Skip to content

Instantly share code, notes, and snippets.

@kkoci
Last active July 2, 2026 17:28
Show Gist options
  • Select an option

  • Save kkoci/b90ca3386a01823907ba19b539a0998c to your computer and use it in GitHub Desktop.

Select an option

Save kkoci/b90ca3386a01823907ba19b539a0998c to your computer and use it in GitHub Desktop.

DealProof TEE Negotiation Market Research

15-Agent Adversarial Swarm — Full Results

Modeled after dmarz's 33-agent πCreds sweep

Scope: Real two-party negotiation markets where TEE attestation is structurally necessary — not just nice to have. Tina's bar applied throughout: only counts if party A's move in round N materially changes party B's move in round N+1.

No competitors to DealProof found. Closest analogs: Enclave Markets (TEE for crypto block trading, no agents, no πCreds), Cybersettle (double-blind insurance claims, no TEE, suspended), Zycus ANA (AI procurement agent, no TEE). None have the full stack: TEE + dual agents + DCAP attestation + πCreds conduct layer + escrow.


MASTER RESULTS TABLE

# Vertical Domain Real Neg. TEE Necessity Volume Pay Reg. Risk Reachability
1 Institutional block trade Finance YES HIGH HIGH HIGH LOW MED (via Flashbots/IC3)
2 OTC derivatives ISDA/CSA Finance YES HIGH HIGH HIGH LOW LOW
3 Leveraged loan bookbuilding Finance YES HIGH HIGH HIGH LOW LOW
4 Creator × brand via manager Creator Economy YES HIGH HIGH HIGH LOW MED-HIGH
5 Music sync licensing Creator Economy YES HIGH MED HIGH LOW MED
6 Programmatic guaranteed deals Creator Economy YES MED HIGH HIGH LOW LOW
7 Competing-offer verification HR YES* HIGH HIGH HIGH LOW HIGH
8 Inter-employer comp benchmarking HR YES HIGH HIGH HIGH MED MED
9 Union collective bargaining HR YES HIGH HIGH HIGH MED LOW
10 Hospital × payer rate negotiation Healthcare YES HIGH HIGH HIGH MED LOW
11 Pharma × PBM rebate Healthcare YES HIGH HIGH HIGH MED LOW (via disruptors)
12 Pharma × payer outcomes contract Healthcare YES HIGH MED HIGH MED LOW
13 AI training data licensing Data YES HIGH MED HIGH LOW MED
14 Healthcare dataset licensing Data YES HIGH MED MED MED LOW
15 Federated learning access Data YES HIGH MED HIGH LOW MED
16 FMCG/CPG trade promotion Supply Chain YES HIGH HIGH HIGH LOW LOW
17 Enterprise SaaS renewal Supply Chain YES MED HIGH HIGH LOW MED
18 OEM × tier-1 supplier cost Supply Chain YES HIGH MED HIGH LOW LOW
19 Bilateral PPA negotiation Energy YES HIGH MED-HIGH HIGH LOW MED
20 Wholesale gas supply contract Energy YES HIGH MED HIGH LOW LOW
21 Voluntary carbon credit Energy YES HIGH MED MED-HIGH LOW MED
22 P&C insurance BI settlement Legal/ADR YES HIGH HIGH HIGH LOW MED
23 EEOC employment settlement Legal/ADR YES HIGH MED-HIGH HIGH MED LOW
24 Commercial arbitration settlement Legal/ADR YES HIGH MED HIGH MED LOW
25 MNO × MVNO wholesale access Telecoms YES HIGH MED-HIGH HIGH LOW LOW
26 Spectrum LSA negotiation Telecoms YES HIGH LOW-MED HIGH MED LOW
27 MNO × neutral host small cell Telecoms YES HIGH MED HIGH LOW LOW
28 DoD sole-source defense contract Gov/Defense YES HIGH HIGH HIGH HIGH LOW
29 BAFO government RFP Gov/Defense PARTIAL MED HIGH HIGH MED LOW
30 OTA defense prototype Gov/Defense YES MED MED-HIGH HIGH HIGH LOW
31 On-chain NFT/in-game auction Gaming YES HIGH HIGH HIGH LOW MED
32 P2E guild scholarship Gaming PARTIAL MED HIGH MED LOW MED
33 Sports data licensing Gaming PARTIAL MED HIGH HIGH LOW LOW
34 Commodity prepayment/RBL Emerging Markets YES HIGH MED-HIGH HIGH MED LOW
35 Cross-border SCF/trade finance Emerging Markets YES MED-HIGH HIGH HIGH MED LOW
36 B2B agent × agent procurement Agentic AI YES HIGH HIGH HIGH LOW MED
37 Agent financial execution Agentic AI YES HIGH MED HIGH LOW MED
38 Multi-agent supply chain coord Agentic AI YES HIGH HIGH HIGH LOW MED
39 Clinical trial site agreements (CTA) Life Sciences YES MED MED HIGH LOW-MED LOW
40 Healthcare staffing / locum tenens Healthcare Staffing YES HIGH HIGH HIGH LOW LOW-MED
41 China MCN / cross-border IP licensing Creator Economy (Cross-border) YES MED-HIGH HIGH MED MED LOW
42 Gig economy rate negotiation (⏰ EU Dec 2026) Platform Labor PARTIAL MED MED MED MED LOW

