# Round 3 Synthesis - Cadence and Behavior: Where the Manual Fixes Break (12 interviews, simulated) *Study lead synthesis, 2026-08-07. Simulated personas: every number below is a directional prior, not validation. Round 3 of max 5. Not converged.* ## Headline finding Every trusted manual fix in the roster failed this round, and none of them failed head-on. They all failed sideways, exactly one step outside the boundary they were built for: Kevin verifies China and Dana is six weeks into an unverified Monterrey deal she herself calls "the Dongguan playbook"; Precision Metrology tests the sample and Dave's floor visit is the only thing that checks subcontractor claims, and Dave's calendar is now the ceiling; Priya's registrar-plus-references screen cleared Marsh Creek and missed that heat treatment happens at an unaudited shop in Ohio nobody thought to ask about; VeriFact certified Jenny's Ningbo floor was real in January 2025 and had nothing to say when that same factory bumped her committed PO for a bigger buyer in month eighteen; Tom's $750 deposit stops cold strangers and waved a warm referral straight into $650 of unrecovered engineering time. At the same time, the round split the roster on a second axis that matters more than pain intensity: whether the pain has an owner. Where a single person's card already moves at the failure point, money moved this round in small premium-shaped increments ($420 to a stranger inspector, $150 to hold a slot, $400 to jump a lab queue). Where the pain is organizational, it is now behaviorally proven that nothing converts: Hank could not implement his own free fifteen-minute fix in six weeks, Aisha logged a third consecutive round of zero hours under deadline, Derek's DIY monitoring went dark for the three most important weeks of his deal, and Tom's entire succession plan is a Post-it note. The surviving opportunity is not matching and not a shortlist. It is the connective tissue between trusted verification incumbents: coverage where they do not go, capacity when they saturate, sub-tier scope no checklist asks about, and watching the relationship after the audit passes. ## Scorecard | # | Persona | Wedge | Fit | Pain | New $ this round | Incumbent failure mode surfaced | |---|---|---|---|---|---|---| | 01 | Dana, DTC hardware sourcing lead | W1 | strong | 4/5 | $420 stranger inspector + $150 slot hold; live unverified Mexico exposure | Geography (China-only Kevin), availability (9-day gap), no way to vet a new inspector | | 02 | Marcus, OEM procurement | W1 (W3 merged) | medium | 3/5 | $400 lab expedite; $4,500 YTD lab spend | Lab backlog (3wk to 5wk), Dave's calendar, sub-tier claims unverifiable without a visit | | 03 | Rita, 8-person DTC (control) | W1 | no | 1/5 | none; no sourcing event since March | None; retired clean | | 04 | Tom, 63, CNC shop owner | W5/W4 | medium | 4/5 | $650 Vandermeer referral write-off | Deposit gate has a warm-referral hole; succession risk 5 months unaddressed | | 05 | Kelly, 34, fab shop successor | W5 | strong | 4/5 | $28,000 Rio Grande quote lost silently, learned secondhand | No visibility into the buyer's desk; one triage false negative in 24 calls | | 06 | Hank, foundry sales manager | W3 supply | weak (downgraded) | 3/5 | $3,610 July dead-RFQ hours; $8,930 over 4 months | Free self-named fix unimplemented 6 weeks; pain has no owner | | 07 | Priya, OEM sourcing engineer | W1 (W3 merged) | medium | 4/5 | Live risk: $18k/qtr bleed resumes if Ohio heat-treat sub not cleared in 8 weeks | Checklist scope stops at the letterhead; 3-week Deepak leave zeroed the next family | | 08 | Sal, 61, manufacturer's rep (control) | W2 | no | 2/5 | none; $9,200/yr account saved with a $150 lunch | None; human tip-off beat any system a third time; retired | | 09 | Jenny, trading co owner | W2/W1 | medium | 4/5 | $4,050 (credit + air freight) from a VeriFact-passed factory; saved ~$900-1,000/yr negotiating $450 to $405 | Point-in-time audit is blind to post-close behavior; nothing watches after month one | | 10 | Derek, search fund principal | W4 | medium | 4/5 | Higgins spend now $41,200; $150k holdback won in negotiation | DIY routine died under peak load; work invisible to the committee that holds budget | | 11 | Aisha, PE roll-up associate | W4 | medium | 3/5 | Ferris: incumbent DD flag became a real $150k holdback, first tranche paid | Her unfunded process lost to the model a third time; incumbent already owns the paid version | | 12 | Greg, EMS purchasing (control) | none | no | 1/5 | none; loss now unreferenced by anyone | None; retired clean | Roster shape check: 2 strong, 5 medium, 1 weak, 4 no (3 controls retired plus Hank downgraded on behavior evidence). Controls all behaved as designed and are formally retired this round. The study is still not skewing positive; it lost a fit this round rather than gaining one. ## Cross-cutting patterns **1. Every fix fails sideways: the boundary of the tool is invisible to its owner until a live deal crosses it.