• Same screen and doors as f2 — hero, URL field, fallback doors — BrainIntakeScreen renders identically for guest and member; only the header and the CTA behavior change (onGuest prop, BrainIntakeScreen.tsx). Promise-list zones (see f2) omitted here as a drawing-once cross-reference, not a real difference.
• Every CTA on the screen (Research →, the Instagram door, the Add-files door) routes to f2b’s account gate instead of acting — no partial read ever starts for a guest.
• Header swaps the balance/name/sign-out cluster for the auth funnel: quiet “Sign in” text link + white “Get early access” pill (useAuth().isGuest, shell RS1/RS3 — no GUEST badge, retired 2026-07-24).
• Guest and empty pages show two columns, two same-size covers — left Business playbook (two-up GLEAM) · right Our read (kicker Example, title What we understood). A fifth matching tome in one carousel is rejected. Tap a playbook cover → f20. Opening a finished example never requires an account; only starting a REAL read does (AC-BRAIN-08). The old rail “Preview template” chips are retired. On the POPULATED page the examples demote to the Example playbooks · 5 → hairline row under “Teach your Brain more” (CPO + Chief Designer 2026-08-16, V3 lock 2026-08-18 — #3608), reopening the same cover carousel as a sheet; a standing second shelf under her books stays rejected.
• URL legend (2026-08-18) lives on this start. Standing caption under the field + live refuse + one Share anything door; V3 Thread under the field, post-read kicker is lantern Your call. Walk: 34-walk · 34-walk-m.

The example shelf is the 4-persona SUBSET with authored playbooks (examplePlaybooks.ts: Retreat host / Bloom Retreats · Workshop host / The Clay Room · Drop maker / Ember Goods · Custom-order / Stitch & Story) plus a drawn example Our-read cover (AC-BRAIN-98). Not the full 8-persona picker roster (1b).

• No intermediate chooser screen — every gated action opens the EXISTING WaitlistModal directly (useGuestGate.tsx); “early access” leads because most visitors are new founders.
• The quiet “Already got private access? Sign in” line under the form is the side door — tapping it swaps to the EXISTING AuthModal in login mode, one level up so it sits above this dialog.
• Same one sheet regardless of which f2a door fired it (Research →, the Instagram door, the Add-files door) — no brain data ever renders behind or through this gate (AC-BRAIN-08).
• Centered dialog per the tap-intent rule’s “short confirm” class — a form, not a destination, so it never becomes a sheet or a split.

• RETIRED 2026-08-16. Current start is the cover shelf at f34. URL caption, live refuse, and the one share door live there — walk at 34-walk. Do not review URL copy on this promise-list frame.
• ONE required ask: a single link field, and the ask is the founder’s own website — not “any link”. The two fallback doors drawn here (Instagram screenshot · add files) are historical. Current product folds them into one Share anything door on f34.
• No per-action price on the surface — confirm-before-charge, plus one plain line saying the run draws from the wallet; the amount appears only at the confirm moment, from props.
• The playbook list is a promise, not a worklist (Peyton's pick, 2026-07-26) — every section named with a .pb-hint descriptor and no CTA, keeping “nothing unfinished is visible” true while answering “what am I getting?”. Research → stays the single action on the screen.

• RESOLVED (B7, #2876) — BrainIntakeScreen.tsx no longer expands ShareInline in place; the mockup and the build now agree: both doors open the ONE share sheet (f15, R6-3) instead. Code repoint happens at build; full detail in parity/parity-brain.md.