*Requires revision loop to pass Tina's bar. Single-round above/below fails.


TOP 6 VERTICALS — DEEP ANALYSIS


VERTICAL 1: INSTITUTIONAL BLOCK TRADE NEGOTIATION

Domain: Financial Markets Parties: Buy-side asset manager × broker-dealer block trading desk (and/or receiving institutional investor)

The Trust Problem

A broker intermediating a block trade is architecturally positioned to see the seller's full position, mandate constraints, and urgency — and has economic incentive to exploit that information to benefit its own principal position or favored buy-side clients.

Documented real-world failure: Morgan Stanley paid $249M to settle SEC/DOJ charges that its equity syndicate desk disclosed confidential block trade information to select buy-side investors, enabling them to short ahead of execution. This persisted three years under NDA. An NDA cannot prevent a trader from making a phone call.

Why TEE Specifically

  • Matching logic runs inside the enclave — broker cannot read either party's full position or reserve price
  • Remote attestation proves the matching algorithm wasn't modified to log positions
  • No party learns "there is a large seller at $X" unless a match exists
  • Structurally impossible for NDA or dark pool ATS to achieve simultaneously

Numbers

  • Global equity block trading: $1.54T notional (2024)
  • Hundreds of thousands of negotiation sessions/year globally
  • Buy-side TCA budget: $1.5M–$5M per $500M block in implicit/explicit costs
  • SaaS fee of $50K–$500K/year easily passes cost-benefit test

Closest Existing Analog

Enclave Markets (crypto only) — Fully Encrypted Exchange running matching engine inside Intel SGX. No traditional equity equivalent. This is the gap.

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 3-8 documented rounds, each party's move changes next
TEE necessity HIGH — Morgan Stanley case proves NDA structurally fails
Volume HIGH — hundreds of thousands/year
Ability to pay HIGH — TCA budget proven
Competitive openness PARTIAL — dark pools exist but don't solve the trust problem
Regulatory risk LOW — regulators actively want better mechanisms here
Reachability MED — warm path via Flashbots/Tina and Andrew/IC3

VERTICAL 2: B2B AGENT × AGENT PROCUREMENT NEGOTIATION

Domain: Agentic AI Parties: Enterprise buyer-side AI agent × supplier-side AI agent

The Trust Problem

When two AI agents negotiate on behalf of their principals, the platform operator running both agents can observe all agent states, intermediate reasoning, and instruction sets. Observable agent behavior lets opponents reverse-engineer budget ceilings and urgency signals. Principals cannot verify what instructions the opposing agent was given.

Foundation: NDAI paper (Stephenson, Miller, Sun, Annem, Parikh, 2025) — Andrew Miller co-authored. Formalizes the agent-TEE trust structure: each agent A_i takes private input x_i (principal instructions), TEE mediates the interaction so no party observes the opponent's instructions. Only the outcome exits.

Live market: Zycus ANA already runs supplier negotiations end-to-end. Gartner: AI agents will intermediate $15T in B2B spending by 2028. 90% of B2B purchases handled by AI agents projected. By 2026, 1 in 5 B2B sellers will face AI buyer agents issuing autonomous counteroffers.

