terminal 1.24 (Terminal.MoveFocusPrevious in its defaults). In textweaver it is the previous cell in a table row; with one pane open Windows Terminal may pass it on, not yet verified, or run table previous column from the palette or give it another key in keymap.toml. The other Ctrl+Alt arrows are not bound by Windows Terminal, but some graphics drivers rotate the screen with them, and a screen reader may keep them for its own table commands.
textweaver's newer chords were checked against the same list and do not clash: Alt+Shift+Q (citations), Alt+Shift+X (explore math), Alt+Shift+Z (syllables), Alt+Shift+J (difficult words), Alt+B (add a reference), F12 and Shift+F12 (notes), Ctrl+Down and Ctrl+Up (paragraphs), and Alt+Shift+PageUp and Alt+Shift+PageDown (RSVP).
Alt+Space opens the window menu. textweaver does not use it.
Ctrl+Shift chords (new tab, close pane, find, scroll) and Ctrl+Alt with digits (switch tabs) do not clash: terminals cannot send textweaver Ctrl+Shift chords, and textweaver's terminal keys use no Ctrl+Alt digits.

To unbind a key in Windows Terminal, open its settings (Ctrl+comma), choose Actions, and remove the key from the action; or add this to the actions list in its settings.json, one line per key:

{ "command": "unbound", "keys": "alt+left" }

Not yet verified which of the two works in your version.

macOS and Linux

See VoiceOver and Orca above.

Keys that may clash

The screen reader key. NVDA, JAWS, and Orca use Insert, or Caps Lock if you choose, as their own modifier key. textweaver uses neither, so your screen reader's commands pass through to it as usual.
Single-letter keys. textweaver's reading keys are single letters and punctuation, such as ., h, and t, like a screen reader's browse mode in a web page. Screen readers do not use their own browse mode in a terminal, so these keys reach textweaver. If a letter does nothing, check that your screen reader is not in a special mode, such as a virtual or review mode.
Dictation and speech recognition. Dictation software types letters, and a letter is a command in textweaver. Press F9 to turn single-key shortcuts off. You hear "Single-key shortcuts off." Chords with Ctrl or Alt, the arrow keys, the function keys, and the command palette (F2) keep working. Press F9 again to turn them back on.
VoiceOver uses Control+Option as its modifier. textweaver's terminal keys use no Ctrl+Alt chords on macOS, so there is no clash by default.
Ctrl+Alt+V chooses a voice, and Ctrl+Shift+V pastes as plain text in edit mode. Neither is a default command of NVDA or JAWS. Most terminals keep Ctrl+Shift+V as their own paste, which reaches textweaver as plain text anyway.
Windows Terminal: see the keys it keeps.

Any key can be changed in keymap.toml. The keyboard reference explains how, and lists what terminals cannot send.

Language

textweaver's own words (messages, lists, help, and settings) are in English, Spanish, French, German, Portuguese, or Arabic. Your documents are read in their own language whatever you choose.

The first run shows a list of languages, each in its own name ("Español", "Français"), your system's language first. Up and Down move, Enter chooses, Escape keeps English.
Later, open the settings screen (Shift+F10), type language, and press Enter or Right on "Interface language". The change is immediate: you hear it confirmed in the new language, then the title line. From the command line: tw settings language es.
The voice follows the language when the speech engine has a voice for it: a Spanish voice for Spanish. When the engine has no voice for the language, the current voice keeps speaking, and textweaver says so. It never goes silent. For example, if your Eloquence has English only, choosing Spanish keeps Eloquence and says that no Spanish voice was found. [speech.voices_by_language] in the settings chooses the voice per language.
Keys keep their written names (Ctrl+S) on the status line and on your Braille display; textweaver's own voice says them in your language.
Answers to yes-or-no questions stay y and n in every language, since they are the keys that answer.
Right-to-left languages. Arabic messages are sent to your screen reader in reading order. On screen, Windows Terminal and the classic console do not support right-to-left text, so sighted helpers may see Arabic reversed there; your screen reader and Braille display are not affected. On Linux and macOS, [interface] rtl reorders it for the screen where the terminal does not.
To check a translation's coverage, en-XA is a test language: every message comes out accented and in brackets, so anything plain was missed.

Braille displays

textweaver draws no braille of its own, in the terminal or in the app. A braille display shows what your screen reader shows: the line at the terminal's cursor, and new text as it appears. This section is written for a 40-cell display, the HumanWare Mantis Q40, with NVDA or JAWS on Windows.

Set up textweaver for the display

Use screen-reader mode: --mode screen-reader, or mode = "screen-reader" under [accessibility]. Hybrid mode lays lines out the same way.
Park the cursor on the status line: cursor = "status" under [accessibility]. The display then rests on the latest message, and your screen reader's read-current-line key repeats it. At a prompt the cursor moves to what you type, and comes back when the prompt closes.

