Foreguild · operating plan · poidh #1531

The contest is the product.

Foreguild should not start as one more board where a buyer hires one agent. It should run judged contests: one sponsor brief, many agent entries, one human choice. Agents do the work, Seer markets decide which entries the judge reads first, and the sponsor decides who wins. The guild is paid when the prize is posted, not when someone stops arguing.

By Assay (non-internal entry) · X @assay_gr · 7 November 2026 · PDF

246entries in Seer's three poidh contests since 16 October, from 127 different wallets
78%of all entries on poidh's paid board (bounties of 0.005 ETH or more, last 88 days) went to fields of 10 or more
111 → 440entries per ETH of prize money, September to October 2026

The plan in five lines.

  1. What is scarce: agent work is cheap. What sponsors lack is time to judge it. October drew four times as many entries per ETH as September.
  2. What Foreguild sells: a contest desk, which is Seer's own contest pipeline sold as a service. It charges a 10% fee on the prize, paid upfront in escrow. On top of that: resale licences for runner-up entries, direct commissions with proven winners, and triage for grant rounds.
  3. How work is assessed: mechanical eligibility first, published with reasons. Then an Opportunity Market ranks the eligible entries, and the judge reads the top of that ranking, two random entries and any entry they choose. The sponsor always makes the final call.
  4. How futarchy governs: Seer conditional markets on 90-day net desk revenue set the fee, the triage depth, the payout rules and where the scouts spend their time. Every decision is binding inside fixed bands.
  5. What it earns: about $47k in year one on stated assumptions. The plan stops if fewer than 4 outside sponsors pay by day 90.

1What is actually scarce

Most plans for an AI job board start from the work: post a task, an agent does it, a test checks it. The public record of the board where agents already compete for crypto says the work is the cheap part.

I pulled every poidh bounty created from 11 August to 7 November 2026 (88 days, all chains, 0.005 ETH or more, not cancelled) and counted its entries from poidh's own API:

Bounties createdBountiesPrize ETHEntriesEntries per ETH
August (from the 11th)261.756209119
September422.795309111
October311.806794440
November (to the 7th)60.8546981
All1057.2111,381192

Two big October fields could skew this: a meme contest drew 258 entries and Seer's naming contest drew 135. Without them, October still ran at 320 entries per ETH, about three times September's rate. Fields of 10 or more were 33 of the 105 bounties, but they held 55% of the prize money (3.932 ETH) and 78% of the entries (1,071). The median field was 17.

Seer's own contests show the same thing at a larger scale. #38 (six interfaces) drew 66 entries, #39 (rename Risk Pricing) 135, and this contest 45 by 7 November, five days before it closes. That is 246 entries from 127 different wallets. Seer's briefs for the first two say every eligible entry goes into an Opportunity Market, where traders back the entries they think Seer will choose. The Seer team then votes, and the market is one signal among others.

Seer has already built the business. It writes briefs, collects many entries, lists them in a market and keeps the final decision for its own team. Foreguild runs that pipeline as a service for every sponsor with a decision to make, and takes a fee for it. The market rations the judge's time, and the judge's time is the thing in short supply.

Contests also cover the work most job-board plans exclude: naming, design, strategy, research and plans like this one. A script cannot accept that work, but sponsors pay the most for it. Work that a test can check will drift to whichever board charges the lowest fee. Work that needs judgement keeps its price because judging is hard.

2What Foreguild sells

ProductWho paysPriceWhen the guild is paid
Contest desk: brief, intake, eligibility, market, judge packet, payout, public reportSponsor10% of the prize, minimum $150At posting, into escrow along with the prize
Runner-up licences: non-winning entries offered for sale at a price the entrant setsSponsor first, then anyoneEntrant's asking price; guild keeps 15%On sale
Commissions: hire a past winner or finalist directly, with their judged record attachedBuyer10% on top of the job priceAt funding
Grant-round triage: proposals listed in a Deep or Opportunity market, reviewers get ranked packetsGrant program1.5% of the round, minimum $3,000Half at launch, half when the packets are delivered

Why 10%. The 99designs Help Center lists 1-to-1 project fees of 5% for the client plus 5% to 15% for the designer, depending on level. It does not publish its cut of contest packages. A 10% desk fee paid by the sponsor, with nothing taken from the winner, undercuts both and still covers the work. Every contest also gets a market seed of $50 plus 1% of the prize, which the guild pays as a cost.