Why TEE Specifically

  • Both agents run inside enclave with instruction sets sealed
  • Agent behaviors (offers, counteroffers, reasoning chains) processed inside enclave — neither principal nor platform operator observes intermediate states
  • Only outcome exits: agreed terms or no deal
  • Hardware attestation lets each principal verify opposing agent ran the code it claims

Numbers

  • $15T B2B spend intermediated by AI agents by 2028 (Gartner)
  • Zycus ANA: production-grade deployment, 1,500+ procurement leaders at 2026 summit
  • 19% of buyers using AI in purchasing report less confidence due to inaccurate AI outputs (Forrester)
  • $10B+ enterprise value at risk from ungoverned AI in go-to-market (Gartner 2026)

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 4-12 rounds, agent moves causally determine counterpart responses
TEE necessity HIGH — platform operator visibility is structural, not behavioral
Volume HIGH (growing fastest of any vertical)
Ability to pay HIGH — Chief Procurement Officer controls budget
Competitive openness OPEN — Zycus has no TEE layer, this is the wedge
Regulatory risk LOW — no specific regulation yet, first-mover window open
Reachability MED — Andrew Miller connection, NDAI paper is the credential

VERTICAL 3: CREATOR × BRAND DEAL VIA TALENT MANAGER

Domain: Creator Economy Parties: Talent manager (representing creator/influencer) × brand in-house influencer marketing team or agency

The Trust Problem

The talent manager structurally holds both parties' private parameters simultaneously:

  • Creator's rate floor (the minimum they'll accept)
  • Brand's budget ceiling (probed through questions like "what's the campaign budget?" before revealing the creator's rate)

Manager is paid 10-25% commission on deal value — incentivized to close at maximum extraction from the brand while keeping the creator just satisfied. Neither side can verify how the manager used their private information.

Specific documented failure modes:

  1. Dual-sided extraction — manager knows both ceiling and floor, extracts maximum spread
  2. Rate card leakage — creator's true floor rate travels through recruiter/manager networks, becomes ceiling in future negotiations
  3. Scope creep exploitation — brand reveals usage rights requirements after creator already showed price flexibility on base rate
  4. Manager self-dealing — routes brand to easiest-to-close creator, not best fit

Why TEE Specifically

TEE accepts each party's private parameters into a sealed enclave, runs matching algorithm, returns only "deal exists at $X for scope Y" — manager sees deal terms without ever learning creator's true floor or brand's true ceiling.

Numbers

  • Global influencer marketing: $32.55B (2025), projected $44B+ (2026)
  • Millions of brand deal negotiations/year at the negotiable tier (micro through macro)
  • Average deal $5K–$25K; 3-6 rounds per negotiation over 5-21 days
  • No incumbent platform addresses the manager dual-information-extraction problem

Named Incumbents (none solve this)

Grin, Aspire, Traackr — marketplace platforms, no negotiation privacy layer The Cirqle — rate negotiation feature with counter-offer history, no enclave Foam.io — talent manager workflow tool, same trust problem

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 3-6 rounds, scope reveals causally change rate in next round
TEE necessity HIGH — manager is architecturally positioned as rent-extracting intermediary
Volume HIGH — millions/year globally
Ability to pay HIGH — brand CMO and talent agency both have budget
Competitive openness OPEN — no incumbent solves this
Regulatory risk LOW — standard contract law
Reachability MED-HIGH — creator economy accessible, cohort has adjacent contacts

VERTICAL 4: COMPETING-OFFER VERIFICATION IN SALARY NEGOTIATION

Domain: HR / Compensation Parties: Job candidate × corporate recruiter/HR business partner

The Trust Problem

Competing-offer bluffing is documented at massive scale. Former Google recruiters confirm offers were rescinded when forged offer letters were detected. The market consequence is adverse selection: candidates who bluff benefit; honest candidates cannot extract the same concessions without bluffing. Employers who demand verification lose candidates with real offers who refuse to share a competitor's confidential document.

Secondary trust problem: When employer requests competing offer letter, they gain intelligence about the competitor's compensation structure — all of which is operationally sensitive. "Companies ask to begin with because they want to know what their competitors are offering."

Why TEE Specifically

  • Candidate's actual offer letter from Company B loads into sealed enclave
  • Employer's salary band for the role loads into same enclave
  • Enclave computes: is the offer genuine? what is the gap?
  • Returns only: "competing offer verified, gap is X%" — employer never sees full offer letter (protecting Company B's confidential comp data), candidate never learns employer's full band ceiling

Numbers

  • 7.6M job openings in US (May 2026 BLS data)
  • 5.2M hires per month in US
  • Tens of millions of professional comp negotiations/year globally
  • Headhunter commission: 20-33% of first-year base, structural incentive to close at employer's preferred number

Critical design note: Must have revision loop to pass Tina's bar. Candidate sees "gap is X%" → revises ask → employer revises offer → iterates to convergence or walk-away. Single-round above/below is a screener, not negotiation.

