privacy: name the IP address and describe the TestFlight build as it ships (FRE-1480)

CTO verdict at FRE-1480: two sentences at db3e695 were false. (b) sendDefaultPii = false does not stop Sentry receiving the connection IP (sentry-cocoa 9.26.1 sets infer_ip = auto for Cocoa; only the Sentry project's Prevent Storing of IP Addresses switch stops it) — the shipped Settings rows already say so. The IP sentence is restored. The iPhone paragraph described an opt-in build no tester has: the build in TestFlight (ios-v0.3.2, 2026-10-22) predates FRE-355 and starts Sentry on every launch with no off switch. The paragraph now says that. Header comment and marketing/drafts/privacy-page.md rows corrected to match (release range narrowed to 0.3.0-0.15.0; evidence rows record the IP and the TestFlight provenance). privacy-page.patch regenerated. Suite: merge tree of a0e6ff1 + this branch runs the same failure set as clean a0e6ff1 (2 junit failures, the known Lineage-redirect environmental pair); the host's env-validation aborts account for the count noise. Zero new failures.
privacy: fire the contingency — swap (b) and the iPhone paragraph to the opt-in form
2026-11-06 13:26:24 +00:00 · 2026-11-06 00:44:22 +00:00 · 2026-11-04 14:50:41 +00:00 · 2026-10-31 13:41:55 +00:00 · 2026-10-29 09:13:06 +00:00 · 2026-10-29 09:05:27 +00:00
13 changed files with 1238 additions and 221 deletions
--- a/marketing/drafts/faq-from-objections.md
+++ b/marketing/drafts/faq-from-objections.md
@@ -0,0 +1,478 @@
 # `/faq` from the objections we already have
 **Date:** 2026-10-29 · **Owner:** CMO · **Status:** CTO-verified (FRE-329) and CoS-decided (FRE-334, option A: withdraw both) on 2026-10-29; **awaiting the sponsor's publish approval** — not approved, not published · **Source issue:** FRE-326
 **Page:** `nook.freno.me/faq` · **Page source:** `freno-dev/src/routes/nook/faq.tsx` · **Site commit read:** `c1c498e` (branch `faq/from-objections-2026-10-29`, unmerged)
 **Product commit read:** The-Nook `f9cbd879` — the exact revision the CTO verified against (0.13.4, build 37)
 Every entry below is traceable to a real question or a real report. Nothing is invented to fill a
 category. No testimonial is used. No price, refund term, or umbrella/cross-promotion answer appears
 here — those stay with FRE-204, the Chief of Staff and the sponsor.
 ---
 ## 1. What I could read, and what I could not
 | Source | What it is | Read on | What it yielded |
 |---|---|---|---|
 | `nook.freno.me/faq` | The live page | 2026-10-29 | Six questions already published: privacy, notch, battery, permissions, macOS 14.2, offline. |
 | FRE-326 description | This board issue | 2026-10-29 | Two purchase-blocking questions named verbatim: *"does it work with Cursor?"* and *"what happens when my agent is idle?"* |
 | NOOK issue corpus | 948 issues imported from Plane, plus the 324-issue Paperclip board | 2026-10-08 → 2026-10-28 | User-observable defect reports. These are reports, not captured asks; they prove an objection is live. |
 | `freno-dev/marketing/positioning.md` | Voice, approved claims, competitor ban | 2026-10-29 | The register and the "never name a competitor" rule. |
 | The-Nook code | `AgentRegistry.swift`, `SessionModels.swift`, `Terminal*`, `hook/setup.go`, `PluginInstaller.swift`, `LicenseService.swift`, `main.swift` | 2026-10-29 | The technical facts each answer is allowed to use. |
 | NOOK-121 | The original `/faq` creation (2026-10-08) | 2026-10-29 | Confirms the six live questions came from the strategy's named objections, not from a support corpus. Refund was deliberately dropped by Mike. |
 ### The gap that shapes this draft
 **Support mail is not readable from this run.** There is no mailbox on the machine; the in-app
 "Report a bug" form `POST`s to `freno-dev/src/routes/api/the-nook/report.ts`, which emails the
 developer and stores nothing queryable. `/privacy` and the repo contain no support corpus.
 **Reply threads are not collected.** `freno-dev/marketing/listening.md` is an empty table; FRE-201
 (the listening job) has not produced a thread yet.
 **Directory comments do not exist yet.** FRE-207 has a submission packet, not live listings; none of
 the directories has a comment thread about The Nook to mine.
 **Consequence:** I cannot rank by observed ask frequency, because no ask corpus exists. What follows
 ranks by **purchase-blocking weight first**, then by **the number of named reports in our own issue
 corpus that belong to the cluster**. The report counts are a proxy for "this objection is live", not
 a count of people who asked. I have not put a percentage on anything.
 ---
 ## 2. The register — clusters, sources, frequency, and whether they block a purchase
 | # | Cluster | The question in the buyer's words | Verbatim source(s), with dates | Frequency evidence | Blocks a purchase? | Disposition |
 |---|---|---|---|---|---|---|
 | 1 | Privacy / what leaves the Mac | "Does The Nook read my prompts or my code?" | Live `/faq`, 2026-10-29; code findings in `main.swift:272-278`, `LicenseService.swift:254`, `docs/sidecar-frames.md` | 1 live question + 3 named code findings | **Yes — the whole decision for this audience** | **Withdraw the live answer; replacement held.** See §4. |
 | 2 | Agent fit / Cursor | "Does it work with Cursor — and can I approve from the notch, or only watch?" | FRE-326, 2026-10-29: *"does it work with Cursor?"*; NOOK-643 (2026-10-20), NOOK-695 (2026-10-20), NOOK-699 (2026-10-20), NOOK-707 (2026-10-21), NOOK-638 (2026-10-20) | 1 board question + 5 reports | **Yes** | **Propose E1** |
 | 3 | Idle / is it actually working? | "What happens when my agent is idle?" | FRE-326, 2026-10-29: *"what happens when my agent is idle?"*; NOOK-281 (2026-10-12), NOOK-415 (2026-10-15), NOOK-417 (2026-10-15), NOOK-823 (2026-10-22), NOOK-898 (2026-10-23) | 1 board question + 5 reports | **Yes** | **Propose E2** |
 | 4 | Install trust / permissions | "What permissions does it need — and what does it write into my agent's config?" | Live `/faq`, 2026-10-29; NOOK-733 (2026-10-21), NOOK-648 (2026-10-20), NOOK-685 (2026-10-20), NOOK-418 (2026-10-15), NOOK-568 (2026-10-17) | 1 live question + 5 reports | **Yes** | **Propose E3** |
 | 5 | Terminal / tmux fit | "Does it work with my terminal, including tmux?" | NOOK-302 (2026-10-13), NOOK-310 (2026-10-13), NOOK-788 (2026-10-22) | 3 reports; no captured ask | **Yes for tmux-first buyers** | **Propose E4** — report-sourced, not an ask |
 | 6 | Notch / display | “Do I need a Mac with a notch?” | Live `/faq`, 2026-10-29; NOOK-290 (2026-10-12), NOOK-291 (2026-10-12), NOOK-386 (2026-10-14) | 1 live question + 3 reports | Compatibility; rarely the blocker | **Rewrite E5** — the pointer-display sentence was never true |
 | 7 | Battery / cost | "How much battery does it cost?" | Live `/faq`, 2026-10-29; NOOK-850 (2026-10-23), NOOK-851 (2026-10-23), NOOK-877 (2026-10-23) | 1 live question + 3 reports | Curious, not a hard blocker | **Rewrite E6 without a number** |
 | 8 | macOS floor | "Why macOS 14.2 or later?" | Live `/faq`, 2026-10-29; NOOK-881 (2026-10-23) | 1 live question + 1 report | Compatibility; a hard no on 14.0/14.1 | **Keep current answer** |
 | 9 | Offline | "Does it work offline?" | Live `/faq`, 2026-10-29; NOOK-494 (2026-10-16) | 1 live question + 1 report | Compatibility; a hard no for offline-only | **Rewrite E8 for the phone exception** |
 | 10 | Trial / licence mechanics | "What am I actually buying — trial, Macs, updates?" | NOOK-682 (2026-10-20), NOOK-684 (2026-10-20), NOOK-60 (2026-10-08) | 2 reports; no captured ask | **Yes** | **Held** — no verbatim ask; do not invent one |
 | 11 | Remote agents / containers | "Can I approve an agent running on a remote box or in Docker?" | FRE-323, FRE-320, FRE-315; `marketing/positioning.md` claims remote SSH | Many reports; the claim is unverified | **Yes — the launch claim** | **Held** — no answer before the demo |
 | 12 | Phone / Watch | "Is there an iPhone app? Do I need one?" | NOOK-939 (2026-10-27), NOOK-294 (backlog) | 2 reports; status unverified | Curious; would be blocking if sold | **Held** — CTO status answer first |
 | 13 | Refund / price | "What's your refund policy? Is $10 the real price?" | FRE-204, NOOK-121 | No readable source; contractual | **Yes** | **Out of scope** — FRE-204, CoS + sponsor |
 ### What changed since the last register (FRE-204)
 - The live page already has six entries; this draft **keeps one** (macOS 14.2), **rewrites four**
  (permissions, battery, offline, and notch — whose pointer-display sentence the CTO found was never
  true), **withdraws one** (privacy), and **proposes three new ones** (E1, E2, E4). It does not re-ask
  the six questions as if they were new.
 - The privacy answer is not merely unverified — it is **false at the current build**. It must come
  down in the same edit, not wait for a nicer replacement.
 - Cursor is now a supported agent row (`AgentRegistry.swift:524`), so "does it work with Cursor?"
  has a factual answer today. It was an open question when the live page was written.
 ### CTO verdict (FRE-329, 2026-10-29)
 The CTO read The-Nook `f9cbd879`, the site at `c1c498e` and the live page, and checked every entry
 against the code. **No entry is cut.** Six change text, exactly as below; E1 and E7 ship as written.
 | Entry | Verdict | What changed in this revision |
 |---|---|---|
 | E1 Agent fit / Cursor | **verified** | nothing |
 | E2 Idle | verified, one sentence tightened | the row-eviction sentence |
 | E3 Permissions / config | **not verified** | three sentences: Login Items toggle, per-event entries, `~/.nook` |
 | E4 Terminal / tmux | **not verified** | one sentence: which terminals take a typed reply |
 | E5 Notch | **not verified** | one sentence: one island per enabled display, not the pointer's display |
 | E6 Battery | verified, one clause tightened | "never runs them in the background" |
 | E7 macOS 14.2 | **verified** | nothing |
 | E8 Offline | **not verified** | one sentence, plus a six-row network list in its gate |
 | §4 privacy | all three findings confirmed; withdrawal is right | `positioning.md` rule 3 comes down in the same change |
 The verdict adds two things to the plan, both now in §4 and §7: the false claim's **source rule** in
 `marketing/positioning.md` — rule 3 and its approved-claims row — must be struck in the same change
 as the withdrawal, or the next page restates it; and the egress inventory's scope widens to cover
 weather, the network panel, the speed test, the changelog, the bug report and Sparkle, plus Sentry's
 `sendDefaultPii` and trace sample rate.
 The verdict's full text and its per-sentence evidence is
 [FRE-329's `faq-verdict` document](/FRE/issues/FRE-329#document-faq-verdict). Nothing in it is
 reproduced here as a new claim; the entry text below is the verdict's own replacement wording, except
 E4, where I took the verdict's first option (keep the Alacritty clause) and narrowed the jump
 sentence, which the verdict itself notes is `open -b` for three of the seven names.
 ---
 ## 3. Proposed entries — page-ready
 Each entry carries the six required things: **claim, audience, channel, owner, date, and what has to
 be true before it ships.** The answer text is the copy to paste; the trace is the evidence a CTO can
 check.
 ### E1 · "Does it work with my agent — and can I approve from the notch, or only watch?"
 > **It depends on the agent, and Cursor is the useful example.** Cursor is watched, not gated: its
 > sessions, finished turns and errors appear in the island, but when Cursor asks for permission the
 > prompt stays in Cursor. That is deliberate — Cursor blocks the action when a permission hook runs
 > and prints nothing, so gating it would override Cursor's own prompts. Other agents do hand the
 > decision to The Nook, so a tool call waits and you tap Allow or Deny from the card. One limit:
 > `cursor-agent -p` one-shot runs send no turn end, so they appear but do not get a completion card.
 - **Question source:** FRE-326, 2026-10-29 (*"does it work with Cursor?"*); NOOK-643, 2026-10-20
  (*"No permission hook is installed, deliberately: Cursor blocks the action when a permission hook
  prints nothing…"*).
 - **Claim:** agent gating is per-agent; Cursor is watch-only.
 - **Audience:** the buyer whose agent is not Claude Code.
 - **Channel:** `/faq`. No FAQ structured data in this pass — see §6.
 - **Owner:** CMO writes; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** at the shipped commit, Cursor still reads `.missing(.approvals)` with the same
  reason, and `.available(.sessions)`, `.available(.completion)` and `.available(.errors)` as stated
  (`AgentRegistry.swift:523-555`, which is where the CTO read it). The separate `/agents` page is
  incomplete for Cursor and the other newer rows; that is a page correction, not a gate for this
  self-contained answer.
 ### E2 · "What happens when my agent is idle?"
 > **Idle means the last turn is over, or the session has not produced anything yet — not that the
 > process is gone.** A running process is liveness, not work: a just-opened agent window can show
 > idle until you type. When an agent finishes a turn with a final message, the row stays green for
 > ten minutes, then goes gray. A stale row is different: it means The Nook's integration cannot
 > report for that session, and the row says so instead of pretending it is idle. Nothing is lost when
 > a row goes idle — the next prompt brings it back and flips it to running. The row itself leaves the
 > island 15 minutes after the turn (5 if the turn ended without a reply), so a long-quiet session
 > reappears on its next event instead of sitting there. If a session that is obviously working says
 > idle, that is a bug; tell us.
 - **Question source:** FRE-326, 2026-10-29 (*"what happens when my agent is idle?"*); NOOK-898,
  2026-10-23 (*"kimi payloads fail to decode, so every Kimi session stays idle"*); NOOK-823,
  2026-10-22 (*"Codex terminal: sessions become idle or disappear prematurely"*); NOOK-417,
  2026-10-15 (*"Never render plain Idle for a session whose integration is dark"*).
 - **Claim:** idle is a statement about the last turn, not a guess at the process; stale is a separate
  state.
 - **Audience:** anyone watching a fleet and wondering whether to intervene.
 - **Channel:** `/faq`.
 - **Owner:** CMO writes; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** `SessionModels.swift:511,552,564-577` still defines `islandStatus` this way (no
  content → idle; complete → ready for 10 minutes → idle; `integrationStale` → stale) and
  `SessionState.swift:35,40` still evicts a finished turn after 15 minutes, 5 when it ended without a
  reply. If an open bug makes a working agent render idle at the shipped commit, the "that is a bug"
  sentence stays but the bug gets a named issue.