• V3 (2026-08-18). Composer is a hairline, not a dashed card. “What this fills” uses Node (you / graft / double) — no emerald “Confirmed”. Read-my-site sparkle is Node · double. SealSpinner on in-flight reads.
• SAVE FIRST, THEN PROPOSE (Peyton ruling 2026-08-01 — the whole point of this frame). Every drop is saved to the Brain corpus (brain_documents, retrievable by assistants) BEFORE any playbook sorting runs. Playbook proposals are a SECOND, optional step on top of a save that already succeeded. “We only take bank transfer in Bali” — knowledge with no playbook section — is now a first-class thing to share, not something that falls out the bottom.
• Today it is the opposite, and material is lost. organize-content.ts:588 returns proposals UNPERSISTED; anything the classifier can’t map to a section hits “Read it — couldn’t sort it yet” or the dropped overflow (:323-378) and is gone. This door has never written the retrievable corpus at all — brain-ingest is the only writer, and nothing here calls it.
• ONE share surface platform-wide — every share affordance routes here with context { target: whole-brain | section:<id> | inventory-row:<id> }; title + top pane render from it, the composer never changes (BrainShareSheet.tsx:175). ShareInline survives INSIDE this sheet only. The first-run header chip’s step 5 (9c) is one more door into this same sheet (/brain?teach=1) — opening it once dismisses the chip; visiting /brain does not.
• Titles are the shipped two, not an invented third (BrainShareSheet.tsx:113-129): scoped section:how-you-show-up → “Teach your Brain — how you show up”; everything else → “Share anything”. “Teach your Brain — Instagram” / “— rates.pdf” have no code path and are corrected throughout this frame — both intake doors (BrainIntakeScreen.tsx:209/:233) call onOpenShare() with no argument, so the Instagram door opens the UNSCOPED sheet.
• 40:60 locked (R6-4): top sticky “what this fills” (caps ~40vh, internal scroll) · bottom composer. Sibling rule: this sheet ingests material; the ask-sheet (f14) collects typed answers — never merged; the FAQ inventory row alone routes there.
• The charge fires on “Save to my Brain”, never on staging or drop. Embeddings are metered (ai_usage · brain_embed · ×2.0), so save-first means every drop now costs — the standing floor holds: ONE screen-level wallet line (present even when empty) + confirm-before-charge, and no per-action “Metered” pill (retired platform-wide, kit 2026-07-25). Voice: we read and fold in — never “scan”/“ingest”/“train”; per read, never monthly.
• Proposals land as f13 confirm cards, provenance “you shared this” — never auto-saved to the playbook. Status: the sheet chrome + composer + “Before we save” confirm shipped-in-artifact · the corpus save, the saved-state panel, and the section-less outcome specified-not-built (this ruling) · the honest-empty EF signal specified-not-built.

• Three real doors drawn with the sheet they open — the full registry lives in “Every door, one sheet” at the end of this frame; every affordance is a plain tap, none draws its own composer.

• Backdrop is the INTAKE page, not the playbook — 15b/15a below show the sheet once a playbook exists (context can be section-scoped by then); this is day one, so context is whole-brain and the 40% pane shows the promise-list zones, not confirmed values.
• The composer is drawn here for the first time — the previous 15e showed a staged-file card and no composer at all, which is not a state the code can produce: BrainShareSheet.tsx:267 mounts ONE ShareInline that is ALWAYS present (hidden only behind the honest-empty panel). Empty composer is what a door actually opens.
• One box, one job: type · paste · drop a file · link a doc all live in the SAME box (never tabs). Placeholder + footer affordances verbatim from ShareInline.tsx:1512 / :1668.
• Primary is disabled until something is staged — nothing to save, nothing to charge.

• Confirm title is “Before we save” (was “Before we read”) and the primary is “Save to my Brain” (was “Organize it”) — the button now names the thing that always happens, not the half that can fail to find a section.
• A staged IMAGE lifts OUT of the composer box and lists once in the confirm block; typed text and doc chips stay IN the box. One listing per item — the duplicate was #1495.
• ONE wallet line, screen-level, inside the confirm block. No per-item pill, no second metered paragraph (kit 2026-07-25, G1-WALLET #2749). A running “N saves · $X so far” tally may join it only once a save has actually fired.
• The per-row “Read it” button is GONE from the staged row: under save-first the row is not individually actionable — one Save covers everything staged, so a second per-row verb would imply two charges. The per-image intent picker (“Read the text off it” / “Use as a brand look instead”) is RETIRED too (Peyton, 2026-08-02): a false choice — the read takes in look, voice, and business at once — and repeating it per file crowded the block. ONE collective line above the list says what we do (images: “We take in the whole picture — how it looks, how you sound, what you sell.” · files: “We read everything you drop and pull what matters into your Brain.”); rows are thumbnail + filename + ✕. Shipped StagedItemRow still renders the picker — retirement rides the build.