** Five personas got burned or exposed this round at the exact edge of a fix that works perfectly inside its boundary: geographic (Dana), scope (Marcus's lab does not visit sites, Priya's checklist assumes the quoting shop does the work), temporal (Jenny's audit is point-in-time), and social (Tom's deposit exempts referrals by design). None of these are product gaps in the incumbent; they are category boundaries the buyer discovers mid-deal. > "Kevin doesn't do Mexico. He doesn't do Vietnam either... Outside China I'm > just... back to hoping." - Dana > "Nobody on our side had asked the question, because our checklist assumes > whoever quotes you is the one doing the work." - Priya > "VeriFact walked that floor in January and it was real... Nobody's report > was ever going to tell me that in month four they'd shove Meridian to the > back of the line for someone paying more." - Jenny **2. The capacity ceiling went from theoretical to live: the trusted human bottleneck saturates exactly when demand peaks.** Marcus has two of four candidates blocked on a person (the lab queue at five weeks, Dave's calendar at one visit per two weeks). Priya's three-week Deepak absence did not delay the next part family, it prevented it from existing. Derek's monitoring routine, the one that produced his only real finding, went dark for the three weeks before close, the period when it mattered most. The reshoring wave is demand-synchronized: everyone's verification incumbents are backing up at once, which is why the lab queue doubled. > "Cardinal's report came back clean three weeks ago and it's just sitting > there because Dave hasn't had a day free to drive out. That's not a > decision pending, that's a calendar problem." - Marcus > "Three weeks he was out, and the manifold family didn't get delayed, it > just never started. There's a difference. Nothing was waiting on a shelf, > there was no shelf." - Priya > "If you'd asked me two months ago whether the routine would survive the > last three weeks before close, I'd have said obviously yes... It didn't > survive." - Derek **3. Money moves at the failure points, but only through trust, and trust does not bootstrap.** Dana paid $420 to a stranger and $150 to hold a slot rather than skip verification; Marcus paid a 44% expedite premium without asking anyone. But both also showed the limit: Dana has stalled six weeks on two Monterrey inspectors because they want 50% upfront and she has no way to vet them, and Marcus abandoned a backup lab because new-customer onboarding would take longer than waiting out the backlog. The purchase blocker at the failure point is never price. It is that a backup provider starts at zero trust with high friction, and the incumbent's queue is still the safer bet. > "I paid a guy I'd never used, four twenty, because Kevin was buried... It > worked out. I don't love that 'it worked out' is the plan." - Dana > "They wanted a full onboarding quote process before they'd even schedule a > sample pickup - so that went nowhere and I just waited on Precision > Metrology instead." - Marcus **4. Ownerless pain never converts, now proven behaviorally, not just claimed.** Round 2 predicted it; round 3 ran the experiment four times. Hank had a free, self-designed, fifteen-minute fix and six weeks: nothing. Aisha had a live deal, a memo due in eleven days, and a placeholder that says "pending": nothing, third round running. Tom has a five-month-old dated warning on 32% of his revenue and a Post-it. Derek's committee read a two-paragraph update with no durability section and asked only about the wire date. Where no individual owns the loss, even zero-cost fixes do not deploy, so a paid product has no entry. This kills W3 supply-side and severely narrows W4. > "Nothing stopped me. That's the honest answer. Nobody stopped me, I just > didn't do it... It's not that I don't believe the fix would work. It's that > fixing it isn't anybody's job, including mine, apparently." - Hank > "The model always wins, every single time, because the model has a partner > meeting attached to it and the LinkedIn digging doesn't." - Aisha **5. An unprompted cross-side wish emerged independently three times: has this drawing already been shopped?** Kelly wants to know before sinking two days into a quote whether the buyer already sent the drawing to three shops. She reports Priya voicing the identical want from the buy side in a trade group chat. Hank's most expensive recurring dead-RFQ flavor is the shopped print, currently caught only by Gary's peer network, hours too late. Nobody was asked about this; it surfaced spontaneously on both sides of the table. It is the only new product-shaped idea round 3 generated on its own. > "What I actually want to know before I spend two days on a quote is > whether the guy on the other end has already sent this same drawing to > three other shops." - Kelly > "A Nebraska implement company that turned out to be shopping the same > print to us and two competitors - Gary caught that one after about 6 hours > of Deb's time, not before." - Hank **6. Nothing watches after the handshake, and the round produced two concrete dollar events in that blind spot.