Sell the content, not the contest

An operating plan for Foreguild, the futarchy-governed AI job board, optimised for revenue. Its first customers are the crypto projects already paying for memes, threads and videos through public contests, every week, in plain sight.

Oliver Dave (@DaveTheCryptoOG), Web3 content creator since 2021, writing for projects under contract and in their public contests. Non-internal entry, 10 November 2026. Drafted with AI assistance. Every figure is marked Fact or Assumption.

Who pays
Crypto projects that already spend on content contests. In one day of X searches on 8 November we found 27 of them running, judging or paying one.
What they get
Foreguild's agents make the project's memes, graphics, thread drafts and short video edits every week. The project approves each piece and posts it from its own accounts. Plans cost $199, $499 or $1,199 a month.
Where the job board is
Every subscription turns into weekly briefs that any outside AI agent can fill. Agents are paid per published piece, with a bonus when it performs.
How futarchy steers
Each week a Seer market forecasts how every candidate post will perform. The client publishes the market's picks, one random candidate and one of its own, and the market settles on real engagement 48 hours later. Company decisions run on conditional markets over 90-day kept margin from renewing clients. One week of the weekly loop ran live on 10 November, and the forecast made before posting picked the winning post (Appendix B).
What it earns
Base case: 51 paying clients and $23,500 monthly revenue by month 12, first profitable month 6, about $10,400 of funding needed to get there. Stop rules at days 60, 90 and 180.

1The customer is already paying

Most plans in this contest say the same thing about demand: agents are cheap, buyers are scarce, and the first job is to find someone who will pay. This plan starts from buyers who are paying already, in public, every day. They are crypto projects buying content through contests.

One day on X

On 8 November 2026 (UTC) we read X's Latest tab for two searches covering meme, thread, content, video, writing and creator contests. We only read posts. We found 27 distinct crypto projects running, judging or paying out a content contest that day Fact. The full list, with a link to each post, is in Appendix A. It is a floor, not a census: contests announced only inside Telegram or Discord groups never reach X search.

What we sawCountDetail
Projects with a live, judging or paid-out contest2724 meme or meme-video, 1 thread, 1 video, 1 creator content
Posted a prize in dollars8$10 to $590, median $100
Paid in the project's own token or in whitelist spots4$BOP, $GWA, XEC, a mint whitelist
Run weekly or in rounds5Chief's Tribe, $CLAUDIA, Astrochimpz, LATER, ivy
Winner pending, or contest questioned in public3Sendu, LATER, Kpurr; plus two entrants publicly complaining about other contests ("no update", "rigged")

The posts say more than the counts. Three of them:

later - snoozy @later_snoozy

Meme contest week 1 is closed … Picking the winner… later. (Tomorrow, promise.)

Judging is the bottleneck, even for a project that runs a contest every week.
View post
breezy klein @0xbreezyklein

Three entries to the Nmeows meme contest at 13:58, 14:05 and 14:12 UTC, fourteen minutes apart, plus entries to two other contests the same day.

Volume is cheap. The project still has to read all of it.
View post
Nmeows @Nmeows_Elysium

Announces three winners and warns that anyone claiming to be Nmeows in their messages should be verified first.

Trust is thin on both sides: entrants fear fake payers, projects fear fake entrants.
View post

What one day does not show

Small memecoins dominate X search. Established projects run larger, rarer contests. From the author's own work since 2021 Field notes: pools of $500 to $2,000 are normal for established projects and $5,000 is occasional; small contests draw 20 to 50 entries and big ones 200 or more; and 15 times a project rugged or simply did not pay its winners. Seer is itself one of these buyers. In its own summary of the last three months, it pledged about $5,000 each to an interface-redesign contest, an AI book contest and this FAO contest, about $1,000 to rename Risk Pricing, and awarded $1,000 to name the job board; this contest alone had 55 plans by 9 November Fact.

The money moved; it did not disappear

On 15 January 2026 X revoked API access for apps that reward users for posting Fact. Kaito shut down Yaps and moved to Kaito Studio, curated campaigns between brands and creators, which opened in beta in March with 16 brand partners Fact. In June, Pump.fun's GO bounty board opened with 225 live bounties and about $115,000 waiting to be claimed, much of it for promotional content Fact. Paying strangers to post is finished. Projects still pay for content, and they are looking for a channel that works.

2Why contests fail the project that pays for them

A contest buys one week of attention and leaks value five ways.

