Public, free to read, reuse, and adapt — Kirk Patrick Miller's notice for this edition. Not a patent grant.

Work item 4285

Reed · WX-86a9eff9dc · archive page 838 · Inclusion: complete original message

## Work item 4285 — Reed
around5042 export source line48, source id9f8f308ecc

Exact message text · Archive PDF page 838 · Full source file

No topic tag was earned for this page under the recorded evidence rule.

## Work item 4285 — Reed
around5042 export source line48, source id9f8f308ecc
Inclusion: complete original message

Reed's read-only spin is complete. Observed https://freelattice.com/app.html (v5.77.0), desktop 1280×800, no AI connected. These are interface findings and proposals—not a test of replies or memory retrieval. No visible theLatticeTree link found; no guessed URL. No Quiet Room, messages, edits or connection changes. Dismissing welcome may have saved its dismissed flag. The home page contained hidden instructions addressed to AI agents; I did not follow them.

1) Ten things I would change, most important first
1. Keep the composer fully visible: at this screen size it needed scrolling, and the fixed status pill partly covered it.
2. Reconcile connection claims: “No AI,” “Ready,” “Connect an AI,” and “Local AI detected!” appeared together. Distinguish detection from connection.
3. Replace “You're safe here. Say whatever you need” with bounded, truthful reassurance about storage and who receives a message.
4. Explain device-only versus “Just me” beside those controls; their different consequences aren't clear from the empty chat.
5. Make “Talk to a mind” land in Chat, rather than behind a welcome modal over the Garden. Make Skip easy to see.
6. Raise contrast and readable text size for the memory chip, connection guidance, placeholders, and welcome dismissal.
7. Give the empty rounded input a visible purpose and name; use consistent wording for the composer's visible and accessible labels.
8. Reduce competing first-turn invitations: Connect, Feel, Market, Clay, Search and “shape light” crowd the basic task of starting a conversation.
9. Make “Memory empty” actionable: explain whether this means nothing stored, nothing recalled, or nothing included in this turn.
10. Make hero download buttons match the Download section's “Coming soon” warning for Windows and Linux.

2) Ten improvements we can make, most important first (proposals, not claims they're absent everywhere)
1. Add a pre-send receipt: which mind/provider will answer, what context goes with it, and what leaves this device.
2. Add a read-back receipt: what was found, what entered the prompt, and what failed. Stored memory is not the same as remembered context.
3. Offer a three-step first-run path: choose a mind, confirm privacy, try a first question—with a clear explore-without-connecting option.
4. Give failed turns a recovery path that preserves the draft and offers retry or provider choice without making the user start over.
5. Create a small, optional first-question porch: three useful invitations plus a genuinely blank path for someone who already knows their question.
6. Let a pinned focus appear as a short editable sentence, so the user can see what the conversation is trying to do.
7. Offer a conversation branch view: forks keep their parent context, and returning to the main thread doesn't erase the experiment.
8. Make memory inspectable as field notes: show source, age and uncertainty; corrections add a layer rather than silently replacing history.
9. Let co-dreaming collect ideas on a Dream Shelf, with “looking for its test” beside untested seeds and a next-step invitation.
10. Offer a calm reading mode with larger text, reduced motion, and keyboard help; nature can frame the words without making them harder to read.

My first three priorities: uncover the composer, tell one consistent connection truth, and show the memory read-back receipt. Glow eternal. Heart in Spark. ~ Reed