Scores

Dimension Score
Real negotiation (Tina's bar) YES* — with revision loop
TEE necessity HIGH — verification without exposing competitor's comp structure is architecturally impossible any other way
Volume HIGH — tens of millions/year
Ability to pay HIGH — CHRO and TA budget
Competitive openness OPEN — Levels.fyi/Glassdoor are public data, this is private verification
Regulatory risk LOW — pay transparency laws are tailwinds
Reachability HIGH — most accessible buyer of any vertical

VERTICAL 5: P&C INSURANCE BODILY INJURY SETTLEMENT

Domain: Legal / ADR Parties: P&C insurer claims adjuster × plaintiff's personal-injury lawyer

The Trust Problem

Each side holds privileged information they cannot disclose:

  • Insurer: internal reserve, adjuster authority limit, bad-faith exposure analysis
  • Plaintiff: true walk-away threshold, privileged case-strength EV, client risk tolerance

ZOPA objectively exists in most cases but is never discovered because neither side can safely reveal their range. Cybersettle proved iterative blind bidding works at scale ($1.8B settled, ~200K claims) but operator technically sees both sides' bids.

Why TEE Specifically

TEE moves bid matching into hardware-attested enclave. Privileged case-strength EVs load as sealed inputs — used in computation without privilege waiver because no human observes them. Architecturally impossible with NDAs: NDAs bind behavior but not technical read access by the operator.

Numbers

  • Millions of BI settlement negotiations/year globally
  • ~95% of PI cases settle pre-trial
  • US personal injury law market: $61.3B revenue (2024)
  • Cybersettle processed 200K+ claims before suspension — market validated

Incumbent Gap

Cybersettle suspended due to patent litigation — market is currently unserved by the closest analog. Window is open.

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 4-10 rounds, new evidence materially shifts subsequent offers
TEE necessity HIGH — privilege preservation during computation is architecturally impossible without TEE
Volume HIGH — millions/year
Ability to pay HIGH — insurer claims budget, billions in leakage annually
Competitive openness OPEN — Cybersettle suspended, market unserved
Regulatory risk LOW
Reachability MED — warm intro needed to P&C claims leadership

VERTICAL 6: FMCG/CPG MANUFACTURER × RETAILER TRADE PROMOTION

Domain: Supply Chain / B2B Commerce Parties: FMCG/CPG manufacturer (P&G, Unilever, Nestlé) × retail buyer (Walmart, Tesco, Carrefour)

The Trust Problem

$500B in annual trade promotions globally. Industry estimates: 35-40% waste — over $175B in nonproductive spend per year. The waste is structurally caused by information asymmetry during negotiation:

  • Manufacturer's private info: cost floor, production capacity, supply chain timing, minimum pass-through for brand objectives
  • Retailer's private info: actual demand forecast, promotional calendar commitments, real pass-through intent vs. represented intent

Forward buying and diverting — documented, industry-named failure modes where retailers buy large quantities at promotional prices then resell at full price or to other retailers, defeating the manufacturer's promotional objective entirely.

Why TEE Specifically

TEE accepts both parties' actual cost floors, demand forecasts, and promotional calendars as sealed inputs. Computes whether a mutually profitable promotion exists at what terms. Returns only the deal — neither side sees the other's raw data. Prevents scope-expansion exploitation by requiring both parties to commit scope before pricing is revealed.

Numbers

  • $500B annual trade spend globally
  • $175B estimated waste annually (35-40%)
  • Trade promotion management software market: $2.64B (2025)
  • 80%+ of consumer goods manufacturers engage in trade promotions
  • 20-25% of gross sales invested in trade spend

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 8-15 rounds over 4-8 weeks
TEE necessity HIGH — $175B waste directly attributable to information asymmetry during negotiation
Volume HIGH — millions of deal-events/year
Ability to pay HIGH — VP Sales / National Account Management budget
Competitive openness PARTIAL — TPM software ($2.64B market) manages spend post-hoc; none address information asymmetry during negotiation
Regulatory risk LOW
Reachability LOW — requires warm intro to major CPG or retail buyer


