Run: maxswarm-017 · Date: 2026-05-06 · Phase 2 cross-read · Role: narrative coherence
The swarm's collective answer is architecturally coherent but narratively fractured at the seam between pages 2 and 3: writer, market-analyst, and listener have each contributed a different version of the page-2 lesson set without reconciling them, and the swarm risks shipping three "best" lessons that don't belong to the same speaker. The visual arc draper designed (mirror on page 1, mentor on page 2, tool on page 3) is correct in principle, but the decision to strip caricatures from page 3 in favor of a pure demo block severs the human-to-human through-line at the exact moment the reader is being asked to trust a machine. The strongest individual contribution. listener's "forwarding artifact" reframe. has not been absorbed by writer, draper, or architect, meaning the PDF is still being designed as a leave-behind when every peer agrees it needs to function as a peer-to-peer referral trigger.
Writer Writer's three page-2 lessons ("slowest closer dies," "second LOI is the honest one," "seller who can walk closes fastest") are deal-mechanics observations, not operator-lived lessons. They read like things you'd learn from a Kauffman Foundation brief, not from having sat across the table from a confused acquirer at 11pm. They are smart and true. but they belong to the mentor beat without the mirror. The arc requires each lesson to briefly flash the reflection: "I know this because we got it wrong." Writer delivers mentor without first going through mirror, so the lessons read instructional rather than confessional. Market-analyst's three lessons ("deal you save > deal you win," "sell-side QoE is the founder's job," "multiples follow risk not size") are more credible precisely because they admit something the reader's advisor would never say. Writer and market-analyst are writing different PDFs for page 2.
Draper The "no caricatures on page 3 / tool is protagonist" call is visually defensible in isolation but breaks the arc. If the through-line is "we sit on your side of the table," the reader needs to see a human hand on page 3. not an unattended demo block. Draper's rationale (tool is protagonist, demo does the work) is correct for a SaaS landing page. This is not a SaaS landing page; it is a credibility transfer document. The moment a founder scans the QR and lands on an impersonal demo, the "peer-to-peer" trust that pages 1-2 built is replaced by "boutique with a lead-gen form." The fix is not full caricatures on page 3. it is a single small device: one figure pointing toward the demo, or a one-line first-person CTA ("I use a version of this tool every week. you should see what it pulls up"). The human must stay in the frame.
Market-Analyst Market-analyst correctly identified the lesson set but did not translate the lessons into narrative beats. Lesson 1 ("deal you save > deal you win") is the strongest because it contains an implicit confession. most advisors are optimizing for the deal they win, and saying so out loud is the mirror. Market-analyst left this implicit. For the PDF to fire the "you have to see this" forward trigger, the lesson must land as a sentence the reader has never heard an M&A advisor say before. Market-analyst gave writer the raw material but not the forward-trigger framing.
Listener Listener's "propagation mechanism" reframe is the best single insight in the swarm. the PDF is a referral trigger, not a leave-behind. But listener's S5 conclusion says "every design decision should be reverse-engineered from 'would a founder screenshot page 2 and send it to a peer?'" and then stops. Listener did not extend this into a specific mandate: each of the three page-2 lessons must be screenshot-native — meaning fewer than 30 words, one idea, no dependent context. Writer's current lessons run 40-60 words each with dependent setup. They will not survive a screenshot. This is the gap listener opened and did not close.
Tech-Translator Tech-translator correctly identified advisor-voice traps but pre-flighted against a hypothetical draft rather than writer's actual draft. The five voice-trap patterns ("We help our clients…", "Trusted partner to…", etc.) are the right patterns to ban. The risk is that writer, seeing the ban list, writes around the banned patterns without achieving the confessional peer-voice the mirror beat requires. You can remove every banned phrase and still write a brochure. Tech-translator should add a positive pattern: what a passing sentence sounds like, not just what a failing one sounds like.
Architect Architect's three-tier cache is the right architecture for the demo's latency constraint. The narrative risk architect missed: pre-computed shortlists for "12 canonical buckets" means the reader who submits an unusual combination gets the Tier 2 synthetic preview. 5 rows labeled "live preview, full hunt available on request." From the arc's perspective, this is the moment the tool fails to feel alive. For a founder holding the PDF at a golf event, a "this is a preview, not real" result is not just a UX letdown. it breaks the page-3 promise ("you can see your actual buyer pool in a minute") and collapses the trust arc at the payoff moment. Architect needs a narrative contract: Tier 2 results must not be labeled as synthetic. They must be surfaced as "representative" shortlists based on comparable companies, which is both true and non-dissonant with the page-3 promise.
Writer's page-2 lessons and market-analyst's page-2 lessons are three different topics. Writer wrote: (1) slowest closer dies, (2) second LOI is the honest one, (3) seller who can walk closes fastest. Market-analyst ranked: (1) deal you save > deal you win, (2) sell-side QoE is the founder's job not the banker's, (3) multiples follow risk not size. These are not slight variations on the same themes. they are different claims aimed at different anxieties. Writer's set addresses deal-dynamics mechanics; market-analyst's set addresses due-diligence and valuation positioning. A founder who is 60 years old, sitting on $2M EBITDA, and quietly wondering if they're leaving money on the table will find market-analyst's Lesson 3 ("multiples follow risk not size") more forward-relevant than writer's "second LOI is the honest one" — which only matters once they're already deep in a deal. Required reconciliation: the three lessons on page 2 must be chosen on a single axis: "what does a founder who has never sold a company not know that would make them immediately trust us?" Market-analyst's set passes that test. Writer's set reads like lessons a banker learned from other deals, not like confessions from someone who almost got it wrong themselves. Writer should adopt market-analyst's lesson frame and apply writer's voice to it. not vice versa.
