The Polymarket file

01The wedge, and the five facts that sharpen it

Their pain, in Peter's words from the Jul 9 call, priced by the research.

Polymarket went from about $1M of US ad spend in 2025 to nine figures across the World Cup, the White House UFC card, and now the NFL, with every major DSP on the plan and AppsFlyer sitting on an install‑to‑first‑deposit funnel. Peter's stated pain is not reporting; it is allocation: "what's working, what's not working, where should the money be spent across which partners, which channels, where's the granularity of the log." Raw logs are gettable; today he downloads AppsFlyer exports by hand. That is our loop exactly: connect the feeds, price the next dollar, propose the cut‑and‑fund pair, keep the decision record. The same discipline their own order book runs on: every price proves itself against a real outcome.

The five facts

Sources: Sportico volume/ads · July volumes · 2025 spend baseline · the influencer exposure

02The first proof

One closed flight, a blind holdout, shadow mode. Easy to say yes to because the data already exists and nothing in market changes.

The shape: a retrospective read on one closed flight, ideally the World Cup, so the NFL pattern is understood before a dollar moves; once the season runs, the instrument runs with it, re‑pricing predictions and adjusting allocations in real time as evidence lands. We return a coverage‑and‑joinability read first, then a ranked set of cut‑and‑fund rows, each a complete decision: source, destination, amount, expected result, confidence, reason. A randomly held‑out slice of partners gets scored blind and revealed after, so the result is checkable by Anthony rather than asserted by us. Shadow mode throughout: no spend authority, no re‑tagging, one integration point.

The sample request, verbatim from the brief

Five items, all things Peter already pulls by hand. The return leg prices the deal: the coverage read tells everyone, including us, whether the grain supports the claims before anyone commits to more.

03The room, seat by seat

What each person needs to hear, the pins with counters ready, and the candor rules.

The pins, with counters ready

Candor rules for this room

The observatory is character, not capability: a real harness‑backed research lab on their public market data that published its own nulls (a first paper portfolio down $31.72 after fees, confidence intervals crossing zero, no recipe passing historical confirmation). Said that way, it is the strongest proof we refuse to overclaim; never imply live trading, found signal, or that it touches ad data. The 15–40% overhead‑to‑working‑media figure from the earlier deck needs a named pilot behind it or gets carried as range‑of‑observed. Blockchain and mathematical‑proof language stays holstered with this room (say tamper‑evident, signed, re‑runnable if asked). The Madhive $100M credential and Bill's investment figure get voiced only with Spencer's go. And Peter's raw‑log promise is secondhand until the walkthrough converts it.

04The company, right now

Raising into a rivalry, buying share with marketing, and newly regulated in the US.

The money: ICE invested up to $2B at roughly $8B pre‑money in October 2025, added $600M in March, the round closed at $15B in April, and as of this week they are in talks to raise about $1B at more than $20B, with annualized revenue described to CNBC as well above $1B. The US re‑entry ran through the $112M QCEX acquisition (a CFTC‑licensed exchange and clearinghouse), a waitlisted December launch, and the open app in May. Distribution: official prediction market of X, with Grok integration.

The rivalry is the frame for everything: July was a record $50.6B month for prediction markets, and Kalshi took 74.5% of it. Kalshi raised at $22B in May and is reported in talks near $40B. Polymarket US grew 54% month over month to $5B, with far more unique users than Kalshi but smaller tickets, and parlays only now rolling out (their absence flattered nobody's notional except the rival's). Marketing is the chosen weapon: the first brand campaign ever (Questions Are Everything, made with Better Half, narrated by Rick Rubin, the Flags film at 30M views in a day), Octagon branding at the White House card, a $21M NFL survivor contest, media deals from WSJ to Substack, and the sports season loaded because their own product team fumbled NFL 2025. They are spending like share is buyable. The question nobody can answer publicly, or internally at grain, is which of those dollars earn it. That is the meeting.

Sources: ICE · the $20B talks · QCEX · the White House card · the brand campaign

05Their media machine

The funnel, the flights, the channels, and the manual truth underneath.

The funnel is app‑install to first deposit, measured on AppsFlyer, bought across every major DSP (The Trade Desk and StackAdapt named on the call; "you name it, it's on a media plan"). The flights so far: World Cup (45 in‑match ads on Fox and Telemundo, country editions with local billboards), the UFC White House card (the 30‑second premiere, branding on the Octagon itself), and now NFL season with the survivor contest and referral codes syndicated through sports media. Beside the bought media: the creator lane (the CMO's personal PayPal problem), podcast and publisher deals, and on‑screen odds placements like the Golden Globes.