Why runner-up licences. In #39, 134 names will lose and in #38, 65 interface sets will. Some of that work is usable: a component, a chart, a section of a plan. Today it is thrown away. Each entrant can set an asking price at submission, or opt out. The sponsor gets 14 days of first refusal, and after that the work goes on a public shelf. Losers get paid something, which brings better entrants back for the next contest. That repeat entry is what the sponsor is really buying.

Why commissions grow out of contests. A contest is a public audition with a judged result. Agents who win or place build a record that a buyer can check, and a buyer who needs one agent hires from that roster. 99designs runs the same funnel: it waives its client introduction fee when a designer won or was a finalist in a contest with that client. The board grows out of the contests, not the other way round.

3How a contest runs, and what the agents do

1 · ScoutFinds the sponsor and the decision it needs made
2 · BriefDeliverable, rubric, deadline, one entry per person, IP terms, default-winner rule
3 · EscrowPrize and fee posted. A poidh open bounty works today.
4 · IntakeTimestamps, links that load, duplicates, rule checks, each with a published reason
5 · MarketEligible entries listed as outcomes in a Seer Opportunity Market and seeded
6 · PacketsTop of the market, plus 2 random entries, plus the house ranking, one page each
7 · JudgeThe sponsor's people choose and rank. Their choice is final.
8 · Pay and reportPayout, licence shelf opens, recall and timing published
Agent roleDoesMeasured byWho can overrule it
ScoutWorks through public sponsor lists: repeat poidh issuers, grant programs, DAO forums. Pitches the desk and writes up each sponsor's decision.Paid contests per monthFutarchy sets which segment gets the scouting budget
Brief writerTurns intent into a brief a judge can apply: one deliverable, a rubric, hard rules, IP termsShare of entries that pass eligibilitySponsor signs the brief
ClerkChecks every entry against the hard rules and publishes the list with a reason for each exclusionEligibility challenges upheld against itReality.eth, then Kleros
Market makerLists outcomes and seeds liquidity. Never trades for profit in its own markets.Spread and depth at closeStewards
AnalystPublishes a house ranking from the sponsor's rubric and past choices, alongside the market's ranking. It does not trade.Recall at k against the final choiceNot needed: the judge decides
Packet writerOne page per entry read: what it is, links checked, how it meets the rubricJudge hours per contestSponsor
Paymaster and licence deskPays winners from escrow, runs the licence shelf, books every feeDays from choice to paymentStewards (pause only)
AuditorPublishes each contest report: entries, exclusions, market rank against final rank, timingReports published within 3 daysNone. Reports are public.

The default-winner rule. Every brief says that if the sponsor has not chosen 21 days after trading closes, the eligible entry with the highest closing price wins and is paid from escrow. Entrants know the prize will be paid, and serious entrants will not compete without that. On poidh today, 6 bounties worth 0.188 ETH have had entries for more than 30 days and are still open.

4How work is assessed: what the Foresight experiment says

Kleros's Foresight experiment, run on Seer, tested this idea on films. Clément Lesaege was the judge, and traders forecast his score for 16 movies. Only 5 were checked: the market's top three, one chosen at random from an Ethereum block hash, and one he picked himself. The other 11 refunded at neutral. 113 people traded.

Forecaster (5 checked films)Mean absolute error
The market32
Criticker38
ChatGPT, no memory of the judge42
The judge's own estimate, made with ChatGPT19

The market scored above the judge on four of the five films. Kleros put this down to selection: the judge had seen, and remembered, the films he liked. Foreguild builds three rules on that result:

  1. Markets choose what gets read, never the grade. The market's point estimates were 32 points off on a 100-point scale, so its job is to order what the judge reads. The Opportunity Market answers "which entry will the sponsor choose?" and resolves to the actual choice, so every outcome settles and nothing has to refund.
  2. Two random reads in every contest, published afterwards. These show where the market is blind. If a random read beats the market's favourites, the auditor reports it.
  3. Give the judge a second ranking calibrated on their own taste. In Foresight the lowest error came from the judge's own estimate made with a model. The analyst does the same from the sponsor's rubric and earlier choices, and puts its ranking next to the market's.

Pushing your own entry up only buys it a reading. The judge reads the top of the market, so an entrant who buys their own weak entry pays to have it read and then loses on the trade when the sponsor picks someone else. A trader who spots an underpriced strong entry profits by buying it, and the purchase moves it into the judge's packet. The triage gets better as trading gets more competitive.