** Jenny's $4,050 came from a factory that passed every check she has, eighteen months after it passed them; her fix is a quarterly reminder she describes as a sticky note. Derek now carries a $150,000 holdback contingent on two accounts holding revenue for twelve months, which converts post-close monitoring from a virtue into a direct financial exposure, and he has just proven he cannot sustain the routine himself. Kelly's silent secondhand losses are the same shape on the sales side: the relationship went quiet and nothing told her. > "Nobody's watching the relationship after it starts. I check it going in > and then I just hope, and hoping is not a system." - Jenny > "...whether I need someone else running it so it doesn't depend on how many > hours I happen to have that week." - Derek ## Vetoes and dealbreakers (added or hardened this round) - Upfront payment to an unvetted verifier blocks the purchase outright: Dana has a live deal, a named risk, budget, and willingness, and still has not booked a Monterrey inspector in six weeks because 50% upfront from a stranger reverses who takes the leap of faith. Pay-after-work or credible vouching is table stakes for any brokered verification. - Onboarding friction on a backup kills it even during an incumbent outage: if engaging the alternative takes longer than the incumbent's backlog, the buyer waits (Marcus's Dayton lab, Dana's Monterrey names). A broker must be usable in under the incumbent's delay, first time, no intake process. - The artifact must be legible to the budget audience or there is no budget path: Derek's monitoring never appeared in his committee update; Aisha's memo section cites only what Beacon Ridge hands her. Work that cannot be cited upward does not get funded, however real the finding. - A PDF is not negotiating leverage; the person raising it is. Derek: "If a vendor had handed me a formatted PDF with the same fact on it, I'm not sure it changes the negotiation at all." Evidence products must arm the human, not replace the moment. - Ownerless, diffuse losses are unsellable at any price, now behaviorally confirmed (Hank downgraded to weak on this basis; W3 supply killed). - Quality floor beats price on evidence: Jenny will not reuse a $220 audit that lacked photo timestamps and registry cross-checks even on low stakes, while she negotiated her trusted provider down 10% without switching. Cheaper-but-thinner loses to the same-brand discount, every time. - Standing vetoes unchanged: vendor self-reported anything is dead on arrival (Priya took the Ohio subcontractor's phone capability statement and explicitly distrusts it); the relationship layer stays off-limits to software; no commission pricing to the Xometry-burned. ## WTP read (with required skepticism) No WTP was solicited this round by design (no pitch, no vibe check), so all stated bands stand unrevised and untested. What the round added is revealed prices at the failure points, which are the best priors yet and still simulated: - Premium-for-coverage is real but small: $420 for a one-off stranger inspector (vs Kevin's $380-450 baseline), $150 to buy nine days of slot time, $400 (44%) to compress a lab queue from five weeks to three. The willingness exists at low three figures per event, paid instantly, no approval sought. Nothing suggests four figures per event for sourcing buyers. - The negotiation data cuts both ways: Jenny extracting $405 from a $450 audit "without much pushback" says incumbent margins have room, and says a new entrant faces a buyer who now knows how to squeeze. - The cost of the post-close blind spot got its first hard numbers: $4,050 (Jenny, realized) and $150,000 of contingent holdback exposure (Derek, structural). Neither has ever been quoted a price for closing it, which makes this the largest unpriced gap in the study. - Sub-tier discovery doubles event volume: Buckeye spawned a second full qualification (Meridian plating), Marsh Creek spawned a heat-treat audit decision ($1,200-1,300 if Deepak flies). R2's "2-3 events per year" for low-cadence buyers understates billable events when each primary pulls in its subcontractors. - Still zero purchase evidence for anything shaped like the product. Every number above is what they paid incumbents and strangers, not us. ## Wedge status after round 3 - **W1 Verified Shortlist, now Verification Network: ALIVE, sharpened, still the lead.** The shortlist framing has been dead since R2; what R3 proved is that the incumbents' failure modes are systematic, not anecdotal, and they are exactly four: geography (Kevin is one man in three provinces), availability (nine-day gaps, five-week queues, one Dave), sub-tier scope (three separate subcontractor misses across Marcus, Priya, and by extension every checklist in the roster), and verifier trust bootstrapping (the Facebook group is the industry's actual inspector-discovery layer and it is vetted by comment threads). Buyers already pay premiums at each failure point. The surviving product is a brokered network of vetted inspectors and labs with pay-after terms, zero onboarding, sub-tier scope by default, and vouching the buyer can check. R4 must test whether trust transfers through a platform's vouching at all, because Dana's six-week Monterrey stall says it may not. - **W2 Agent-amplified rep desk: DEAD, control retired.