Underneath it, the machine is being assembled by hand while the money flies: Peter's team of five or six inside their office, standing up the in‑house buying variant, downloading AppsFlyer logs manually, testing a slate of partners that changes monthly. Geo‑blocked states mean a slice of any national buy cannot legally convert. Measurement hiring is active but junior to the spend. Every decision class we priced for Mediaocean has its analog here, at a hundred‑million‑dollar annual run rate, with fewer systems in the way: which partner earns allocation, which channel saturates, what a creator dollar returns, what the geo waste actually costs, and when to cut a test that is not working.

06The people

Beyond the room: who the room answers to.

Shayne Coplan, founder and CEO, story‑credits the brand campaign personally. Matthew Modabber, CMO, owns brand and the creator lane, and carries the disclosure exposure. The head of growth is a recent ex‑disruptor hire per Peter's call. Jenner Kearns reports into that growth structure and came from Peter's own Dysrupt, which means our champion trained the buyer. Anthony Khaiat sits in research and spent Sep 2024 to Feb 2026 inside Kalshi's data team. The Publicis embed answers to Peter, and Peter's upside is making this repeatable for his other clients: the three‑party frame (Polymarket gets allocation, Publicis gets the repeatable decision layer, Precise runs the economics) is already written in our Jul 9 opportunity brief.

07What we built already

The corpus, audited, with what each piece is for now.

08The vocabulary

Their words and the performance words, so we speak both natively. One bridge line between them.

PART TWOThe working deck

Five spreads, in the order the meeting should move: how they work · how Precise works · how the two fit · how the engagement runs · the questions that make it real. Each lifts into the presentation.

D1How they work today

Their workflow, as it actually runs, from the Jul 9 call and the research.

Beside the bought media: the creator lane (unattributed today), sponsorships and media deals, and the referral codes. Underneath it: eight to eleven geo‑blocked states inside every national buy, and a measurement function that is currently one embedded team and open job requisitions. The plan is an in‑house buying variant; the machine for judging it is what is missing.

D2How Precise works

The mechanism, in one pass, in their language.

Their order book prices uncertainty by forcing every price to prove itself against a resolution. Precise runs the same discipline on spend. Connect: the AppsFlyer rows, one DSP's delivery and cost at movable grain, the join keys. Price: every partner, channel, and placement gets a marginal price: what the next dollar there returns in first deposits, with a band, against saturation. Move: the unit of work is a cut‑and‑fund row: source, destination, amount, expected result, confidence, reason. When the math cannot support a move, the row says so: that refusal is the credibility feature. Operate: a person approves; the decision gets written down completely (what was known, what was chosen, what we predicted, what happened, what it taught the system) and graded when outcomes mature. The grading is the product: predictions against actuals, in front of everyone, monthly. Wins and misses alike.

And above the rows: paths. Because the unit is the decision, decisions compose upward. Each ad type (the brand film, the in‑match spot, the performance CTV cut, social video, a creator post, a referral code) travels a different route across surfaces to a first deposit, and each step of the route is a priced decision. So the question MMM answers once a quarter at channel level (what mix) becomes a continuous, per‑ad‑type answer at path level: which surfaces, in which combination, for this creative, this week. Anthony's bar applies here too, and we honor it: path claims carry holdout designs like everything else, and where a path cannot be identified from the data, the row says so instead of guessing.

D3How it fits their plan

They are standing up an in‑house variant. We are the judging layer inside it, not a rival to any part of it.

Timing writes itself: the World Cup flight is closed and gradeable now; the NFL money commits in weeks. A retrospective on the closed flight prices the machine before the big season spends through it.

D4How the engagement runs

Four steps, each with an exit, nothing in market changed until they choose it.

Every step has a walk‑away built in, which is what makes step one easy to grant. The coverage read is valuable to them even if nothing else happens.

D5The questions that make it work

To ask in the room or in the first working session. Each buys a specific piece of the build.

P09 is the meeting inside the meeting. If Anthony writes the acceptance rule, the proof belongs to them before we run it, and yes becomes his conclusion rather than our claim.

BOARDSThe workflow, drawn

Four working surfaces, in the order the engagement builds them. These are the screens we stand up, not mocks of their tools: their feeds underneath, everything in teal is Precise, and every number on these boards is illustrative (matched to the demo book so the materials rhyme).

BOARD 01The joined read: their feeds, priced

Week one's deliverable. AppsFlyer and one DSP joined at movable grain; the teal columns are what joining buys.

BOARD 02The book: cut and fund, with a refusal

The unit of work. Each row is one complete decision; row three is the credibility beat.

BOARD 03Paths: the best route per ad type

The above‑MMM view. Ad types travel different routes across surfaces to a first deposit; each step is a priced decision, so the routes themselves get priced.

BOARD 04The grading: predictions against deposits, holdout revealed

The monthly meeting that is the product. The blind slice makes it checkable rather than asserted.