[accessibility]
mode = "screen-reader"
cursor = "status"

What a 40-cell line shows

Status, title, list, and prompt lines are written to put their key fact in the first 40 cells. Tests check about thirty terminal lines from a walk through a document in screen-reader mode with cursor = "status" (crates/textweaver-tui/tests/it/braille.rs), every prompt and menu name in all six languages, and some sync and font messages; the rest follow the message style guide without a test yet.

Messages start with what happened: "Opened essay.", "Page 12, line 400: ..." A question comes before what it is about: "Exported essay.html. Open it? y or n." and "Open web link? y or n." then the address.
The title line starts with where you are, then the reading state: "Line 12 of 400, 3%, Reading". In a PDF it names the page: "Page 12 of 30, 40%". The mode, rate, and engine follow, then the document's name. In screen-reader and hybrid modes the line starts at the first cell, with no padding.
Lists say the place first: "3 of 12, Chapter two, level 2". The list's first line is its place and title: "3 of 12, Outline, 12 headings". In screen-reader and hybrid modes a list covers the whole window with no border, so no border or document text comes before an item.
Prompts show a short label, then what you type: "Find: chapter", "Go to page: 12". A long label is cut to its first part on that line ("Export settings to file"); the whole label is said, and shown on the status line, when the prompt opens.
Pages in a PDF. Say Position (Shift+W) starts with the page: "Page 12 of 30. Line 400 of 2000, 20 percent." Go To (Ctrl+G) takes a page: in a PDF a plain number is a page, and line 12 is a line. p 12 and page iv work in any document with pages, and the printed page label wins, so p 1 is the page printed "1" even after pages i to x. A PDF with no headings lists its pages in the outline (Alt+O).
Keys keep their written names (Ctrl+S) on the display; symbols such as arrows and box drawing are not used as the only signal anywhere.
A line that runs past 40 cells still reads correctly when you pan; only the start is guaranteed.

NVDA braille settings to try

In NVDA's Settings, Braille category. Every item here is not yet verified on a Mantis Q40.

Tether braille: "automatically" or "to focus", so the display follows the terminal's cursor, which cursor = "status" keeps on the status line. "To review" leaves the display where you moved it. To verify.
Show messages and Message timeout: NVDA's own messages (such as "focus mode") cover the line for a few seconds. "Use timeout" with a short timeout, or "Disabled", keeps textweaver's status line in view. To verify.
Word wrap: on, so a line longer than 40 cells breaks between words when you pan. To verify.
Focus context presentation: "fill display for context changes" is NVDA's default; try it if the start of a line is not what you expect. To verify.
Interrupt speech while scrolling: on, if you read by braille and do not want speech to go on while you pan. To verify.

JAWS braille settings to try

In Settings Center (Insert+6), Braille group. Every item here is not yet verified on a Mantis Q40.

Braille mode: Line mode shows the line at the cursor, which is what textweaver's layout is designed for. Structured mode adds control type words. To verify.
Braille follows active cursor: on, so the display follows the terminal's cursor (the PC cursor), which cursor = "status" keeps on the status line. To verify.
Flash messages: JAWS shows its own messages for a while (Flash message time). A shorter time, or messages off, keeps textweaver's status line in view. To verify.
Word wrap: on, so panning breaks between words. To verify.
Status cells: off on a 40-cell display, so all 40 cells show the line. To verify.

Checklist for the Mantis Q40

Open a document in screen-reader mode with cursor = "status". Are "Opened" and the document's title in the first 40 cells?
Press Alt+O. Does "3 of 12" come before each heading's text as you move?
Open Find (Ctrl+F). Does the display show the label, then what you type?
Press Shift+W, then go to page 12 of a PDF (Ctrl+G, 12, Enter). Do the position and the page come first?
Try the NVDA and JAWS settings above. Which ones work, and which do you keep?

What has been tried

Tried in testing: a 40-cell Mantis Q40 through NVDA and JAWS on Windows, reading the status line, messages, and moving by unit in the terminal reader, before this layout; and, in the app, the document control, the settings dialog, and edit mode, across two sessions. The layout above waits for the checklist.
Not tried: other cell widths, other display models, Orca's braille on Linux, and VoiceOver's on macOS.
The BRF writer (tw convert --to brf, see Converting documents) is a separate feature: a grade 1, or grade 2 with the liblouis feature, braille file you save and read on a notetaker or emboss, not the live display output above. A BRF from a document with math writes the math in Nemeth or UEB mathematics; see Math in braille files.
If your combination behaves differently from this, add it to the checklist below and let the project know what you found.