• The saved panel is durable, not a toast, and it sits ABOVE the proposals so the founder reads “kept” before she reads “sort it?”. A toast is precisely the QA-Bug-6 failure this ruling closes: wallet debited, tick shown, nothing findable afterwards.
• It carries its own proof — the filename and a line of the kept text — so “Saved” is never an unverifiable claim at the moment it is made. Correction (2026-08-01, verified against origin/main): the read path DOES exist — brain-ops ships a list-items action over the list_brain_documents RPC, consumed today by useBrainItems.ts (already mounted in PlaybookPage.tsx:176 for the “sources:” line), and remove-item.ts deletes a whole document group, user_id-scoped, with no orphaned chunks. What was missing was a founder-facing LIST, not the API: DocumentsCard was retired as collateral when its only mount went dead (useBrainComprehension.tsx:4-14).
• The destination now exists — “Other things your Brain knows” (Peyton-locked 2026-08-01), seated at the foot of the playbook in 1d. One section, read · retitle · delete; no search, folders, tags or versioning (that is the document-manager failure mode). Every note carries a one-line title · source · date — one fact = one row, never a blob, because an untitled blob is a junk drawer and a titled row is correctable (ADR-023). BrainItem.title is nullable, so the untitled row is a REAL state and is drawn as one.
• HARD LOCK it must not break (useBrainComprehension.tsx:4-14): “ONE gap analysis per page; known = the playbook itself, missing = the gap queue.” A playbook SECTION listing notes is legal under it; a page-level roll-up of what the Brain knows is NOT — that recreates the deleted ComprehensionOverview. No coverage/completeness framing, no “your Brain knows N things across M sources”, no second inventory competing with the playbook. Row grammar reuses the surviving VoiceExemplarList.tsx pattern rather than inventing one.
• House knowledge, never customer rows. “Bank transfer only in Bali” belongs here; a guest list, a booking, or a passport number does not — that is the Records Layer, a different surface with a different contract.
• “Your assistants can use this now, whether or not it lands in your playbook” is the honesty line — it is what makes the section-less case (15h) a success rather than a shrug.
• Confirm grammar verbatim from f13 (✓ Use · fix · discard), one card per target section; adopting lands the line with provenance “you shared this” (D2 — a draft is a STATE, never a source chip). Declining every proposal changes nothing about the save.
• Top-pane row ticks emerald on land (C12, positive outcome only) — and when “How you show up” lands, f12’s walled-platform line resolves. The composer resets to empty and stays mounted; we never auto-close on save.

• Today’s single “Read it — couldn’t sort it yet” panel conflated two different outcomes. Save-first separates them, and the separation is the whole fix.
• 15h · saved, nothing proposed — a SUCCESS. The founder sees the identical saved panel from 15b — same emerald tick, same weight, same words — plus ONE quiet line: “Nothing here belongs in a playbook section — that’s fine.” Identical weight is the mechanism: a different-looking panel is what makes a founder read it as failure. Exit offered, never demanded: “Put it in a section yourself →” (f13).
• 15i · couldn’t save — a real failure, and the only one. Neutral monochrome, never red (an honest miss, not a system error), the founder’s material still in the box, and the money stated plainly: “Nothing was charged.” Retry is the primary exit.
• The code delta: shareHonestEmpty (BrainShareSheet.tsx:230-258) currently renders “Read it — couldn’t sort it yet” and HIDES the composer underneath. Under save-first it must (a) never fire when the corpus save succeeded — that is 15h, a saved state — and (b) stop hiding the composer, since a save that landed has nothing to retry.
• No state in the ladder dead-ends; every exit is named.

• NEW (PRD debt) · save-first ordering — Given a founder submits anything through the share door, when the submit runs, then the material is written to the retrievable Brain corpus BEFORE any playbook classification runs, and a classification that maps nothing never prevents, reverses, or hides that save.
• NEW (PRD debt) · saved is stated durably — Given a corpus save has landed, when the sheet re-renders, then a persistent in-sheet panel names what was saved (filename and/or a line of the kept text) and states that assistants can use it now — never a toast, and never only a section tick.
• NEW (PRD debt) · section-less is not a failure — Given a save landed and produced zero playbook proposals, when that outcome renders, then it uses the SAME saved panel as the proposals case plus one neutral explanatory line, carries no error styling and no red, and offers (never requires) a manual “put it in a section” exit.
• NEW (PRD debt) · every drop is a charge, disclosed once — Given save-first makes each submit a metered embed (ai_usage · brain_embed · ×2.0), when the share sheet is open, then exactly ONE screen-level line states that saving draws from the wallet, it is present even with nothing staged, no charge fires before an explicit confirm, and no per-action metered pill is rendered.
• NEW (PRD debt) · a failed save costs nothing — Given the corpus save fails, when the failure renders, then the founder’s material is preserved in the composer, a retry is offered, and the state states that nothing was charged (the embed must not settle ahead of a durable write).
• NEW (PRD debt) · saved material has a home — Given a founder has saved material to the corpus, when she wants to see or remove it, then a founder-facing listing of saved material exists. Blocking gap: none exists today — brain-ops has no list-documents action and no Brain surface reads brain_documents. Until it does, the saved panel is the ONLY evidence a save happened, and material she can never see or delete is a privacy and trust problem as much as a UX one.

