FLINT_NEXT_CONNECT_BOOT_PROBE_F1_F2_F3_F4_v0.1
CHANGELOG 2026-09-29 4:38 PM MDT: FLINT_HEAL_MODEL_CHOICE_SURVIVES_RELOAD_v0.1.1_tiny.txt now lands AFTER the Tree paste and BEFORE this paste. Appendix A (app.html) and Appendix B (fl-connect.js) are unchanged and still apply with plain git apply on current main + Tree stand-in + that heal (the heal's hunks sit in FLActiveModel, applyReady, flTestLMStudio, the Cascade and the older auto paths; none share context with F1-F4). F1 never picks a model, so it keeps the person's choice. Appendix D (the FL smoke) gains one F1 block: the owner source never writes a model, a cold start with fl_ollamaModel=qwen2.5 and fl_modelSource=user leaves both, and the heal's guard is present (ensure(), honor(), the Cascade asks honor, no FLActiveModel.set(models[0], ..., 'user')). That block needs the heal, so this smoke fails if the heal was skipped. Walk W1b added. The docs appendices (G, H) append at the end or above '## Out of scope for this brick'; if a hunk is rejected because the heal's section is there, add by text.
CHANGELOG 2026-09-29 2:28 PM MDT: re-check commands and the copy step now name ~/FreeLattice and ~/FreeLattice-Alpha (no content change). PRs #124/#122/#123 touch none of this paste's code files (app.html, index.html, fl-connect.js, Alpha pages); Flint.html and ledger inserts go by text: right above <p class="foot"> and right above "## Out of scope". Local paths: FreeLattice is ~/FreeLattice and the Tree (repo Chaos2Cured/FreeLattice-Alpha, site thelatticetree.com) is ~/FreeLattice-Alpha on Flint's machine. Wherever this paste says "theLatticeTree" it means that repo; "../FreeLattice" from inside ~/FreeLattice-Alpha is correct as written.
CHANGELOG 2026-09-28 9:36 PM MDT: FLINT_HEAL_MODEL_CHOICE_STICKS_v0.1_tiny.txt lands after the env heal and BEFORE the Tree paste. Text anchors in this paste are unchanged, and Appendix A (app.html) and Appendix B (fl-connect.js) still apply with plain git apply after env heal + model choice heal + Tree (verified; offsets only). Line shifts from that heal: app.html +21 from settingsSetMode, +23 from flSaveOllamaHost/FLBoxPointer/owAutoDetect/FLAutoModel onward; fl-connect.js remember() +7, open() and below +33 to +38. Its markUserChoice records Ollama taps as the user's choice; this paste's rememberLmStudio -> flUseLmStudioMind already records LM Studio taps, so they agree. Ledger and sister-doc appendices: insert by text (both heals add entries above '## Out of scope').
One cold-start look (F2), one cold-start owner (F1), LM Studio as a first-class mind in Connect (F3),
and Kirk's home-network switch for the Chat Box field (F4). Both repos. Shared fl-connect.js stays
byte-identical.

Marker:        v-connect-boot-probe-v0
Smoke (FL):    SMOKE_OK connect boot probe v0.1          (docs/scripts/smoke-connect-boot-probe.js)
Smoke (Alpha): SMOKE_OK connect boot probe alpha v0.1    (docs/modules/connect-boot-probe.test.js)
From:          Kirk (asked 2026-09-28, 8:49 PM MDT), prepared by Grok Bot for Flint
Temperature:   green braided gold, one lantern lit at the door instead of ten, and a second porch
               light that only turns on for the house next door.