 ### E3 · "What permissions does it need — and what does it write into my agent's config?"
 > **The app itself runs unprivileged.** Fan control and charge limiting go through a small helper
 > that macOS requires to run as root for hardware access; the app registers it on first use and takes
 > you to System Settings › General › Login Items & Extensions, where you switch it on once. For each
 > agent it adds what that agent needs — a hook group per event it listens to, or one plugin link —
 > inside the agent's own config file or directory. It keeps its own files under `~/.nook` — the hook
 > binary and one plugin copy per agent — and adds only its own entries to that agent's config. A hook
 > entry you added yourself in the same file survives install and uninstall, and it only touches the
 > records and switches it created. Uninstall removes exactly what it wrote — and switches off what it
 > turned on — with no leftover links or orphaned files.
 - **Question source:** live `/faq`, 2026-10-29 (*"What permissions does it need?"*); NOOK-418,
  2026-10-15 (a custom agent's install target is never shown to the user); NOOK-733, 2026-10-21 (a
  Codex hook installed but not trusted, so it never ran); NOOK-648, 2026-10-20 and NOOK-685,
  2026-10-20 (uninstall leftovers); NOOK-568, 2026-10-17 (helper re-registration tore down an
  approved helper).
 - **Claim:** reversible install; the helper is the only root step; uninstall is exact.
 - **Audience:** the buyer who has a hand-tuned `~/.claude/settings.json` or `~/.codex/hooks.json`.
 - **Channel:** `/faq`.
 - **Owner:** CMO writes; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** `hook/setup.go`'s strip functions and `PluginInstaller.uninstall` still remove
  only their own records at the shipped commit (`removeIfOurs`, `stripOurGroups`/`groupIsOurs`); the
  helper is still registered through `SMAppService.daemon` with a Login Items toggle and no password
  prompt (`PrivilegedHelper.swift:62-112,204-208`); the custom-agent target is still shown and
  accepted before install (NOOK-418); the only host switch uninstall flips is Amp's own hook-enable
  flag (`PluginInstaller.swift:379-385`) — and the answer now says so.
 ### E4 · "Does it work with my terminal, including tmux?"
 > **Yes — it recognises Terminal.app, iTerm2, Warp, Ghostty, kitty, WezTerm and tmux. Four of those
 > take a typed reply; the rest are watched.** A session is tied to the terminal that started it. Jump
 > lands in the right window for Terminal.app, iTerm2, kitty and any tmux pane — in tmux, the right
 > pane; on Warp, WezTerm and Ghostty it brings the app forward instead. Replies type straight back
 > into Terminal.app, iTerm2, kitty and any tmux pane; on Warp, WezTerm and Ghostty you answer in the
 > app's own window, and Alacritty takes no typed reply from another app at all.
 - **Question source:** no captured ask. Underlying reports: NOOK-302, 2026-10-13 (*"a tmux session
  shows only 'tmux', and its outer terminal comes from whoever started the tmux server"*); NOOK-310,
  2026-10-13 (*"jump and send address tmux panes on the default server, ignoring the session's tmux
  socket"*).
 - **Claim:** seven terminal families and tmux panes included. Remote panes are not claimed here;
  the remote launch story is still held.
 - **Audience:** anyone whose workflow is tmux-first.
 - **Channel:** `/faq`.
 - **Owner:** CMO writes; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** `TerminalAncestry.programs` still recognises those seven names; `TerminalSend.attempts`
  still builds a delivery channel only for tmux, herdr, kitty, iTerm2 and Terminal.app and refuses
  Alacritty by name; `TerminalJump` still raises the app rather than the window for Warp, WezTerm and
  Ghostty. This entry is **report-sourced, not ask-sourced**; if a captured ask never arrives, it is
  still publishable because the reports are real, but it ranks below E1–E3 until one does.
 ### E5 · "Do I need a Mac with a notch?"
 > **No. On a notched MacBook the island sits in the notch. On machines and displays without one, it
 > sits at the top centre of the display — the same pill with the same slots.** You get one island on
 > every display it is turned on for (all connected displays on first launch), each anchored to the top
 > of its own screen; Settings › Displays trims that list, and the pointer only decides which display's
 > floating controls are live. Everything inside the island — agent cards, fan control, the stats strip
 > — works identically either way.
 - **Question source:** live `/faq`, 2026-10-29; NOOK-290, 2026-10-12 and NOOK-291, 2026-10-12
  (non-notch display geometry bugs), NOOK-386, 2026-10-14 (empty pill tucks away on a non-notch
  display).
 - **Claim:** a notch is optional; external displays work.
 - **Audience:** Mac mini / Studio / external-monitor buyers.
 - **Channel:** `/faq`.
 - **Owner:** CMO keeps the question and rewrites the answer; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** `IslandDisplayManager` still owns one island per enabled display, `DisplayStore`
  still snapshots every connected display on first launch (`DisplayStore.swift:52-59`), and the
  non-notch geometry tests still pass. Nothing tests that first-launch snapshot today — that is a test
  gap to file, not a gate on this answer.
 ### E6 · "How much battery does it cost?"
 > **It is a native app, not a background agent.** It watches the agents you already run and never
 > runs them in the background, so their draw is the number that matters. We do not have a published battery
 > measurement for The Nook alone; when we do, it will be a measurement, not an estimate. If it costs
 > more than it should on your Mac, that is a bug — tell us.
 - **Question source:** live `/faq`, 2026-10-29; NOOK-850, 2026-10-23 (*"the idle app does work
  nothing asked for (CPU wins from a live profile)"*), NOOK-851, 2026-10-23 (a 30 fps timeline ticks
  the fan icon at idle), NOOK-877, 2026-10-23 (compact Nook still costs too much CPU at rest).
 - **Claim:** native app, not a background agent; **no numeric battery claim**.
 - **Audience:** the buyer checking whether this is another menu-bar battery sink.
 - **Channel:** `/faq`.
 - **Owner:** CMO writes; CTO verified the no-number wording 2026-10-29.
 - **Date:** CTO verified the no-number wording 2026-10-29; publish on approval, target 2026-11-05.
 - **Must be true:** either a measurement exists and is quoted, or the no-number sentence ships. Do
  not restore "the heaviest thing it does is a sensor read" — the profile findings above contradict
  it.
 ### E7 · "Why macOS 14.2 or later?"
 > **The audio mixer listens to each app's own sound stream, and Apple only shipped that API in macOS
 > 14.2 (Sonoma).** The app links against it, so 14.0 and 14.1 cannot launch it.
 - **Question source:** live `/faq`, 2026-10-29; NOOK-881, 2026-10-23 (*"A macOS 14.0/14.1 Mac cannot
  launch the app at all: dyld dies on a strongly linked CATapDescription overlay symbol"*).
 - **Claim:** the floor is 14.2 because of the audio tap API, not a policy choice.
 - **Audience:** anyone on an older macOS.
 - **Channel:** `/faq`.
 - **Owner:** CMO keeps; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** `Package.swift:12` and `LSMinimumSystemVersion` still say 14.2, and
  `DeploymentFloorTests` still refuses an older floor.
 ### E8 · "Does it work offline?"
 > **The Mac app does.** The island, agent cards, permission cards and fan control work with no
 > network; the hooks talk to the app over a local socket. The network is used for update checks,
 > licence activation, crash reports and the optional trial email, plus a few things only when you use
 > them: weather (asked for through our relay, with the location you chose), the network panel's
 > IP-and-ISP lookup, the speed test, the changelog in Settings, and a bug report if you send one. The
 > phone link is the exception: when you are away from your Mac, the phone reaches it through our
 > relay, so that part needs a connection.
 - **Question source:** live `/faq`, 2026-10-29; NOOK-494, 2026-10-16 (*"iOS: the main list gives no
  sign the phone is offline from the Mac"*).
 - **Claim:** the Mac app is offline-first; the phone link is not.
 - **Audience:** the buyer with a disconnected or privacy-sensitive setup.
 - **Channel:** `/faq`.
 - **Owner:** CMO writes; CTO verified 2026-10-29.
 - **Date:** CTO verified 2026-10-29 ([FRE-329](/FRE/issues/FRE-329)); publish on approval, target 2026-11-05.
 - **Must be true:** the only network uses at the shipped commit are the update check, licence
  activation, crash reporting, the optional trial email and the phone relay, plus exactly these five
  user-triggered ones — `/v1/weather` on the relay (`SidecarRelayAPI.swift:182-196`),
  `NetworkInfoService.swift:60-88,276`, `SpeedTest.swift:36,57`, `ChangelogModel.swift:102`,
  `BugReportClient.swift:72-94` — and `/privacy` lists the same set. This entry shares its gate with
  the privacy correction below.
 ---
 ## 4. The correction that is not a new entry: privacy
 **Live answer, quoted from `nook.freno.me/faq` on 2026-10-29:**
 > "No. The hooks and plugins that connect your agents to The Nook send session metadata only — which
 > agent is running, which session, which event happened, which terminal it lives in. No prompt text
 > and no code content leaves your device. Two things do leave the device, and we'd rather say so
 > plainly: anonymous usage telemetry, and your email address if you choose to give it at trial start.
 > That's the whole list."
 **Why it cannot stay up:** it is contradicted by the current build.
 1. **Sentry starts unconditionally with `sendDefaultPii = true`** (`Sources/NookApp/main.swift:272-278`).
   That is crash and error reporting with personal identifiers, not "anonymous usage telemetry".
 2. **The licence/trial call sends `IOPlatformUUID` and the Mac's device name** (`LicenseService.swift:254`,
   `:305`). Precision from the verdict: the identifiers ride on activation, not on every launch — a
   licensed Mac's launch call is `/status` with the key only (`:290-295`).
 3. **With the phone link on, prompt and reply text transit `nook-relay.freno.me`, sealed end to end**
   (`docs/sidecar-frames.md`). The relay can route the payload but cannot read it; the live answer
   still says no prompt text leaves the device.
 `/privacy` discloses the hardware UUID, device name and Stripe, but **not** Sentry and **not** the
 relay. So there is no replacement sentence I can write that is both true and consistent with the
 privacy page today.
 **Proposed:** remove the privacy entry from `/faq` in the same edit that ships the other entries,
 rather than replace it with another sentence I cannot stand behind. The replacement is held until
 the Security Engineer's egress inventory and the `/privacy` update land.
 **Decided 2026-10-29 — option A, withdraw both.** The Chief of Staff ruled on
 [FRE-334](/FRE/issues/FRE-334) (`decision` document) that the `/faq` answer and `positioning.md`
 rule 3 with its approved-claims row come down **in one commit**, and that no privacy claim is written
 anywhere until FRE-333 lands and the CTO verifies it. Three consequences for the sequence below: the
 withdrawal is no longer one of the sponsor's options, it is settled company policy; the publish card
 asks one thing, publish the eight entries with the privacy answer withdrawn; and if an answer ever
 ships the seven entries with the false answer still up, that is not a variant to absorb — it stops
 and goes back to the CoS, because it would publish something the CoS has ruled out.
 **The rule that produced the falsehood comes down with it.** The CTO's verdict traces the live answer
 to `freno-dev/marketing/positioning.md` rule 3 (“Honest privacy scoping”) and to the matching
 approved-claims row, which state the defensible claim as “No prompt or code content leaves the
 device” with the backing “envelope carries session/event/terminal metadata only”. That backing is true
 of the hook→app envelope and says nothing about what leaves the device, so if rule 3 stands while the
 FAQ answer comes down, the next page restates the claim. Rule 3 and that row are struck in the same
 commit as this patch (§7 step 3; the same edit also fixes the stale “macOS 14+” row in that table,
 which NOOK-881 retired).
 - **Claim:** none until the inventory exists — deliberately.
 - **Audience:** the buyer for whom this is the whole decision.
 - **Channel:** `/faq` (withdrawal), `/privacy` (replacement), `marketing/positioning.md` (the rule).
 - **Owner:** CoS for the withdrawal decision — **done 2026-10-29, option A**; Security Engineer for the
  egress inventory and the disclosure text; CTO for any build change and for verifying the replacement;
  CMO for the replacement copy and the rule-3 strike.
 - **Date:** withdrawal decision **done 2026-10-29** (was targeted 2026-10-31); inventory 2026-11-05;
  replacement copy 2026-11-09, and it needs both the CTO's verification and the sponsor's approval —
  rule-3 strike in the same commit as the page patch (§7 step 3).
 - **Must be true:** every egress in the inventory is either in the answer or gone from the build, and
  `/privacy` lists the same set. The inventory's scope is wider than the draft first drew it — it must
  cover `main.swift`, `LicenseService.swift`, `Info.plist`, `ChangelogRelease.swift`,
  `SidecarController.swift`, **plus** `SidecarRelayAPI`/`WeatherService` (`/v1/weather`),
  `NetworkInfoService`, `SpeedTest`, `ChangelogModel`, `BugReportClient` and Sparkle's appcast — and
  on the Sentry side not only that it starts but `sendDefaultPii` and the trace sample rate. Until
  then, do not publish a privacy answer.
 ---
 ## 5. Held — clusters with no entry yet
 | Cluster | Why it is held | Owner and date |
 |---|---|---|
 | Trial / licence mechanics ("trial, Macs, updates") | No captured ask. NOOK-682 and NOOK-684 are defect reports, not a buyer question. The answer is factual (14 days, 3 Macs, Sparkle updates), but the task says do not invent a question to fill the slot. | CMO to capture an ask; CTO confirms `LicenseService.swift:43`, `:262` by 2026-11-03. |
 | Remote agents / containers | The launch claim is unverified (FRE-315, FRE-322, FRE-323, FRE-320). No FAQ answer before the demo. | CTO evidence; CMO copy the day it exists. Revisit 2026-11-12. |
 | Phone / Watch | Status is unverified; the landing page and FAQ currently disagree. | CTO status answer by 2026-11-03; CMO copy after. |
 | Refund / price | Contractual, and explicitly out of this issue. | FRE-204, CoS + sponsor. |
 | Custom agents | No captured ask; the `/agents` page already says a custom agent can wire itself in. | No entry until asked. |
 ---
 ## 6. What the page edit would change
 **File:** `freno-dev/src/routes/nook/faq.tsx` (site `master` at `134f41e`; the branch is cut from
 local `c1c498e`, which removed the remote claim from the landing page, not from the FAQ).
 | Action | Entry |
 |---|---|
 | **Remove** | “Does The Nook read my prompts or my code?” — until the egress inventory and `/privacy` are true |
 | **Rewrite (correction only)** | “Do I need a Mac with a notch?” → E5 (per-display island, not the pointer's display) |
 | **Keep, with the exact text above** | “Why macOS 14.2 or later?” |
 | **Rewrite** | “What permissions does it need?” → E3 (adds the config/uninstall answer) |
 | **Rewrite** | “How much battery does it cost?” → E6 (no number) |
 | **Rewrite** | “Does it work offline?” → E8 (phone exception, six-row network list) |
 | **Add** | E1 (agent fit / Cursor) |
 | **Add** | E2 (idle) |
 | **Add** | E4 (terminal / tmux) |
 | **Edit** | The file header comment that cites `positioning.md` rule 3 → “Privacy is deliberately absent until the egress inventory and /privacy make an honest answer possible.” |
 | **Edit** | The `PageHead` description, which still advertises the privacy question → the new eight-entry list |
 Order on the page: E1, E2, E3, E4, E5, E6, E7, E8. The mailto link, the accordion behavior and the
 page title stay unchanged.
 **Not in this pass:** FAQPage JSON-LD. The page has none today; adding structured data is an SEO
 change, not an objection-mining change. It should be its own reviewed edit — and E1 no longer promises
 it.
 **Patch artifacts:** `freno-dev/marketing/drafts/faq-page.patch`, regenerated for this revision, and
 `freno-dev/marketing/drafts/positioning-privacy-withdrawal.patch` for the rule-3 strike step 3 needs
 to land with it (both proposals, neither applied). The page file and `positioning.md` are untouched
 until the gates below are satisfied; `faq-page.patch` re-checked against `freno-dev` master
 `c1c498e` on 2026-10-29 and applies cleanly.
 ---
 ## 7. Publish gates and sequence
 | Step | Owner | Date | What settles it |
 |---|---|---|---|
 | 1. CTO verifies the claims and evidence pointers in E1–E8 | CTO | **done 2026-10-29** | [FRE-329](/FRE/issues/FRE-329), verdict in its `faq-verdict` document: E1/E7 verified as written, E2/E6 verified after one sentence each, E3/E4/E5/E8 corrected above, **no entry cut**. Applied in this revision. |
 | 2. Human approves this revision | Sponsor | target 2026-10-31 | The draft moves from `marketing/drafts/` to `marketing/approved/`, or the confirmation card on FRE-326 is accepted. The card asks one question now — publish the eight entries with the privacy answer withdrawn — because 2b is settled. |
 | 2b. Withdrawal of the false privacy answer and of `positioning.md` rule 3 | Chief of Staff | **done 2026-10-29** | [FRE-334](/FRE/issues/FRE-334) `decision`: **option A, withdraw both, one commit**, no privacy claim anywhere until FRE-333 lands and the CTO verifies it. Recorded here so the page, the rule and the card cannot disagree. |
 | 3. Apply the page patch, strike rule 3 + its approved-claims row, commit on `freno-dev` | CMO | 2026-11-03 | Both patches apply together to `freno-dev` master `c1c498e` (checked 2026-10-29, in a throwaway worktree, as one change set); the diff contains only the FAQ entries, the page comment/description, the `positioning.md` rule-3 strike and the 14.2 row. **The site's suite has not been run — this host has no `bun`** — so step 3 still has to run it before the commit. One commit, so the page and the rule move together. |
 | 4. Merge to `freno-dev/master` (the publish) | Sponsor or Chief of Staff | 2026-11-05 | A human merge; Vercel deploys. No agent pushes to master. |
 | 5. Verify the live page | CMO | 2026-11-05 | `https://nook.freno.me/faq` returns 200 and shows the eight entries with no privacy answer; the text matches the approved draft, sentence for sentence. |
 **If the CTO had rejected an entry:** cut it from the patch and ship the rest — it did not reject any.
 **If the sponsor does not approve:** nothing is applied; the draft stays in `drafts/`.
 **If the support mailbox becomes readable:** re-sort the register by observed asks before step 3;
 the top three may change, and the page order should follow the asks, not this draft's proxy ranking.
 **If a `Must be true` line stops being true** before step 3 — a fix regresses, a line moves — the
 entry goes back to the CTO, not onto the page.
 ---
 ## Appendix A — verbatim source notes
 | Date | Source | Verbatim |
 |---|---|---|
 | 2026-10-29 | Live `/faq` | "Does The Nook read my prompts or my code?" / "Do I need a Mac with a notch?" / "How much battery does it cost?" / "What permissions does it need?" / "Why macOS 14.2 or later?" / "Does it work offline?" |
 | 2026-10-29 | FRE-326 description | “does it work with Cursor?” / “what happens when my agent is idle?” |
 | 2026-10-29 | [FRE-329](/FRE/issues/FRE-329) CTO verdict | “Four entries ship as written, two ship after a wording fix… two (E3, E4) currently say something false to a buyer.” Every correction in §3 is that document's own replacement sentence. |
 | 2026-10-08 | NOOK-121 | "Add /faq from real objections" — shipped the six live topics; refund dropped by Mike. |
 | 2026-10-20 | NOOK-643 | "agents: Cursor via hooks.json" — "No permission hook is installed, deliberately: Cursor blocks the action when a permission hook prints nothing…" |
 | 2026-10-21 | NOOK-707 | "agents: Cursor desktop chats bind to the shared extension-host pid and never end" |
 | 2026-10-23 | NOOK-898 | "hook: kimi payloads fail to decode, so every Kimi session stays idle" |
 | 2026-10-22 | NOOK-823 | "Codex terminal: sessions become idle or disappear prematurely" |
 | 2026-10-15 | NOOK-417 | "Never render plain Idle for a session whose integration is dark" |
 | 2026-10-15 | NOOK-415 | "Sessions whose integration predates the bridge credential freeze as Idle forever" |
 | 2026-10-21 | NOOK-733 | "codex: a non-interactive session never runs an installed hook (0.163.0)" |
 | 2026-10-20 | NOOK-648 | "opencode: uninstalling a pre-package install leaves its old nook.js link dangling" |
 | 2026-10-20 | NOOK-685 | "tests: InstallerTests.clineHookFiles writes through the managed TaskComplete symlink and then expects that path to survive uninstall" |
 | 2026-10-15 | NOOK-418 | "security: a custom agent's install target is never shown to, or approved by, the user" |
 | 2026-10-17 | NOOK-568 | "helper: launch-time re-registration tears down an approved helper and leaves it unregistered" |
 | 2026-10-13 | NOOK-302 | "terminal: a tmux session shows only 'tmux', and its outer terminal comes from whoever started the tmux server" |
 | 2026-10-13 | NOOK-310 | "terminal: jump and send address tmux panes on the default server, ignoring the session's tmux socket" |
 | 2026-10-22 | NOOK-788 | "pane: a remote reply inside a private tmux server resolves to nothing, and a pane id can be sent without its socket" |
 | 2026-10-12 | NOOK-290 | "Compact island: weather slot clips mid-glyph on non-notch displays (no text cap)" |
 | 2026-10-12 | NOOK-291 | "Compact island: non-notch pill is 36 pt wider than its slot row (padding double-counted in pillWidth)" |
 | 2026-10-14 | NOOK-386 | "island: hide the empty pill under the notch (or above the top edge) and keep a narrower hover trigger" |
 | 2026-10-23 | NOOK-850 | "Nook: the idle app does work nothing asked for (CPU wins from a live profile)" |
 | 2026-10-23 | NOOK-851 | "island: a 30 fps timeline ticks the fan icon even when the fan is idle" |
 | 2026-10-23 | NOOK-877 | "island: compact Nook still costs too much CPU at rest (round two, A/B per fix)" |
 | 2026-10-23 | NOOK-881 | "A macOS 14.0/14.1 Mac cannot launch the app at all: dyld dies on a strongly linked CATapDescription overlay symbol" |
 | 2026-10-16 | NOOK-494 | "iOS: the main list gives no sign the phone is offline from the Mac" |
 | 2026-10-20 | NOOK-682 | "licensing: a new user can be blocked at launch instead of getting the 14-day trial" |
 | 2026-10-20 | NOOK-684 | "licensing: parseISO rejects the server's fractional-second trial timestamp, so an expired server trial resets to 14 days" |
 | 2026-10-08 | NOOK-60 | "Capture email at trial start" |
 ## Appendix B — frequency method
 - **Live questions:** one per entry that quotes the live page. There are six, and they are the only
  questions on the page today.
 - **Board questions:** the two examples in FRE-326. They are not customer quotes; they are the
  board's own words for the purchase-blocking objections this issue was created to answer.
 - **Report counts:** the number of distinct NOOK issues I read whose subject or resolution directly
  produced the objection, listed by identifier and date in §2 and Appendix A. I did not use a raw
  keyword count: "agent" appears in 344 issues and "privacy" in 177, which says nothing about what
  buyers ask. The counts are deliberately small and inspectable.
 - **No ask frequency, no percentages.** The first thing a real support corpus changes is the order,
  not the wording.
 ---
 *Draft ends. Nothing here is published. The page file stays unchanged until the sponsor approves this
 revision and a human merges it.*
--- a/marketing/drafts/faq-page.patch
+++ b/marketing/drafts/faq-page.patch
@@ -0,0 +1,81 @@
 diff --git a/src/routes/nook/faq.tsx b/src/routes/nook/faq.tsx
 index 03d135a..bc3adef 100644
 --- a/src/routes/nook/faq.tsx
 +++ b/src/routes/nook/faq.tsx
@@ -1,8 +1,7 @@
 /**
  * `/faq` on the Nook subdomain — answers to the objections that actually
 - * stop a purchase. Privacy scoping follows marketing/positioning.md rule 3:
 - * "no prompt or code content leaves the device" (telemetry and the trial
 - * email do leave — never claim "no data leaves the device").
 + * stop a purchase. Privacy is deliberately absent until the egress
 + * inventory and /privacy make an honest answer possible.
  */
 import { For } from "solid-js";
 import { createSignal, Show } from "solid-js";
@@ -17,34 +16,46 @@ interface FaqEntry {
 const FAQ: FaqEntry[] = [
   {
 -    question: "Does The Nook read my prompts or my code?",
 +    question:
 +      "Does it work with my agent — and can I approve from the notch, or only watch?",
     answer:
 -      "No. The hooks and plugins that connect your agents to The Nook send session metadata only — which agent is running, which session, which event happened, which terminal it lives in. No prompt text and no code content leaves your device. Two things do leave the device, and we'd rather say so plainly: anonymous usage telemetry, and your email address if you choose to give it at trial start. That's the whole list."
 +      "It depends on the agent, and Cursor is the useful example. Cursor is watched, not gated: its sessions, finished turns and errors appear in the island, but when Cursor asks for permission the prompt stays in Cursor. That is deliberate — Cursor blocks the action when a permission hook runs and prints nothing, so gating it would override Cursor's own prompts. Other agents do hand the decision to The Nook, so a tool call waits and you tap Allow or Deny from the card. One limit: cursor-agent -p one-shot runs send no turn end, so they appear but do not get a completion card."
   },
   {
 -    question: "Do I need a Mac with a notch?",
 +    question: "What happens when my agent is idle?",
     answer:
 -      "No. On a notched MacBook the island sits in the notch. On machines and displays without one, it anchors to the display your cursor is on, so a desktop setup with external monitors gets the same treatment. Everything inside the island — agent cards, fan control, the stats strip — works identically either way."
 +      "Idle means the last turn is over, or the session has not produced anything yet — not that the process is gone. A running process is liveness, not work: a just-opened agent window can show idle until you type. When an agent finishes a turn with a final message, the row stays green for ten minutes, then goes gray. A stale row is different: it means The Nook's integration cannot report for that session, and the row says so instead of pretending it is idle. Nothing is lost when a row goes idle — the next prompt brings it back and flips it to running. The row itself leaves the island 15 minutes after the turn (5 if the turn ended without a reply), so a long-quiet session reappears on its next event instead of sitting there. If a session that is obviously working says idle, that is a bug; tell us."
   },
   {
 -    question: "How much battery does it cost?",
 +    question:
 +      "What permissions does it need — and what does it write into my agent's config?",
 +    answer:
 +      "The app itself runs unprivileged. Fan control and charge limiting go through a small helper that macOS requires to run as root for hardware access; the app registers it on first use and takes you to System Settings › General › Login Items & Extensions, where you switch it on once. For each agent it adds what that agent needs — a hook group per event it listens to, or one plugin link — inside the agent's own config file or directory. It keeps its own files under ~/.nook — the hook binary and one plugin copy per agent — and adds only its own entries to that agent's config. A hook entry you added yourself in the same file survives install and uninstall, and it only touches the records and switches it created. Uninstall removes exactly what it wrote — and switches off what it turned on — with no leftover links or orphaned files."
 +  },
 +  {
 +    question: "Does it work with my terminal, including tmux?",
     answer:
 -      "The Nook is a small native UI, not a background agent — it watches the agents you're already running; it doesn't run anything itself. The heaviest thing it does is read and write your Mac's fan and charging controllers, which is a lightweight sensor read. If you're running enough coding agents to need it, those agents dwarf the app's own draw."
 +      "Yes — it recognises Terminal.app, iTerm2, Warp, Ghostty, kitty, WezTerm and tmux. Four of those take a typed reply; the rest are watched. A session is tied to the terminal that started it. Jump lands in the right window for Terminal.app, iTerm2, kitty and any tmux pane — in tmux, the right pane; on Warp, WezTerm and Ghostty it brings the app forward instead. Replies type straight back into Terminal.app, iTerm2, kitty and any tmux pane; on Warp, WezTerm and Ghostty you answer in the app's own window, and Alacritty takes no typed reply from another app at all."
   },
   {
 -    question: "What permissions does it need?",
 +    question: "Do I need a Mac with a notch?",
 +    answer:
 +      "No. On a notched MacBook the island sits in the notch. On machines and displays without one, it sits at the top centre of the display — the same pill with the same slots. You get one island on every display it is turned on for (all connected displays on first launch), each anchored to the top of its own screen; Settings › Displays trims that list, and the pointer only decides which display's floating controls are live. Everything inside the island — agent cards, fan control, the stats strip — works identically either way."
 +  },
 +  {
 +    question: "How much battery does it cost?",
     answer:
 -      "The app itself runs unprivileged. Fan control and charge limiting go through a small helper component that macOS requires to run as root for hardware access — the app installs it on first use and you approve it once with your password. The hooks live in each agent's own config directory; you can inspect or remove every file it writes."
 +      "It is a native app, not a background agent. It watches the agents you already run and never runs them in the background, so their draw is the number that matters. We do not have a published battery measurement for The Nook alone; when we do, it will be a measurement, not an estimate. If it costs more than it should on your Mac, that is a bug — tell us."
   },
   {
     question: "Why macOS 14.2 or later?",
     answer:
 -      "The audio mixer listens to each app's own sound stream, and Apple only shipped that API in macOS 14.2 (Sonoma)."
 +      "The audio mixer listens to each app's own sound stream, and Apple only shipped that API in macOS 14.2 (Sonoma). The app links against it, so 14.0 and 14.1 cannot launch it."
   },
   {
     question: "Does it work offline?",
     answer:
 -      "Yes. Everything The Nook does — the agent grid, permission cards, fan and thermal control — works with no network at all, and your agents' hooks talk to the app over a local socket. The network is only used for update checks, license activation, and the optional trial email."
 +      "The Mac app does. The island, agent cards, permission cards and fan control work with no network; the hooks talk to the app over a local socket. The network is used for update checks, licence activation, crash reports and the optional trial email, plus a few things only when you use them: weather (asked for through our relay, with the location you chose), the network panel's IP-and-ISP lookup, the speed test, the changelog in Settings, and a bug report if you send one. The phone link is the exception: when you are away from your Mac, the phone reaches it through our relay, so that part needs a connection."
   }
 ];
@@ -56,7 +67,7 @@ export default function NookFaqPage() {
       <EdgeCacheHeaders maxAge={300} />
       <PageHead
         title="Frequently asked questions"
 -        description="Does The Nook read your prompts? Do you need a notch? Battery cost, permissions, the macOS 14.2 requirement, offline use — answered straight."
 +        description="Which agents can you approve from the notch? What does idle mean? Terminal and tmux support, permissions, the macOS 14.2 requirement, offline use — answered straight."
         ogTitle="The Nook — FAQ"
       />
       <SubdomainHeader />
--- a/marketing/drafts/positioning-privacy-withdrawal.patch
+++ b/marketing/drafts/positioning-privacy-withdrawal.patch
@@ -0,0 +1,40 @@
 diff --git a/marketing/positioning.md b/marketing/positioning.md
 index 1bc4973..26d6b04 100644
 --- a/marketing/positioning.md
 +++ b/marketing/positioning.md
@@ -53,13 +53,20 @@ pages, marketing copy, comparison tables, meta descriptions, emails, FAQs, all o
 Copy that says "13 agents" or "no data leaves the device" must have a test backing it.
 If a claim cannot be pointed at code or a test, it does not go in copy.
 -### 3. Honest privacy scoping
 +### 3. No privacy claim until the egress inventory exists (withdrawn 2026-10-29)
 -Telemetry and the trial email **do** leave the device. The defensible claim is:
 +The claim that stood here — "No prompt or code content leaves the device" — is **false at HEAD**
 +and must not be restated in any copy until it is verified and replaced. It was backed by an
 +argument about envelope metadata, which says nothing about what leaves the machine. Three things
 +at HEAD contradict it: crash reporting initialises with `sendDefaultPii = true`; the licence call
 +sends the hardware UUID and the device name; the phone relay carries sealed prompt and reply
 +payloads. `/privacy` names the hardware UUID and the device name and says nothing about crash
 +reporting with PII or about the relay.
 -> **No prompt or code content leaves the device.**
 -
 -Never "no data leaves the device".
 +So: write **no** privacy claim at all — not softened, not partial. "Telemetry and the trial email
 +do leave the device" is true and insufficient; it is not a substitute. The replacement goes in
 +only after FRE-333 lands an egress inventory and `/privacy` text that match what the app does.
 +See FRE-326 (the `/faq` withdrawal) and FRE-333 (the inventory).
 ### 4. Sell what works today
@@ -87,8 +94,7 @@ nothing posts itself.
 | Permission gate answered from a card | awaitPolicyTable + bridge directive path (`hook/normalize.go`, `serialize.go`) |
 | Fan / thermal / charge control | `helper/` (privileged helper, SMC, no kext) |
 | One-time purchase, no subscription | $10 → $15 at 1.0; Stripe checkout |
 -| macOS 14+ | 27 source files use the `@Observable` framework |
 -| No prompt or code content leaves the device | envelope carries session/event/terminal metadata only (see `docs/agent-interface.md`) |
 +| macOS 14.2+ | the CoreAudio overlay symbol the mixer uses only exists from 14.2 (NOOK-881) |
 ## Held claims (not publishable)
--- a/marketing/drafts/privacy-page.md
+++ b/marketing/drafts/privacy-page.md
@@ -0,0 +1,92 @@
 # `/privacy` — replacement copy, FRE-354 corrections applied 2026-11-04
 **Status:** applied on branch `faq/from-objections-2026-10-29` (commit after `bb6cf27`), per the CTO's FRE-354 verdict — "apply the corrections above and the page is signed." The two copy decisions the verdict delegated to the CMO are taken: the **fuller iPhone paragraph** (its push-token evidence row is recorded in the evidence table below and routed to FRE-333's owner to append to the signed inventory), and the **full (c) block with the LAN/tailnet sentence**, which pulls the corresponding Section 3 clause in. Nothing here is published; the sponsor's merge + push on card [`392a8465`](/FRE/issues/FRE-326#interaction-392a8465) is still the deploy step.
 **Sequencing fact, updated 2026-11-06:** the contingency has **fired**. `v0.15.1` (tagged 2026-11-04, first tag containing FRE-355's merge `6e74b261`) and `v0.15.2` are live on the appcast (`freno.me/api/TheNook/appcast.xml`, newest item 0.15.2). Shipped behaviour when enabled: crash/error reports, session pings (`enableAutoSessionTracking` untouched), traces at 0.1, and **Sentry also receives the connection's IP address** (`sendDefaultPii = false` is the SDK default and does not suppress it; only the Sentry project's *Prevent Storing of IP Addresses* switch does — FRE-1480). Paragraph (b) and the iPhone paragraph are therefore swapped to the opt-in form in this commit, with two corrections against the pre-written swap text: pings and traces are named (the contingency's own completeness warning), and the IP sentence is kept. Releases 0.3.0 through 0.15.0 started Sentry on every launch with no off switch; the page describes the current release.
 **Patch:** `marketing/drafts/privacy-page.patch` — applies cleanly to `c1c498e` (`git apply --check`, verified 2026-10-29; parse-checked with esbuild, not the site suite).
 **Evidence base:** [FRE-333 `egress-inventory`](/FRE/issues/FRE-333#document-egress-inventory) rev 1, eleven destinations, file:line per row, read at `The-Nook` `3c3ba7ee`; and the same document's §G, which proposed a page text. This patch is the §G text **corrected against the inventory** — the differences are listed below, each with the row that forces it.
 **Why this exists:** [FRE-326](/FRE/issues/FRE-326) + [FRE-334](/FRE/issues/FRE-334#document-decision) recorded option A — the privacy claim is withdrawn everywhere, and no privacy claim returns until the inventory exists, `/privacy` matches it, the CTO verifies both, and the sponsor approves the replacement copy. The inventory landed on 2026-10-29. This is the page half.
 | Field | Value |
 |---|---|
 | Claim this page carries | None. It is a disclosure page, not a claim: it states what the app sends and what it does not. The one favourable sentence is (j), which is the CTO's FRE-322 §2 code fact and which he may still decline. |
 | Audience | Someone deciding whether to pay $10 for a tool that reads their prompts. They read this page *before* the FAQ. |
 | Channel | `nook.freno.me/privacy`, one page |
 | Owner | CMO (copy + commit), CTO (verification), sponsor (approval of the replacement copy), Chief of Staff (publish switch) |
 | Date | Prepared 2026-10-29. Applied 2026-11-04 on the branch with the FRE-354 corrections; the contingency (b)/iPhone swap applied 2026-11-06. `Last Updated` on the page now reads November 6, 2026 — re-check it against the actual deploy date at push. Not a public date. |
 | What has to be true before it ships | 1) CTO signs the page against `The-Nook` HEAD, paragraph by paragraph — [FRE-354](/FRE/issues/FRE-354), due 2026-11-02. 2) Sponsor accepts the replacement copy. 3) The Sentry paragraph matches the build being deployed — see *Contingency*. 4) It deploys in the same push as the FAQ withdrawal; a page that discloses Sentry next to a FAQ that still says "anonymous usage telemetry" is worse than either alone. |
 ## What changed against the live page
 The live page (fetched 2026-10-29T09:2xZ and again 09:4xZ, HTTP 200, `Last Updated: August 26, 2026`, unchanged) says the app "runs locally on your device and sends only the minimal data required for trial and licensing" and lists four things. Against the inventory that is wrong in two directions: it **omits** Sentry (crash reports with IP address, every launch, no off switch), the relay and its install ID, the licence key crossing to the relay, sealed phone payloads, weather coordinates going to the relay *and* to Apple, the GeoIP third parties, the speed test, the changelog fetch, bug-report contents and update checks. It also calls the hardware UUID "masked", which the code contradicts — it sends the raw `IOPlatformUUID`.
 The replacement is organised by *connection*, not by *data type*, because that is how a reader checks it: eleven rows in the inventory, ten of them reachable from the page, one (Sentry) prominent because it is the only one a reader can turn off entirely and still miss in older builds. On the current Mac release (`v0.15.1`+) it is opt-in and off by default; the iPhone build in TestFlight (`ios-v0.3.2`, 2026-10-22) predates it. See *Contingency*.
 ## Differences from FRE-333 §G, and what forces each
 §G is the Security Engineer's proposed text; this patch is that text corrected against §G's own inventory. Six deliberate differences, none of them stylistic:
 | # | §G says | This patch says | Why |
 |---|---|---|---|
 | 1 | Opening: "processed on your device and are never sent to us" | Opening drops "never sent to us"; (c) and (j) state the two paths agent data does take | Sealed frames still transit our relay (§A.5, §B), and a remote machine carries them by design (§C). §G's own §B says the relay routes them. "Never sent to us" is the class of sentence that got the FAQ answer withdrawn. |
 | 2 | (d) Weather: relay registration implied only with Phone | (d) adds that Weather registers **even with Phone off, and on a licensed Mac that registration presents the license key** | §A.3 + §D.2. `SidecarController.weather()` routes through the same `keystore.relayCredential(licenseKey:)` as the phone path (`SidecarController.swift:269-279`, `SidecarRelayAPI.swift:298-312`). §G under-disclosed it. |
 | 3 | (a) trial/activate/status only | (a) adds `/deactivate` (key + UUID) and the trial email (email + UUID, sent once) | §A.2, `/D.4`. §G omitted both endpoints. |
 | 4 | (i) "The email you give at trial start or checkout delivers your license key" | (i) checkout delivers the key; a trial email is one reminder before the trial ends | `LicenseService.submitTrialEmail` + `OnboardingExtrasStep` say the trial email is an expiry reminder; the key is issued at purchase. §G's sentence promised a key that the trial email does not deliver. |
 | 5 | (b) Sentry: opt-in, "off unless you turn it on" | (b) is the opt-in form, applied 2026-11-06: off unless enabled in Settings, and it names session pings and the trace sample; **the IP sentence, kept** | The contingency fired with `v0.15.1`. Corrections against the pre-written swap text: pings/traces named because they still flow when enabled; IP kept because `sendDefaultPii = false` does not stop Sentry receiving the connection IP (FRE-1480); the Contingency text already said so. |
 | 6 | No (j) | (j) carries the CTO's FRE-322 §2 code fact, with the remote-machine exception | Not required for truth; it is the one favourable sentence, and the CTO may decline it. Delete the block if he does. |
 Two facts the inventory corrected and the page does not repeat wrongly: Sparkle's automatic checks are **off by default** here (`UpdateManager` owns the schedule; FRE-322 §2.4 said "on by default" and is stale on this row), and the UUID is **raw, not masked**.
 ## Paragraph-level evidence
 | Paragraph | What it now says | Evidence at `3c3ba7ee` |
 |---|---|---|
 | (a) Licensing/trial | raw UUID + device name; every launch while trialing; key-only once licensed; deactivate sends key + UUID; trial email sends email + UUID once; trial/status fail open, activation needs a connection once | `LicenseService.swift:300-313` (`/trial`), `:271-284` (`/status`), `:247-262` (`/activate`), `:286-302` (`/deactivate`), `:219-231` (`/trial-email`) |
 | (b) Crash reporting | opt-in, **off by default**; when enabled: crash/error reports, session pings, traces at 0.1; **Sentry also receives the IP address** (`Options.swift:267-275` at 9.26.1: `sendDefaultPii = false` is the default and does not suppress it); no prompts or code | `Sources/NookApp/main.swift` `startIfEnabled` — `sendDefaultPii = false` (`policy.sendDefaultPii`), `tracesSampleRate = 0.1` while on (`CrashReporting.onTracesSampleRate`); `CrashReporting.registerDefaults` pins `enabledKey: false`; evidence read 2026-11-06, IP row per FRE-1480 |
 | (c) Phone relay | off by default; relay is the default path, LAN/tailnet direct pairing available; install ID + license key at registration; sealed payloads; relay sees device IDs/names/platforms, timing and volume, not content | `SidecarRelayAPI.swift:144-153`, `SidecarPayload.swift:484-490`, `RemoteEndpoint.swift:14-24`, `docs/sidecar-frames.md` (sealing), inventory §B |
 | (d) Weather | 0.1° grid snapped **on the Mac**; registers with the relay even with the phone off, licence key included on a licensed Mac; Apple sees place lookups and position | `WeatherForecast.swift:25-30`, `SidecarController.swift:269-279`, `SidecarRelayAPI.swift:298-312`, `WeatherLocation.swift:14-77`, `:89-143` |
 | (e) Network panel | third-party IP lookup sees your IP; speed test uploads zeros only | `NetworkInfoService.swift:82,276`, `SpeedTest.swift:10-11,36` |
 | (f) Update checks | ship **off**; turning them on fetches `freno.me`, app version + IP | `UpdateManager.swift:48-55` (`automaticallyChecksForUpdates = false`, default key false) |
 | (g) Changelog | fetches release notes from `freno.me` | `ChangelogRelease.swift:75-78` |
 | (h) Bug reports | user-sent only; machine snapshot listed in full | `BugReportClient.swift:44-94`, `BugReportSnapshot.swift:17-36` |
 | (i) Email/payment | checkout email delivers the key; trial email is a reminder; Stripe | `LicenseService.swift:219-231`, `OnboardingExtrasStep.swift:86-110` |
 | (j) What does not leave your Mac | hook and bridge are not a cloud service; helper has no networking; two paths carry agent data across the network, both user-switched: remote machines over SSH or the tailnet, and the sealed phone relay | `hook/transport.go:143` (`resolveBridge`), `BridgeBindPlan.swift:9-23`, `RemoteEndpoint.swift:14-24`, inventory §B + §C |
 | iPhone companion (TestFlight beta) | the TestFlight build (`ios-v0.3.2` = `2e322da5`, 2026-10-22) predates FRE-355: unconditional start, `sendDefaultPii = true`, traces 1.0, no off switch (`ios/AgentNookMobile/App.swift:13-20`); the opt-in code exists at `v0.15.2` but no `ios-v*` tag carries it; other traffic is the phone connection in (c) plus the APNs push token given to the relay | `RelayLink.pair` `RelayLink.swift:104-118`, `updatePushToken` `:226-232` — row added per FRE-354, TestFlight provenance per FRE-1480; append to the FRE-333 inventory |
 ## Contingency — the Sentry paragraph is the only one that moves
 **FIRED 2026-11-06 — applied on the branch.** `v0.15.1` is the first release containing FRE-355 and is live on the appcast. The swap below was applied with the corrections recorded in the sequencing paragraph at the top of this document (pings and traces named; IP sentence kept — the pre-written text below had it, and dropping it was wrong; FRE-1480). The original text is kept for the record:
 (b) as committed describes today's build. The inventory's recommendation (§E.1) is an opt-in setting, default off, dropping `sendDefaultPii` and cutting `tracesSampleRate`, on Mac **and** iOS. If that ships, (b) is replaced in the same commit that changes the behaviour:
 > **(b) Crash and error reporting — off unless you turn it on.** If you enable crash reporting in Settings, The Nook sends crash and error reports to Sentry, our crash-reporting processor. Reports include the app version, macOS version, Mac model and what the app was doing when it failed, and Sentry also receives your IP address. They never include your prompts or code.
 **Sequence call, and who owns it.** Shipping (b)-with-no-off-switch is honest, but it puts "no off switch" in front of a buyer during launch week. The better order is the build change first, then this page — that is a build/priority call, not mine: it is proposed on the board as [FRE-355](/FRE/issues/FRE-355) (Senior Engineer, date proposed 2026-11-06, CTO to confirm), and if the CTO says the build change cannot land before the page, the page ships with today's text. Publishing the truthful-today version is never *blocked* by that choice; the *preferred* order is the other way.
 **This is why (b) must not be softened on the way out.** If the opt-in setting ships but `tracesSampleRate` or session tracking stays on, (b)'s opt-in form is incomplete again — it names only "crash and error reports" while performance traces and session pings still flow. The paragraph and the build change move together or neither does. (FRE-354 rewrote this warning: the original named a macOS username that sentry-cocoa never sends.)
 ## If the CTO declines paragraph (j)
 (j) is the only favourable sentence on the page and the only one carrying a CTO verdict rather than an inventory row. It is [FRE-322 §2](/FRE/issues/FRE-322#document-verdict) — *"verified as a code fact, not publishable yet… publishing it before the disclosure work lands would put a true sentence next to false ones."* The disclosure work is this page. If the CTO still declines it, delete the whole (j) block; nothing else on the page depends on it, and the page stays complete and true.
 ## Open questions for the CTO's read
 1. **The iOS app.** §D.10: the iOS target starts the same Sentry DSN with the same `sendDefaultPii` (`ios/AgentNookMobile/App.swift:12-20`). This page says "runs on your Mac". Is any iOS build in users' hands? If yes, the page needs one sentence (or the build needs the same opt-in); if no, say so and the page stays Mac-only.
 2. **Weather registration without a licence key.** §E.3 suggests decoupling weather registration from the licence key. If that ships before the page, (d) loses its licence-key clause; if after, the page is true today.
 3. **(j)** — publish, or cut.
 ## Apply checklist — what the apply commit must do, in order
 1. Re-read the live page and the inventory's HEAD; if either moved, re-derive (a)–(j) before applying. The inventory is pinned to `3c3ba7ee`.
 2. Bump `Last Updated` to the **deploy** date — the patch carries 2026-10-29 as prepared, and that line is the one place the date must change at apply.
 3. If the Sentry opt-in build change is in the same release, swap (b) for the *Contingency* paragraph — one hunk.
 4. Apply `privacy-page.patch` and the FRE-326 `faq-page.patch` + `positioning-privacy-withdrawal.patch` in the **same** commit, so no page states the old claim while another corrects it.
 5. Run the site's suite — `bun test` — which this host cannot do (no `bun`; `node_modules` present, `package.json` test script is `bun test`). A parse/transpile check with esbuild is **not** that suite and must not be recorded as one.
 ## Rules this page obeys
 - No competitor named (category nouns only, and none needed here).
 - No claim without a row in the inventory. Where the inventory has no row, the page says nothing.
 - No roadmap promise: the page never says a feature or fix is "coming", so the Sentry paragraph does not promise an off switch.
 - No price, no discount, no date, no launch language.
 - It is a correction of an existing page, not new marketing; nothing was added to sell anything.
 - It does not add a `/faq` privacy answer. That answer's return is gated separately ([FRE-326](/FRE/issues/FRE-326) `privacy-return-path`).
--- a/marketing/drafts/privacy-page.patch
+++ b/marketing/drafts/privacy-page.patch
@@ -0,0 +1,237 @@
 diff --git a/src/routes/nook/privacy.tsx b/src/routes/nook/privacy.tsx
 index da0cc1d..db27d8c 100644
 --- a/src/routes/nook/privacy.tsx
 +++ b/src/routes/nook/privacy.tsx
@@ -1,9 +1,38 @@
 /**
  * The Nook privacy policy — `nook.freno.me/privacy`.
  *
 - * Reflects the actual data The Nook sends when licensing: a masked hardware
 - * UUID + device name for trial/activation limits, the email used for purchase,
 - * and Stripe payment processing.
 + * Every statement on this page is traced to the egress inventory at
 + * `The-Nook` HEAD (FRE-333, `document-egress-inventory`): eleven internet
 + * destinations, file:line evidence per row. When the app's egress changes,
 + * this page changes in the same commit — not later.
 + *
 + * The crash-reporting paragraphs ((b) and the iPhone companion) describe the
 + * build users run today: crash reporting is opt-in, off by default, on
 + * releases v0.15.1 and later — FRE-355 (merge `6e74b261`) first shipped in
 + * `v0.15.1` (2026-11-04), and the live appcast serves `v0.15.2`. When enabled,
 + * Sentry receives crash and error reports, session pings, and performance
 + * traces at a 0.1 sample rate (`tracesSampleRate = 0.1`), and it also receives
 + * the connection's IP address (`sendDefaultPii = false` is the SDK default and
 + * does not suppress it — sentry-cocoa 9.26.1 sets `infer_ip = auto` for Cocoa,
 + * and only the Sentry project's *Prevent Storing of IP Addresses* switch stops
 + * that; FRE-1480). Releases 0.3.0 through 0.15.0 started Sentry on every
 + * launch with `sendDefaultPii = true` and no off switch; this page describes
 + * the current release. The iPhone paragraph describes `ios-v0.3.2`
 + * (2026-10-22, the build in TestFlight) until an `ios-v*` tag carries
 + * FRE-355's opt-in code.
 + *
 + * The *Contingency* in `marketing/drafts/privacy-page.md` fired with `v0.15.1`
 + * and is applied in this commit, with two corrections against the pre-written
 + * swap text: the shipped build still sends session pings and traces while
 + * enabled, so they are named (the contingency's own completeness warning), and
 + * the IP sentence is kept — the pre-written swap text already had it, and
 + * `sendDefaultPii = false` does not stop Sentry receiving the IP (FRE-1480).
 + *
 + * Corrected against the CTO's /privacy verdict (FRE-354): no macOS username
 + * from Sentry (sendDefaultPii sets the IP only), session pings named, (a)
 + * fail-open scoped, (c) relay-path qualified with the direct-pair sentence,
 + * (j) "not a cloud service" with the tailnet path, the iPhone paragraph added
 + * (TestFlight beta is in users' hands), and section 3 says "content".
  */
 import { buildMainSiteUrl } from "~/lib/subdomain-url";
 import { PageHead } from "~/components/PageHead";
@@ -21,39 +50,139 @@ export default function NookPrivacyPolicy() {
       <SubdomainHeader />
       <div class="min-h-screen px-[8vw] py-[10vh]">
         <div class="py-4 text-xl">The Nook&apos;s Privacy Policy</div>
 -        <div class="py-2">Last Updated: August 26, 2026</div>
 +        <div class="py-2">Last Updated: November 6, 2026</div>
         <div class="py-2">
 -          Welcome to The Nook (&apos;We&apos;, &apos;Us&apos;, &apos;Our&apos;).
 -          Your privacy is important to us. This policy explains what we collect,
 -          why, and how it is handled. The app itself runs locally on your device
 -          and sends only the minimal data required for trial and licensing.
 +          The Nook runs on your Mac, and the iPhone and Apple Watch companion
 +          is in TestFlight beta. Your prompts, your code and your agent
 +          sessions are processed on your own devices. This page lists every
 +          connection the apps make over the internet — and the local-network
 +          and tailnet paths the phone can use — what each one sends, and
 +          whether you can turn it off.
         </div>
         <ol>
           <div class="py-2">
             <div class="pb-2 text-lg">
 -              <span class="-ml-4 pr-2">1.</span> Data We Collect
 +              <span class="-ml-4 pr-2">1.</span> What leaves your devices
             </div>
             <div class="pl-4">
               <div class="pb-2">
 -                <div class="-ml-6">(a) Device hardware UUID:</div> When you start
 -                a trial or activate a license, The Nook sends a hardware UUID
 -                from your Mac. It is used solely to enforce the 14-day trial and
 -                the 3-device activation limit, and to prevent resetting the
 -                trial by reinstalling the app.
 +                <div class="-ml-6">(a) Licensing and trial:</div> When you start
 +                a trial or activate a license, The Nook sends your Mac&apos;s
 +                hardware UUID and device name to our licensing server
 +                (freno.me). The UUID enforces the 14-day trial and the 3-device
 +                activation limit and stops a trial being reset by reinstalling
 +                the app; the device name is stored with the activation so the
 +                Mac can be identified in your license records. While you are on
 +                the trial, this is sent at each launch. Once you are licensed,
 +                each launch re-checks only that your key has not been revoked.
 +                Deactivating a Mac sends the license key and that Mac&apos;s
 +                UUID. If you add an email during the trial, that request
 +                carries the email and the UUID, and it is sent once. The trial
 +                and status checks fail open: with no connection, an unexpired
 +                trial or a stored license keeps working, because license
 +                verification itself is offline. Activating a new license needs
 +                a connection once.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(b) Crash and error reporting — off unless
 +                you turn it on:</div> Crash reporting is off by default and
 +                runs only if you enable it in Settings. When it is on, The Nook
 +                sends crash reports, error events, hang reports, session pings,
 +                and a sample of performance traces to Sentry, our
 +                crash-reporting processor. Each report carries the app version,
 +                the macOS version, the Mac model and what the app was doing
 +                when it failed. Sentry also receives your IP address. They
 +                never include your prompts, your code or your session content.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(c) Phone connection:</div> The phone
 +                connection ships off and is yours to enable. With the default
 +                relay path, turning it on registers your Mac with our relay
 +                (nook-relay.freno.me) using a random install ID, and a licensed
 +                Mac presents its license key once at registration. Everything
 +                your Mac and your phone then exchange — session updates,
 +                approvals, replies — is encrypted end-to-end: the relay routes
 +                it and cannot read it. On the relay path, the relay can see
 +                which devices are connected — their IDs, names and platforms —
 +                when, and how much traffic they exchange.
 +              </div>
 +              <div class="pb-2">
 +                You can also pair over your LAN or tailnet instead; those paths
 +                carry the same sealed traffic and do not use the relay.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(d) Weather:</div> If you add the Weather
 +                module, your Mac asks our relay for forecasts using coordinates
 +                rounded to a 0.1° grid — about 11 km — and the rounding happens
 +                on your Mac before anything is sent. The Weather module
 +                registers with the relay for this even if you never turn on the
 +                phone connection, and on a licensed Mac that registration
 +                presents the license key, exactly as the phone connection&apos;s
 +                does. &quot;My Location&quot; asks macOS for your
 +                approximate position, and naming or searching for a place uses
 +                Apple&apos;s Maps service, so Apple sees those lookups. Neither
 +                feature sends us your precise location.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(e) The Network panel:</div> While the
 +                Network panel&apos;s connection section is on screen, The Nook
 +                asks a third-party IP-lookup service (get.geojs.io, with
 +                ipwho.is as backup) what your public IP address and provider
 +                are — that is how the panel can show them to you. Those services
 +                necessarily see your IP address. The speed-test button measures
 +                against Cloudflare&apos;s public speed endpoint and uploads only
 +                zeros, never your data.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(f) Update checks:</div> Automatic update
 +                checks ship off. If you turn them on, or press &quot;Check for
 +                Updates&quot;, The Nook fetches the update feed from freno.me
 +                and the request carries your app version. Like every request on
 +                this page, it also arrives from your IP address.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(g) Changelog:</div> Opening Settings ›
 +                Changelog fetches the release notes from freno.me.
 +              </div>
 +              <div class="pb-2">
 +                <div class="-ml-6">(h) Bug reports:</div> only what you choose
 +                to send. If you write a bug report, you decide what it says.
 +                Bug reports also include a machine snapshot — macOS version, Mac
 +                model, CPU, memory, free disk, locale, and your displays&apos;
 +                names and resolutions — so the problem can be reproduced. The
 +                optional contact email is used only to reply to you.
               </div>
               <div class="pb-2">
 -                <div class="-ml-6">(b) Device name:</div> The Nook sends your
 -                Mac&apos;s name so you can recognize the device in your license
 -                activations. It is not used for any other purpose.
 +                <div class="-ml-6">(i) Your email and payment:</div> The email
 +                you give at checkout delivers your license key and looks up your
 +                order. An email you add during the trial is used for one
 +                reminder before the trial ends. Payments are processed by
 +                Stripe; we never see or store your card details.
               </div>
               <div class="pb-2">
 -                <div class="-ml-6">(c) Email:</div> The email you provide at
 -                checkout is used to deliver your license key and to look up your
 -                order.
 +                <div class="-ml-6">(j) What does not leave your Mac:</div> The
 +                hook and the bridge that connect your agents to The Nook are
 +                not a cloud service: the hook dials the bridge on your Mac, and
 +                the bridge only listens. Two paths carry agent data across the
 +                network, and both are switched on by you: if you add a remote
 +                machine, the hook on that machine dials this Mac over SSH or
 +                your tailnet; and if you use the phone connection, the session
 +                updates, approvals and replies described in (c) travel through
 +                the relay, sealed end to end. The privileged helper that
 +                controls fans and charging has no networking code at all.
               </div>
               <div class="pb-2">
 -                <div class="-ml-6">(d) Payment:</div> Payments are processed by
 -                Stripe. We never see or store your card details.
 +                <div class="-ml-6">The iPhone companion.</div> The iPhone and
 +                Apple Watch app is in TestFlight beta. The build in TestFlight
 +                starts Sentry on every launch, with no setting that turns it
 +                off: it sends crash, error and hang reports, session pings,
 +                performance traces, the app version, the iOS version, the
 +                iPhone model and your IP address. Those reports never include
 +                your prompts, code or session content.
 +                Its other network traffic is the phone connection in (c) — the
 +                relay, or a direct LAN/tailnet link — plus, if you allow
 +                notifications, the push token it gives the relay so Apple can
 +                deliver them.
               </div>
             </div>
           </div>
@@ -63,9 +192,11 @@ export default function NookPrivacyPolicy() {
               <span class="-ml-4 pr-2">2.</span> How We Use It
             </div>
             <div class="pb-2 pl-4">
 -              Device identifiers, your email, and license records are used only
 -              to operate the trial, deliver purchases, and enforce the 3-device
 -              license. We do not use them for advertising or sell them to anyone.
 +              Device identifiers, install IDs, your email and license records are
 +              used only to operate the trial and the licenses, to run the relay
 +              and the weather service, to answer bug reports, and to deliver
 +              purchases. We do not use them for advertising and we do not sell
 +              them to anyone.
             </div>
           </div>
@@ -74,9 +205,13 @@ export default function NookPrivacyPolicy() {
               <span class="-ml-4 pr-2">3.</span> Data Security
             </div>
             <div class="pb-2 pl-4">
 -              License keys are signed and verified with Ed25519. The connection
 -              to our servers uses TLS. Your license key is stored in the macOS
 -              Keychain on your Mac.
 +              License keys are signed and verified with Ed25519 and stored in
 +              the macOS Keychain. The session content your Mac and phone
 +              exchange is encrypted end-to-end with keys your devices generate
 +              at pairing; our relay holds no key that can read it, though it
 +              can read the routing metadata described in (c). Every internet
 +              connection described on this page uses TLS; the direct
 +              LAN/tailnet paths use the same end-to-end encryption.
             </div>
           </div>
--- a/marketing/drafts/remote-ssh/copy.md
+++ b/marketing/drafts/remote-ssh/copy.md
@@ -0,0 +1,76 @@
 # Remote SSH copy — HELD, do not publish
 **Status:** removed from the live site on 2026-10-29 in the FRE-323 correction.
 **Owner:** CMO. **Verified by:** CTO, on the FRE-320/FRE-205 work.
 **Restore only when:** FRE-320/FRE-205 ship a genuinely one-click install, the CTO
 verifies that flow, and a recording shows a remote agent joining the island.
 Until then this file is the only home for these sentences.
 ## What was removed
 ### 1. Landing collage tile
 Source: `src/components/nook/FeatureCollage.tsx` (removed 2026-10-29).
 ```
 kicker: Remote
 title:  Agents on other machines
 body:   Run agents on Linux boxes and SSH servers — one click installs the
        nook-hook, and their sessions join the same island.
 ```
 The mock attached to it showed two session rows whose agent chips named
 `build-box` and `gpu-rig`.
 ### 2. Landing feature breakdown section
 Source: `src/components/nook/FeatureBreakdowns.tsx` (removed 2026-10-29).
 ```
 kicker: Remote
 title:  Any SSH box, one click away
 body:   Pick a host from your ssh config or your tailnet and click once — the
        nook-hook installs itself and its agents appear beside your local ones,
        indistinguishable.
 bullets:
 - One click: connect, push the hook, configure agents, check it can reach this Mac
 - Hosts mirror this Mac's installed integrations automatically
 - Remote sessions are ordinary rows; the agent chip names the machine
 ```
 The section's demo animated a `gpu-rig` host moving from "Not set up" through
 `connecting` → `pushing the hook binary` → `configuring agents` → `checking the
 host can reach this Mac` → `Live`, then showed a remote session row.
 ### 3. Approvals bullet
 Source: `src/components/nook/FeatureBreakdowns.tsx` (removed 2026-10-29).
 ```
 - Works across every hooked agent, local or remote
 ```
 ### 4. Feature chip
 Source: `src/components/nook/IslandExtras.tsx` (removed 2026-10-29).
 ```
 One-click SSH setup
 ```
 ## What replaced it
 Nothing. The Remote tile and section were removed rather than replaced; the
 existing Approvals cell already carries the verified local claim:
 > Approve an agent's tool call from your Mac's notch, and the agent carries on.
 ## Restore checklist
 1. FRE-320/FRE-205 have shipped the one-click install and removed the
   `~/.ssh/config` / `ssh-copy-id` friction described in FRE-320.
 2. The CTO has verified the shipping flow and the evidence is on FRE-320.
 3. A recording exists showing a remote agent joining the island.
 4. CMO restores the copy above, then owns the launch plan for it — with an
   owner and date per item — before anything goes back on the site.
--- a/marketing/drafts/remote-ssh/rationale.md
+++ b/marketing/drafts/remote-ssh/rationale.md
@@ -0,0 +1,28 @@
 # Why the remote claim is parked
 The site was selling a setup path a buyer does not get.
 FRE-320 measured today's actual path (2026-10-29): the one-click flow is
 app-UI-only, needs a hand-edited `~/.ssh/config` for any host not already in ssh
 config or on the tailnet, needs a one-time `ssh-copy-id` because the app runs ssh
 with `BatchMode=yes`, and has no manual fallback for minting the per-host
 credential. FRE-315 and FRE-322 §3 correction 4 marked the remote claim
 unverified and demo-gated. The CTO named it the largest launch risk because it
 was already in front of buyers.
 So the landing page now says only what the product can back today: local agents,
 the card, and the notch. The remote story is not deleted — it is parked here with
 the FRE-320/FRE-205 work so it can return the day the one-click install ships and
 is recorded.
 ## What has to be true before it returns
 | Requirement | Owner | Date |
 |---|---|---|
 | One-click install ships without hand-edited `~/.ssh/config` or `ssh-copy-id` | FRE-320/FRE-205 engineering | When FRE-320/FRE-205 ships |
 | CTO verifies the shipping flow and records the evidence on FRE-320 | CTO | Same day, before any copy moves |
 | A recording shows a remote agent joining the island | Sponsor/CMO | Before restore |
 | Restore copy and rebuild the launch plan with an owner + date per item | CMO | Day the three rows above are true |
 Nothing in this folder is publishable. `marketing/approved/` remains the only
 place a human publishes from.
--- a/marketing/positioning.md
+++ b/marketing/positioning.md
@@ -9,7 +9,7 @@ Primary source: `~/Code/the-nook/docs/marketing-strategy.md`.
 ## The sentence
 > **Your Mac and your agents, under control.** The Nook puts your whole coding-agent
-> fleet in the notch — thirteen agents, local or over SSH — so you see what is running,
+> fleet in the notch — thirteen agents — so you see what is running,
 > unblock it in one tap, and keep the machine from overheating while it works.
 ## The wedge
@@ -64,8 +64,14 @@ Never "no data leaves the device".
 ### 4. Sell what works today
 Phone approvals are deferred until a shipped iOS app exists — no "(coming soon)"
-parentheticals. The launch story is remote SSH: agents running on other machines,
+parentheticals. The launch story is the approval card: an agent blocked on a
-visible and unblockable from the Mac's notch.
+permission prompt, unblocked from the Mac's notch, and work resumes.
 Remote SSH is **held**, not approved. The one-click install has not shipped and
 has not been recorded (FRE-320, FRE-205), and the current path needs hand-edited
 `~/.ssh/config` and a one-time `ssh-copy-id` (FRE-320). Its copy is parked in
 `marketing/drafts/remote-ssh/` until the CTO verifies the shipping flow and a
 demo exists.
 ### 5. Humans hold every account
@@ -78,13 +84,18 @@ nothing posts itself.
 | Claim | Backing |
 |---|---|
 | Thirteen hooked agents, one glance | `AgentSpec.all` — claude, codex, opencode, pi, omp, muse, hermes, openclaw, zai, gemini, cline, qwen, deepseek |
 | Remote agents over SSH | `hook/setup.go` remote install path; `NOOK_HOST` TCP transport |
 | Permission gate answered from a card | awaitPolicyTable + bridge directive path (`hook/normalize.go`, `serialize.go`) |
 | Fan / thermal / charge control | `helper/` (privileged helper, SMC, no kext) |
 | One-time purchase, no subscription | $10 → $15 at 1.0; Stripe checkout |
 | macOS 14+ | 27 source files use the `@Observable` framework |
 | No prompt or code content leaves the device | envelope carries session/event/terminal metadata only (see `docs/agent-interface.md`) |
 ## Held claims (not publishable)
 | Claim | Why it is held | What has to be true first |
 |---|---|---|
 | Remote agents over SSH, one click to install | The one-click path is app-UI-only and needs hand-edited `~/.ssh/config` plus a one-time `ssh-copy-id`; no demo exists (FRE-320). FRE-315 and FRE-322 hold it unverified and demo-gated. | FRE-320/FRE-205 ship a genuinely one-click install, the CTO verifies it, and a recording shows a remote agent joining the island. Copy parked in `marketing/drafts/remote-ssh/`. |
 ## Voice
 - Short, concrete, no hype adjectives. Numbers over adjectives ("thirteen agents",
--- a/src/components/nook/FeatureBreakdowns.tsx
+++ b/src/components/nook/FeatureBreakdowns.tsx
@@ -1,6 +1,5 @@
 import {
  For,
  Show,
  createEffect,
  createSignal,
  onCleanup,
@@ -21,7 +20,6 @@ import { Campfire, type CampfireState } from "./Campfire";
 import {
  ApprovalCardMock,
  SessionRowMock,
  Glyph,
  CLAUDE,
  CODEX
 } from "./ModuleMocks";
@@ -37,7 +35,7 @@ import {
 * Detailed feature breakdowns: alternating copy + live-demo sections.
 * Each demo stages the real interaction loop: attention flares, approval
 * resolves, fans throttle, the charge ceiling holds, meetings surface,
- * the camera wakes, a remote host goes live. Newer demos live in
+ * the camera wakes. Newer demos live in
 * MoreDemos.tsx; app-anatomy pieces in ModuleMocks.tsx.
 */
@@ -807,112 +805,6 @@ function CameraDemo() {
  );
 }
 /* ── Demo 7: Settings ▸ Remote hosts — one click to Live ──────────── */
 /** RemoteMirror's progress steps, in order. */
 const SETUP_STEPS = [
  "connecting",
  "pushing the hook binary",
  "configuring agents",
  "checking the host can reach this Mac"
 ];
 function RemoteHostDemo() {
  /* -1 not set up · 0…3 setting up · 4 live */
  const [step, setStep] = createSignal(-1);
  const [pressed, setPressed] = createSignal(false);
  useDemoLoop(11000, [
    [0, () => (setStep(-1), setPressed(false))],
    [1400, () => setPressed(true)],
    [1700, () => (setPressed(false), setStep(0))],
    [2700, () => setStep(1)],
    [3700, () => setStep(2)],
    [4700, () => setStep(3)],
    [5800, () => setStep(4)]
  ]);
  const live = () => step() === 4;
  const working = () => step() >= 0 && step() < 4;
  const chip = () =>
    live()
      ? { color: "rgb(48,209,88)", label: "Live" }
      : working()
        ? { color: "rgba(255,255,255,0.55)", label: "Setting up" }
        : { color: "rgba(255,255,255,0.35)", label: "Not set up" };
  return (
    <div class="relative mx-auto w-full max-w-[360px] space-y-3">
      {/* the host's card in Settings ▸ Agents ▸ Remote hosts */}
      <ModuleCard>
        <div class="space-y-2 text-left">
          <div class="flex items-center gap-2.5">
            <div class="min-w-0 flex-1">
              <p class="text-[13px] font-semibold text-white">gpu-rig</p>
              <p class="truncate text-[11px] text-white/45">tailnet peer</p>
            </div>
            <Show when={working()}>
              <span class="h-3 w-3 animate-spin rounded-full border-[1.5px] border-white/25 border-t-white/80" />
            </Show>
            <span class="flex items-center gap-1.5 text-[10.5px] font-medium text-white/85">
              <span
                class="h-[7px] w-[7px] rounded-full transition-colors duration-300"
                style={{ background: chip().color }}
              />
              {chip().label}
            </span>
            <Show
              when={live()}
              fallback={
                <span
                  class="rounded-md px-2.5 py-1 text-[11px] font-semibold text-white transition-all duration-200"
                  style={{
                    background: ACCENT,
                    opacity: working() ? 0 : 1,
                    transform: pressed() ? "scale(0.92)" : "scale(1)"
                  }}
                >
                  Set up
                </span>
              }
            >
              <Glyph name="refresh" size={11} class="text-white/50" />
              <span class="rounded-md bg-white/10 px-2 py-1 text-[11px] font-semibold text-[#ff6961]">
                Remove
              </span>
            </Show>
          </div>
          <p class="h-4 text-[11px] text-white/50">
            {working() ? SETUP_STEPS[step()] : " "}
          </p>
        </div>
      </ModuleCard>
      {/* its agents arrive as ordinary rows, the machine named in the chip */}
      <div
        class="transition-all duration-500"
        style={{
          opacity: live() ? 1 : 0,
          transform: live() ? "translateY(0)" : "translateY(6px)"
        }}
      >
        <SessionRowMock
          title="trainer · fp16"
          prompt="benchmark fp16"
          state="Running python bench.py…"
          status="running"
          agent={CODEX}
          host="gpu-rig"
          origin="iTerm2"
          age="now"
        />
      </div>
      <FakeCursor
        phase={pressed() ? "press" : step() === -1 ? "idle" : "live"}
        at={{ x: 88, y: 16 }}
      />
    </div>
  );
 }
 /* ── Demo 8: campfire states ───────────────────────────────────────── */
 function CampfireDemo() {
@@ -976,7 +868,6 @@ export default function FeatureBreakdowns() {
          "Approval cards pin above the session list; questions answer inside their row",
          "Allow for session waves through the rest of that session's requests",
          "Answer from iPhone or Apple Watch too, now in TestFlight beta",
          "Works across every hooked agent, local or remote"
        ]}
        reversed
      >
@@ -996,20 +887,6 @@ export default function FeatureBreakdowns() {
        <TerminalReplyDemo />
      </BreakdownSection>
      <BreakdownSection
        kicker="Remote"
        title="Any SSH box, one click away"
        body="Pick a host from your ssh config or your tailnet and click once — the nook-hook installs itself and its agents appear beside your local ones, indistinguishable."
        bullets={[
          "One click: connect, push the hook, configure agents, check it can reach this Mac",
          "Hosts mirror this Mac's installed integrations automatically",
          "Remote sessions are ordinary rows; the agent chip names the machine"
        ]}
        reversed
      >
        <RemoteHostDemo />
      </BreakdownSection>
      <BreakdownSection
        kicker="The campfire"
        title="A hearth that tells the truth"
--- a/src/components/nook/FeatureCollage.tsx
+++ b/src/components/nook/FeatureCollage.tsx
@@ -1,12 +1,6 @@
 import { For, createSignal, onCleanup, onMount, type JSX } from "solid-js";
 import { IslandPill, STATUS, CALM, CRITICAL, ModuleCard } from "./IslandMock";
-import {
+import { ApprovalCardMock, DieLine, CLAUDE } from "./ModuleMocks";
  ApprovalCardMock,
  SessionRowMock,
  DieLine,
  CLAUDE,
  CODEX
 } from "./ModuleMocks";
 /**
 * Feature collage: five cells sold as the app's actual UI. The hero cell
@@ -105,35 +99,6 @@ function FansPanelMock() {
  );
 }
 /**
 * Remote agents as the panel draws them: ordinary session rows whose agent
 * chip names the machine ("Claude Code · build-box").
 */
 function RemoteMock() {
  return (
    <div class="space-y-1.5">
      <SessionRowMock
        title="api · main"
        prompt="fix the flaky test"
        state="Running pytest -x…"
        status="running"
        agent={CLAUDE}
        host="build-box"
        age="4m"
      />
      <SessionRowMock
        title="trainer · fp16"
        prompt="benchmark fp16"
        state="Done · replied 1m ago"
        status="ready"
        agent={CODEX}
        host="gpu-rig"
        age="12m"
      />
    </div>
  );
 }
 /* Deterministic usage wander for the traces. */
 const wander = (n: number, k: number) =>
  0.42 + 0.22 * Math.sin(n * 0.45 + k * 1.7) + 0.12 * Math.sin(n * 1.3 + k);
@@ -256,19 +221,6 @@ export default function FeatureCollage() {
            </InkStage>
          </CollageCell>
          {/* Remote agents */}
          <CollageCell
            kicker="Remote"
            title="Agents on other machines"
            body="Run agents on Linux boxes and SSH servers — one click installs the nook-hook, and their sessions join the same island."
          >
            <InkStage class="w-full">
              <div class="w-full px-3 text-left">
                <RemoteMock />
              </div>
            </InkStage>
          </CollageCell>
          {/* Gauges */}
          <CollageCell
            kicker="Your Mac"
--- a/src/components/nook/IslandExtras.tsx
+++ b/src/components/nook/IslandExtras.tsx
@@ -740,7 +740,6 @@ const CHIPS_B = [
  "tmux · iTerm2 · kitty",
  "Markdown replies",
  "Tailscale hosts",
  "One-click SSH setup",
  "iPhone widgets",
  "Watch-face widgets",
  "Screenshots to clipboard",
--- a/src/routes/nook/faq.tsx
+++ b/src/routes/nook/faq.tsx
@@ -1,8 +1,7 @@
 /**
 * `/faq` on the Nook subdomain — answers to the objections that actually
- * stop a purchase. Privacy scoping follows marketing/positioning.md rule 3:
+ * stop a purchase. Privacy is deliberately absent until the egress
- * "no prompt or code content leaves the device" (telemetry and the trial
+ * inventory and /privacy make an honest answer possible.
 * email do leave — never claim "no data leaves the device").
 */
 import { For } from "solid-js";
 import { createSignal, Show } from "solid-js";
@@ -17,34 +16,46 @@ interface FaqEntry {
 const FAQ: FaqEntry[] = [
  {
-    question: "Does The Nook read my prompts or my code?",
+    question:
      "Does it work with my agent — and can I approve from the notch, or only watch?",
    answer:
-      "No. The hooks and plugins that connect your agents to The Nook send session metadata only — which agent is running, which session, which event happened, which terminal it lives in. No prompt text and no code content leaves your device. Two things do leave the device, and we'd rather say so plainly: anonymous usage telemetry, and your email address if you choose to give it at trial start. That's the whole list."
+      "It depends on the agent, and Cursor is the useful example. Cursor is watched, not gated: its sessions, finished turns and errors appear in the island, but when Cursor asks for permission the prompt stays in Cursor. That is deliberate — Cursor blocks the action when a permission hook runs and prints nothing, so gating it would override Cursor's own prompts. Other agents do hand the decision to The Nook, so a tool call waits and you tap Allow or Deny from the card. One limit: cursor-agent -p one-shot runs send no turn end, so they appear but do not get a completion card."
  },
  {
    question: "What happens when my agent is idle?",
    answer:
      "Idle means the last turn is over, or the session has not produced anything yet — not that the process is gone. A running process is liveness, not work: a just-opened agent window can show idle until you type. When an agent finishes a turn with a final message, the row stays green for ten minutes, then goes gray. A stale row is different: it means The Nook's integration cannot report for that session, and the row says so instead of pretending it is idle. Nothing is lost when a row goes idle — the next prompt brings it back and flips it to running. The row itself leaves the island 15 minutes after the turn (5 if the turn ended without a reply), so a long-quiet session reappears on its next event instead of sitting there. If a session that is obviously working says idle, that is a bug; tell us."
  },
  {
    question:
      "What permissions does it need — and what does it write into my agent's config?",
    answer:
      "The app itself runs unprivileged. Fan control and charge limiting go through a small helper that macOS requires to run as root for hardware access; the app registers it on first use and takes you to System Settings › General › Login Items & Extensions, where you switch it on once. For each agent it adds what that agent needs — a hook group per event it listens to, or one plugin link — inside the agent's own config file or directory. It keeps its own files under ~/.nook — the hook binary and one plugin copy per agent — and adds only its own entries to that agent's config. A hook entry you added yourself in the same file survives install and uninstall, and it only touches the records and switches it created. Uninstall removes exactly what it wrote — and switches off what it turned on — with no leftover links or orphaned files."
  },
  {
    question: "Does it work with my terminal, including tmux?",
    answer:
      "Yes — it recognises Terminal.app, iTerm2, Warp, Ghostty, kitty, WezTerm and tmux. Four of those take a typed reply; the rest are watched. A session is tied to the terminal that started it. Jump lands in the right window for Terminal.app, iTerm2, kitty and any tmux pane — in tmux, the right pane; on Warp, WezTerm and Ghostty it brings the app forward instead. Replies type straight back into Terminal.app, iTerm2, kitty and any tmux pane; on Warp, WezTerm and Ghostty you answer in the app's own window, and Alacritty takes no typed reply from another app at all."
  },
  {
    question: "Do I need a Mac with a notch?",
    answer:
-      "No. On a notched MacBook the island sits in the notch. On machines and displays without one, it anchors to the display your cursor is on, so a desktop setup with external monitors gets the same treatment. Everything inside the island — agent cards, fan control, the stats strip — works identically either way."
+      "No. On a notched MacBook the island sits in the notch. On machines and displays without one, it sits at the top centre of the display — the same pill with the same slots. You get one island on every display it is turned on for (all connected displays on first launch), each anchored to the top of its own screen; Settings › Displays trims that list, and the pointer only decides which display's floating controls are live. Everything inside the island — agent cards, fan control, the stats strip — works identically either way."
  },
  {
    question: "How much battery does it cost?",
    answer:
-      "The Nook is a small native UI, not a background agent — it watches the agents you're already running; it doesn't run anything itself. The heaviest thing it does is read and write your Mac's fan and charging controllers, which is a lightweight sensor read. If you're running enough coding agents to need it, those agents dwarf the app's own draw."
+      "It is a native app, not a background agent. It watches the agents you already run and never runs them in the background, so their draw is the number that matters. We do not have a published battery measurement for The Nook alone; when we do, it will be a measurement, not an estimate. If it costs more than it should on your Mac, that is a bug — tell us."
  },
  {
    question: "What permissions does it need?",
    answer:
      "The app itself runs unprivileged. Fan control and charge limiting go through a small helper component that macOS requires to run as root for hardware access — the app installs it on first use and you approve it once with your password. The hooks live in each agent's own config directory; you can inspect or remove every file it writes."
  },
  {
    question: "Why macOS 14.2 or later?",
    answer:
-      "The audio mixer listens to each app's own sound stream, and Apple only shipped that API in macOS 14.2 (Sonoma)."
+      "The audio mixer listens to each app's own sound stream, and Apple only shipped that API in macOS 14.2 (Sonoma). The app links against it, so 14.0 and 14.1 cannot launch it."
  },
  {
    question: "Does it work offline?",
    answer:
-      "Yes. Everything The Nook does — the agent grid, permission cards, fan and thermal control — works with no network at all, and your agents' hooks talk to the app over a local socket. The network is only used for update checks, license activation, and the optional trial email."
+      "The Mac app does. The island, agent cards, permission cards and fan control work with no network; the hooks talk to the app over a local socket. The network is used for update checks, licence activation, crash reports and the optional trial email, plus a few things only when you use them: weather (asked for through our relay, with the location you chose), the network panel's IP-and-ISP lookup, the speed test, the changelog in Settings, and a bug report if you send one. The phone link is the exception: when you are away from your Mac, the phone reaches it through our relay, so that part needs a connection."
  }
 ];
@@ -56,7 +67,7 @@ export default function NookFaqPage() {
      <EdgeCacheHeaders maxAge={300} />
      <PageHead
        title="Frequently asked questions"
-        description="Does The Nook read your prompts? Do you need a notch? Battery cost, permissions, the macOS 14.2 requirement, offline use — answered straight."
+        description="Which agents can you approve from the notch? What does idle mean? Terminal and tmux support, permissions, the macOS 14.2 requirement, offline use — answered straight."
        ogTitle="The Nook — FAQ"
      />
      <SubdomainHeader />
--- a/src/routes/nook/privacy.tsx
+++ b/src/routes/nook/privacy.tsx
@@ -1,9 +1,38 @@
 /**
 * The Nook privacy policy — `nook.freno.me/privacy`.
 *
- * Reflects the actual data The Nook sends when licensing: a masked hardware
+ * Every statement on this page is traced to the egress inventory at
- * UUID + device name for trial/activation limits, the email used for purchase,
+ * `The-Nook` HEAD (FRE-333, `document-egress-inventory`): eleven internet
- * and Stripe payment processing.
+ * destinations, file:line evidence per row. When the app's egress changes,
 * this page changes in the same commit — not later.
 *
 * The crash-reporting paragraphs ((b) and the iPhone companion) describe the
 * build users run today: crash reporting is opt-in, off by default, on
 * releases v0.15.1 and later — FRE-355 (merge `6e74b261`) first shipped in
 * `v0.15.1` (2026-11-04), and the live appcast serves `v0.15.2`. When enabled,
 * Sentry receives crash and error reports, session pings, and performance
 * traces at a 0.1 sample rate (`tracesSampleRate = 0.1`), and it also receives
 * the connection's IP address (`sendDefaultPii = false` is the SDK default and
 * does not suppress it — sentry-cocoa 9.26.1 sets `infer_ip = auto` for Cocoa,
 * and only the Sentry project's *Prevent Storing of IP Addresses* switch stops
 * that; FRE-1480). Releases 0.3.0 through 0.15.0 started Sentry on every
 * launch with `sendDefaultPii = true` and no off switch; this page describes
 * the current release. The iPhone paragraph describes `ios-v0.3.2`
 * (2026-10-22, the build in TestFlight) until an `ios-v*` tag carries
 * FRE-355's opt-in code.
 *
 * The *Contingency* in `marketing/drafts/privacy-page.md` fired with `v0.15.1`
 * and is applied in this commit, with two corrections against the pre-written
 * swap text: the shipped build still sends session pings and traces while
 * enabled, so they are named (the contingency's own completeness warning), and
 * the IP sentence is kept — the pre-written swap text already had it, and
 * `sendDefaultPii = false` does not stop Sentry receiving the IP (FRE-1480).
 *
 * Corrected against the CTO's /privacy verdict (FRE-354): no macOS username
 * from Sentry (sendDefaultPii sets the IP only), session pings named, (a)
 * fail-open scoped, (c) relay-path qualified with the direct-pair sentence,
 * (j) "not a cloud service" with the tailnet path, the iPhone paragraph added
 * (TestFlight beta is in users' hands), and section 3 says "content".
 */
 import { buildMainSiteUrl } from "~/lib/subdomain-url";
 import { PageHead } from "~/components/PageHead";
@@ -21,39 +50,139 @@ export default function NookPrivacyPolicy() {
      <SubdomainHeader />
      <div class="min-h-screen px-[8vw] py-[10vh]">
        <div class="py-4 text-xl">The Nook&apos;s Privacy Policy</div>
-        <div class="py-2">Last Updated: August 26, 2026</div>
+        <div class="py-2">Last Updated: November 6, 2026</div>
        <div class="py-2">
-          Welcome to The Nook (&apos;We&apos;, &apos;Us&apos;, &apos;Our&apos;).
+          The Nook runs on your Mac, and the iPhone and Apple Watch companion
-          Your privacy is important to us. This policy explains what we collect,
+          is in TestFlight beta. Your prompts, your code and your agent
-          why, and how it is handled. The app itself runs locally on your device
+          sessions are processed on your own devices. This page lists every
-          and sends only the minimal data required for trial and licensing.
+          connection the apps make over the internet — and the local-network
          and tailnet paths the phone can use — what each one sends, and
          whether you can turn it off.
        </div>
        <ol>
          <div class="py-2">
            <div class="pb-2 text-lg">
-              <span class="-ml-4 pr-2">1.</span> Data We Collect
+              <span class="-ml-4 pr-2">1.</span> What leaves your devices
            </div>
            <div class="pl-4">
              <div class="pb-2">
-                <div class="-ml-6">(a) Device hardware UUID:</div> When you start
+                <div class="-ml-6">(a) Licensing and trial:</div> When you start
-                a trial or activate a license, The Nook sends a hardware UUID
+                a trial or activate a license, The Nook sends your Mac&apos;s
-                from your Mac. It is used solely to enforce the 14-day trial and
+                hardware UUID and device name to our licensing server
-                the 3-device activation limit, and to prevent resetting the
+                (freno.me). The UUID enforces the 14-day trial and the 3-device
-                trial by reinstalling the app.
+                activation limit and stops a trial being reset by reinstalling
                the app; the device name is stored with the activation so the
                Mac can be identified in your license records. While you are on
                the trial, this is sent at each launch. Once you are licensed,
                each launch re-checks only that your key has not been revoked.
                Deactivating a Mac sends the license key and that Mac&apos;s
                UUID. If you add an email during the trial, that request
                carries the email and the UUID, and it is sent once. The trial
                and status checks fail open: with no connection, an unexpired
                trial or a stored license keeps working, because license
                verification itself is offline. Activating a new license needs
                a connection once.
              </div>
              <div class="pb-2">
-                <div class="-ml-6">(b) Device name:</div> The Nook sends your
+                <div class="-ml-6">(b) Crash and error reporting — off unless
-                Mac&apos;s name so you can recognize the device in your license
+                you turn it on:</div> Crash reporting is off by default and
-                activations. It is not used for any other purpose.
+                runs only if you enable it in Settings. When it is on, The Nook
                sends crash reports, error events, hang reports, session pings,
                and a sample of performance traces to Sentry, our
                crash-reporting processor. Each report carries the app version,
                the macOS version, the Mac model and what the app was doing
                when it failed. Sentry also receives your IP address. They
                never include your prompts, your code or your session content.
              </div>
              <div class="pb-2">
-                <div class="-ml-6">(c) Email:</div> The email you provide at
+                <div class="-ml-6">(c) Phone connection:</div> The phone
-                checkout is used to deliver your license key and to look up your
+                connection ships off and is yours to enable. With the default
-                order.
+                relay path, turning it on registers your Mac with our relay
                (nook-relay.freno.me) using a random install ID, and a licensed
                Mac presents its license key once at registration. Everything
                your Mac and your phone then exchange — session updates,
                approvals, replies — is encrypted end-to-end: the relay routes
                it and cannot read it. On the relay path, the relay can see
                which devices are connected — their IDs, names and platforms —
                when, and how much traffic they exchange.
              </div>
              <div class="pb-2">
-                <div class="-ml-6">(d) Payment:</div> Payments are processed by
+                You can also pair over your LAN or tailnet instead; those paths
-                Stripe. We never see or store your card details.
+                carry the same sealed traffic and do not use the relay.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(d) Weather:</div> If you add the Weather
                module, your Mac asks our relay for forecasts using coordinates
                rounded to a 0.1° grid — about 11 km — and the rounding happens
                on your Mac before anything is sent. The Weather module
                registers with the relay for this even if you never turn on the
                phone connection, and on a licensed Mac that registration
                presents the license key, exactly as the phone connection&apos;s
                does. &quot;My Location&quot; asks macOS for your
                approximate position, and naming or searching for a place uses
                Apple&apos;s Maps service, so Apple sees those lookups. Neither
                feature sends us your precise location.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(e) The Network panel:</div> While the
                Network panel&apos;s connection section is on screen, The Nook
                asks a third-party IP-lookup service (get.geojs.io, with
                ipwho.is as backup) what your public IP address and provider
                are — that is how the panel can show them to you. Those services
                necessarily see your IP address. The speed-test button measures
                against Cloudflare&apos;s public speed endpoint and uploads only
                zeros, never your data.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(f) Update checks:</div> Automatic update
                checks ship off. If you turn them on, or press &quot;Check for
                Updates&quot;, The Nook fetches the update feed from freno.me
                and the request carries your app version. Like every request on
                this page, it also arrives from your IP address.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(g) Changelog:</div> Opening Settings ›
                Changelog fetches the release notes from freno.me.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(h) Bug reports:</div> only what you choose
                to send. If you write a bug report, you decide what it says.
                Bug reports also include a machine snapshot — macOS version, Mac
                model, CPU, memory, free disk, locale, and your displays&apos;
                names and resolutions — so the problem can be reproduced. The
                optional contact email is used only to reply to you.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(i) Your email and payment:</div> The email
                you give at checkout delivers your license key and looks up your
                order. An email you add during the trial is used for one
                reminder before the trial ends. Payments are processed by
                Stripe; we never see or store your card details.
              </div>
              <div class="pb-2">
                <div class="-ml-6">(j) What does not leave your Mac:</div> The
                hook and the bridge that connect your agents to The Nook are
                not a cloud service: the hook dials the bridge on your Mac, and
                the bridge only listens. Two paths carry agent data across the
                network, and both are switched on by you: if you add a remote
                machine, the hook on that machine dials this Mac over SSH or
                your tailnet; and if you use the phone connection, the session
                updates, approvals and replies described in (c) travel through
                the relay, sealed end to end. The privileged helper that
                controls fans and charging has no networking code at all.
              </div>
              <div class="pb-2">
                <div class="-ml-6">The iPhone companion.</div> The iPhone and
                Apple Watch app is in TestFlight beta. The build in TestFlight
                starts Sentry on every launch, with no setting that turns it
                off: it sends crash, error and hang reports, session pings,
                performance traces, the app version, the iOS version, the
                iPhone model and your IP address. Those reports never include
                your prompts, code or session content.
                Its other network traffic is the phone connection in (c) — the
                relay, or a direct LAN/tailnet link — plus, if you allow
                notifications, the push token it gives the relay so Apple can
                deliver them.
              </div>
            </div>
          </div>
@@ -63,9 +192,11 @@ export default function NookPrivacyPolicy() {
              <span class="-ml-4 pr-2">2.</span> How We Use It
            </div>
            <div class="pb-2 pl-4">
-              Device identifiers, your email, and license records are used only
+              Device identifiers, install IDs, your email and license records are
-              to operate the trial, deliver purchases, and enforce the 3-device
+              used only to operate the trial and the licenses, to run the relay
-              license. We do not use them for advertising or sell them to anyone.
+              and the weather service, to answer bug reports, and to deliver
              purchases. We do not use them for advertising and we do not sell
              them to anyone.
            </div>
          </div>
@@ -74,9 +205,13 @@ export default function NookPrivacyPolicy() {
              <span class="-ml-4 pr-2">3.</span> Data Security
            </div>
            <div class="pb-2 pl-4">
-              License keys are signed and verified with Ed25519. The connection
+              License keys are signed and verified with Ed25519 and stored in
-              to our servers uses TLS. Your license key is stored in the macOS
+              the macOS Keychain. The session content your Mac and phone
-              Keychain on your Mac.
+              exchange is encrypted end-to-end with keys your devices generate
              at pairing; our relay holds no key that can read it, though it
              can read the routing metadata described in (c). Every internet
              connection described on this page uses TLS; the direct
              LAN/tailnet paths use the same end-to-end encryption.
            </div>
          </div>
Author	SHA1	Message	Date
fixture	fd91b0d3ef	privacy: name the IP address and describe the TestFlight build as it ships (FRE-1480) CTO verdict at FRE-1480: two sentences at `db3e695` were false. (b) sendDefaultPii = false does not stop Sentry receiving the connection IP (sentry-cocoa 9.26.1 sets infer_ip = auto for Cocoa; only the Sentry project's Prevent Storing of IP Addresses switch stops it) — the shipped Settings rows already say so. The IP sentence is restored. The iPhone paragraph described an opt-in build no tester has: the build in TestFlight (ios-v0.3.2, 2026-10-22) predates FRE-355 and starts Sentry on every launch with no off switch. The paragraph now says that. Header comment and marketing/drafts/privacy-page.md rows corrected to match (release range narrowed to 0.3.0-0.15.0; evidence rows record the IP and the TestFlight provenance). privacy-page.patch regenerated. Suite: merge tree of `a0e6ff1` + this branch runs the same failure set as clean `a0e6ff1` (2 junit failures, the known Lineage-redirect environmental pair); the host's env-validation aborts account for the count noise. Zero new failures.	2026-11-06 13:26:24 +00:00
CMO (FrenoCorp)	db3e695ecf	privacy: fire the contingency — swap (b) and the iPhone paragraph to the opt-in form v0.15.1 (2026-11-04) is the first release tag containing FRE-355's merge 6e74b261, and the live appcast serves v0.15.2. The page's own Contingency made the always-on (b) conditional on no release shipping FRE-355; that condition no longer holds, so the always-on text would be false on the build the page deploys with. Swap applied with two corrections against the pre-written contingency text, both measured in the shipped starter (Sources/NookApp/main.swift, CrashReporting.onTracesSampleRate): - session pings and performance traces (0.1) still flow while enabled, so they are named — the contingency's own completeness warning; - sendDefaultPii = false means no IP is attached, so the swap text's 'Sentry also receives your IP address' is dropped, not kept. iPhone paragraph corrected the same way (iOS gates identically: ios/AgentNookMobile/App.swift startIfEnabled, SettingsView default off). Last Updated -> November 6, 2026. privacy-page.patch regenerated (c1c498e..branch); draft sequencing/evidence rows updated.	2026-11-06 00:44:22 +00:00
fixture	4d6550f0d1	privacy: apply the FRE-354 verdict corrections to the /privacy replacement FRE-354 ruled the page 'verified after corrections — do not apply as it stands'. This applies all of them, plus the two copy decisions the verdict delegated to the CMO (fuller iPhone paragraph; full (c) block with the LAN/tailnet sentence and the matching section-3 clause). - (b) no macOS username (sendDefaultPii sets the IP only), session pings named; still the always-on form: no release tag contains FRE-355's opt-in merge (6e74b261; v0.15.0 predates it), so the page stays true on the build users run. Contingency swap owed with the first release that ships FRE-355. - (a) device-name sentence corrected (no activations UI exists); fail-open scoped to trial/status, activation needs a connection once. - (c) relay qualified as the default path; relay sees IDs/names/platforms; direct LAN/tailnet pairing sentence added. - (j) 'not a cloud service' instead of two absolutes FRE-322 flagged; tailnet path named. - iPhone companion paragraph added (TestFlight beta is in users' hands, same Sentry config, push token disclosed). Push-token evidence row recorded for FRE-333's inventory. - Opening, section heading ('What leaves your devices') and section 3 ('content', not all traffic) per the verdict; Last Updated bumped. - Draft evidence table and contingency warning rewritten per the verdict (the old warning named a username Sentry never sends). Suite on this tree: 441 pass / 3 skip / 53 fail / 2 errors — failure and error sets byte-identical to clean master `d44133f` (458/3/53/2; the delta is master's remote-claims.test.ts, 17 passing tests this branch's base predates). Local only; the push is the sponsor's (card 392a8465).	2026-11-04 14:50:41 +00:00
Paperclip Agent	bb6cf27821	faq: publish the eight objection entries; withdraw the false privacy answer Sponsor-approved on FRE-414 card 57e1d01e (faq-publish = approve, 2026-10-31T13:19:10Z), per the FRE-334 option A ruling: the eight /faq entries from the objections draft (faq-draft rev c48396ea, CTO-verified in FRE-329) ship, and the live privacy answer comes down in the same commit - it says no prompt or code content leaves the device, which the FRE-333 egress inventory and FRE-359 verdict B proved false at HEAD. Same commit carries the /privacy replacement (privacy-page.patch, rev 4), corrected against the FRE-333 inventory: the page no longer calls the hardware UUID masked, discloses the relay registration a Weather panel performs even with Phone off (licence key included on a licensed Mac), /deactivate, and the one-time trial-email request. positioning.md needs no change here: master's 2026-10-29 reconciliation already withdrew rule 3 and the approved-claims privacy row; this branch precedes it and the human merge brings the reconciled file in. Suite: bun test on this tree - 458 pass / 3 skip / 53 fail / 2 errors, failure set identical to clean master `d44133f` (environmental: rate limiting, CSRF, IP handling, lineage redirect). No merge by any agent; deploy waits on the push credential (FRE-414 card 5c0eb58e).	2026-10-31 13:41:55 +00:00
Paperclip Agent	e46c0a2677	marketing: correct the /privacy copy against the FRE-333 inventory FRE-333's egress inventory landed after this patch was first written, and its own §G proposed a page text. This is that text corrected against the inventory it cites, six differences, each forced by a row: - the opening no longer says prompts are "never sent to us" (sealed frames transit our relay; a remote machine carries them by design) - (d) Weather states that it registers with the relay even with Phone off, and that a licensed Mac presents its license key there - SidecarController.weather() goes through the same relayCredential(licenseKey:) as the phone path - (a) adds /deactivate (key + UUID) and the trial email (email + UUID, once) - (i) stops saying a trial email delivers a license key; it is one reminder before the trial ends, the key is issued at checkout - (b) keeps today's build (on every launch, no off switch, IP + macOS username); the opt-in form is the Contingency swap, so the page is true on the build it deploys with - (j) carries the CTO's FRE-322 §2 code fact, with the remote-machine exception; he may still cut it Page file untouched, nothing published. Patch re-checked with git apply --check and parse-checked with esbuild; the site's own suite is `bun test` and this host has no bun, so that stays an apply-time obligation. Verification is FRE-354; the Sentry opt-in is proposed as FRE-355.	2026-10-29 09:13:06 +00:00
CMO	2c917e63e8	marketing: prepare the /privacy replacement as a reviewable patch (not applied) The live page says the app 'sends only the minimal data required for trial and licensing' and calls the hardware UUID masked. Against the FRE-333 egress inventory that is wrong in two directions: it omits Sentry (every launch, sendDefaultPii = true, no off switch), the relay and its install ID, the licence key crossing to the relay, sealed phone payloads, weather to the relay and to Apple, the GeoIP lookups, the speed test, the changelog fetch, bug-report contents and update checks; and the UUID it sends is raw. Reorganised by connection, ten reachable rows from the inventory, one favourable sentence (the FRE-322 §2 hook/bridge fact) that the CTO may still decline. Sentry paragraph describes today's build with the opt-in swap held in the rationale file. Page file untouched; the patch applies cleanly to `c1c498e`.	2026-10-29 09:05:27 +00:00
Paperclip Agent	871e47dee9	faq: mirror the CoS decision into the drafts copy of the plan Byte-identical to The-Nook drafts/faq-from-objections.md at 2c1915e3. FRE-334 returned option A, so step 2b is done and the publish card has one option, not three. Source issue: FRE-326. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:13:46 +00:00
Paperclip Agent	f20389e64e	faq: keep the drafts copy of the plan in step with The-Nook Byte-identical to The-Nook drafts/faq-from-objections.md at b86a980b. The plan now names the rule-3 patch and FRE-334, the Chief of Staff's withdrawal ask. Source issue: FRE-326. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:11:35 +00:00
Paperclip Agent	534c1ccd86	marketing: prepare the rule-3 strike as a reviewable patch (not applied) The CTO verdict on the /faq entries (FRE-329) traced the false privacy answer to marketing/positioning.md rule 3 and its approved-claims row, and said the rule has to come down in the same change as the /faq withdrawal or the next page restates the claim. This adds the exact patch for that, saved next to the page patch so both go in as one commit when the gates clear. It is not applied: the withdrawal decision is the Chief of Staff's (FRE-326 step 2b), and the replacement copy waits on FRE-333's egress inventory. It also carries the stale "macOS 14+" row, which NOOK-881 retired: the app cannot start below 14.2, and the site has said 14.2+ since `134f41e`. Rule 3's new text is corrected against /privacy as it reads today: that page does name the hardware UUID and the device name, and says nothing about crash reporting with PII or the relay. Source issues: FRE-326, FRE-329, FRE-333. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:11:12 +00:00
Paperclip Agent	4a18beeaff	faq: apply the CTO verdict to the eight entries Six entries change text, exactly as FRE-329's replacement sentences: E2 keeps the row for 15 minutes, E3 names the Login Items toggle and ~/.nook instead of "one entry, no writes outside the directory", E4 says which terminals take a typed reply, E5 says one island per enabled display, E6 says it never runs agents in the background, E8 lists the five user-triggered network uses. E1 and E7 are unchanged; no entry was cut. The page patch is regenerated from the corrected entry text, so the draft and the patch cannot drift apart. The gate table now records the verdict (done 2026-10-29) and puts the privacy withdrawal plus the positioning.md rule-3 strike on the Chief of Staff with a date, because the verdict found rule 3 is the source of the false claim. Nothing is applied to the page: src/routes/nook/faq.tsx is untouched on this branch. Source issue: FRE-326. Verdict: FRE-329. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:08:15 +00:00
Paperclip Agent	3d41d6cd45	faq: make E3 one page paragraph Keep the draft's paste-ready answer identical to the page patch. Source issue: FRE-326. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:03:45 +00:00
Paperclip Agent	477bb807fc	faq: tighten three claims in the objections draft E1 self-contained (no /agents completeness promise), E4 drops the held remote-pane sentence, E8 names crash reporting. Page patch regenerated; still not applied. Source issue: FRE-326. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:03:09 +00:00
Paperclip Agent	7ea654069c	faq: draft the entries from the objections we have The draft maps every proposed /faq entry to a live question or a named NOOK report, withdraws the privacy answer that the current build contradicts, and holds the questions we cannot answer yet. The page patch is included but not applied; it waits on the CTO verdict and the human move to approved/. Source issue: FRE-326. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-10-29 06:01:40 +00:00
Paperclip Agent	c1c498e37f	site: remove the unproven remote claim from the Nook landing page FRE-323. The one-click remote install is app-UI-only and needs hand-edited ~/.ssh/config plus a one-time ssh-copy-id (FRE-320); FRE-315 and FRE-322 hold the claim unverified and demo-gated. - drop the Remote tile and its mock from the feature collage - drop the "Any SSH box, one click away" breakdown, its demo and bullets - drop "Works across every hooked agent, local or remote" - drop the "One-click SSH setup" chip - move the copy to marketing/drafts/remote-ssh/ and mark remote held in marketing/positioning.md No replacement claim was added. Deploy waits on a push credential for the GitHub repo Vercel builds.	2026-10-29 05:55:26 +00:00