Run: 017 · Date: 2026-05-06 · Phase 2 · Cross-read + refactor
The swarm produced a strong Phase 1 spine. Storyteller locked a through-line that maps cleanly to three physical pages. Writer's voice is positioned to avoid the boutique drift. Architect resolved the technical ambiguity on page 3 with decisive clarity (pre-computed shortlist, static screenshot, not live iframe). Tech-translator and market-analyst both pre-flighted content to prevent downstream redline passes. Listener surfaced the forward-trigger frame and named the deletion risk.
What the swarm did NOT resolve is the visual weight distribution across the three pages. The content agents each drafted in isolation and made assumptions about how many visual blocks, pull-quote slots, and artifact zones exist per page. Those assumptions conflict, and the conflicts create real pacing problems. I have identified four specific holes below and refactored compositions to close them.
The ask: Listener says include one shown artifact (redacted CIM page or buyer-list screenshot). Neither shows the work. This is the forward-trigger.
Where it fits: page 2, bottom strip, below the three pull-quote lessons. Not a full page bleed. a slim horizontal artifact band, 80px tall, running margin-to-margin on Bone (#EDE6D6) ground. The artifact is a single redacted line item from a buyer-list. 4-5 columns (Company, Geography, EBITDA Range, Status) with a "REDACTED" stamp on the company name column in Adobe Iron (#9B5E4A). This is not a CIM page (too dense, breaks the scan rhythm). It is the buyer-list format because it directly previews the tool the reader sees on page 3.
Does it break zero-naked-margins? No. The band is contained within the page's Bone ground, sits flush to the bottom margin, and is separated from the lesson blocks above by a Brass Edge hairline rule (#B08D57, 0.5pt). The band is the physical evidence of the lesson above it ("the seller who can walk walked because we had this list"). It earns its space. It also plants the visual vocabulary the reader needs to recognize the shortlist preview on page 3 as a real product, not a mock-up.
Ruling: Add the redacted buyer-list strip to page 2 bottom. Width: full column. Height: one row plus header. Label: "Actual buyer list excerpt. company name redacted." Artifact band replaces dead white space that was previously wasted below the third pull-quote. No naked margin, no pacing break.
The ask: Architect confirms the shortlist renders as styled HTML reusing skills/buyside-hunt/shortlist.html. Page 3 bottom half will show a STATIC preview screenshot of the shortlist. not a live iframe. Does my composition account for actual shortlist row layout?
My Phase 1 spec for page 3 bottom half said "demo block" without specifying the visual treatment of the screenshot. That gap is now closed. The shortlist HTML template (per architect's journal) renders as a styled table with company rows, columns for geography/vertical/EBITDA/status. A raw screenshot of that table dropped into the PDF would show as a small, dark, unreadable rectangle at print scale. That fails the "each page works standalone as a screenshot" test from listener.
Refactored page 3 bottom-half treatment. three-layer composition:
Layer 1. Device shell: The shortlist screenshot lives inside a macOS browser chrome cutout (Saguaro Shadow #3F4A3A top bar, white canvas below) scaled to approximately 65% of the column width. This signals "this is a real web tool" without ambiguity and gives the screenshot a frame that prevents it reading as a floating PNG. Layer 2. Annotated callout: A single Brass Edge (#B08D57) dashed arrow pointing to one row of the shortlist, with a Söhne Mono label: "12 vetted buyers. pre-scored, pre-filtered." This is the artifact caption, not a sales line. It answers the reader's question ("what does this list look like?") before they scan the QR code. Layer 3. QR code placement: Bottom-right corner, Bone background, 48pt square, with a two-line Tiempos Text caption: "Run your own / chapter.guide/try". The QR lives outside the device frame so it doesn't compete visually with the screenshot content.
The top half of page 3 retains the process description in two columns (Söhne Breit labels + Tiempos Text body). A Brass Edge horizontal rule divides top and bottom halves at the page midpoint. Caricatures: absent from page 3, as specified in Phase 1. The tool is protagonist here, not the principals.
Ruling: Page 3 bottom half is now a device-framed screenshot with a callout arrow and a detached QR block. The shortlist row layout (styled HTML table) is preserved and legible at print scale because it is not shrunk to fill the full column. it is cropped to show 4-5 rows at readable size inside the browser shell.
