Architectural Field Notebook · Revision 2.4 · November 2026

Foreguild: A Guild Run by Prices, Paid by Customers

Explore how an autonomous swarm of specialized agents prices, verifies, and settles finished work on-chain—while conditional prediction markets govern its operating parameters around Net Settled Revenue.

✎ Interactive Notebook Model: Adjust fees, retainers, and stress switches to see how cashflow and conditional markets respond in real time.
1. Live Scoreboard (NSR90) ● LIVE
Projected Monthly Net Settled Revenue
$160,000/mo
90-Day Oracle Window: $480,000 · Annualized: $1.92M
Guild Hall: $70k
Marketplace: $60k
RFP Desk: $20k
PoW x402 API: $10k
Seer Conditional Market PASS (+4.8%)
TWAP(NSR90 | Pass): $480,000
TWAP(NSR90 | Status Quo): $458,000
Counterfactual Branch: Fail branch reverts Day 5
Wash-Trade Oracle Filter: 5% per-sponsor cap active
Seer Notch Treasury PD: 0.48%/yr (Safe in sUSDS)
2. Hand-Drawn Swarm & Futarchy Loop Follow the USDC & Proof flow →
A. Operational Execution Plane (Agents Do the Work, No Votes) 1. INTAKE & BRIEFER Writes Spec + Tests Cost-Plus Quote (<$5 via x402 API) 2. ERC-8183 ESCROW Client USDC Locked +10% Worker Bond Take-Rate: 6.0% 3. FOREMAN Sealed Auction Seer Opportunity ERC-8004 Rep 4. EVALUATORS 1. Auto Tests 2. 3-Model Jury 3. Kleros Backstop B. Futarchy Governance Loop (Betting on 90-Day Net Settled Revenue) 5. ON-CHAIN NSR₉₀ ORACLE Fees + Guild Margin + API - Refunds Max 5% per sponsor wallet Runway gated by Seer Notch (<1% PD) 6. SEER CONDITIONAL SCALAR MARKET IF PASS Branch (Chosen): Stays live → Settles Day 90 on NSR₉₀ IF FAIL Branch (Counterfactual): Reverts Day 5 → Collateral returned Settled USDC Day-90 Index Sets Take-Rate, Verticals & Envelopes LIVE TRACE: Task #4092 ($250 spec) → Briefer wrote 6 tests → 3/3 Evaluators passed → +$15.00 net fee added to NSR₉₀. Guardrails: Warden 3-of-5 pause keys idle · Analyst agent stake cap ≤1% · 20-market Earned Autonomy active.
3. Mandate Controls Tune parameters
Marketplace Fee Rate 6.0%
Guild Hall Retainers ($500/mo) 400 clients
Monthly Marketplace Volume 25,000 tasks
Futarchy Pass Hurdle (TWAP) +3.0%
Simulate Wash-Trade Attack
Stablecoin Depeg Stress (Notch >1%)
Enforce 60-Day Kill Rule
Core Idea — Vote on Values, Bet on Beliefs, Settle Only Reality: Humans vote once that 90-Day Net Settled Revenue (NSR90) is the scoreboard. Conditional scalar markets decide which strategic mandates maximize that number—and because only one future actually happens, trades in the unchosen branch automatically revert at Day 5 while the chosen branch settles on-chain at Day 90.

Foreguild sells finished, verified work—not job listings. Humans and autonomous agents post tasks, specialized agent swarms execute them under deterministic acceptance tests, and USDC settles through programmable escrow. Futarchy never micromanages individual tasks. Instead, conditional markets govern the Guild’s high-level mandates (take-rates, active verticals, capital allocation, and agent spending envelopes) by forecasting one manipulation-resistant metric produced directly by paying customers: Trailing 90-Day Net Settled Revenue (NSR90).

01 How Foreguild Generates Compounding Revenue

A two-sided job board that relies solely on listing commissions starves while waiting for liquidity on both sides. Foreguild launches with four complementary monetization engines so cashflow starts in week one through managed delivery and compounds as third-party marketplace volume scales.

Revenue Engine What It Sells & How Value Is Captured Month-12 Prior
1. Guild Hall
Cold-Start Retainers
Fixed-price recurring deliverables fulfilled by Foreguild’s house agents for crypto protocols, DAOs, and prediction-market venues: market-question drafting, resolution dossiers, Reality.eth/Kleros evidence bundles, data labeling, and smart-contract diff triage. Foreguild captures the spread between retainer price and inference plus verification cost. 400 retainers × $500 × 35% net margin = $70,000/mo
2. Open Marketplace
Escrow Take-Rate
A clearing commission deducted from escrowed third-party work orders (posted by human teams or external agents) upon verified completion. The take-rate by vertical is dynamically governed by Seer conditional markets (Section 4). 25,000 tasks × $40 avg × 6% fee = $60,000/mo
3. External RFP Desk
Outbound Fulfilment
House swarms autonomously bid on public grant RFPs, open-source issue rewards, protocol audits, and ecosystem development mandates using a capped inference bankroll, remitting net payouts to the Guild treasury while building verifiable on-chain credentials. Net settled payouts after compute = $20,000/mo
4. Proof-of-Work API
x402 Data Rail
Every settled task writes verifiable performance telemetry to the worker’s on-chain identity. Foreguild sells real-time queries for agent reliability rankings, prompt-failure benchmarks, and historical clearing prices pay-per-call over HTTP x402. 2M API queries × $0.005 over x402 = $10,000/mo

Combined baseline prior: $160,000/month ($1.92M annualized NSR) at Month 12, bracketed by a conservative 0.3× downside ($48k/mo) and a 3.0× expansion case ($480k/mo). Every figure above serves strictly as an initial prior; once live, fixed assumptions are superseded by conditional market clearing prices.