VERTICAL 7: CLINICAL TRIAL SITE AGREEMENT (CTA) NEGOTIATION

Domain: Life Sciences / Pharma Parties: Pharmaceutical sponsor (or CRO acting as sponsor's agent) × clinical research site (academic medical center or community site) Source: Dedicated follow-up sweep — identified as missed vertical from Tier 1

The Trust Problem

Two structural asymmetries define CTA negotiation:

FMV opacity: Sponsors maintain internal FMV (Fair Market Value) databases of per-patient rates accepted at comparable sites across all their trials. Sites negotiate blind — no access to what competing sites accepted. Anti-kickback statutes and the Physician Payments Sunshine Act legally require payments to be at FMV, yet the process for establishing FMV is documented as "an opaque subject with rather unclear guidelines." Sites systematically undervalue their services or over-negotiate to compensate for uncertainty.

CRO instruction opacity: When a CRO negotiates on behalf of a sponsor, the site cannot determine whether a CRO position represents the sponsor's actual limit or the CRO's own risk-averse posture. Sites frequently report that CROs refuse to escalate contested terms to the actual sponsor, causing extended stalemates. The CRO is structurally positioned as a commercially-interested intermediary — Pattern 1 from the swarm's structural finding.

Documented Contested Terms (Multi-Round, Not Template-Resolvable)

From ACTA workgroup empirical prioritization of 28 CTA terms:

  • Indemnification / limitation of liability
  • Intellectual property (especially Phase I)
  • Publication rights (sponsor wants 18-month embargo; site wants independent right)
  • Subject injury / patient care costs
  • Per-patient payment / FMV budget line items
  • Overhead / indirect cost rates (AMCs want 30%+; sponsors resist)
  • Timeline milestones / enrollment commitments

Round Structure (Tina's Bar: MET)

  • Average CTA negotiation: 64 days industry-wide, >100 days at AMCs
  • 4-8 documented redline exchanges on legal terms, plus separate iterative budget rounds
  • Duke internal process: 5 business days (first action) → 7 business days per redline exchange → escalation triggers at 45 and 55 business days
  • Causally linked moves confirmed: site proposes higher per-patient rate → sponsor counters with reduced overhead percentage → site adjusts publication clause → sponsor lengthens review period. Each round causally determines the next.

The $800K/Day Figure — Verified

  • Tufts CSDD: $40K/day direct cost of trial delays
  • $500K–$800K/day in commercial opportunity cost (lost revenue for each day a drug reaches market late)
  • Mayo Clinic with AI (TCN): reduced CTA cycle from 6-9 months to 25 days
  • ACTA standard template: saves 48-57 days vs. no template
  • MDIC MCTA for medical devices: cycle time worsened from 133 to 164 days (2019-2024) — existing solutions are not solving the problem

Why TEE Specifically — The FMV Wedge

Standard NDA and audit clause cannot give a site access to the sponsor's FMV database without transferring competitive intelligence about every other site's accepted rates.

A TEE-backed FMV verification tool would:

  • Accept sponsor's internal FMV database as sealed input
  • Accept site's proposed per-patient rate as sealed input
  • Compute: is this proposal within FMV range for this protocol type/geography/phase?
  • Output only: "in range / out of range" with hardware attestation
  • Neither party sees the other's raw data; sponsor's FMV database is never exposed; site's negotiating floor is never revealed

This is a standalone wedge product — doesn't require replacing TCN or Ledger Run, doesn't require full CTA negotiation automation. It solves the specific legally-mandated FMV compliance problem that every existing platform handles poorly.

Named Incumbents

  • The Contract Network (TCN) — AI-powered CTA negotiation, Mayo Clinic partnership, venture-backed. Focused on process/workflow speed, no TEE. Reduced Mayo's cycle from 6-9 months to 25 days.
  • Ledger Run — CTA negotiation platform, early-stage. Process focus, no sealed computation.
  • ICON, Parexel, Covance/Labcorp — CROs that negotiate CTAs as bundled service. Structurally conflicted — they ARE the CRO instruction opacity problem.
  • ACTA / TransCelerate / MAGI — template standardization consortia. Cover ~80% of standard terms; indemnification, IP, publication rights, and FMV remain contested regardless.

TEE-specific gap: No incumbent provides sealed FMV computation or attested negotiation records. EMA ALCOA++ requirements for tamper-proof audit trails would be exceeded by TEE-attested negotiation logs.