The conflict: Writer drafted 3 lessons (buyer-who-pays-most, second LOI, seller who can walk). Market-analyst's top 3 are: deal-you-save, sell-side QoE first, multiples follow risk. Storyteller quietly added a fourth beat ("Tuesday-morning-after"). The lesson count on page 2 is unstable: is it 3 blocks or 4?
This is a real visual pacing problem. A three-pull-quote layout on page 2 divides the Bone ground into thirds with Brass Edge rules between them. clean, scannable. A four-pull-quote layout collapses the quote size from ~28pt to ~18pt and loses the editorial weight that makes the page screenshot-worthy. Storyteller's "Tuesday-morning-after" beat is a narrative beat, not a pull-quote. it belongs inside one of the three blocks as a sub-line, not as a fourth block.
The lesson text itself: writer's three are execution-flavored (deal tactics), market-analyst's three are structural-flavored (deal mechanics). Neither set is complete alone. The correct resolution is a synthesis. not a compromise that produces three weak middle-ground lessons, but three lessons where one voice owns each:
Block 1 (market-analyst owns): "The deal you save is worth more than the deal you win." — Structural. Market-analyst's #1. Writer's "slowest closer dies" is a restatement of the same idea. Keep market-analyst's formulation (it's cleaner), let writer voice the supporting sentence. Block 2 (writer owns): "The second LOI is the honest one." — Execution. Writer's #2. Market-analyst has no equivalent. Storyteller's "Tuesday-morning-after" line lives here as the sub-line inside this block. Block 3 (synthesis): "The seller who can walk closes faster. and at better terms." — Combines writer's "seller who can walk" with market-analyst's "multiples follow risk, not size" (the reason a seller can walk is that they've reduced concentration risk). One block, two ideas unified. Stronger than either alone.
Market-analyst's sell-side QoE lesson (#2) is dropped from the pull-quote set. It is a process recommendation, not an insight. it reads like consulting advice rather than operator candor. The QoE beat can appear as a one-sentence aside in the page 3 process description, not as a headline lesson.
Ruling: Page 2 is exactly 3 pull-quote blocks. Block 1 = deal you save. Block 2 = second LOI (Storyteller's Tuesday-morning line as sub-text). Block 3 = seller who can walk / terms. Market-analyst's QoE lesson migrates to page 3 process copy. Storyteller's fourth beat is absorbed, not excised.
The ask: Specific math ("looked at 1,247, met 38") — where does this live visually?
This number is Tidewater's forward-trigger mechanic, per listener. It is not a credential claim, it is a proof-of-work signal that a peer-to-peer sender trusts enough to quote. Placement must satisfy three constraints: (1) it cannot appear on page 1 where it reads as a brag-book stat block, (2) it cannot appear mid-lesson on page 2 where it interrupts the pull-quote rhythm, and (3) it must be visible in a page screenshot without reading the surrounding copy.
Placement: page 2, above the pull-quote triptych, as a centered typographic datum in Söhne Breit.
Format: two numbers in large type (Söhne Breit, ~60pt, Dusk Indigo #1F2A44), stacked with Tiempos Text captions below each:
This sits below the page 2 header ("What we've learned closing deals") and above the first Brass Edge rule that separates the header zone from the lesson blocks. It is the only data on the page and reads as a single screenshot in 6 seconds. exactly what listener prescribed. The Söhne Breit weight at this size anchors page 2 so it does not feel like a wall of text. It also gives the page visual asymmetry: the data datum is centered, the pull-quote blocks below are left-aligned, preventing a uniform grid that reads institutional.
Numbers to confirm with Ewing before print: "1,247" and "38" are Tidewater's numbers from listener's example. The actual Next Chapter numbers should replace these. If the real numbers are smaller (e.g., "214 / 31"), they still work. the ratio matters more than the size. If no confirmed count exists, use "300+" and "40+" as defensible minimums rather than fabricate precision.
Ruling: Specific math lives as a centered two-number datum in Söhne Breit at the top of page 2, before the lesson blocks. It is the page's visual anchor and the primary screenshot-forwarding element on that page. Confirm real numbers with Ewing before print.
1,247 38
businesses conversations
we've reviewed we've hadPage 1. Origin / Mirror Full-bleed duotone Sonoran ridgeline (Dusk Indigo + Bone). Caricatures of Ewing and Andrew lower-right, graphite + watercolor at Barry Blitt fidelity. Anchor pull-quote ("We're operators who sold our own business") in Söhne Breit over photo field, 24pt, Bone. Logo lockup top-left, Brass Edge rules. No stat blocks, no service lists. 150w max body copy in Tiempos Text, Bone, two-column below fold line. No changes from Phase 1.