Real AC ids stop at AC-BRAIN-75 in docs/product/platform/brain/prd-pl-brain.md. Nothing above is assigned an id here on purpose — ids are minted in the PRD, never in an artifact.

• 375 carries the full ladder (empty → staged → saved+proposals → saved/nothing proposed → couldn’t save), desktop carries three (empty → staged → saved+proposals). Mobile is the fuller tier on purpose — it is the primary one, and the two terminal states are the ones this ruling changes.
• 768 is NOT drawn, and that is a stated decision rather than an omission. BrainSplitHost drops the 40% page pane entirely below modalTokens.slide.split.engageBreakpoint (1024px), so 768 renders the 375 composition at a wider measure — one column, stacked digest, same ladder. There is no third distinct composition to draw. If the build ever moves the engage breakpoint below 1024, this frame owes a real 768 tier.

• whole-brain auto-sorts into sections; everything still lands as drafts behind D4 confirm — never silently saved.

• Intake + research/inventory doors (screenshot, add-files, walled/failed rows — f2, f5, f1b/f1c/f1e) → whole-brain or that inventory row. The AC-BRAIN-40 known-unreadable-host intercept (paste blocked before the crawl fires, platform named inline, matching door opens) is drawn once, at f2’s own inline error variants — not redrawn here.
• “How you show up” screenshot/reviews doors (f12) → section: how-you-show-up.
• Playbook “Teach your Brain more” rows + detail-sheet EDIT “add” (f1c/f1d, f13, f16) → whole-brain or that section.
• Every other inline ShareInline is deleted or converted to a → f15 door; the FAQ inventory row alone routes to the ask-sheet (f14) instead.

• Five page-level states, both tiers: empty · researching · partial · complete · error — full state-flow diagram (arrives-from/leaves-by per state) now lives in prd-pl-brain.md §“Playbook state flow.” A state map, never a duplicate: row grammar lives in f9, sheets in f13/f14, checklist owned by 1b (below — the standalone research-checklist frame folded in here, Peyton catch 2026-07-30). No state draws its own 768 tier.
• Couldn’t-read inventory renders as a counted rail row here (count + jump) in 1c/1d — full row grammar and resolution flow are f5’s. 1e is the one exception: when the read actually failed, the rows sit inside the error banner itself (Option B), and the rail keeps only the count — one block of bad news, not a summary pointing at a list.
• “Teach your Brain more” is a door → f15, context whole-brain — it never expands a composer in place.
• The five-error contract (which error draws where, and its fix path) lives once, in parity/parity-brain.md.

• The live checklist (rows verbatim — same sources, same order, same three honest states: done/active/waiting) sits where the hero lived in f2; the step-2 gate this checklist waits on rises as its own self-opening 40:60 sheet (Peyton 2026-08-01, spec-brain-identity-gate-sheet.md) — drawn in full at 1f, seated below in this frame’s own sequence; the “while we research” persona picker rides alongside as the one-tap “while we wait” moment, never auto-decided. Playbook shell below stays unchanged from f2 (every section still shows only its promise line). Consent language stays visible in the checklist caption (draws-from-your-wallet, already confirmed at f2, restated here so it’s never out of sight).
• After confirm the checklist stops stacking history (Peyton decision 5, 2026-08-01 · NOT built): one confirmed line, the active step, and any row that still asks something of her (the walled-source “share a screenshot →” door). Rejected: a strict two-line collapse — it removes her only available action during the wait. Drawn at 1b′ below, both tiers.
• Never a page-count row (“Site (4 pages)”) — sources, not pipeline stages; live counts ride a source row’s META instead (· 6 pages, · 2 of 4 sites — decision 11, 2026-08-02, streamed by shipped BrainReadingProgress.tsx/AC-URLX-08). A walled source never gets bespoke copy, it points at the ONE share sheet (f15) — these walled/failed rows are live projections of the same registry drawn in full at f5, one mechanism rendered mid-research here.
• Identity-ambiguous branch (375 only; desktop identical) — merged from the retired f3. When we can’t confirm this is the founder’s business, step 2 fails soft: “We couldn’t confirm this is your business — type your business name and we’ll keep going →”. Research continues from the typed name; a source we still can’t attribute lands as an f5 row (confirm or share), never silently guessed.
• No jump-nav here, on purpose — nothing exists yet to jump to; the playbook shell shows only promise lines until a draft lands (→ 1c). The example shelf stays under those lines — persona-agnostic, not gap-dependent.

