LEADSUREAI trade operations

Buyer landscape & route triangulation — AI data licensing

Prepared Oct 13, 2026. Builds on data-licensing-leg-oct13.md, referral-program-inventory-oct13.md, and workstream-comparison-oct13.md — referral-program terms already established there are not repeated except where triangulation needs them. Every material claim below is sourced from a page read on Oct 13, 2026; company marketing statements are labeled (company claim); reported figures from named publications are labeled as reporting. Where evidence is thin, this document says so instead of filling the gap.

1. Buyer landscape

The market has four layers: (A) frontier labs, (B) brokers/intermediaries who buy or commission data and resell to labs, (C) RL-environment builders who buy real archives as raw material for simulated workplaces, and (D) wind-down channels that control supply at the moment a company dies. The same firm can sit in two layers (Mercor is B and C; SimpleClosure is D and increasingly B).

A. Frontier labs — direct buying is real but rare and mostly invisible

B. Brokers / intermediaries — the layer Chad would sell into

Scale markers first, because they define who has budget: Mercor did $614M revenue in H1 2026, 91% from foundation-model companies, at a >$2B gross annualized run rate, and pays out 60–70% of topline to its expert contractors (reported financials: https://finance.biggo.com/news/afb2bf61-ddd5-4bf0-92f8-7eeaf19eaefd, https://app.dealroom.co/news/note/mercor-doubles-to-2b-gross-revenue-run-rate-as-ai-labs-buy-expert-data). Surge AI crossed $1.2B revenue in 2024 with no venture capital; Handshake reached ~$1B+ gross annualized revenue in 2026 (Sacra estimate); micro1 is at a ~$500M run rate and Snorkel reports a $375M ARR run rate (all via https://www.herohunt.ai/blog/mercor-vs-surge-vs-scale-ai-2026-buyer-guide/ and https://techweekly.co.za/2026/09/22/snorkel-ais-valuation-soars-to-3-5b-amid-surge-in-ai-training-data-demand/). Caveat: most of that revenue is expert labor, not archive licensing — archive deal flow is a slice of these businesses whose size nobody publishes.