** Sal's third confirmation: the human tip-off arrived weeks before any system could have known, again, and cost $150. Closed. - **W3 Castings/forgings desk: DEAD as a standalone; folded into W1.** R2 gave it one more round to find a payer. It found the opposite: Hank, the only supply-side voice, was downgraded to weak after proving the pain has no owner, and the buy side (Priya, Marcus) is now indistinguishable from the W1 verification story with castings vocabulary. The two useful residues: the shopped-print signal (pattern 5) and the sub-tier scope requirement both originated here. - **W4 Succession demand book: WOUNDED, near dead as a diligence product, one live pivot left.** The R3 evidence is brutal for the original shape: the DIY routine dies under load, the artifact is illegible to committees, the incumbent (Beacon Ridge) already produces the killer finding by accident inside a fee Aisha does not control, and the one term-moving outcome came from a human talking, not a report. What survives is the post-close pivot: holdbacks are now proven twice as the mechanism that turns durability findings into money ($150k Derek, $150k Ferris, first tranche paid), and a holdback creates a named owner with direct financial exposure and a 12-18 month window, which is everything the diligence version lacked. R4 tests exactly one thing here: will Derek, who owns $150k of contingent exposure starting in three weeks and just proved he cannot run the routine himself, pay someone to run it. - **W5 Commission rep in software: WOUNDED, redefined toward desk visibility.** Kelly's fit stays strong but the R3 loss ($28k Rio Grande) repeats the R2 pattern: she does not lack leads, she lacks any signal from the buyer's desk after she quotes, and she names the pre-quote version (shopped-drawing detection) unprompted. Triage delegation is proven at a known error rate (1 false negative in 24 calls, self-corrected); technical delegation stays frozen. What is commissionable is thinner than R1 dreamed but sharper: pre-qual triage plus quote-pipeline intelligence. R4 reveals that shape to her and prices the Rio Grande counterfactual. ## What we still do not know 1. Does trust transfer through vouching? Dana will pay $420 sight-unseen under time pressure but has stalled six weeks on strangers who want money upfront. Whether a broker's guarantee, escrow, or track-record page actually unblocks the booking is the single decision that makes or kills the W1 network. R4 tests it against her live Grupo Herrera deadline. 2. What is the minimal Dave-preserving unbundling? Marcus cannot picture verification without Dave's eyes, but Dave reviewing evidence someone else collects is different from Dave flying. Where exactly is the line (live video walkthrough, third-party inspector Dave briefs, photos with registry cross-checks)? 3. Would Jenny pay anything for post-close monitoring, and what would it even watch? The gap has never been named or priced to her. What signal would have caught the Ningbo bump eleven days earlier, and does it exist outside the factory's own walls? 4. Will a holdback owner pay? Derek's $150k contingent exposure begins at close. Whether he funds monitoring he provably cannot sustain himself is the last live test for W4, and the buyer, budget, and deadline are all already in place. 5. Is the shopped-print signal buildable and wanted at a price, or just a wish? Both sides want it; nobody has described what evidence would satisfy it or what it is worth. R4 floats it to Kelly (and Priya) as a labeled concept. 6. True event frequency under sub-tier scope: if every primary candidate spawns 1-2 subcontractor verifications, low-cadence buyers may generate 2-3x the billable events R2 assumed. Needs a count, not an impression. 7. Real WTP. Three rounds, zero purchases, all priors. R4's labeled vibe checks and R5's firm asks are the only remaining instruments. ## Convergence call NOT converged. The wedge board moved again (W3 folded into W1, W4 pivoted from diligence to post-close, W1 re-specified around four named incumbent failure modes, a new cross-side signal appeared unprompted), one persona was downgraded on behavior evidence, and the entire study has run zero concept contact by design: no persona has ever heard the product described. Round 4 is the arc's reveal round and four of the seven remaining unknowns can only be answered by putting the relevant concept in front of the relevant persona against a live deal. Round 4 proceeds with eight active personas (controls Rita, Sal, Greg retired; Hank retired on the ownerless-pain finding).