Every arc, every page, one place. Each arc gets a file (the working instrument), a research annex (the raw depth), and the shared session pages. The files re‑room after every meeting: each file is rebuilt around what was actually said, so the version you open is the current state of the account. Status below is live as of the dateline.
| Arc | Where it stands | The pages |
|---|---|---|
| Mediaoceanthe substrate under the estate · three workstreams, one machine | Thursday Aug 6 · 10:30 · their office 120 Broadway with Markman, Bill, Kane, Carafello. Run sheet, scenario rooms, and questions ready; strawman gets torn apart at the internal session Wednesday. | the file · the walkthrough · research annex |
| Polymarketprice the spend like you price the world · through Publicis | intro call · Wed Aug 5 · 5pm Straight after the internal session. Peter embedded and championing; the wedge is allocation across nine‑figure spend; first proof is the World Cup retrospective with a blind holdout. | the dossier · research annex |
| iHeartwhat the listen was worth · recommendations, not writes | follow‑up · Fri Aug 7 · 1:30 POC granted out loud on Jul 30 (Backer and Pilcer, Unified); Friday is where the walkthrough, the S3 extract, and the reverse demo get their dates. | the file · research annex · the deck |
| Page | What it is for | Where |
|---|---|---|
| Master deck 01 | The one surface that runs the week: the Ford v Ferrari opener, the deployment talk, the run of show, each arc's questions with its sealed calls, and the library. A call is a decision made on the record: what was known, what was chosen, what was predicted at what confidence, and, once scored, what happened and what that taught; sealed means committed before the answer comes back. The room makes those calls together; transcripts feed back; the deck becomes the run book. | /sealed |
| The session screen | Load on the shared screen Wednesday 2–5 at Spring Place: tabs per arc, the discussion questions, keyboard 1–5. The room talks; the recording runs from minute zero; the Polymarket call follows at 5. | /screen |
| The session kit | The printable version: the questions per page, what Claude pulls from the transcript, and the deck loop written down. | /session |
The Jul 30 transcript is the best call we have on tape. What went well was not luck; it was seven moves, and they are now the call template.
| Move | What happened, on the record | The rule it becomes |
|---|---|---|
| 1 | Matt walked the live dashboard, not slides: governance module, constraint packs, per‑segment recommendations, real events logging | Demo the working thing. Credibility compounds from product, never from decks |
| 2 | We let them talk and collected their words verbatim: the backlog ("never ending"), the headless requirement, Tetris and the loudest person | Harvest vocabulary and pain in their words; the follow‑up quotes them to themselves |
| 3 | When Pilcer said no new interface, Adam bent live: "we could also issue it as a browser plugin… a go‑over on top of whatever you need" (15:38) | Bend to their workflow in the room, not in the proposal. Objection to accommodation in one breath |
| 4 | The proof mechanics were theirs: Pilcer proposed the S3 extract and the reverse demo; Josh scoped "recommendations without rights to any campaign changes" | Let them design the yes. A proof someone designs themselves never gets renegotiated |
| 5 | The ask sat beneath every approval threshold: closed orders, read‑only, no credentials. Josh's own verdict: "seems like an easy test" | Size the first ask under the level where anyone has to ask permission |
| 6 | The deck went out the same afternoon, built from the call, quoting the call | The same‑day artifact. Bill named this pattern too: the deck seven minutes after the meeting was the thing he repeated to his own room |
| 7 | The recording ran, so all of the above is quotable, checkable, and re‑roomable | File the transcript. Every unrecorded call is a gap in the estate |
Replication is mechanical now: the deck loop runs these seven as its front half, and the question tables (below) carry the harvest into the next conversation.
The thesis under all three arcs, said once, plainly.
The opportunity for Precise concentrates in a specific class of company: the very large, old‑world‑adjacent media enterprises. iHeart and Mediaocean are the archetypes: decades of workflow, enormous flow ($3.9B of revenue on one, $200B of other companies' money through the other), legacy estates that are the business rather than a liability, and cultures that genuinely buy from vendors: iHeart most of all, an older media company that has always run on partners. This is a different sandbox from the one the new‑media crowd fights over: we are not chasing Trade Desk's market, and we are not selling to companies that build everything themselves. It is the sandbox where MadHive made its money: enterprise, broadcast‑adjacent, relationship‑driven, and patient once you are in.
The goal, in one sentence: enterprise deployment of the Boombox capability substrate, activating legacy software as governed capabilities across a network downstream. Every arc is a variation of that sentence. At Mediaocean the estate is Prisma and the toolkits; at iHeart it is the traffic desks, Triton, and the rules pipeline; the network downstream is their customers, who inherit priced decisions without buying anything new. Polymarket is the deliberate exception that proves the machinery on fast data with a modern stack, so the enterprise class can see it working at speed.
The test for every next target reads straight off the class: very large, old‑world adjacent, a legacy estate that runs the business, vendor‑receptive culture, and a quarter‑end somewhere in the building where decisions are made by hand. Companies that pass all five get an arc on this desk.
Every arc runs on the same discovery pattern, and it is the house pattern: questions that turn a customer call into scoped work, with the four interrupts always running underneath: is there a real problem, can the work be done, are they allowed to move, and what is the next artifact. Each file carries its own numbered table (Q01–Q16 for Mediaocean, P01–P13 for Polymarket, I01–I12 for iHeart), and every question is an instrument: its answer moves a wall forward in the forecast (a wall is a constraint the plan has to clear), unblocks a next step that a flagged risk was holding shut, or converts a verbal yes into a date and an owner. This is also how the network approach gets built without ever being pitched: each answered question is granted access, granted data, or a named relationship, and the graph assembles from asked‑and‑answered rather than from slides. The files re‑room on the answers; that is the whole loop.
Each arc runs the same method: deep research (Exa plus the transcripts) becomes a working file; the file carries assumptions, questions with owners, boards that show the fit, and rooms, the step‑by‑step walkthroughs from where the account stands to the close; a five‑lens adversarial pass checks it before anyone external sees derived material; everything publishes here in the same tan house style, with a theme toggle and no search indexing. After every meeting the transcript comes back, the file re‑rooms the same day, and requested decks get built from the manifest. The files are working when they change.