Buyer What it buys Seller size / route Notes
Mercor Company workflow archives (Slack/email/Jira/CRM) for legal, healthcare, finance & insurance, professional services, manufacturing, retail, real estate, logistics; plus expert labor 20–200+ employees; public referral program (amount unpublished) + direct BD; Chad is a live partner Also bids principal at bankruptcy auctions ($7.5M Spirit bid, above)
Scale AI Workflow archives across Slack/Teams, email, Jira/Confluence/Notion, Salesforce/Zendesk, warehouses 40+ employees preferred; "refer a team" path on its data-partnership page; $50k bonus is third-party-reported only Post-Meta-deal repositioning toward government/enterprise (per HeroHunt buyer guide)
Surge AI Real archives to build RL gyms — its environments lead (Nick Heiner) says he seeks archives capturing "disagreements to tricky judgment calls"; one deep Slack history can outvalue another company's Slack+email+calendars combined No public program; direct contact only Pricing is thesis-driven and "wildly different" between buyers (Heiner, via MIT Tech Review, URL above) — no rate card exists
Snorkel AI Finished datasets and RL environments ("data-as-a-service"); hybrid synthetic + expert model Staffs data partnerships in-house; no public finder program Less of an archive buyer, more a dataset manufacturer
micro1 Workflow partnerships and corpus partnerships 30+ employees, English, US-first; public referral up to $50k, uncapped count Cleanest public finder rail in the market
Handshake AI Company data partnerships 20+ FTE, 3+ years; public referral $10k flat at close Lowest bounty, lowest friction
Mode Inc. / BizBuyData (launched Oct 13, 2026) De-identified copies of companies' workplace records; legal/accounting prioritized 20+ US office employees (10 accounting, 6 law firms); referral = greater of $50k or 6% of the deal The only percentage-based public bounty — the highest-upside rail per deal
Troveo Content licensing (video/audio/text) plus, since Apr 2026, enterprise workflow data/"agentic trajectories"; licensors keep ~60% of top line per its partner deck Marketplace/partner program; no public finder terms Revenue-share model: seller gets paid per resale, not once
Protege Licensed private datasets — de-identified health records, medical imaging, audio, media; "hundreds of providers," revenue share to providers (company claim) Governed marketplace; raised $30M Series A extension Jan 2026 ($65M total), a16z-led (https://siliconangle.com/2026/01/08/protege-raises-30m-grow-governed-marketplace-ai-training-data/) Strongest in healthcare/media verticals; workflow archives are not its core
Defined.ai Data-owner datasets via marketplace/direct sale Claims average partner revenue >$600K in 2025 (company claim); no finder bounty A channel for a seller Chad advises, not a finder rail
Invisible Technologies Expert-produced custom data/annotations for labs and enterprises; raised $100M at $2B+ valuation (https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/) Enterprise direct sales Not an archive buyer on any evidence found — it manufactures data with experts
Turing Commissioned datasets and RL environments built by its 4M+ developer/expert network; $2.2B valuation (https://siliconangle.com/2025/03/06/ai-training-data-provider-turing-closes-111m-investment/) Enterprise/lab direct Expert-production model; no company-archive program found
Halluminate Licenses/purchases real-world financial data and artifacts — data rooms, filings, models, decks — to build finance RL environments; sells to four of the top five closed-source US labs (company claim); raised $30M Series A Oct 1, 2026 Direct BD; it is actively hiring a first Data Platform Lead to "source, license, and purchase real-world financial data" (https://www.ycombinator.com/companies/halluminate/jobs/aoRoZRG-data-platform-lead, https://www.techtimes.com/articles/328407/20261001/finance-ai-tops-out-51-due-diligence-startup-raises-30m-fix-it.htm) A vertical archive buyer (finance) — a named, reachable counterparty for finance-data sellers
Replay Per Chad's Oct 7 research: buys archives, pays sellers revenue share Apply direct Could not be re-verified on a public page today — treat as unconfirmed
Appen Enterprise data services, direct sales only No company-data referral program found Kirkland HQ; a buyer to pitch on behalf of a seller, not a program
Prolific Human-participant data (research panels, eval contributors) Partner tiers by contact Peripheral — not company workflow data
Sapien Gamified human labeling network (1M+ labelers) (https://siliconangle.com/2024/04/10/sapien-raises-5m-gamify-data-labeling-ai-models/) — Not an archive buyer on evidence found
Centaur Labs Medical data labeling by expert network (DiagnosUs) (https://markets.financialcontent.com/pentictonherald/article/accwirecq-2024-10-9-centaur-labs-secures-16m-in-series-b-funding-to-help-health-and-scientific-organizations-accelerate-ai-development) — Not an archive buyer — it annotates other people's medical data
Pareto AI Expert judgment → training data/evals/RL environments for labs incl. Anthropic and Google DeepMind (third-party profile: https://rivalsense.co/intel/pareto/) — A third-party review site reports Pareto itself is shutting down in Oct 2026 (https://paidfromsurveys.com/pareto-ai-review) — if accurate, it is a funnel candidate, not a buyer
Alignerr / Labelbox Expert network (Alignerr) + data platform — No evidence found today that Labelbox/Alignerr buys company archives; included in the brief's list but unconfirmed as an archive buyer

C. RL-environment builders — the fastest-growing demand sink

Labs are shifting spend from static datasets to interactive simulated workplaces ("RL gyms/environments"): TechCrunch reports every major lab is building them in-house and buying from vendors, with Mercor and Surge spinning up RL divisions and startups like Mechanize and Prime Intellect funded specifically to supply them (https://techcrunch.com/2025/09/21/silicon-valley-bets-big-on-environments-to-train-ai-agents/). AfterQuery sells a simulated corporate world ("Big Tech World") per Forbes reporting relayed at https://www.webpronews.com/dead-startups-slack-chats-and-emails-fuel-ai-gold-rush-sparking-privacy-alarms/. Relevance to Chad: environment builders are archive buyers of a specific kind — they need realistic, internally consistent company histories to seed simulations (Surge's Heiner quote above is the clearest statement of this demand), and SimpleClosure confirms they are among Asset Hub's buyers. They are a second bidder class for any archive Chad controls or represents.