Volume

  • Hundreds of thousands of CTA negotiations globally per year
  • US alone: ~50,000-80,000 industry-sponsored trials active at any time, each requiring a CTA per site
  • High-value, high-friction subset: AMC negotiations, Phase I-II trials with contested IP terms
  • MEDIUM volume (thousands of genuinely contested multi-round negotiations per year at the tier where TEE adds value)

Ability to Pay

  • Sponsor-side: VERY HIGH. $40K/day direct + $500K-$800K/day commercial opportunity cost. A $5K-$50K per-negotiation TEE fee is trivial vs. delay cost.
  • Site-side (AMC): MEDIUM. Budget-constrained research administration offices.
  • Primary buyer: sponsor-side VP Clinical Operations or Chief Procurement Officer.

Scores

Dimension Score
Real negotiation (Tina's bar) YES — 4-8 documented rounds, causally linked moves
TEE necessity MEDIUM — FMV gap is structural and unsolvable without sealed computation; but existing AI tools gaining traction on process side without TEE
Volume MEDIUM — hundreds of thousands of CTAs/year; contested multi-round subset is thousands/year
Ability to pay HIGH — sponsor-side delay costs make fee trivial
Competitive openness PARTIAL — TCN/Ledger Run own workflow; TEE-specific angle has no incumbent
Regulatory risk LOW-MEDIUM — EMA ALCOA++ is a tailwind; HIPAA/GDPR apply to patient data not CTA terms
Reachability LOW — 18-36 month enterprise sales cycle; 6-12 month procurement process; no obvious landing without pharma warm intro

The Honest Founder Assessment

Strongest FOR argument: FMV attestation is a standalone wedge with clear ROI, regulatory tailwind, and no incumbent. Doesn't require replacing TCN. Both sides benefit simultaneously (sponsor avoids kickback exposure; site avoids underpayment). Can be built without existing pharma relationships.

Strongest AGAINST argument: The buyer is a pharma VP of Clinical Operations with an existing CRO relationship, a 6-12 month IT/GCP validation requirement before any new software touches the contracting process, and deep caution about unknown vendors. Solo founder without pharma relationships faces 18-36 months to first paying customer. Not a near-term bet — a validated secondary vertical to activate when a pharma warm intro materializes.



VERTICAL 8: HEALTHCARE STAFFING / LOCUM TENENS RATE NEGOTIATION

Domain: Healthcare Staffing Parties: Staffing agency × hospital (bill rate side) + staffing agency × clinician (pay rate side) Source: Dedicated follow-up sweep — identified as missed vertical from Tier 1

The Trust Problem

The staffing agency is architecturally positioned as a dual-information intermediary — exactly Pattern 1 from the structural finding:

  • Agency knows the hospital's bill rate ceiling (what the hospital is willing to pay)
  • Agency knows the clinician's pay rate floor (what the clinician will accept)
  • Agency's margin is the spread between the two — economically incentivized to maximize opacity on both sides

Documented margin structure: Locum tenens agency markups 40-50% of bill rate. Travel nursing agency take 30-45% of bill rate. Neither hospital nor clinician can verify the actual spread without seeing the other party's private terms.

Specific documented failure modes:

  1. Agency margin opacity — hospital pays $200/hr bill rate; clinician receives $110/hr; agency takes $90/hr spread. Neither side knows the other's number.
  2. False urgency via credentialing manipulation — agency controls information flow on credentialing status; can create artificial urgency by delaying or accelerating credential processing as a negotiating tactic.
  3. Exclusivity gaming — agency claims exclusivity to prevent hospital from checking competing agencies; hospital cannot verify whether clinician is simultaneously being placed elsewhere.
  4. MSP/VMS intermediation — Managed Service Provider layer adds another intermediary between hospital and agency, further obscuring bill rate and margin information.

Tina's Bar (MET — in non-commodity segment)

  • Round 1: hospital opens requisition at target bill rate; agency tests clinician supply
  • Round 2: agency reports low fill probability; hospital raises bill rate or changes terms
  • Round 3: agency decides how much rate movement to pass through to clinician; revises offer
  • Round 4: clinician counters on pay, travel, housing, schedule; agency either compresses margin or seeks more from hospital

3-6 rounds documented for premium locum tenens and crisis/shortage travel assignments. Hospital's move in round N causally determines agency's offer to clinician in round N+1.

Why TEE Specifically

Standard NDA binds agency behavior but cannot prevent mental use of both sides' private information. Audit clauses require disclosure to verify — but disclosure destroys the confidentiality that makes the system work.