Persona-picker chips are DB-driven (mf_personas selector via the persona_roster view, ADR-022) — this drawing reflects the ready roster as of 2026-07-30 (Peyton catch, corrects the invented “Retreat & workshop host” / “Studio or space owner” chips). Never hardcode-extend this list in the artifact; the DB is the source.

• The gate moves from an inline card to the shipped sheet — ModalWrapper presentation="sheet" with the .sheet.co 40:60 geometry, the same one f14 uses. No new component. The inline card was a ~600px column that shoved the whole page down; what ships today (BusinessIdentityGate.tsx, #2339) keeps every behavior below — only the seat changes.
• It opens itself. The free read finishing is a SYSTEM event — there is no trigger button. Checklist step 2 is the resting state and the re-entry if she dismisses the sheet (✕ / backdrop / Esc), never the thing that opens it.
• Pane mapping: basics in the 60%, plan in the 40%. The four identity facts and the action row are the work (.co-main); the five-line “once you confirm, we’ll map” is read-only reassurance — the single biggest thing pushing content down in the card — and moves beside the decision (.co-pane). Semantic note (open, confirm before build): f14’s documented pane job is “where does this sit”; using it for “what you’ll get” is a defensible extension of that contract, not a precedent it already covers.
• A share door inside “Not quite.” If we read the wrong business entirely, typing a better name is a weak clue — the ONE ShareInline composer EMBEDS in the editor under a short lead — Wrong business? Show us the right one. (Peyton, 2026-08-02 — supersedes the door-as-link + example-nouns copy: drop a link, file, image, or text with zero extra taps). Verbatim component reuse (f15 grammar; the composer never changes by context) — a recorded SCOPED EXTENSION of the R6-3 “ShareInline survives inside BrainShareSheet only” lock: the identity-gate editor is the second sanctioned host; no second composer exists. Material saves on arrival (save-first, shipped). Decision 9 (Peyton, 2026-08-02): dropped material triggers a re-find, not a blind ride-along — the primary becomes Find my business ✦, the free-class peek re-resolves from what she shared, and the sheet re-presents Here’s what we found — is this you? (1f″) for a fresh That’s me — run ✦ BEFORE anything paid. Typed fixes skip the re-find — her correction IS the confirmation (Confirm & run ✦). The paid bundle fires exactly once, after the final confirm; the re-find is the free-class peek plus the already-metered share-save, capped ×2, then “Start over — no charge” (stays nested here). f27 stays deferred as the guard for POST-read re-runs.
• Everything shipped stays verbatim: #2339 one-question rule · P0-3 tap-required before the paid bundle · #2749 wallet subline · #2352 confirmed name = the exact web-search string · T1172 double-tap guard · editing is ONE shared toggle (the pencil on any row flips all three fields; the value itself is the button; Contact is not editable) · “Not quite” opens the same basics editor and, once open, becomes cancel — it never proceeds to the paid run. AC-BRAIN-75’s delta (editable “You do” + optional founder-name query line) has shipped, incl. the ✎ on the founder-name row (BusinessIdentityGate.tsx:244) this artifact previously omitted. Shipped labels still read “Name” / “Your name · optional” — the 2026-08-02 renames below ride the build.
• Follow-ups (Peyton, 2026-08-02): the NAME / YOUR NAME collision is fixed from both ends — row 1 reads Business name, the founder row reads Founder full name · optional (“full” earns its place: the field is a web-search string, and a bare first name searches uselessly). Contact stays read-only — it is evidence we read the right site, not a read input (the confirm payload never carries it); a wrong contact drops straight into the embedded composer (see the door ruling above — the examples-in-copy variant from earlier that day is superseded). Button ladder (same day, decision 9): face + re-proposal That’s me — run ✦ · editor typed-fix Confirm & run ✦ · material staged Find my business ✦ — the re-find sits between staging and the run (1f″). Rejected: an editable contact row (a field with no consumer, re-opens the #2977 three-editable-lines scope guard) and a dedicated contact-row door (a second door into the same correction path).
• Decision 9 pulls ONE capped pre-read iteration of that loop forward (share → re-find → re-confirm, ×2 cap); the fuller POST-read correction loop stays DEFERRED at f27, not a build target. The automatic identity-ambiguous branch (we couldn’t confirm at all) stays 1b’s, AC-BRAIN-44.