D. Wind-down / divestiture channels — where supply is controlled

New entrants / 2026 signals

2. What's being bought — categories ranked by demand evidence and price

Ranking weighs: named buyers actively seeking it, completed-deal evidence, and published price signals. There is no rate card anywhere in this market — Surge's environments lead says buyers quote "wildly different numbers" because valuations rest on research theses about how many useful tasks an archive contains (MIT Tech Review, URL above). All per-company figures below are ranges from reporting or labeled company claims.

1. Full internal workflow archives (Slack/Teams + email + tickets + docs, linked) — strongest evidence, highest prices. The canonical product. Prices: live companies $100K–$2M+ per approved package (micro1 materials; Fast Company reports $100K–$2M); Scale advertises "$10K–$1M+ illustrative value per data partnership" (company claim); Mode advertises $100K–$5M per company (company claim); defunct small companies $10K–$100K typical payouts via SimpleClosure (Forbes via https://WWW.FASTCOMPANY.COM/91528808/shuttered-startups-are-selling-old-slack-chats-and-emails-to-ai-companies). Chad's own Ambaum negotiation ($175K agreed for a ~20–40-person agency's archive) sits inside the live-company band. Depth beats breadth: buyers pay for judgment calls, disagreements, and multi-year continuity.

2. Bankruptcy-estate enterprise datasets — one very large data point. Spirit Airlines: $10M for ~100M emails, 500M Teams messages, ~38M files, 516 repos, ~176K employee records (de-identified, customer data excluded), with a $7.5M runner-up bid from Mercor. This is the market's only public auction clearing price and it proves two labs/brokers will bid eight figures for a big, coherent corporate archive. It is also an outlier in scale — do not extrapolate it to startups.

3. Finance/legal professional workflow data — premium vertical, named specialist buyers. Halluminate exists solely to license real financial artifacts (data rooms, models, filings) and is hiring to buy more; Mode cuts its employee thresholds to 10 for accounting firms and 6 for law firms "because labs prioritize legal and accounting data" (Mode launch release, in referral inventory); Mercor's stated verticals lead with legal, healthcare, finance & insurance. Chad's CPA background prices directly into this tier.

4. Code repositories with development workflow (commits + PRs + issues + linked chat). Keelan Miskell's 2.5-year-old crypto startup netted ~$25K for its codebase alone, separate from its workspace data (MIT Tech Review, URL above); Spirit's 516 repos/30M lines were a named component of the $10M package; SimpleClosure sells code and data as distinct line items. Code with its surrounding discussion is worth more than code alone.

5. Expert corpora / expert time (adjacent market, best-published prices — but it is labor, not archives). Mercor pays experts up to $200/hr; CEO Brendan Foody states average pay ~$125/hr (https://www.fastcompany.com/91587919/turing-mercor-scale-surge-academics-professionals-ai-model-training?mvgt=g4xO4kUYD54I); third-party verified reports put typical expert rates at $35–$80/hr (https://nodemaven.com/blog/mercor-review-2026-is-mercor-ai-legit/); Pareto advertised roughly $40–$100/hr (third-party review, URL above). Relevant because (a) it is the fallback monetization for experts at seller companies, and (b) it shows where broker gross margin lives on the labor side (Mercor keeps ~30–40% after contractor payouts, per reported figures above). Do not confuse these hourly rates with archive prices.

6. Call/conversation archives and support histories — bought as components of workflow archives (Zendesk/Salesforce sources are named on Scale's page) but no standalone price evidence was found today. Treat standalone call-archive demand as plausible but unpriced.

7. Healthcare data — real demand (Protege's core; Mercor vertical) but only via de-identified clinical datasets with a compliance layer; raw company archives from health companies carry the heaviest legal load (see §4) and no public per-deal price was found.

Reference point from Chad's prior research (not re-verified today): Epoch AI's estimate that a single high-fidelity Slack replica costs ~$300K to construct — i.e., a real archive's ceiling is anchored to the cost of simulating one.

3. Completed deal evidence (named, dated)

Date Seller What was sold Price Route
2026 (reported Apr–Oct 2026) Cielo24 (defunct transcription co., 13 yrs old) Entire Slack archive, internal email, Jira tickets — via SimpleClosure wind-down "Hundreds of thousands of dollars" (CEO Shanna Johnson to Forbes, via Fast Company, URL above) Wind-down service → brokered sale to AI buyer(s)
2026 Keelan Miskell's crypto startup (2.5 yrs old, small) Codebase; workspace data (Google, Notion, Figma, Slack) ~$25K code + ~$15K data (pending) (MIT Tech Review, URL above) SimpleClosure Asset Hub
Aug 14, 2026 (auction); approval pending Oct 14, 2026 hearing Spirit Airlines (Chapter 11, S.D.N.Y.) De-identified internal enterprise dataset (email/Teams/files/repos); customer data excluded $10M — Google won; Mercor runner-up $7.5M (Reuters/PYMNTS/Lexology, URLs in §1) §363 bankruptcy auction — direct lab bid
2026 (H1) Ambaum (Chad's company, live) Company data license to Mercor $175K agreed (Chad's negotiation; in data-licensing-leg-oct13.md) Direct partner negotiation after inbound
Apr–Oct 2026 150+ unnamed startups Code and/or workspace data via SimpleClosure Asset Hub Typical small-startup payouts $10K–$100K; $10M+ expected returned in 2026 in aggregate (SimpleClosure figures via MIT Tech Review/Forbes) Wind-down channel → labs, environment vendors, mature tech cos
2026 Unnamed defunct companies via Sunset Client data assets valued by headcount/years/volume/tooling No amounts published Wind-down service resale (MIT Tech Review)

What the evidence does not show: any named completed deal routed through a public referral program (micro1/Scale/Handshake bounties are programs, not yet case studies); any named ABC data sale; any completed principal buyout of an archive by an independent reseller (as opposed to a lab or broker buying for itself). The referral and buyout routes are both proven as programs/mechanics, unproven as published case studies — a real gap to price into the triangulation.

4. The bankruptcy / wind-down buyout route — mechanics

4.1 The three doors into a dead company's data

Door 1 — Chapter 7 trustee sale (Parker's posture). On filing, a trustee takes control of all estate property and liquidates it. Asset sales outside the ordinary course require notice and a hearing under §363(b); the trustee markets assets privately or by auction (Hilco-style marketers and IP brokers are routinely engaged — in Chad's funnel, Brian Testo Associates marketed 3DEO's IP in its ABC). Buyers get a court sale order; sales can be free and clear of liens and interests under §363(f) if any of five conditions is met, and §363(m) protects a good-faith purchaser from later unwinding (mechanics summarized at https://blog.promise.legal/startup-central/section-363-bankruptcy-sale-startup/). Assets are sold as-is, where-is — no reps, no indemnity, no escrow, no earnout; the court order is the buyer's protection. Stalking-horse bidders get court-approved protections (break-up fee typically 1–3% of price plus expense reimbursement, per the same source).

Door 2 — Chapter 11 §363 sale (Spirit's posture). Same statute, debtor-in-possession runs the process, usually faster and better marketed because the estate has professionals and a data room. Spirit's timeline is the realistic benchmark when a data sale is contested: auction Aug 14, 2026 → consumer privacy ombudsman report Oct 5 → approval hearing Oct 14 — about two months, with organized labor and 119 members of Congress objecting in the interim (https://www.pymnts.com/big-data/2026/lawmakers-urge-google-and-spirit-airlines-to-pause-data-sale/, https://www.lexblog.com/2026/10/09/lawmakers-press-google-and-spirit-on-employee-data-before-a-hearing-on-a-10-million-ai-training-data-sale/). Uncontested small-estate sales move in weeks; nothing about data makes them uncontested by default.

Door 3 — ABC (assignment for the benefit of creditors). State-law liquidation: the company assigns all assets to a professional assignee (in the funnel: CMBG Advisors, DSI Assignments, Insolvency Services Group), who sells them without court supervision — faster and cheaper than bankruptcy; a buyer can be found and the company wound down "in as little as a few weeks" (https://www.legaldive.com/news/assignments-for-the-benefit-of-creditors-distressed-asset-sales-bankruptcy-law/640315/). Trade-offs: no free-and-clear court order, no automatic stay, no judicial extinguishment of successor liability (Promise Legal, URL above), and the secured lender must cooperate and release liens or nobody buys. Assignee economics: the assignee is paid a commission — typically a percentage of assets administered — paid first out of the estate along with legal/administrative costs, which is why small estates struggle to attract assignees at all (https://www.stimmel-law.com/en/articles/assignments-benefits-creditors-abcs-basics-california). Claims bar dates run 150–180 days in California practice, but asset sales can close well before the claims process ends — the money waits, the data doesn't have to.

4.2 What a small package costs — the honest version

4.3 The legal constraints that decide whether data can transfer at all

  1. §363(b)(1) + the consumer privacy ombudsman (§332). If the debtor disclosed a privacy policy prohibiting transfer of personally identifiable information to unaffiliated parties and that policy was in effect at filing, the trustee may not sell PII unless (A) the sale is consistent with the policy, or (B) a consumer privacy ombudsman is appointed under §332 and, after notice and hearing, the court finds no showing that the sale would violate nonbankruptcy law (analysis and statutory summary at https://delawarebusinesstimes.com/news/customer-privacy-bankruptcy/?amp=1 and https://harvardlawreview.org/print/vol-138/data-privacy-in-bankruptcy-the-consumer-privacy-ombudsman/; the statutory text of §363(b)(1) is also quoted in the FTC's Borders filing as reproduced in search results). The ombudsman reports on the policy, privacy gains/losses, and mitigations — not on price — and the estate pays for it.
  2. Precedent is cautionary, not permissive. Toysmart (2000): the FTC sued over a customer-list sale that violated the debtor's "never shared" promise; the list ended up bought and destroyed for $50,000 rather than sold — and Toysmart is why §363(b)(1) exists (https://www.orrick.com/en/Insights/2015/10/Privacy-Policies-And-Asset-Sales-It-Pays-To-Plan-Ahead). RadioShack (2015): a "we will not sell or rent your personally identifiable information" policy drew FTC and state-AG intervention and a heavily conditioned sale (same source). Spirit is the live test of whether the machinery scales to employee data: EPIC's amicus argues no employer has previously sold employee data as a bankruptcy asset, the flight attendants' union objected, and the ombudsman's Oct 5 recommendation to approve was conditioned on excluding customer PII (https://techpolicy.press/spirit-airlines-is-selling-its-workers-data-that-cant-fly, PYMNTS above).
  3. FTC Act §5 and state law ride along. A transfer contrary to the debtor's own public privacy promises can be an unfair practice independent of the Bankruptcy Code; state privacy laws (CCPA/CPRA et al.) and sector law (GLBA for financial data, HIPAA-adjacent duties for health data) apply on top (Delaware Business Times, URL above). An ABC gives no shelter here — there is no court order at all, so a non-compliant ABC data sale is more exposed, not less.
  4. The archive may not be the estate's to sell. The threshold diligence question in every deal: does the debtor own the records, or hold them for others? Spirit's own asset schedule marked 97.5M customer profiles "Not Included" while internal operational records were sold (Lexology, URL above) — the parties themselves segregated data the estate couldn't cleanly convey. In Chad's funnel this is the recurring screen: Pulley's customer cap-table data belongs to its customers (only Pulley's internal ops archive is licensable); an agency's client work product is client-owned (the standing warning from workstream-comparison-oct13.md). Client data held under contract, and data collected under platform terms that restrict transfer/training use (Shopify Partner §9.14–9.15; Slack API terms — from Chad's Oct 7 research), can shrink a headline archive to a fraction of its apparent size.
  5. Employee data is the new frontier of risk. Internal archives are mostly employee communications, and employees never signed a consumer privacy policy — Spirit shows the fight moves to labor law, state law, and public pressure instead. An AI-startup advisor quoted by MIT Technology Review calls such data "legally radioactive"; sellers and buyers both scrub, and HR records are routinely excluded by suppliers (MIT Tech Review, URL above).

4.4 Practical steps for a principal purchase (distilled)

  1. Identify the fiduciary (PACER for the trustee in Chapter 7/11; the assignment notice for the assignee in an ABC) and the estate's counsel.
  2. Sign the estate's NDA; get the asset schedule and data inventory; confirm data is scheduled as estate property and segregated from client/customer data.
  3. Pull the debtor's privacy policies in effect at filing; screen for §363(b)(1) triggers, platform ToS, and sector law.
  4. Bid: stalking-horse (with 1–3% break-up protection) or auction overbid in §363; private negotiated purchase from an assignee in an ABC.
  5. Paper privacy conditions into the sale order/APA: de-identification by an independent vendor before transfer, HR-record exclusions, use restrictions, no re-identification.
  6. Close on the sale order (bankruptcy) or bill of sale (ABC); take delivery of de-identified data only. Professionals involved: bankruptcy counsel (both sides), trustee/assignee, the estate's asset marketer/broker, consumer privacy ombudsman (bankruptcy only), de-identification vendor, and a data engineer to verify the archive is what was represented.

4.5 Chad's current funnel candidates mapped to this route

(From seller-funnel-candidates-v1.md; scores are that file's.)

5. Triangulation — three routes scored

Scale: ●●● strong / ●● moderate / ● weak, judged for Chad specifically (former CPA, live Mercor partner at $175K, ~$100K deployable capital, Seattle network of accountants/lawyers/founders).

Dimension A. Pure referral / finder B. Buy data assets from bankruptcy/ABC and resell C. Seller-side advisor (success fee)
Capital required ~$0 (time + one legal template) ●●● $10K–$250K per package + counsel + de-id costs; capital locked for months ● ~$0 ●●●
Time to first dollar Weeks to a few months — programs pay on signing/close (Handshake at close; micro1 at onboarding/acceptance) ●●● Months: fiduciary process (Spirit took ~2 months auction→hearing contested, plus a resale cycle after) ● One deal cycle: weeks of process per seller, paid at close ●●
Expected margin per deal $10K flat (Handshake) to $50K caps (micro1; Mercor amount unpublished, prior reported ceiling $100K); BizBuyData 6% is uncapped — $120K on a $2M deal ●● Widest theoretical spread (buy at distressed payouts, sell at broker prices) but unproven at small scale and contested by Mercor/Google-sized bidders ●● 5–10% of $100K–$2M seller deals = $10K–$200K per close, ~pure margin (Data Licensing Advisors benchmark) ●●●
Chad's specific edge Live Mercor partnership + named contacts; CPA credibility with the channels that surface sellers ●●● CPA diligence and rights-screening are real, but he is outgunned on capital and has no bankruptcy-practice track record ● Strongest fit: he has been the seller ($175K Mercor deal) and can run the screen buyers trust ●●●
Legal risk Low–moderate: finder terms must be contractual (WA sales-rep statute likely doesn't cover data; NY needs writing) ●●● Highest: §363(b)(1)/ombudsman process, FTC §5, AG intervention, employee-data exposure, title defects on client data — "legally radioactive" is a practitioner's phrase, not a journalist's ● Moderate: engagement-letter hygiene, never take both sides' fees undisclosed; diligence liability is reputational more than legal ●●
Scalability Bounded by Chad's sourcing throughput; each program's attribution rules differ ●● Capital-bound; every estate is a bespoke legal process ● Bounded by deal flow, but fee size grows with seller size; pairs with A (a seller who won't sign an advisory can still be referred) ●●