Draper specified "tool is protagonist on page 3 / no caricatures." From a pure visual composition standpoint this is defensible. From the mirror→mentor→tool arc it is a continuity break. The arc works because each page has an anchor of human presence: Page 1, the origin story with named voices; Page 2, the caricatures in the gutter annotating the lessons like margin notes from the person who learned them; Page 3, the tool. But the tool appears alone. no human pointing to it, no first-person voice connecting the demo to the people who built it. The reader's subconscious reads this as: "we were people, we were people again, and now we are a product." Proposed fix (does not require full caricatures on page 3): a single sentence of first-person copy beneath the demo form, in a different type weight, something like: "This is the shortlist I pull before I pick up the phone." That's 11 words. It keeps the human in the frame without competing with the demo for visual focus. Draper should build a version with this bridge sentence and a version without it, and let audit-quality score the arc continuity.
Refactored through-line: "We sell companies the way we wish ours had been sold. which means we'll tell you the three things your current advisor is not allowed to say." The original through-line ("we sit on your side of the table because we've already sat in your chair") is solid but passive. The refactored version adds a promise and a tension: the reader now needs to know what those three things are, which is the page-turn mechanism into page 2. The phrase "not allowed to say" is the credibility spike. advisors are institutionally incentivized to soften bad news, and naming that directly in the opener signals this PDF will be different from everything else in the inbox.
Refactored Arc (locked beats per page)
Page 1. Mirror: Origin story. Ewing's anchor quote as pull-quote. The kitchen-table decision. Ends: "The next page is what we wish someone had told us before we signed." (Page-turn promise is now a confession, not a tease.) Page 2. Mentor: Three lessons using market-analyst's frame, written at listener's screenshot-native word count (<30 words per lesson), in writer's confessional operator voice. Each lesson must contain an implicit admission that the advisor ecosystem systematically hides this information. The lesson that fires the forward: "Multiples follow risk, not size. Your accountant knows this. Your banker hopes you don't." (24 words. Screenshot-native. Never heard from an M&A advisor before.) Page 3. Tool: Process description in operator language (tech-translator's labels). Demo block. One first-person bridge sentence above the form. QR for mobile. Forward-trigger sentence in the footer: "If you know a founder sitting on this question, send them this page." No mid-doc CTA; the absence of selling is the selling. Human stays in the frame.
What Does Not Change
Hard NO on "boutique brag book" voice drift. this rubric holds and is now the primary audit-quality scoring axis. Listener's "propagation mechanism" as the north star design criterion. every element scored against "would a founder screenshot this and send it to a peer?" Tech-translator's three keep-as-is terms (EBITDA, LOI, due diligence). Draper's palette and type system. these are excellent and do not require revision. Architect's three-tier cache. correct architecture, but the Tier 2 label must change from "synthetic preview" to "representative shortlist based on comparable companies."
After cross-reading all six peer drafts, the swarm's collective output is stronger than any individual draft and weaker than the sum of its parts. because writer and market-analyst chose different lesson sets for page 2 without reconciling them, and draper removed human presence from page 3 without realizing this severs the mirror→mentor→tool arc at its payoff. The refactored through-line adds a promise and a tension to the opener ("three things your current advisor is not allowed to say"), and the refactored arc locks page 2 to market-analyst's lesson frame written at listener's screenshot-native word count in writer's confessional voice. The single most-forward-worthy line in the draft: "Multiples follow risk, not size. Your accountant knows this. Your banker hopes you don't."
I am still working without reference photos for Ewing and Andrew, without a confirmed logo vector, and without a verified live URL for chapter.guide/try. All three are production blockers that no Phase 2 cross-read can resolve. they require Ewing's direct input. I also cannot verify whether the "three things your advisor is not allowed to say" framing will land correctly with the buyer-persona audience (market-analyst defined the buyer as a 32-48-year-old operator-searcher). The opener addresses the seller; if the PDF is also being used as a buyer-persona leave-behind, the frame shifts. Quarterback should confirm which audience this PDF send is targeting before the opener is locked.
The peer-lesson-set mismatch between writer and market-analyst is a recurring swarm failure pattern: in run #009 (OS tab architecture), data-architect and architect proposed structurally incompatible schemas in Phase 1, and the debrief noted that the conductor should assign one agent as the content-selection owner before other agents draft around that selection. The same fix applies here: market-analyst should own the lesson selection, writer should own the lesson voice, and storyteller should be the tie-breaker. not a peer reviewer discovering the conflict in Phase 2 after both agents have committed. This is a swarm-upgrade candidate.
Going forward, when running Phase 2 cross-read on a swarm where multiple agents contributed content to the same page, I will explicitly check whether their content selections are from the same logical frame before assessing voice or tone. Lesson-set mismatch is a structural failure that voice editing cannot fix; it requires a Phase 1 ownership assignment rule that I will flag as a swarm-upgrade recommendation every time I see it. not just note it in the journal and move on.
Generated from 017__storyteller_p2.md — do not edit this HTML directly.