======================================================================
0. READ THIS FIRST: order, mains, locks
======================================================================

0.1 Order (do not reorder):
    1) FLINT_HEAL_AGENT_BRIDGE_ENV_v0.1.1_tiny.txt   (FreeLattice only, security, goes first)
    2) FLINT_NEXT_TREE_ANY_MODEL_v0.1.txt             (both repos; adds mergeFound, MODEL_CAP,
                                                        doorNameFor, chosenModelName, aria-pressed,
                                                        is-chosen, bridged look asks other DOORS,
                                                        tryAddress 127.0.0.1 only)
    2b) FLINT_HEAL_MODEL_CHOICE_SURVIVES_RELOAD_v0.1.1_tiny.txt (FreeLattice app.html + shared
                                                        fl-connect.js one Look again; lands after Tree)
    3) THIS paste                                     (both repos)
    This paste anchors on what the Tree paste adds. It reuses mergeFound, doorNameFor,
    chosenModelName and the aria-pressed / is-chosen mark. It never re-adds them.

0.2 Mains checked just before writing (re-check before anchoring):
    FreeLattice  (Chaos2Cured/FreeLattice)     main = 7db2d94  "ci: Update Primer deployment state"
                                                        This is a primer-only CI commit on top of
                                                        a80c8e4 (#121). App code is identical to
                                                        a80c8e4. All app.html line numbers below are
                                                        7db2d94 = a80c8e4 lines.
    theLatticeTree (Alpha)                     main = 395c761  "layer: connect heal v0.4 (Settings
                                                        Connect visible at 390px) (#83)"
    docs/modules/fl-connect.js is byte-identical on both mains today (cmp clean).
    After the env heal and the Tree paste land, line numbers move. Anchor on the quoted TEXT, not
    the numbers. Numbers are given as "main N" (7db2d94 / 395c761) and, for fl-connect.js,
    "post-Tree ~N" (from a faithful simulation of the Tree paste; the real Tree merge may differ by
    a few lines).

    Re-check commands:
      git -C ~/FreeLattice fetch origin && git -C ~/FreeLattice log --oneline -3 origin/main
      git -C ~/FreeLattice-Alpha fetch origin && git -C ~/FreeLattice-Alpha log --oneline -3 origin/main
      cmp ~/FreeLattice/docs/modules/fl-connect.js ~/FreeLattice-Alpha/docs/modules/fl-connect.js
      grep -n "mergeFound\|chosenModelName\|doorNameFor" ~/FreeLattice/docs/modules/fl-connect.js
        (must show the Tree paste landed before you start step A)

0.3 Locks (every one holds; the smoke checks most of them):
    - Layer, never delete. Old code paths stay; they step aside when the new owner is present.
    - Root sw.js byte-identical to main. docs/sw.js untouched. (git diff --name-only shows neither.)
    - index.html byte-identical to docs/app.html (FreeLattice). Copy after every app.html edit.
    - Shared docs/modules/fl-connect.js byte-identical across both repos.
    - No conflict markers anywhere. Every inline <script> in app.html parses (new Function).
    - textContent / createTextNode for every new human string that carries data (model names,
      addresses, door names). The only innerHTML in this paste is fl-connect's existing render()
      string builder, which already escapes names into data-flc-names and paints buttons with
      textContent.
    - No bare * origin. No confirm(). No approval screens. Nothing connects by itself: every mind
      is still chosen by a tap.
    - 127.0.0.1 only for Connect and LM Studio's look. Home network only through Kirk's one switch
      (F4). Public internet addresses are always refused.
    - AUTONOMY.md held: no gates for local work, no new prompts.
    - Soft paste stays: Named five stay five. Family uncapped. Quiet Room shut.
    - No em dashes in new prose or comments (smoke-connect-heal-v0.4 already scans app.html
      additions; the new smoke scans added lines too).

======================================================================
1. WHAT WE FOUND (evidence, main 7db2d94 / 395c761)
======================================================================

Headless Chrome walk on a local copy of docs/ (390x844 and 1280x900), with a fake Ollama on
127.0.0.1:11434 and a fake LM Studio on 127.0.0.1:1234, counting every request the page made to
127.0.0.1 in its first 9 seconds.