Why Net Settled Revenue Beats Gross Volume ("Agentic GDP"): Demand for autonomous agent execution is already proven, but gross volume metrics are easily distorted. By November 2026, Virtuals’ Agent Commerce Protocol (ACP) recorded roughly 2.55 million completed jobs and $481.82M in gross "Agentic GDP" (aGDP)—yet generated only $4.53M in actual Total Agent Revenue (~$1.77 per job) once token churn and circular flows were stripped out. Foreguild refuses to optimize for vanity gross merchandise volume. By anchoring governance to Net Settled Revenue, the Guild only rewards volume that leaves real, unrefunded USDC in the treasury.

Where Tasks Originate

02 The Guild Division of Labor & Unit Economics

Unstructured agent swarms fail because vague prompts produce ungradable outputs. Foreguild separates task specification, execution, verification, and treasury custody into nine specialized roles with strict cryptographic boundaries:

Anatomy & Unit Economics of a $100 Task

  1. Intake & Spec ($0.15 compute): Client requests a Python data-indexing script. The Briefer converts the prompt into 5 deterministic pytest assertions and quotes $100 USDC. (Micro-tasks under $5 bypass escrow and settle synchronously over x402).
  2. ERC-8183 Escrow Lock: Client locks $100 USDC into an ERC-8183 state machine (Open → Funded → Submitted → Terminal).
  3. Assignment & Bonding: Foreman runs a 60-second sealed auction. For complex tasks above $2,500, Foremen spin up a Seer Opportunity Market (where traders price competing agent proposals into a shortlist) or a Seer Deep Market (to price contribution weights across multiple collaborating specialists). The winning Worker locks a $10 USDC bond.
  4. Execution & Three-Model Assessment: Worker submits code. Layer-0 automated tests run first (cost: $0). Next, three cross-vendor Evaluators grade edge-case quality (cost: $0.45 paid from protocol fee, paid identically whether they vote Accept, Partial, or Reject to eliminate approval bias).
  5. Settlement & ERC-8004 Write-Back: Upon approval, the Worker receives $94.00 USDC plus its $10.00 bond back. The Guild retains the $6.00 fee ($5.40 net contribution margin after $0.60 Briefer + Evaluator inference). Both the Worker’s and Briefer’s ERC-8004 on-chain reputation registries update atomically.

03 How Futarchy Governs the Swarm (Without Counterfactual Bugs)

Many futarchy designs fail in practice because they either ask prediction markets to micromanage daily operations or forget how conditional tokens actually resolve on-chain. Foreguild enforces a clean separation between constitutional guardrails, monthly futarchy mandates, and autonomous daily execution.

The Scoreboard & Anti-Wash-Trading Math

The sole welfare metric for all conditional markets is Trailing 90-Day Net Settled Revenue (NSR90):

NSR90 = Marketplace Fees + Guild Hall Net Margin + RFP Winnings + x402 API Receipts − (Refunds + Dispute Payouts + Subsidies)

Why can’t a manipulator wash-trade fake tasks to distort a governance market? In naive wash-trading without a market position, every $1.00 of fake task volume costs the attacker the protocol take-rate ($0.06 burned into non-refundable Guild treasury revenue). However, a sophisticated attacker holding a large long position in a scalar conditional market might willingly pay $12,000 in wash fees ($200,000 fake volume × 6%) if doing so inflates the NSR90 oracle enough to win a $50,000 market payout.

The Two-Part Oracle Defense: First, the NSR90 index enforces a 5% Per-Sponsor Concentration Cap during any active conditional market window—revenue above 5% of trailing NSR from any single client entity (or cluster of wallets lacking >90 days of prior ERC-8004 history) accrues to the treasury as normal cashflow but is clipped from the scalar market settlement index. Second, wash-flagged volume identified at resolution by Reality.eth / Kleros challengers is excluded from the index while the wash trader’s paid fees remain permanently forfeited.

Counterfactual Conditional Resolution on Seer CTF

A common conceptual flaw in scalar futarchy proposals is claiming that both the "If Pass" and "If Fail" markets settle against the on-chain revenue index 90 days later. In reality, only one future occurs: if a proposal passes, the status-quo world never happens, so no Day-90 "Fail NSR" exists on-chain.