LeakWhat happens
Judging timeSomeone on the team reads 20 to 200 entries, and winners slip by days.
Farmed entriesThe same accounts submit to every contest, often several times, increasingly with unedited AI output.
Reach lands elsewhereThe winning post sits on the entrant's account, often small, not on the project's own feed where its holders are.
Nothing compoundsNext week starts from zero. Nobody learns the brand's voice, and the best performer may never enter again.
TrustEntrants fear the prize won't be paid; projects fear impersonators and fake engagement.

Some plans in this contest propose selling the contest itself as a service, with a market choosing which entries a judge reads. That repairs the first leak and leaves the other four. Contests are the workaround projects use because steady, on-brand content is hard to buy. Foreguild sells the steady content.

3What Foreguild sells

PlanPrice a monthEach monthReplaces
Spark$19912 memes or graphics, weekly performance noteA $50 weekly meme contest
Pulse$49930 memes or graphics, 4 thread drafts, 4 short video editsA $125 weekly contest
Campaign$1,19960 memes or graphics, 12 thread drafts, 12 short videos, launch-day kitA $300 weekly contest, or one $1,200 monthly

Prices and contents are starting points Assumption; futarchy moves them inside fixed bands (section 5).

Two smaller products

Open board. Humans and AIs post one-off tasks, such as an explainer video, a launch kit or a translation of a thread, and Foreguild keeps 15% of the task price. This is the general job board the brief describes, but it is not the base the business depends on.

Contest screening. Projects that still want a community contest can hand Foreguild the entries. The screener removes duplicates, farmed and rule-breaking entries and returns a ranked short list for 10% of the pool, minimum $49. It is the doorway to a subscription: a project that sees its own contest sorted in an hour is the easiest Spark sale there is.

Rules that hold for every client

The project posts from its own accounts. Foreguild never posts for a client, never pays anyone to post and never buys engagement, so it sits outside X's ban on reward-for-posting apps. No content predicts a price or promises a return. Clients pay monthly in USDC up front, and creators are paid weekly from that balance, so no creator ever waits on a project to honour a prize.

4How a week runs, and what the agents do

BriefBrand clerk writes the week's slots from the brand kit
CreateOutside creator agents submit up to three candidates per slot
ScreenRules, originality, facts, no price talk
ForecastSeer market and house model rank every candidate
ApproveClient taps yes and posts from its own account
ScoreEngagement percentile at 48 hours, bots filtered
PayCreators paid per published piece plus bonus
AgentDoesMeasured byWho can overrule it
ScoutFinds projects publicly running content contests (Appendix A, run daily) and offers a two-week trialTrials that become paying clientsFutarchy picks the target segment
Brand clerkBuilds a brand kit from the site, docs and past posts, plus a list of claims never to makeHow often the client edits a pieceClient signs the kit
Creator agentsOpen to any agent that registers a payout wallet. They pick slots and submit candidatesPublish rate and performance percentileClient chooses what posts
ScreenerBrand safety, near-duplicate check against the web and past entries, facts checked against docs, no financial promises, image rightsPieces the client rejects for a rule breakClient
House forecasterRanks candidates from the client's own history and publishes it beside the marketDid the best performer sit in its top 3?Not needed: it only advises
Market clerkLists each week's candidates on Seer, seeds liquidity, submits results to Reality.ethSpread and depth at closeStewards; disputes go to Kleros
ScorerComputes each post's engagement percentile from analytics the client shares, filters botsScores overturned on disputeReality.eth, then Kleros
PaymasterPays creators weekly in USDC from the client's prepaid monthDays from scoring to paymentStewards can pause only
Account agentWatches for churn and runs the monthly review with each clientRenewal rateClient

Where the work comes from and how it is assigned

Subscriptions are the main source of tasks: each plan turns into a fixed number of weekly slots, so supply of work is known in advance. One-off tasks on the open board and contest screening jobs come on top. In the first month every registered creator agent may submit to any slot. After that, slots are offered first to agents with the best trailing performance percentile. Agents the audience rewards get more of the work, and that is the main way governance directs what the swarm does day to day.

Spending, disputes and people

Fees sit in a treasury in a yield-bearing dollar token. Fixed costs are capped by the budget in section 6, and any spend over $2,000 goes to a futarchy decision. Creators can dispute a score; clients can dispute a screening failure. Both go to Reality.eth with a bond, then Kleros Court. Three groups of people stay in the loop: the client, who approves every post (if nobody approves, nothing posts); three stewards on a multisig, who can pause, take down content and suspend an agent but cannot move money outside the budget; and Kleros jurors.

5How futarchy governs: two loops

Loop A, every week: which posts to publish

This loop copies the design of Kleros Foresight, which runs on Seer: 16 movies, one judge, a market forecasting his rating of each, and only five actually watched, namely the market's top three, one at random and one of his own choice. Unwatched movies redeemed at neutral value.

Code plansNeed a test script to accept work. Only fits work a script can check.
Contest-desk plansNeed a human judge to read the short list. Slow and costly per decision.
ForeguildThe audience is the judge. Every published post resolves in 48 hours, with no one grading it.

For each client each week, about 20 candidates become scalar markets on one question: if published, what engagement percentile will this post reach against the client's last 30 posts? A percentile, as in Foresight, makes a small account and a large one comparable. The client publishes the market's top picks, one random candidate and one of its own choice, so a candidate the market ignores can still prove it wrong. Unpublished candidates redeem at neutral. Appendix B runs this loop once, live, on the author's own account. At 48 hours the scorer posts the result to Reality.eth; anyone can challenge with a bond and escalate to Kleros.

The market decides what is shown first. It never decides what is posted, and it never pays a creator directly. That matters because Seer's markets are often thin: another entrant measured a median traded volume of about $7 across Seer's conditional markets on Gnosis Fact. So each client-week gets a small seed of $5 to $20, the house forecaster fills in when trading is thin, and the market's ranking is shown to the client as advice until its record beats the house model.

Gaming. A creator could bot engagement on a client's post to earn the bonus. Engagement is read from the client's own analytics, weighting bookmarks, profile visits and replies from established accounts over raw likes; outliers are reviewed; the bonus is capped at 25% of base pay; and a creator caught botting is removed and loses unpaid earnings through a Kleros case.

Loop B, as needed: how the company is run

One measure, KM-90, kept margin over 90 days: subscription revenue from clients in their second month or later, minus creator pay, compute and market seeds. It rewards clients who stay, so a wave of trial sign-ups or memecoins that die after a month cannot inflate it, and revenue from a client wallet linked to a creator wallet is excluded.

Decided by futarchy (binding inside the band)BandStarts at
Price of each plan±40% of start$199 / $499 / $1,199
Creator pay per published meme or graphic$2 to $8$4
Performance bonus share0% to 25%15%
The scout's target segmentMemecoins, DeFi and infrastructure, NFT and gaming, AI agents, Web2 brandsMemecoins and DeFi
Free trial length0 to 21 days14 days
Any spend above $2,000Up to 10% of treasuryNone

Each proposal gets a pair of Seer conditional scalar markets on KM-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 last 72 hours of a seven-day trading window. For the first 90 days results are advisory; after that they bind inside the bands, and a change outside a band also needs the stewards. Not decided by futarchy: what gets posted (the client), disputes (Kleros) and takedowns (stewards).

A worked decision. Should Pulse rise from $499 to $599? More revenue per client pushes KM-90 up; more cancellations push it down. Traders price both worlds. If KM-90 under "raise" beats KM-90 under "keep" by 3%, new Pulse clients pay $599. At day 90 both markets settle on the margin actually kept, and whoever priced the trade-off badly pays for it.

6Revenue, with numbers you can check

Per client per monthSparkPulseCampaign
Price$199$499$1,199
Creator pay (posts $4, threads $12, videos $18)−$48−$240−$600
Screening and compute−$8−$20−$40
Market seeds−$22−$43−$86
Payments and support (3%)−$6−$15−$36
Contribution$115 (58%)$181 (36%)$437 (36%)

All of these are assumptions Assumption. New clients are assumed to split 50% Spark, 35% Pulse and 15% Campaign, giving a blended $454 a month and $186 of contribution. Open-board fees and contest screening add an assumed $12 per active client per month.

ScenarioMonthly churnClients M6Clients M12Revenue M12Year 1 revenueYear 1 netFirst profitable monthFunding needed
Low15%921$9,536$54,695−$18,71012$18,772
Base12%2151$23,545$134,123$6,1126$10,351
High8%40117$54,558$287,107$53,2544$8,157

Fixed costs run $3,000 to $4,000 a month for the first six months and $4,000 to $7,500 after, covering hosting, models, monitoring, and a part-time human for client onboarding and steward duty. The base case needs 3 new paying clients in month 2 rising to 10 a month by month 11, roughly one every three days against a market where we counted 27 running contests in a single day. Churn is high on purpose, because memecoins die.

51paying clients at month 12
$23,545revenue in month 12
$134,123year-one revenue
6first profitable month

The interactive calculator (in the HTML version) and the tables use the same formula, published as model.py alongside this plan.

7The first 90 days

DaysWorkDone when
1 to 14Build the brand clerk, screener and scorer, with three in-house creator agents. Dry-run on five projects from Appendix A using their public posts, running the Appendix B loop for each.Five brand kits and one full week of candidates per project
15 to 45Ten free two-week trials, offered to projects seen running contests. Open slots to outside creator agents. First weekly markets on Seer, advisory only.Ten trials running, first 40 posts scored
46 to 90Convert trials. Launch contest screening. First Loop B decision: trial length.15 paying clients

Stop rules, written now. Day 60: fewer than 4 paying clients, stop and return any prepaid months. Day 90: fewer than 10 paying, or fewer than half of month-one clients renewing, cut back to contest screening only. Day 180: if the house model's top 3 catches the best performer more often than the market's in over 60% of weeks, stop seeding Loop A markets.

8Risks

RiskWhat we do
Fake engagement decides bonusesClient-side analytics, quality-weighted signals, bonus cap, outlier review, removal through Kleros
X changes its rules againNobody is paid to post and nothing posts through Foreguild. Scoring uses the client's own analytics, so it does not depend on X's API, and the content works on Telegram and other platforms too
AI slop damages a client's brandScreener plus human approval on every post. A client that rejects most of a week gets a credit, which hits KM-90, so the markets feel it
Memecoins churn fastKM-90 only counts renewing clients, and futarchy can move the scout toward DeFi, infrastructure, AI agents or Web2 brands
Scam projects buy contentThe scout checks contract, liquidity and team history before a trial. No price promises in any content; stewards can end a client
Thin marketsMarkets rank and advise; the house model covers gaps; the day-180 rule stops Loop A if it adds nothing

9What this plan does not claim

Foreguild does not exist and no client has signed. The only market run so far is the one-week demo in Appendix B. The census is one day of read-only X searches and misses Telegram and Discord. The field notes come from the author's own work and are not audited. Every price, cost, conversion and churn figure is an assumption, and the stop rules say what would prove them wrong. Engagement can be faked; the plan limits that but cannot remove it.

Appendix A. The 27 contests, 8 November 2026

X Latest tab, read-only, two searches. Prizes as posted; blank means none was stated in the post we read.

ProjectFormatPrize as postedStatus that dayPost
Chief's Tribe (@Crypto_Chief3)meme$10 in SOL + custom PFPweekly; winner pick pendingpost
$CLAUDIA (@anabology / SXSI.com)video$420.69 / $69.42 / $100 community favouriteround 2 open to 13 Novpost
Sendu (@Sendu_on_Ely)meme$100 poolclosed; under reviewpost
Astrochimpz (@Astrochimpz)memenot statedrunningpost
Bop & BONK (@TimxBop)meme1,000,000 $BOP across 3 winnerswinners announcedpost
$SOLUNA (@soluna_coin)memenot statedrunningpost
George (community-run)meme5 x $10 USDCjust launchedpost
NFT mint 'Art & Meme contest'meme/videowhitelist spotswinners announcedpost
Kpurr (@kpurr_elysium)memenot statedrunningpost
LOBIN (@lobincoin)meme$250 / $150 / $100 + $50 bonusclosed 8 Nov; winners Tuesdaypost
UniWall (@uniwallhook)creator content$50winner announcedpost
PBUG (@pbugcoin)memenot statedrunningpost
HATS (@tinyhatsfun)memenot statedrunningpost
BROOZ (@BROOZ_MEME)memenot statedrunningpost
QUANT (@TheTrueQuant)memenot statedrunningpost
ivy (@ivydotmeme)memegiveawaysrunning; new contest plannedpost
Gorkmemenot statedrunningpost
BANDIT_ONCHAIN (@BanditOnchain)thread$100 pool, 5 winnerslaunchedpost
Nmeows (@Nmeows_Elysium)meme/videonot statedwinners announcedpost
GWA (@AdilLoser1)meme/video1K-5K $GWArunningpost
The Roamals (@The_Roamals)memenot statedopens 9 Novpost
LATER (@later_snoozy)memenot statedweek 1 closed; week 2 openpost
PAWG OGmeme$100 poolrunningprofile
Chillbara (@ChillbaraSupra)memenot stated10 finalists pickedprofile
Venice (@venicethedino)memenot statedwinners announcedpost
CurveRadar ($CRDR)memenot statedwinners announcedpost
Cain / XEC (@caincurrency)meme1,000,000 XECone week leftpost

Appendix B. Live demo: one week of Loop A on a real feed

On 10 November 2026 the author ran one Foreguild week on his own X account, with himself as the client. The market was created first, at about 10:03 UTC, and the five candidates were posted after it, between 10:10 and 10:58 UTC.

The market: Seer categorical market on Base, 0xBF11…CbF8. "Which of these 5 #Foreguild demo posts by @DaveTheCryptoOG, posted on X on 10 November 2026, will show the highest view count on X at 12:00 UTC on 12 November 2026?" Fact

The house forecast, committed on 9 November before any post existed: D, then A, E, C, B. It is published here so it can be scored against the market and the result.

For a single demo we used one categorical market (which post wins) instead of one scalar market per post as in section 5, so that the whole week settles on one public answer.

PostCandidatePosted (UTC)House rankViews, 12 NovActual rank
DText: 50+ plans, one market, the team decides10:1015151
AMeme: prediction market vs decision market10:222515
EQuestion: what should a market predict next?10:383723
CMeme: reading "futarchy" for the first time10:494714
BExplainer graphic: futarchy in three boxes10:5751082
Post A: two panels, a crowd watching odds on who wins, then a fork in the road with a forecast for each choice
A
Post C: a confused face reading the word futarchy, then a relaxed face reading vote on what you want, bet on how to get it
C
Post B: a proposal splits into a market if yes and a market if no; the higher forecast wins
B

Result

Post D won, with 515 views against 108 for the runner-up Fact. Views were read from X at 11:16 UTC on 12 November, 44 minutes before the market's reading time; a gap of more than four to one leaves no doubt about the winner. The author reported the answer, Post D, on Reality.eth at about 12:10 UTC on 12 November; it becomes final if no one challenges it within Reality.eth's challenge period of about three and a half days.

The house forecast picked the winner and placed E and C correctly, third and fourth, but swapped A and B: the explainer graphic came second and the decision-markets meme came last. One influence is visible and disclosed: Seer's own account liked Post D and replied to it on 10 November, which put it in front of Seer's followers. That is the week's lesson for Foreguild. Who engages with a post moved it more than its format did, so Loop A scores each post against the client's own baseline, and the house model should learn from who responds, not only from what was posted. No liquidity was added, so the market served as a public commitment made before the posts went up, not as a price signal.

What this shows, and what it does not

It shows the whole weekly loop working with today's tools: candidates made by AI, a Seer market listed before publication, real posts, and an answer any reader can check on X. It does not show demand. One small account, five posts, no outside client and no liquidity are a working demonstration, not evidence for the revenue model in section 6.

Disclosure. Seer's mobile create form offered no opening-date field, so Reality.eth reporting opened when the market was created. The question fixes the reading time at 12:00 UTC on 12 November, and the author will challenge any answer reported before then.

Appendix C. Sources

  1. Seer on X, FAO Operating Plan Contest, 5 November 2026, and Poidh bounty Base #1531 (55 claims read 9 November 2026). poidh.xyz/base/bounty/1531
  2. Kleros, "16 Movies, 1 Judge, and a Prediction Market: Introducing the Kleros Foresight", 4 March 2026. blog.kleros.io
  3. KuCoin News and The Block on X revoking API access for reward-for-posting apps, 15 to 16 January 2026. kucoin.com, theblock.co
  4. BlockEden, on Kaito Studio's March 2026 beta with 16 brand partners. blockeden.xyz
  5. Cointelegraph, Pump.fun bounty platform with 225 live bounties and $115,000 unclaimed, 5 June 2026. cointelegraph.com
  6. Seer documentation, futarchy markets on Gnosis. seer-3.gitbook.io
  7. Seer on X, 6 November 2026, pledge of $25,000 toward Open Risk Pricing for DeFi, built with Kleros and Credora.
  8. Seer on X, 10 November 2026, "In the last three months, Seer has…" (list of contests and pledges).
  9. Seer market depth as measured in Poidh claim 9480 (548 conditional markets on Gnosis, median traded volume about $7).
  10. Census of 8 November 2026: Appendix A, one link per project.