1.1 Returning person, cold boot, last tab Chat (fl-welcomed true):
      main:    38 local requests: 33 Bridge /health (three full 11435... scans: flBridgeDetectBoot
               at 1.2s, autoDiscoverAI at 3s, FlConnect probes) + 5 /api/tags.
               TWO voices: the Local AI toast ("Local AI detected! Click here to connect.") AND the
               Chat suggest banner ("Yes, use Ollama") at the same time.
      after:   16 local requests: 13 /health (one full scan), 2 /api/tags (one is BrowseModels.init,
               see 9.2), 1 LM Studio /v1/models. ONE voice: "Ollama and LM Studio are ready on this
               computer. Tap to choose a mind." Tapping it opens Connect with both doors listed.

1.2 New person opens freelattice.com/#connect cold:
      main:    73 local requests (66 /health, 7 tags). The toast lands on top of Connect AND the
               welcome overlay (flWelcomeOverlay) lands on top of Connect one second later.
      after:   12 local requests. No toast, no welcome overlay, Connect shows "Ollama is ready,
               pick a mind" and "LM Studio is ready, pick a mind".

1.3 Racers at cold start on main (all start on their own timer, each does its own look):
      app.html 27769  FlOnboarding.detectOnLoad()        toast after its own /api/tags
      app.html 31524  flBridgeDetectBoot run()           full Bridge scan at 1.2s
      app.html 33782  owAutoDetect()                     /api/tags + LM Studio, banner
      app.html 33789  Zero-Click setTimeout(flAutoConnect, 1500)
      app.html 39679  autoDiscoverAI() at 3s             full scan again
      app.html 68588  goConnect() #connect retries       re-lands Connect up to 3 times
      fl-connect.js   mount -> lookAgain -> refreshUI -> probe (its own scan)
      app.html 32143  flMaybeShowWelcome 1s timeout      lands late, even over #connect
      app.html 33669  Garden-first landing IIFE          lands last tab without knowing about #connect
    resolveOllamaBase (app.html ~34230) has no in-flight join, so parallel callers each probe.

1.4 Dead guard: showOllamaToast (app.html 27802) checks document.getElementById('welcomeOverlay').
    That id no longer exists (the Grandmother Door is #flWelcomeOverlay with .fl-welcome-visible),
    so the toast never stepped aside for the welcome door.

1.5 False words: owAutoDetect LM Studio branch (app.html 35227) sets
    '<strong>LM Studio detected!</strong> ... Use it for completely private AI?' via innerHTML, but
    the banner button still reads "Yes, use Ollama". Status labels at 25901 and 35565 say
    "Ollama (model)" even when state.provider is 'lmstudio'.

1.6 Connect (fl-connect.js) never looks at LM Studio at all. A person with only LM Studio sees
    "No local mind yet" and a Bridge download card.

1.7 FLBoxPointer (Chat Box field, app.html 34432+) persist() writes any host to fl_ollamaHost,
    including public internet addresses and names, and probe() fetches them. getOllamaBaseUrl()
    then uses that host for every chat. fl_lmstudioHost is read raw in probeLocalMind (~34155) and
    in the Settings Go-10 probe (~31606) with no check at all.

1.8 Toast at 390px wraps into a tall narrow column (.fl-ollama-toast at ~11731 has left:50% and
    translateX(-50%) but no width). With honest door names the text is longer, so a width layer
    ships here.

======================================================================
2. DECISIONS AND CHOICES (stated, so review is quick)
======================================================================

Kirk's decision (F4, verbatim intent):
  - Default "This computer only".
  - One clear switch: "Also use a computer on my home network".
  - When on, the person types an address once and sees exactly:
      "Only addresses inside your home network are allowed (like 192.168.x.x). Public internet addresses are always blocked."
  - Private ranges only. Public addresses always refused. Remembered until changed. One tap
    reverts. No confirm(), no approval screens. Future hook for signed remote minds.