• Row anatomy is real f9 grammar (confirmed/draft/empty mixed); nothing here duplicates the detail sheets (f13/f14).
• Inventory line — RETIRED from the rail (Option B, 2026-07-26, applied globally): a founder mid-confirmation with an unresolved read failure never sees a rail summary for it — the row lives ONLY in the 1e banner while genuinely unread, plus its persistent full listing at f5. The rail's health group here carries "Needs a refresh" alone; counting rows the banner or f5 already show would rebuild the two-places split Option B removes.

• 375 mobile: mirrors the desktop's settled two zones, sources line, and Our-read row — abbreviated per the platform's existing mobile-collapse convention (full row set → desktop drawing below / anatomy → f9).
• Residual inventory line: none drawn. By this state every couldn't-read source has resolved. The rail's “1 missing” item is a different kind of gap (a section not yet researched) and stays as-is.
• Jump-nav is seated here (under the page title, sticky-under-masthead) — jump-nav sits on every populated page.

• ONE registry, three renderings: live as 1b’s walled/failed rows · persistent in the readiness-readout (f1’s rail teaser → 5a below) · cited in the Our-read brief (f18 “what we still can’t see”), never forked into playbook chapter names.
• Walled (Instagram) and login-gated (rates.pdf) rows never show “Try again” — retrying can’t pass a wall; sharing is the only real fix.
• Resolved rows keep their line, tick emerald (C12), drop their fix — the count decrements in view, never silently disappears.
• Sibling rule: the FAQ row routes to f14’s ask-queue (typed answers); everything else → f15 (shared material). Retry/re-read pricing stays unfloored pending finance (#2690) — no $ drawn here.

• “Try again” is reserved for a genuine our-side read failure — drawn with a distinct example (pricing page) so it’s never confused with walled/login rows, whose only real fix is sharing. Exhausted = a wording change on the same row, still a door; no countdowns, no self-sharpening claims.

• The locked flat .banner.warn strip (connects.html f7 / pulse.html) — glyph ring, title + one-line description, amber-tint CTA. The old boxed card (rounded-xl amber glass + 40px AlertTriangle tile + outlined button) is RETIRED. Shipped 2026-07-30 (#2921) — DriveFileScopeBanner.tsx now renders the flat strip (15px glyph ring, amber tint on the strip never the copy, compact CTA); its own comment names the boxed card retired. This frame is a match, not a target.
• CTA = the banner’s own .cta treatment, NEVER an embedded ConnectionStatusButton — REQ-C2-01 (connects PRD) is row-scoped; full ruling in connects.html 7b. Verb ledger holds: Reconnect Google. CTA opens IncrementalScopeModal with DRIVE_FILE_SCOPE — the sheet itself, the confirming/hand-off states, and the return are drawn below in 26b (previously undrawn anywhere).
• Placement per PRD §2.4: always above the relevant content, never inside a card — here above the playbook (BrainWorkspace). It is the ONLY thing that may sit between the jump-nav and “Your playbook” since the per-assistant capability panel retired from /brain (f30 folded 2026-08-13).
• One component, three surfaces: connects.html f7/7b (connects context) · this frame (knowledge) · wizard Personalize (wizard context — assistant-flows artifact does not depict it yet; flagged as mockup debt, not skipped).

• Reuses the shared .sheet/.sheet-handle/.sheet-title/.sheet-context chrome already established above (frame 13’s confirm sheet) — never a new sheet shape for this frame.
• The hand-off is annotated, not drawn — Google’s own consent screen is out of scope for this artifact; the mockup shows only the moment of departure and the moment of return.
• One component, three contexts (connects / wizard / knowledge) sharing this exact body — this frame documents the knowledge-context copy only; the connects context (“File selection”) is drawn in connects.html 7c/7d/7e, verbatim from DriveFileScopeBanner.tsx’s own string tables. Wizard context artifact remains mockup debt per frame 26’s own note.

• Findings land as dimmed .draft-block rows on the sections they belong to (f9’s row anatomy verbatim) — no confirm screen.
• ✓ Use per row; full confirm/fix/discard lives in the section’s detail sheet (D4, f13).
• Retires at build: the transient post-read OurReadSheet.tsx is superseded — the report slide-over (f18) is the ONE Our-read surface.