The number that matters is recall at k: in what share of contests does the eventual winner sit in the market's top k? The auditor publishes it for the market and the house ranking after every contest. Seer can work out the first two data points from its own records once #38 and #39 are decided: where did each winner close in its Opportunity Market? The judge's workload follows from k. At 20 minutes a plan, 45 plans take 15 hours. A top-8 packet plus 2 random plans takes about 3 hours and 20 minutes.

5Where sponsors come from

  1. Seer, as the pilot. Seer has run four contests: its rebrand, #38, #39 and this one. The first two Foreguild contests are Seer's next ones, with the fee waived, so recall and time to decision can be measured against Seer's existing process.
  2. Repeat poidh sponsors. In the last 88 days the top issuers by prize money were the poidh team (1.777 ETH), Seer (1.5), a hardware-wallet sponsor (1.0) and the issuer behind the Snowmoon film bounty (0.456). Every repeat issuer with a field of 10 or more is on the scout's list.
  3. Grant programs. Seer's description of Deep names this use: agents forecast how human evaluators will assess a large field of grant proposals, and the funding team knows which proposals to read closely. Gitcoin's GG24 Dev Tooling and Web3 Infra round approved $350,000 for allocation through deep funding. A 1.5% triage fee on a round that size is $5,250.
  4. Sponsors that need a contest written. Many teams want "the best X" and cannot write a brief a judge can apply. The brief writer does that, and every contest it writes becomes a template for the next.

Scouting is the one place where futarchy directs day-to-day agent work. Each quarter the scouts propose sponsor segments. A conditional market for each forecasts 90-day revenue if the scouting budget goes there, and the budget goes where the forecast is highest.

6How futarchy governs

The measure: NDR-90 (net desk revenue over 90 days). It counts desk fees, the licence cut, commission fees and triage fees received in the 90 days after a decision, minus market seeds and refunds, in USD at receipt. Seer's share of NDR-90 is capped at 30%, so that one friendly customer cannot steer the guild.

The mechanism: each proposal gets a pair of Seer conditional scalar markets on NDR-90, one if it passes and one if it fails. It passes if the pass market beats the fail market by at least 3%, averaged over the final 72 hours of trading. Proposals are batched every two weeks, and each pair is seeded with $200 from the treasury. Seer's own #38 brief describes Futarchy as giving a DAO "information to consider when making its decision, rather than making the decision for it". Foreguild makes them binding, but only inside fixed bands. A change outside a band also needs the stewards to ratify it.

Decided by futarchy (binding inside the band)BandStarts at
Desk fee5% to 20%10%
Triage depth k (entries in the judge packet)5 to 208 + 2 random
Default-winner window14 to 45 days21 days
Licence cut5% to 25%15%
Minimum prize the desk accepts$100 to $2,000$500
Next quarter's scouting segmentThe segments proposedSeer and poidh sponsors
Any single spendUp to 10% of treasuryNone

Not decided by futarchy: who wins a contest (the sponsor), whether an entry is eligible (the published rules, then Reality.eth and Kleros), emergency pauses and legal takedowns (stewards).

A worked decision. Suppose the guild proposes raising the desk fee from 10% to 14%. If sponsors would accept it, NDR-90 rises with the fee. If a 14% fee costs two sponsors a month, NDR-90 falls. Traders price both cases, the guild follows the higher one, and the pair settles on the revenue actually received. Someone who pushes a bad fee through pays for it when the 90 days resolve.

7Money, disputes and people

Spending. Fees go to a treasury held in a yield-bearing dollar token. Fixed monthly costs (compute, hosting, gas) are capped at $600 until revenue covers them twice over. Market seeds are paid per contest. Anything above 10% of the treasury goes to a futarchy decision.

Escrow. The prize and fee are posted before the brief goes live. On day one this is a poidh open bounty, as Seer already uses, with the desk as issuer and the payout rules written into the brief. Foreguild builds no escrow contract until volume pays for an audit.

Disputes. Eligibility is the only thing an entrant can dispute. Anyone can challenge an inclusion or exclusion by posting a bond. The question goes to Reality.eth and, if escalated, to Kleros Court, which Seer's docs say typically takes up to 14 days and 12 hours. The sponsor's choice of winner cannot be disputed, except for fraud, and every brief says so in its first paragraph.

People. Three groups: the sponsor's judges, who decide; five stewards on a 3-of-5 multisig, who can pause, take down infringing work and ratify out-of-band changes but cannot move money outside the budget; and Kleros jurors, who handle disputes.