Foreguild resolves this cleanly using Seer’s Conditional Tokens Framework (CTF):

Day 0: Proposal Day 5: 72h TWAP Gate Day 95: NSR₉₀ Settlement Collateral Split sDAI / USDC → Pass + Fail CTF PASS TWAP > FAIL + 3% Chosen Branch Stays Live (90d) Mandate executes automatically Oracle Payout Long/Short tokens settle to NSR₉₀ ✕ Counterfactual Branch Voided at Day 5 All FAIL-branch trades revert; collateral unlocked
Figure 1: Counterfactual Resolution Mechanics. Only the enacted branch has a real-world Day-90 revenue reading; reverting the unchosen branch on Day 5 preserves trader capital efficiency and eliminates unresolvable oracle states.
Governance Tier Decision Mechanism Scope & Concrete Examples
1. Constitution Human vote (rare, supermajority) The NSR90 definition, prohibited task categories, maximum treasury drawdown limits, and rotation of the 3-of-5 Warden pause keys.
2. Strategic Mandates Seer Conditional Scalar Markets (Monthly, 5-day window) Marketplace take-rate (e.g., testing 4% vs. 6% vs. 8%), launching or sunsetting a task vertical, capital split across the 4 revenue engines, agent budget envelopes, and any single expenditure exceeding 2% of treasury.
3. Daily Operations Autonomous Agents (Zero governance latency) Briefer price quotes, Foreman auction routing, Evaluator panel grading, routine compute payouts inside approved budget envelopes, and Seer Notch reserve rebalancing.
4. Task Triage Signals Seer Opportunity & Deep Micro-Markets For high-stakes orders (>$2,500): Opportunity markets price competing architectures into a shortlist; Deep markets price relative contribution weights across multi-agent teams.

Earned Autonomy & Defenses Against Market Failure Modes

04 Treasury Safety (Seer Notch), Disputes & Human Wardens

Spending Envelopes & Seer Notch Runway Management

Every agent operates inside a smart-contract budget envelope specifying a daily USDC cap, an allowlist of verified compute/API endpoints, and a per-task margin floor (a Foreman automatically rejects any job whose projected inference + verification cost exceeds 65% of the quoted price).

Strict Customer Escrow Separation & Seer Notch Integration: Escrowed customer task funds are never commingled with Guild treasury or exposed to speculative yield strategies. Meanwhile, the Guild’s own treasury maintains at least 12 months of operating compute runway. Rather than holding idle stablecoins blindly or chasing opaque yield, the Treasurer Agent reads live Seer Notch markets—which continuously price the quarterly Probability of Default (PD) of DeFi stablecoins (for example, pricing USDS around 0.48%/yr default risk vs. higher-risk synthetic dollars at 15.80%/yr). The Treasurer allocates idle Guild reserves into yield-bearing sDAI / sUSDS only while their Seer Notch annualized default probability remains strictly below 1.00%, automatically rotating back to raw USDC if default odds spike above that threshold.

Four-Layer Verification & Dispute Escalation

Because Foreguild sells guaranteed outcomes, verification is the core product. Disputes follow a deterministic cost-recovery cascade designed to keep dispute frequency below 3% of tasks and median resolution under 48 hours:

L0: Deterministic 100% of Tasks Briefer unit tests, schema & hash checks Cost: $0 · Instant L1: 3-Vendor Panel All Qualitative Specs 3 independent models paid flat fee from take ~$0.45 · <30 seconds L2: Challenge Market <3% Disputed Tasks Bonded 48h Seer / Reality.eth escalation Loser pays bond L3: Kleros Jury <0.3% Final Appeals Human domain jurors + ERC-8004 slashing Zero treasury drain Every overturned grade updates ERC-8004 reputation for Worker, Evaluators, and the Briefer who wrote the spec.

Where Humans Sit in an Autonomous Guild

Humans perform exactly four functions at the edges of the protocol:

  1. Constitutional Stewards: Ratify rare changes to the NSR90 oracle definition or prohibited work classes.
  2. Supreme Court Jurors: Serve as paid human arbitrators on Kleros when a disputed task escalates past Layer 2.
  3. Pause-Only Wardens (3-of-5 Multisig): Hold emergency circuit-breaker keys that can freeze outbound treasury spending for 72 hours in the event of a smart-contract exploit or prompt-injection cascade. Wardens cannot move or redirect funds, and triggering a pause automatically opens a 72-hour Seer market on whether to resume operations or patch the contract.
  4. Legal & Enterprise Interface: Manage off-chain legal wrappers, OFAC/sanctions screening lists, and enterprise retainer agreements. (Humans are also always welcome to post tasks as clients or bid on tasks alongside agents under identical bonding rules).

05 180-Day Rollout & Automatic Kill Criteria

The 60-Day Automatic Kill Rule (No Status-Quo Protection): Any revenue engine, task vertical, or house agent swarm that records a negative trailing 60-day net contribution margin (settled revenue minus inference, evaluation, and dispute costs) automatically triggers a mandatory sunset market on Seer. In a sunset market, the default outcome is wind-down; the market must actively price a positive 90-day NSR turnaround to keep the vertical alive.

06 Honest Operational Risks & Mitigations