The app

textweaver also has an app, the graphical version, written entirely in Rust with AccessKit for screen readers. It shares documents, keys, settings, notes, and voices with the terminal reader, and it has been checked with NVDA, JAWS, and a braille display on Windows. The textweaver app covers it in full: starting it, the file chooser, edit mode, how messages reach your screen reader (a live region, or UI Automation notifications), and the spoken word's highlight. ADR-0027, ADR-0028, and ADR-0033 record what was checked.

Checklist: try each mode with JAWS and NVDA

Run through this once with NVDA and once with JAWS, in Windows Terminal and, if you like, again in the classic console. Use a throwaway state folder so your own settings are untouched: add --home %TEMP%\tw-check to each command (in PowerShell, --home $env:TEMP\tw-check). Two sample files have what is needed: fixtures/sample.md has a table, and fixtures/o/math-sample.md has math.

First run

Delete the folder %TEMP%\tw-check if it exists. With your screen reader running, start tw --home %TEMP%\tw-check fixtures/sample.md.
Expected: the question "NVDA is running. Use hybrid mode ...? y or n" (JAWS: "JAWS is running."). Your screen reader reads it from the status line, and textweaver speaks it too.
Press y. Expected: "Hybrid mode. ..." is read once, by your screen reader. Quit with Ctrl+Q, then y, and start again: no question this time.

Hybrid (`--mode hybrid`)

Screen reader settings: dynamic content (NVDA) or screen echo "all" (JAWS) on; typing echo on.

Press Alt+. (next sentence). Expected: your screen reader reads the sentence preview once; textweaver is silent.
Press . (say the sentence). Expected: textweaver reads the sentence in its voice; your screen reader says nothing more.
Press Space. Expected: textweaver reads on; the status line does not change while it reads. Press Space again: "Paused" from your screen reader.
Press Right. Expected: your screen reader says the word; textweaver is silent.
Press Ctrl+F and type ada. Expected: your screen reader echoes the typing; textweaver does not. Press Escape.
Press t to reach the table, then .. Expected: textweaver narrates the row with its headers.
Open fixtures/o/math-sample.md (Ctrl+O) and press . on a formula. Expected: textweaver says the math in words. Then press Alt+Shift+X, Down, and Right. Expected: each part of the formula is said and highlighted; Escape leaves.

Screen reader (`--mode screen-reader`, or `--no-speech`)

Press Alt+., then .. Expected: your screen reader reads the preview, then the sentence; textweaver never speaks.
In fixtures/o/math-sample.md, press . on a formula. Expected: your screen reader reads the math in words, not dollar signs and caret symbols.
Press Space. Expected: your screen reader reads the document a sentence at a time, and the cursor moves along. Press + or - to change the pace, Space to pause and resume, Escape to stop.
With say_all = "voice" under [accessibility], press Space again. Expected: textweaver's voice reads instead.

Self-voicing (`--mode self-voicing`)

Turn your screen reader to speech on demand, or turn dynamic content (NVDA) or screen echo (JAWS) off.
Press Alt+. and Space. Expected: textweaver speaks everything, once.

Quiet screen and the status-line cursor

Add quiet_screen = true and cursor = "status" under [accessibility] in %TEMP%\tw-check's settings.toml, and start in self-voicing mode with your screen reader's dynamic content or screen echo on.
Press Space. Expected: the title line's "line ... of ..." stays still while textweaver reads, and your screen reader does not chatter.
Press Escape, then Alt+., then your screen reader's "read current line" (NVDA+Up, Insert+Up). Expected: it repeats the message on the status line.

Keys

With the default keys: h and Shift+H move by heading, l by list, k by link, q by block quote, Backspace goes back, and 1 to 6 with and without Shift move by heading level, also on a non-US keyboard layout. Add preset = "classic" under [keyboard] and start again: . moves by sentence and q asks to quit.
In Windows Terminal, press Alt+Left after a jump. Expected: textweaver goes back, or Windows Terminal takes the key. Note which, and whether it changes with more than one pane.
Press Alt+Shift+A three times. Expected: hybrid, screen reader, and self-voicing, each announced once.

Please note which steps did not behave as expected, with the screen reader, its version, and the terminal.

If something goes wrong

Everything is said twice. Use hybrid mode (Alt+Shift+A), or see Avoid hearing things twice.
Your screen reader does not read the status line. Turn on its reading of new text in terminals: NVDA's "report dynamic content changes" (NVDA+5), or JAWS's screen echo (Insert+S).
Nothing is said at all. In screen-reader mode textweaver is silent by design; check that your screen reader reads new text. The title line shows the mode.
The question about hybrid mode never came, or came wrongly. See The first run with a screen reader and TEXTWEAVER_SCREEN_READER.
A key does nothing. See Windows Terminal keys that clash and Troubleshooting, "A key does nothing".
No speech from textweaver. Check the title line: silent means --no-speech or no engine. See Troubleshooting.