• ONE .pb-row grammar for every section and state — title + chip + 1–3-line summary + “detail →”; the old section-vs-summary-row fork is dead.
• Summary by type: prose 1–2 sentences · lists one roll-up line · Q&A coverage “N of M answered”, full list in the sheet. Coverage dots render ONLY while coverage is mixed (●◐○ shows shape); at complete the count alone says it — count + six filled dots was the same fact twice (say-it-once, Peyton 2026-08-11 + 2026-08-13).
• Summary-line contract (chief-designer / Fable, 2026-08-13 — Peyton: the faded lines had no definition). Four classes; the write path produces the line to fit, the row never improvises. SENTENCE (About, sound, revenue, who you serve, where headed) = her confirmed lead sentence, verbatim. FACTS (where & how, look, how you run) = ≤3 facts joined by ·. ROLL-UP (what you sell) = count + price range, never names. COVERAGE (FAQs) = “N of M answered”. Safety net only: line-clamp-2 on .pb-value — no hard character cap in the UI. Color stays --fg-70 (D7 reaffirmed); the “faded” feeling was undefined content, not the token. NEVER on the index: full FAQ answers, offering enumerations, verbatim policies. Those live in f33 (the book) and the detail sheets (one chapter).
• Two zones, locked order; all 10 sections render always, even empty or unconfirmed — 10, not 11: “How you describe it” is a LAYER inside each offering’s detail under “What you sell”, never a row in this stack. A still-drafted row keeps ✓ Use inline — the one exception to read-only.
• Confirmed values at --fg-70 (D7), 26px row breathing room; right rail: “What’s missing” then “Preview template”, prices from props.
• Row grammar v2 — the “say it once” pass (chief-designer, 2026-08-13; Peyton readability ruling). A populated row is exactly THREE things: title (.pb-section-title 15/500) · the founder’s words (.pb-value 14/300 fg-70) · ONE door (detail →, or (sound)/(look) on Branding). Nothing else by default. Provenance rides each ZONE head once as a chip-origin (V-11’s letter — the page-level “sources:” line RETIRES from this frame; its inventory job lives in the Teach sheet). A row marks only what DIFFERS from its zone default: + you told us on What you sell, ✓ you confirmed beside the sound line (the honesty mark — the full “sounds like you — because you confirmed it does” sentence lives in f22, said once, where the voice is), ◐ thin on Where you’re headed (the one genuinely thin row — verdicts render by exception only). Retired from the page: 8 per-row chip-origins · the complete-state FAQ dot string · Branding’s three quiet meta-lines (readiness detail → f22/f24) · the “optional” chip once the row holds content (it still renders on the EMPTY row, where it informs the skip decision) · the per-row “N to answer →” entry word (gaps have one home: the rail + ask flow) · the rail’s two explainer notes (“never counted”, “can’t be typed”) — the ✦ CTA’s own sub-line already says the section only fills by research. The word “still” appears zero times.
• Sub-section pass (Peyton 2026-08-17). Branding renders two labeled sub-sections — How you sound / How you look (.pb-sub-title, one register below the section title) — each closed by the STANDARD detail → door; the (sound)/(look) paren doors RETIRE. “Customer questions & policies” splits the same way: retitled What customers ask with Questions + Policies subs. Title pass: “Where & how you operate” → Where you operate. Index dividers are V3 Thread (U12), not a plain hr.
• V3 collapsed row — kit foundations Playbook (Brain) (2026-08-18, Peyton). Same drawing as kit design-foundations.html “Platform · Proposed — Playbook (Brain)”. A chapter is one .pb-row.pbk: Edge (2px left bar) · mark · title + one meta line · detail → on the right. The left Edge is the same solid 2px bar on empty and filled rows — it is the row’s spine, not a filled-only decoration. Dotted Edge is only when the double wrote it (doc or draft). The mark is what changes: Seal = confirmed (external). Node you = she typed it (internal). Node graft = empty. Lantern + dotted Edge + .need wash = a draft waiting (the one gold moment). Thread is the zone rule only — under “Your playbook” and the internal-zone seam. Empty day-one carries the f34 example-cover carousel on the page (Preview template rail retired) and the 2026-08-17 jump chips; populated demotes examples to Example playbooks · 5 →. Jump chips: About · Where you operate · What you sell · Why you · Questions · Branding / Revenue · Who you serve · How you run · Where headed. How you show up is not a playbook row or chip.
• 2026-08-17 refresh: populated index is a shelf of two books (playbook + Our read). 9b’s white “Read the whole playbook” button is retired. How you show up is not a playbook row — it lives in Our Read.