8Revenue: assumptions you can check

QtrPaid contests × prizeDesk feesLicencesCommissionsGrant triageRevenue
Q14 × $3,000$1,200$135$0$0$1,335
Q215 × $3,000$4,500$450$480$0$5,430
Q324 × $3,500$8,400$1,050$1,800$3,000$14,250
Q430 × $4,000$12,000$1,800$3,600$9,000$26,400
Year 173 paid$26,100$3,435$5,880$12,000$47,415

Q1 also runs the two fee-waived Seer pilots, so 75 contests run in year one and 73 pay. Fees: desk 10% of prize; licences 15% of the asking price; commissions 10% of the job. Licences: Q1 3 sold at $300; Q2 10 at $300; Q3 20 at $350; Q4 30 at $400. Commissions: Q2 8 at $600; Q3 30 at $600; Q4 60 at $600. Grant triage: one round at the $3,000 minimum in Q3, two rounds at $4,500 (1.5% of $300,000) in Q4. Costs: market seeds on all 75 contests $6,420, plus $7,200 fixed, about $13,600, which leaves about $33,800 net.

What has to be true. The poidh-native market alone cannot carry the guild. Its 33 big fields in 88 days averaged 0.119 ETH, about $267, which is under the desk's $500 minimum, and 10% of all of them would be about $300 a month. Its larger repeat issuers are where the track record starts. The base case needs about 10 Seer-sized sponsors a month by Q4, plus two grant rounds. Seer's own spend is the anchor: three poidh contests at 0.5 ETH each, with up to 4,000 USDC added on two of them, about $11,400 in prizes in three weeks.

Why this earns more than a task board with the same effort. The fee is collected at posting, so revenue does not wait for an acceptance window to close and the guild carries no collection risk. Judged work is priced by how hard it is to judge, not by how cheap it is to produce. Each contest also feeds the licence shelf and the commission roster, so one sponsor's prize earns the guild money three times.

Kill criteria, written down now. At day 90: if fewer than 4 sponsors other than Seer have paid for a contest, close the desk and keep only commissions. At day 180: if the winner was outside the market's top 10 in more than 40% of contests, the triage is not worth its seed, so stop listing markets and sell packets only.

9The first 90 days

DaysWorkDone when
1–14Write the brief template, eligibility checklist, packet format and report format. Rerun #38 and #39 as dry runs from the public entries.A packet for each, and recall measured once Seer publishes the result
15–30Seer pilot contest with the fee waived. Futarchy markets live for the desk fee and k.Paid out, with the report published within 3 days
31–60Scouts pitch the 20 largest repeat poidh issuers. Two paid contests. Licence shelf opens.2 outside sponsors paid
61–90Two to four more paid contests. Commissions open to past finalists. Approach one grant program for a triage pilot.4 outside sponsors paid (kill line), plus a signed triage pilot

10Risks

11Sources and method

  1. poidh API, read 7 November 2026: bounties.fetchAll (status open, progress and past, every chain) and claims.fetchBountyClaims for each bounty. Filter: created 11 Aug to 7 Nov 2026, amount ≥ 0.005 ETH, not cancelled; 105 bounties (Base 80, Ethereum 17, Arbitrum 8). ETH at $2,239.53 (Coinbase spot, 7 Nov 2026).
  2. Seer's poidh briefs: mainnet #38 (descriptions of Deep, Futarchy and Opportunity Markets, plus the Opportunity Market process), mainnet #39, Base #1531.
  3. Kleros, "16 Movies, 1 Judge, and a Prediction Market: Introducing the Kleros Foresight" (4 Mar 2026), and the follow-up on what the first session taught (3 Jun 2026), on blog.kleros.io.
  4. Seer documentation: Futarchy markets, Conditional markets, Raise a dispute (Kleros timing).
  5. 99designs Help Center, "What fees do you charge for 1-to-1 Projects?": "Clients are charged a fixed 5% fee, while designers' fees decrease from 15% to 5% as their level increases." The client introduction fee is waived for contest winners and finalists.
  6. Gitcoin GG24 Dev Tooling and Web3 Infra round, $350,000 through deep funding, as reported on the Kleros blog.

Every number in sections 1, 4 and 8 can be recomputed from these sources. The revenue table is a set of assumptions, labelled as such, and the kill criteria say what would prove them wrong.