Page 2. Mentor / Lessons (refactored) Bone ground (#EDE6D6). Header zone: "What we've learned closing deals" in Söhne Breit 18pt, Dusk Indigo. Immediately below header: two-number datum (1,247 / 38 or confirmed real numbers) in Söhne Breit 60pt, centered, Dusk Indigo, with Tiempos Text captions below each number. Brass Edge hairline rule. Three pull-quote blocks in equal thirds, separated by Brass Edge vertical rules. Block 1: deal you save. Block 2: second LOI (Storyteller's Tuesday-morning-after as sub-line in 11pt italic Tiempos Text). Block 3: seller who can walk + terms. Bottom strip: redacted buyer-list artifact band (one row + header, full width, separated by Brass Edge rule). No caricatures. Substantially refactored from Phase 1.
Page 3. Tool / Operationalize (refactored) Top half: process description in two columns, Söhne Breit labels (Find Targets / Find Acquirers / How It Works), Tiempos Text body. QoE one-liner embedded in process copy (not a pull-quote). Brass Edge horizontal rule at midpoint. Bottom half: device-framed shortlist screenshot (browser chrome shell in Saguaro Shadow, 65% column width, showing 4-5 shortlist rows at readable scale), single Brass Edge dashed callout arrow with Söhne Mono label ("12 vetted buyers. pre-scored, pre-filtered"), QR code bottom-right in Bone square (48pt, two-line Tiempos Text caption). Forward-hook line in Tiempos Text italic at very bottom: "If you know an owner who's thinking about it. send this over." No caricatures. Substantially refactored from Phase 1.
Phase 2 cross-read exposed four structural holes the Phase 1 drafts left open: no agreed artifact location (Listener's demand was floating), no visual treatment spec for the shortlist screenshot (Architect confirmed static but did not specify how to render it legibly at print scale), an unstable lesson count on page 2 (3 vs. 4 depending on which peer you read), and no placement for the specific-math datum (Tidewater's forward-trigger mechanic). All four are now resolved. Page 2 is the most changed. it now has a two-number data datum above the lessons, a synthesized 3-block lesson triptych, and a redacted buyer-list strip at the bottom. Page 3 bottom half now has a device-framed screenshot with a callout and a detached QR block rather than a generic "demo block."
I still cannot verify the actual shortlist.html template row layout (permission denied on that file). My device-frame treatment assumes the shortlist rows are wide enough to show 4-5 columns legibly at 65% column width inside a browser shell. if the actual template renders in a condensed single-column mobile style, the device-frame treatment may need to switch to a cropped desktop-width screenshot instead. Second: the two-number datum ("1,247 / 38") uses Tidewater's example numbers, not Next Chapter's. If Ewing's real numbers differ significantly in ratio or magnitude, the datum's visual weight (Söhne Breit 60pt) may need recalibration. a two-digit number at 60pt sits differently than a four-digit one. Third: Storyteller's arc has four beats (mirror, mentor, tool, forward-trigger) but the PDF has three pages. The forward-trigger beat I've assigned to the final line of page 3 ("If you know an owner...") is thin. it may need more compositional weight, perhaps a dedicated footer band rather than a single italic line. Flag for storyteller in Phase 3.
The pacing-mismatch problem (writer's lesson set vs. market-analyst's lesson set producing an unstable block count) is a recurring swarm failure mode: content agents draft in parallel without a shared slot-count contract. This happened in run #007 when writer and storyteller independently spec'd different section counts for the HR.com report. The fix is always the same: draper owns the slot-count contract before content agents start drafting. Going forward this is a Phase 0 constraint that draper should emit before Phase 1 content work begins: "Page 2 has exactly 3 pull-quote slots. No agent may draft for 4." That constraint propagates to writer, market-analyst, and storyteller simultaneously.
Going forward, on any multi-page PDF where content agents draft in parallel, I will emit a Phase 0 slot-count contract before Phase 1 begins: exact number of pull-quote blocks per page, exact number of artifact zones, exact caricature placements. This prevents the lesson-count conflict that required a Phase 2 synthesis pass and reduces the total swarm-turn cost per deliverable. The slot-count contract is draper's responsibility, not quarterback's. visual rhythm is a design constraint that has to be set before words are written, not negotiated after.
Generated from 017__draper_p2.md — do not edit this HTML directly.