• Head chrome is two bands, not four: “Your playbook · N% of required”, then the shelf. On populated (9b / 9b-m): tap the Stark cover (opens f33). The door hugs its label (inline-flex, min 44px, left-aligned) — not a full-bleed bar (Peyton 2026-08-14 Chrome: the LOOK full-width door was too large). Quiet line under it: “The live playbook · last change …”. Then the hairline group has TWO rows, not three: “✦ Our read — what stands out & what changed” · “Teach your Brain more”. Our read is the analyst note (Option B), not a second playbook. Peyton 2026-08-13: the first draw buried the door as a 13px sibling of Our read, so it vanished on the snapshot. Same slot (Fable Option A — own place in the header, not a chapter row), but it has to read as a button. Empty day-one (9a) has no book yet, so the button does not render. Jump-nav and both zone heads stay per their 2026-07-17 locks.
• First-run header chip sits in the shell, not in the playbook (Peyton 2026-08-16): after Pulse, 9a (thin playbook) carries step 4 — “Every assistant reads this page”. After the required set is ready, 9c carries step 5 — “Drop what’s new — or re-read the site” — one door to f15. 9b is the default after they open Teach once (chip gone). Visiting /brain does not dismiss either step. Our read and research are not numbered chips. Never “learns” / “gets smarter.” Desktop (≥1024, 2026-08-16; line rescoped v4.1 same day): keeps the single center line, kicker becomes n of 5, and the chip carries its own progress line — a chip-wide .12 track 7px under the caption, lit white + glow to the required-%. v4 lit the header’s existing bottom hairline instead; Peyton on production: at 1440 that read as a bar under the whole chrome — “just the progress line, not the full bar.” The header hairline stays dark; the determinate line is chip anatomy (same two-layer edge as the band, at caption scale). Tier rule: desktop treatment at lg (1024px) and up only; everything below — tablet included — uses the 375 band as a second header row (was sm/640, which left tablets a center chip reading as detached). 375 (2026-08-16, third pass — Peyton rejected both the stacked caption and the inline line): progress BAND — a full-bleed tone-step zone (--surface-el, the active-nav “you are here” tone) as a second header row; its bottom edge is a real determinate line, lit white + quiet-active glow to the required-%, against a .12 track. One line inside the band: living mark 9×12 (assemble once, then rest) · n of 5 · words · →; whole band is the 44px tap. Amends the no-fill caption lock — band, not pill: full-bleed, zero radius, borderless, nothing pressable-looking. Still not a pill, still monochrome (Preview keeps the only amber).
• 375 first: 9a-m / 9b-m / 9c-m (#f9-m) are full-page twins — jump-nav is the v4.63 trigger + sheet (f10-m), the rail collapses to the ONE “What’s missing · N →” row above the playbook (BR1). The two-column example shelf (Business playbook + Our read) sits on the page on 9a-m only (day-one pitch). Populated 9b-m / 9c-m demote examples to the Example playbooks · 5 → hairline (AC-PLAYBOOK-16 — no standing shelf under her books). 9a-m / 9c-m carry the first-run progress band. 9a-away-m on the same section is the Assistants away line (Make them sound like you · 88%) so the cramped live shot can be scored here. Replaces the “no separate 375 tier” gap formerly flagged on the parts cut.
• Spec-vs-code (2026-08-15 — supersedes the stale 2026-08-13 note): LOOK (#3459) + the door/paint pass (#3484) are on live Pages. Shipped PlaybookPage.tsx now renders row grammar v2 — the shipped index matches 9b/9b-m: title · her words · one detail → door (Branding keeps (sound)/(look) only), provenance on the zone heads with exception marks only, no pencils / per-row ProvenanceChips / AskEntryWord / page-level “sources:” line, the hug-content Read the whole playbook button under “Your playbook” (→ /brain/book), the quiet “The live playbook · last change …” line, and the hairline Our read · Teach group. Remaining drawing debt was on THIS file, not the live page: the sibling full-page twins (f1/f4/f11/f12/f16/f17/f31 and their 375 gallery copies) drew pre-v2 chrome until they were synced to 9b/9b-m on 2026-08-15. Still specified-not-built: f18’s four-block analyst-note copy (the research-mint engine) — a content hold behind the Our-read door, not index chrome; do not restyle it.
• “Product listing copy” standalone row RETIRED from this contract (2026-08-04); renamed “How you describe it” (Peyton, 2026-08-10) per the 2026-07-31 merge ruling (AC-LISTING-28 ↗) — it lives inside each offering’s detail at f28, never as a sibling playbook row. Founder-facing label only: the listing{} key, the brain-generate branch id 'listing', channel_registers.listing, and the spec/PRD term “general product listing” all stay unchanged — renaming the data model to match the label is a defect, not consistency. Code note: the LIVE page still renders the pre-merge specified-not-built row (PlaybookPage.tsx `listingCopy` anchor, #2978) until the #3028 build lands — that residual is build lag, not contract; builders draw from f28.