TEE enables:

  • Hospital's true bill rate ceiling into sealed enclave
  • Clinician's true pay rate floor into sealed enclave
  • Agency's proposed spread into sealed enclave
  • Output: "spread is within verified fair-band range" with attestation
  • Neither hospital nor clinician sees the other's private number; agency cannot lie about the spread without the attestation failing

Numbers

  • US healthcare staffing market: $45.48B (2025)
  • 8% of US physicians actively working locums; 27% have prior locum experience (CHG 2025)
  • Travel nursing: peak $24B post-COVID, stabilizing at $15-18B
  • Volume: HIGH — hundreds of thousands of placement negotiations/year in contested segment
  • Value at stake: even 10% spread compression on $45B market = $4.5B in redistributed value

Named Incumbents (structural conflict)

  • AMN Healthcare, CHG Healthcare, Barton Associates, CompHealth, Weatherby, Cross Country — they ARE the margin opacity problem. Will resist, not adopt.
  • Boutique low-margin agencies (Locumpedia, competitive challengers) — already use transparency as differentiation; would benefit from third-party verifiable proof. Entry point.
  • Hospitals / health systems frustrated with MSP/VMS opacity — most natural buyer.

Entry Point

Sell to hospitals and boutique challenger agencies — NOT to AMN/CHG. Hospital buyer: VP of Workforce or Chief Nursing Officer. Budget: contingent labor / temporary staffing line. Value proposition: verifiable proof that agency spread is within fair-band range, reducing hidden intermediation cost.

Scores

Dimension Score
Real negotiation (Tina's bar) YES — non-commodity segment, 3-6 documented rounds
TEE necessity HIGH — trusted-middleman failure, sealed computation is the only mechanism
Volume HIGH — hundreds of thousands/year in contested segment
Ability to pay HIGH — hospitals face acute vacancy costs, already heavy contingent labor spend
Competitive openness LOW-MED — entry via hospitals and boutique challengers, around incumbents
Regulatory risk LOW — no specific regulation; price transparency tailwinds in healthcare generally
Reachability LOW-MED — needs healthcare staffing warm intro; hospital procurement cycle 3-6 months
Incumbent conflict HIGH — AMN/CHG profit from opacity; adversarial relationship

CHINA MCN / CROSS-BORDER IP LICENSING (Secondary Vertical — Research Only)

Domain: Creator Economy (Cross-border) Status: Gist only — not added to primary product brainstorm. Distribution too hard for UK/US solo founder without China-based partner.

Trust problem is real and well-documented:

  • Yin-yang dual contracts: public contract shown to platforms/brands; private side letter captures true economics. SPC actively prosecutes these.
  • Revenue misreporting: Inner Mongolia streamer claimed 260K RMB; tax investigation revealed 15M+ RMB hidden over two years.
  • Rate card opacity: MCNs negotiate with many brands simultaneously; brands cannot verify whether they are getting competitor rates.
  • Cross-border enforcement gap: US/EU judgments generally not enforceable in Chinese courts; ex ante verifiable computation more valuable than better contracts.

Volume: 40,000+ MCNs in China, $30B+ market, global MCN market projected $153B by 2034. HIGH volume.

Why not primary: Distribution into China MCN ecosystem requires local partners, guanxi, Chinese language capability. Platforms (Douyin, Kuaishou) hold the data and could internalize TEE before a solo founder can monetize. Realistic entry only via Western brands or cross-border agencies — slow, indirect path.

Activate when: A China-adjacent distribution partner, cross-border agency relationship, or Western brand customer with China operations materialises.


GIG ECONOMY RATE NEGOTIATION (Parked — Timing Risk)

Domain: Platform Labor Status: Parked. Market is 1-2 years from existing at scale. Revisit Q1 2027.

Trust problem is real:

  • Algorithmic rate-setting opacity: HRW 155-page May 2025 report documents six of seven major platforms use opaque algorithms; workers don't know pay until after completing job.
  • GMB June 2025 dynamic pricing report: "dynamic pricing benefits Uber more than drivers" — even formal union recognition doesn't solve the data asymmetry.
  • The TEE argument: sealed algorithmic fairness attestation proves algorithm is fair without revealing the algorithm. Exactly right technically.