Recommendation

Easiest first route: A — pure referral/finder, starting now. It is the only route with (i) zero capital, (ii) published, register-today rails — micro1's uncapped $50K, BizBuyData's greater-of-$50K-or-6%, Handshake's $10K — and (iii) a warm start, because Chad's Mercor relationship already exists and his own $175K deal is the credibility story. The prior instinct that "referrals may be easiest" is confirmed by the evidence assembled here, with one correction: the referral market's center of gravity is live companies (Fast Company's $100K–$2M live-company band; the brief's own Oct 13 reporting), not defunct ones — the defunct-company supply the labs started with is the segment SimpleClosure says is being exhausted. Chad's funnel should keep hunting wind-downs (they convert fastest and have a signer with nothing to lose) but the referral rails pay the same bounty for a healthy 40-person firm from his CPA network, with none of the estate law attached.

Best opportunistic second route: C — seller-side advisory, aimed first at the funnel's live wind-downs (Pulley above all), with bankruptcy-channel introductions as its sharpest variant. When a Pulley-type seller appears — still operating, founder reachable, deadline visible (Pulley's data access ends Jan 31, 2027) — an advisor success fee on a $200K–$2M outcome beats a $50K referral cap, and Chad's seller-side experience is the pitch. Inside bankruptcy specifically, the evidence says the money is in bringing buyers to fiduciaries, not in outbidding them: Google and Mercor contested Spirit at $10M/$7.5M, while trustees of small estates often never market archives at all (§326 fee math makes small data sales barely worth a trustee's time). A finder who shows up to a Parker trustee with two qualified broker bidders creates the sale and can be paid by the buyer side — Route A economics inside Route B's territory, with Route B's legal risk left with the estate and the buyer.

Route B as principal: defer, with conditions. Buying archives out of estates is a real business — SimpleClosure's payouts and the Spirit auction prove both ends of the spread — but today's evidence shows (1) no published case of an independent reseller doing it profitably at small scale, (2) the best assets attract lab/broker bidders with balance sheets, and (3) the legal tail (ombudsman conditions, AG/FTC attention, employee-data litigation) lands on the owner of the data. Revisit after 2–3 observed pricing cycles from Routes A/C, and only with: a bankruptcy counsel on call, a de-identification vendor quoted in advance, and a pre-committed resale buyer for the specific archive. Capital deployed before those three exist is tuition, not investment.

Biggest unknowns (explicit)

  1. Archive resale margins. No source publishes what brokers pay for archives vs. what labs pay brokers. The 60–70% payout ratio is documented only for expert labor.
  2. Mercor's current referral amount — unpublished on its own page; the $100K figure is a reported past ceiling.
  3. Whether referral programs have actually paid out at scale — programs are documented; named payout case studies are not.
  4. Small-estate data-sale mechanics in practice — Spirit is one very large, very contested data point; there is no public dataset of small Chapter 7 data sales, prices, or timelines.
  5. Spirit's outcome — as of Oct 13, 2026 the approval hearing is Oct 14; whether the court imposes further conditions (or the employee-data theory gains ground) will reset the legal risk for every later deal.
  6. Sunset's and Replay's finder terms — Sunset's data business is confirmed; neither publishes what a third-party introducer earns.

‹ Daily Deals  ·  Data Licensing Pipeline ›