Precise × Polymarket · Working document

First questions, and the path from first read to earned authority

August 2026 · most of this is a ten-minute pass for whoever owns the warehouse and the platform accounts · the methodology brief · the log-fidelity ladder · the session deck

Sixteen questions. None of them require pulling data yet; they tell us together what the first read can see, and they surface the two or three places where a vendor conversation is needed before anything else.

Why this is worth a week of anyone's timeOne example of what a first read produces, from facts you already hold: national campaigns deliver into every state, and you can only take a deposit in some of them. How much of the buy landed where a deposit is impossible? Nobody's dashboard prints that number today, the read produces it in the first pass from questions 7, 10, and 11 below, and every dollar it finds is found money, reallocatable the same week.

AThe outcome side

Deposits are the ground truth everything gets measured against, so we start by pinning down exactly what a deposit record looks like.

  1. Where does the authoritative deposit record live?Internal warehouse, AppsFlyer event, or both, and which one wins when they disagree.
  2. What does one deposit row carry?Timestamp, amount, first-versus-repeat flag, and whatever user or device key it carries.
  3. How do web signups and app signups meet?Whether a person who converts on mobile web and later installs is one record or two.
  4. How far back is the history clean?The first read wants one trailing window; knowing the depth tells us how much calibration history we can build immediately.
  5. Which outcomes beyond first deposit matter enough to weight?Deposit amount, repeat funding, retention. We can weight several; you declare the weights.

BThe media side

The reads are only as deep as the logs. This block finds out which partners already give you depth and which need to be asked.

  1. The active platform list.Paid social, mobile DSPs, desktop DSPs, search, linear, sponsorships, creator work, whatever is live or recently live.
  2. Which platforms already provide log-level or raw exports?Some make this an easy grant. Some need it demanded at the next negotiation, and the ready-to-paste ask is at /logs.
  3. AppsFlyer raw export access.Whether raw-data delivery is already switched on, and where it lands.
  4. The segment and audience lists in use, per platform.Names and sources are enough; this is what the decomposition prices.
  5. Rough spend by platform for a recent month.Ranges are fine. This sizes where the first read can matter most.

CThe constraints

Recommendations that ignore your real constraints are noise, so the constraints go into the model, not around it.

  1. Where can you operate?The live-state map, so geography is a first-class variable rather than a surprise.
  2. What is fenced off?Legal, brand, or category constraints on where and how you can buy.
  3. Who approves a budget move today, and how fast?The loop runs at the speed of this answer; knowing it sets the cadence of the reads.

DThe working setup

  1. Who is the data access owner on your side?One name. We send the NDA and DPA the same day.
  2. Where do you want the reads delivered?Dashboard, a morning email, Slack, or straight into your own tooling by API.
  3. A window and a slice.One recent trailing window for the first read, and a slice you are comfortable holding out so you can grade us from day one.

The suggested path

01 · first readDecompose what already ranLogs plus the deposit table for one trailing window. Contribution against cost, below the level the platforms report at. You check it against what you already believe before anyone acts on anything.
02 · recommend modePredictions on the recordReads on a regular cadence. Every recommended move carries its predicted result and band, written down first. Your team makes the moves. Every prediction gets graded against the holdout slice, in writing.
03 · earned authorityAutomate what the record supportsWhere the grades hold, the moves you choose become API calls instead of emails, and the same loop widens toward which market goes in front of which person. The pace is yours; the track record is the permission.
The goal you named on the call, most of spend allocated without a human in the loop, is a track-record problem before it is a technology problem. Step 03 is that goal, reached at whatever pace the grades justify.

Precise · August 2026 · answers, partial answers, and "we need to check" are all useful; send them to Spencer, and methodology questions to Matt, Adam, or Jordan.