Why parked:

  • EU Platform Work Directive national transposition deadline: December 2, 2026. As of July 2026, most member states still drafting implementing legislation.
  • The product reframes from "negotiation platform" to "compliance attestation tool" — different buyer (platform legal/compliance team), different sales motion (B2B enterprise SaaS), same 6-12 month procurement cycle problem.
  • US market 2-3 years behind (Prop 22 upheld; federal bill uncertain).
  • UK bilateral negotiation (Uber-GMB, Deliveroo-GMB quarterly meetings) is live but small scale — the negotiation is collective, quarterly, and the data tool GMB actually uses is standard statistical analysis, not TEE.

Timing advantage if you move now: EU member states still drafting legislation. A founder who builds the algorithmic fairness attestation tool today and engages with the legislative drafting process could be the first compliant solution when mandates activate December 2026. That's a 6-month window.

Activate when: EU national transposition laws take effect AND a specific platform compliance team engagement materialises.


STRUCTURAL FINDING: THE THREE PATTERNS

After adversarial analysis across 38 candidates, real two-party TEE negotiation clusters into three structural patterns:

Pattern 1: The Commercially-Interested Intermediary

An intermediary holds both parties' private parameters simultaneously and has economic incentive to exploit the information asymmetry.

Examples: Talent manager (creator × brand), PBM (pharma × payer), block trade broker (asset manager × receiving investor), headhunter (candidate × employer), platform operator (agent × agent)

Why TEE is necessary: NDA binds behavior but cannot prevent mental use of information. TEE removes the intermediary from the information channel structurally — not just legally.

This is the largest and most immediately addressable pattern.

Pattern 2: The Live Data That Cannot Be Disclosed

The private information is a live, continuously-changing data feed that cannot be disclosed even for verification without permanently transferring competitive intelligence.

Examples: Network telemetry (MNO × MVNO), actuarial cost model (payer × hospital), agent instruction set (B2B agent × agent), cost curve (PPA generator × offtaker)

Why TEE is necessary: Audit clauses and NDAs can govern documents. They cannot govern live data without disclosing it. TEE is the only mechanism that can process live data from two competing parties and return a verified result without either party seeing the other's operational data.

Pattern 3: Legally-Protected Information That Must Be Used

Private information is protected from disclosure by law (attorney-client privilege, TINA cost certification, HIPAA), but must be USED in a joint computation to resolve a negotiation.

Examples: Case-strength EV in PI settlement (attorney-client privilege), litigation funder floor in arbitration, TINA-certified cost data in DoD contracts, PHI-adjacent data in healthcare rate setting

Why TEE is necessary: Information enters the enclave, computation runs, only the result exits — privilege never waived. Human mediators approximate this via shuttle caucus but cannot provide cryptographic attestation of non-leakage.

CONNECTION TO DEALPROOF STACK

Swarm Finding DealProof Component
Both agents need sealed instruction sets TDX enclave on Phala Cloud
Neither principal can verify opposing agent's code DCAP on-chain attestation — verifiable before payload sent
Negotiation conduct needs auditing not just output πCreds auditor layer — SCAE/ACPP threat coverage
Settlement needs to be trustless Solidity escrow on Ethereum Sepolia
Multi-round memory integrity Contexto memory hashes

The key insight from the swarm: DealProof's architecture was built for the B2B agent × agent case. Every other vertical is the same primitive applied to a different trust gap. The stack doesn't change — only the domain-specific parameters change.


COMPETITIVE LANDSCAPE — NO DIRECT COMPETITORS

Player What They Have What They're Missing
Enclave Markets TEE for crypto block trading No negotiation agents, no πCreds, no escrow, no DCAP
Cybersettle Double-blind insurance settlement No TEE (operator sees both bids), currently suspended
Zycus ANA AI procurement negotiation agent No TEE, platform operator sees everything
SUAVE (Flashbots) Confidential execution for DeFi Not negotiation, no agent layer, no πCreds
Phala Network TEE infrastructure Infrastructure provider, not application layer
Oasis Network Confidential compute Infrastructure provider, not application layer

DealProof is the only product combining: TEE execution + dual autonomous agents + DCAP on-chain attestation + πCreds conduct credentials + on-chain escrow.


Generated from 15-agent adversarial swarm. Tier 1: 15 domain research agents (Sonnet). Tier 2: 6 adversarial passes (Tina's bar, TEE necessity, incumbent killer, volume reality, regulatory minefield, buyer existence). Tier 3: Opus synthesis. Total candidates evaluated: 38. Survivors after full adversarial battery: 6 primary, 9 secondary.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment