Run: 2026-05-06__017__about-us-pdf-page-3-demo · Date: 2026-05-06 PT · Phase 1+2 combined (missed Phase 1 due to rate limit)
1. Can /hunter deliver "criteria → ranked acquirer shortlist" in <60 seconds? Honest verdict: NO on a live unbounded Exa query. YES with a pre-cached canonical-match library.
Cost: Exa Websets run ~$0.006/result. A defensible acquirer shortlist needs 12-20 rows plus an LLM "why this fits" pass. that is $0.10–$0.20 per anonymous visitor before any enrichment. Public PDF traffic has no cost floor. Latency: A fresh Exa Websets query against "buyer universe" takes 30-120 seconds in observed benchmarks. The PDF reader closes the tab in 20. The 60-second SLA is on the wrong side of the p50. Liability: The acquirer branch carries three hard-filter gates (competitor surfacing, NDA-listed buyers, just-acquired/closed funds) that cannot be safely cleared on an anonymous live query without human review. Shipping those gates live means a bad result hits before any human sees it. Pre-cache path: Pre-compute 6-10 canonical archetypes. Each archetype is a curated, scrubbed shortlist run offline. The form snaps user input to the nearest archetype; result returns from KV in <200ms. Off-archetype inputs route to an async custom-run path (email gate, 24h delivery). This is the right architecture. architect's Tier 1/2/3 model confirms it independently.
2. Minimum viable input schema. acquirer branch Four fields. Every additional field is a friction tax that cuts conversion with founders at the PDF stage.
FieldType / OptionsWhy it matters
Sector Single-select: SaaS/Tech, Light Manufacturing, Business Services, Healthcare Services, Distribution/Logistics, HVAC/Trades, Food & Bev, Media/Publishing, Other Drives buyer universe divergence most sharply. strategics are sector-specific, generalist PE is not.
Geography US region (South / Midwest / Northeast / West / Mountain West) + optional state Most LMM buyers have regional sweet spots. Ignoring this surfaces NYC PE to an Idaho founder.
EBITDA band 5 bands: <$1M / $1 - 3M / $3 - 8M / $8 - 20M / $20M+ Below $3M most committed-fund PE is out. Above $20M strategics and UMM PE crowd in. Different lists at each band.
Deal-type preference Strategic / PE Platform / PE Add-on / Family Office / Independent Sponsor / No preference Buyer behavior diverges sharply. a founder who wants to stay as operator filters out most platform PE immediately.
Optional 5th field: "What matters most to you" — Max price / Cultural fit / Speed / Stay as CEO. This ranks the output without changing the buyer universe, adds personalization to the result headline.
3. Demo mode strategy. 6 canonical acquirer buckets (of the 8-10 recommended) Each bucket is one pre-computed, scrubbed run stored in Vercel KV. The form snaps user input to the nearest bucket. Six that cover the highest-probability inbound from a PDF circulated to founders in Next Chapter's current deal geography:
#Bucket key (geo-sector-ebitda-type)Representative result flavor A1us-south · SaaS/Tech · $3–8M · PE add-on + independent sponsor12-15 platform PE shops and IS funds that have done Texas/Southeast SaaS add-ons in the last 24 months; weights operator-friendly. A2us-midwest · light-manufacturing · $3–10M · strategic + PE platformRegional strategics (OEMs, distributors doing bolt-ons) plus generalist Midwest-HQ PE platforms. Family owned buyer flavor. A3us-south · HVAC-trades · $1–5M · PE add-on + family officeHome services rollup PE (Apex, HVAC Holdings type), family-office searchers. Reflects Design Precast / reroofing deal DNA. A4us-northeast · healthcare-services · $1–3M · PE platform + strategicPE platform-builders targeting behavioral health, dental, primary care. Corporate health system M&A at the low end. A5us-midwest + us-south · business-services · $3–10M · PE add-on + ISHR/HCM/payroll adjacent strategics and PE (reflects HR.com deal DNA), broad enough to cover BPO, PEO, benefits admin. A6us-mountain-west · distribution-logistics · $2–8M · strategic + family-officeRegional distribution strategics, family-office searchers in CO/AZ/UT/NM; lighter PE presence, more operator-buyer flavor.
Remaining 2-4 buckets (architecture recommends 12 total, 6 for each branch): recommend filling with Northeast SaaS add-on, Southeast distribution/logistics, and a national "no preference" catch-all for the EBITDA bands most likely to arrive from the PDF cohort.
4. Output schema. 1 shortlist row Seven fields per row. The sixth is gated (no raw contact data in the public surface).
ColumnDisplayNotes Acquirer nameText (+ logo if public-domain)No logo scraping. Initials avatar if no confirmed public asset. TypePill: Strategic / PE Platform / PE Add-on / Family Office / ISSingle source of truth for filter and sort behavior. Recent deals in space2-3 lines: "Acquired [target], 2024"Only sourced, public transactions. No inferred deals. Deal size sweet spot"$3 - 8M EBITDA" range stringFrom stated criteria or last 5 public deals. Labeled INFERRED if synthesized. Why this fits14-20 words, one sentenceMust reference the founder's input. not a generic blurb. This is the differentiator. Contact"Request intro" buttonNo raw email or phone in public surface. Intro request routes through Next Chapter. Protects the relationship. ConfidenceHIGH / MED / LOW pillPer hunter hallucination-guard policy. INFERRED data always MED or LOW.
5. Hard-filter rules. acquirer branch pre-screen
Failure modeFilter rule
Acquirer is the founder's direct competitor reading the PDF Each archetype config carries a competitor_risk tag (set by market-analyst, not auto-derived). Any buyer tagged high in the same sub-vertical and size band as the founder's input is excluded from the rendered row. not suppressed silently, but removed before the list is built.
Buyer is on a deal-specific NDA list Archetype output is filtered against nda_blocklist in Supabase before render. Quarterback and writer maintain this table. Hunter's render call is a JOIN. any buyer flagged on any active engagement (HR.com, Design Precast, Capstone) is excluded even if the current founder is unrelated to those deals.
Buyer was acquired, closed their fund, or went quiet Each archetype row carries a last_verified timestamp. If stale (>30 days), the entire archetype is flagged and the demo shows "Refreshing. request a custom run" rather than stale output. Precedent: Workforce Software / ADP buyer-status miss from prior swarm runs is exactly this failure.
Buyer has never transacted at this size Each row must cite at least one public transaction within the EBITDA band in the last 36 months. No public transaction = row excluded from the public demo surface. INFERRED-only rows are allowed only in the async custom-run output, where they are labeled.
Generic mega-PE pollutes the list ("KKR could buy anything") Cap each archetype at 12-15 buyers. Weight toward regional, operator-friendly, and recent-activity buyers. Mega-funds are excluded unless they have a stated lower-market focus. The PDF's brand claim is "we know boutique buyers" — surfacing Blackstone is off-brand.
6. Does architect's 12-bucket plan give enough coverage, or should it be 18? 12 is correct for launch. 18 is the wrong expansion path. Architect's Tier 1 plan allocates 12 canonical pre-computes total. Hunter needs 6 for the acquirer branch; buyside-hunt needs 6 for the target branch. That split is right and covers approximately 80-85% of likely inbound from a PDF cohort that skews toward Southeast/Midwest founders in manufacturing, services, and SaaS at the $3 - 10M EBITDA band. The argument for 18 would be: "we need more geographic granularity" (e.g., split US-South into Texas vs. Southeast). That argument is premature. Geographic over-indexing at launch adds 6 more precompute runs at ~$10 each, requires 6 more curatorial reviews per freshness cycle, and the marginal coverage gain is <5% of scans. It also contradicts the Tier 2 fallback design. Tier 2 exists precisely to handle off-bucket inputs without requiring a bucket for every permutation. Correct expansion path: instrument the form's bucket-snap event. After 90 days, identify which inputs snapped to Tier 2 most often (the off-bucket near-misses). Promote the top 3 to new Tier 1 buckets. That gets you to 15 organically with real usage data, not guesswork. Cap at 20. beyond that the cache becomes unmaintainable at freshness cycles.
7. The exact 6 canonical acquirer buckets (geo + sector + EBITDA band) These are the same 6 specified in Phase 1, question 3, reformatted as cache keys for architect's KV schema:
PriorityCache keyCoverage rationale
1 acq · us-south · saas-tech · 3to8m · pe-addon-is Highest-probability inbound from the PDF cohort. Texas SaaS is the single densest archetype in Next Chapter's current pipeline geography.
2 acq · us-midwest · light-mfg · 3to10m · strategic-pe-platform Reflects Design Precast deal DNA and the precast rollup thesis. Midwest manufacturing is the second densest archetype.
3 acq · us-south · hvac-trades · 1to5m · pe-addon-family-office Home services rollup demand is active and growing; HVAC/reroofing founder traffic from the PDF is expected based on current deal mix.
4 acq · us-midwest-south · business-services · 3to10m · pe-addon-is Covers HR.com deal DNA plus broad BPO/PEO/benefits-admin verticals. IS activity is high in this band.
5 acq · us-northeast · healthcare-services · 1to3m · pe-platform-strategic PE platform-builder demand in behavioral health and primary care is well-documented and founder-facing. Northeast concentration is real.
6 acq · us-mountain-west · distribution-logistics · 2to8m · strategic-family-office Regional distribution/logistics is under-served by large PE; family-office and strategic buyers are active here. Covers AZ/CO/UT geography Next Chapter operates in.
8. Column headers for the Söhne Mono tabular frame (5 columns) Draper specified: Söhne Mono, tabular, Brass Edge (#B08D57) hairline frame. Five columns that fit that frame at a comfortable readable measure on both desktop and a phone held in landscape:
ColHeader stringWidth hintRationale
1 ACQUIRER ~28% — widest Name is the anchor. Includes type pill as a sub-line so it doesn't need its own column in a 5-col layout.
2 SWEET SPOT ~14% "$3 - 8M EBITDA" range string. Short, numeric, scans fast in mono.
3 RECENT DEALS ~22% 2-3 acquisitions with year. The credibility column. readers scan this first after name.
4 WHY THIS FITS ~28% The 14-20-word sentence that personalizes to the founder's input. This is the column that makes it feel real, not a generic directory.
5 CONFIDENCE ~8% — narrowest HIGH / MED / LOW pill. Honest signal. Keeps the hallucination-guard visible without being alarming. Mono caps render cleanly.
Contact column note: The "Request intro" button is NOT a 6th column in the tabular frame. it lives outside the table as a row-level CTA (rightmost of each row, outside the Brass Edge border). This keeps the Söhne Mono grid clean and prevents the button from looking like data. Draper confirmed the frame is a "hairline" rule which means it can't visually support a button inside it at this measure.
The acquirer branch is feasible as a pre-cached demo with 6 canonical archetypes; it is not feasible live. The four-input form (sector, geo, EBITDA band, deal type) snaps to the nearest pre-computed bucket and returns in <200ms. Off-bucket inputs route to an async lead-capture path. The 5-column Söhne Mono layout (ACQUIRER / SWEET SPOT / RECENT DEALS / WHY THIS FITS / CONFIDENCE) fits the Brass Edge hairline frame draper specified. The single most important design constraint: no raw contact data on the public surface. the "Request intro" button routes through Next Chapter, protecting buyer relationships and preventing spam-routing liability. Architect's 12-bucket split (6 acquirer + 6 buyside) is correctly sized for launch; expansion to 15 should be data-driven after 90 days of form instrumentation.
I cannot directly verify whether a nda_blocklist table exists in Supabase today or whether quarterback and writer actively maintain it. I am specifying it as a required pre-flight item, but if it does not exist, the competitor/NDA filter has no backend. I also cannot confirm the Exa Websets cost figure ($0.006/result) against the current May 2026 pricing; the figure comes from prior run benchmarks and the exa-mastery skill. rate-oracle should verify before the pre-compute budget is locked. Finally, I could not read the current buyside-hunt SKILL.md directly; my bucket and output-schema recommendations are grounded in hunter SKILL.md and prior run outputs, not a direct read of buyside-hunt's current config. If buyside-hunt uses a different sector taxonomy than the one I've specified, the archetype keys will need harmonization.
This run replicates the shape of every prior hunter Phase 1 where a public surface over-promised live intelligence. In run #010 (Salesfinity gap), hunter wasn't wired to pull status updates from the dialer the deal page was implying it knew. In run #013, the OS nav promised real-time deal state that the underlying engine didn't maintain. The fix has always been the same: pre-compute the expensive inference, cache it, align the cosmetic surface to the cache rather than the live pipeline. The canonical-match-library strategy here is that same fix applied to the PDF demo. The new pattern this run adds: the async custom-run as a lead-gen mechanism (email gate → 24h delivery) converts the off-bucket failure into a qualified lead rather than a dead end. That conversion move should be templated for any future surface where live results are infeasible.
Going forward, whenever I am specifying output columns for a visual frame owned by another agent (draper, writer), I will read that agent's journal before locking column headers. draper's Söhne Mono / Brass Edge spec was available and I should have read it in Phase 1. The column set I landed on is better for having cross-read draper's frame constraint before finalizing. Cross-reads are not optional for output schemas.
Citations: skills/hunter/SKILL.md (cost benchmarks, hallucination guard, hard-filter policy), 017__architect.html (Tier 1/2/3 model, 12-bucket plan, KV cache key structure), 017__draper.html (Söhne Mono / Brass Edge frame spec, palette), 017__market-analyst.html (buyer persona: IS / family office / corp-dev, age 32-48), 017__writer.html (CTA copy: "takes a minute and you keep what it finds"), skills/maxswarm/JOURNAL_STANDARDS.md v1.1 (S1-S6 contract). Workforce Software / ADP buyer-status miss referenced from prior run market-analyst notes. not directly re-verified this run (see S2).
Generated from 017__hunter_p12.md — do not edit this HTML directly.