What we’ve modeled.
Questions we have been asked, the data each took, the model, and what it showed. Written as method: no client, no figure that identifies one.
The work, on the record
-
01
Analysis
Written up
Payment patterns on a collectibles platform
What it showed The accounts to hold before the labels printed.
Tell us the question →How it was done
- The question
- Which accounts to stop before items ship.
- The data it took
- The platform’s shipment and payment exports.
- The model
- Each account’s payment history read against the platform’s own rules. One move the first version missed was added after a hand read: a payment card blocked as suspected fraud, then smaller amounts, then spent to the card’s limit.
- The forward test
- Most accounts the desk put at verify or above were later banned by the platform; almost none of those it read as fine.
- What we learned
- What an account holds can exceed what it deposited without that being a fraud signal; and payment exports need a stable payment id, or merging two of them double-counts.
-
02
Scenario
Written up
Carrier spend against billed
What it showed Where charged and billed diverged, by carrier and by service.
Tell us the question →How it was done
- The question
- Does what the carriers charged match what the customer was billed, shipment by shipment?
- The data it took
- Carrier invoices and the operator’s billing records.
- The model
- Each shipment’s carrier cost matched to its billed postage, with weight and service level; a reconciliation flags unbilled and double-counted lines.
-
03
Scenario
Written up
A collection by disposition
What it showed The order changes the result more than any single item does.
Tell us the question →How it was done
- The question
- What a collection returns under different disposition orders.
- The data it took
- A custody baseline of graded collectibles, each with its storage location.
- The model
- Sell-through, buy-back and hold side by side, every assumption dated.
-
04
Read
Read
Platform economics, from the chain
What it showed 353,332 packs paid in the week; most sold straight back on both (97.7 percent on one, 90.0 percent of card value on the other); 1,613 cards left one vault, in 344 shipments.
The Platform Record →How it was done
- The question
- On a pack platform, how many packs are paid for, how many go straight back to the platform, and how many cards actually leave the vault?
- The data it took
- Two platforms’ public programs, contracts and treasury wallets on chain, for one complete week, 11 to 17 October 2026, as the Brich House Platform Record’s October 2026 edition reads them.
- The model
- Packs paid counted per platform; packs sold straight back as a share of packs paid; cards burned for physical shipment counted as what left the vault. Only wallet-paid packs are valued on chain, so card checkouts are left out of every ratio.
- What the record held
- The other platform’s chain cannot show redemptions, so what left its vault is an upper bound, not a count. Neither platform publishes an independent count of the cards behind its tokens. No wallet is shown or stored.
-
05
Analysis
Signed
Who pays when a pull is charged back
What it showed The network defaults pay the affiliate between two weeks and ninety days after the sale; the cardholder has 120 days, longer where delivery comes later, so the same money can leave twice.
Read the note →How it was done
- The question
- When a pull bought by card is disputed, has the affiliate who sent the buyer already been paid?
- The data it took
- The published payout calendars of three affiliate networks, the dispute windows the processors publish for Visa and Mastercard, and the two networks’ monitoring programs, each read on a dated public page.
- The model
- Two clocks side by side: the days from a sale to the affiliate’s payout, and the days from a sale to the last day the cardholder can dispute it. Then the monitoring levels a platform’s monthly disputes are counted against.
- What the record held
- Every calendar, window and level is the publisher’s own, linked with the date it was read. The note names no platform and no merchant, and reads the rules as mechanics.
-
06
Read
Read
What’s inside, devices read from FCC records
What it showed 6 covered, 8 clean as far as shown, 6 unknown; the host maker alone settled five of the six covered reads.
What’s inside →How it was done
- The question
- Can a device’s public FCC record say who made what is inside it, and does any of it sit on a US list?
- The data it took
- Twenty devices’ public FCC records: every application under each FCC ID, the exhibit lists, the labels, the internal photos, the test reports and the manuals. Cameras, Wi-Fi access points and drones.
- The model
- The host maker from the grantee code; the modules the record declares; chip markings read from the internal photos in three passes; then the host and every component maker checked against the FCC Covered List, Section 889 and the 1260H list.
- What the record held
- Where the record is silent, the page says so: shields, unlabelled parts and confidential exhibits can hide a component, and an unknown is never written as clean.
-
07
Read
Last run
Every change to the six US lists
What it showed The newest change, 13 October 2026: the FCC rule on logic-bearing components from Covered List producers took effect, with marketplaces to show the FCC ID from 2027.
The tracker →How it was done
- The question
- What changed on the US ban and sanctions lists, on what date, and what does it mean for someone who owns, buys or ships the equipment?
- The data it took
- Each publisher’s own files and notices: the FCC Covered List, Section 889, FASCSA orders as SAM.gov shows them, the 1260H list, the Commerce Entity List and the OFAC SDN List. Checked daily.
- The model
- Every change logged with its date, the publisher’s notice and one line on what it means for owners; scheduled changes listed ahead of their dates; a list that cannot be read on the day is marked not verified.
- What the record held
- Individuals on the SDN List are counted, never named. On the last run the FASCSA orders could not be read, and the tracker says so rather than showing an old copy.
-
08
Analysis
Revised
Cargo corridors
What it showed Losses more than doubled in the second quarter of 2026 while recorded events fell: fewer thefts, bigger ones, and fictitious pickups held flat while other methods fell.
The corridor desk →How it was done
- The question
- On the New Jersey corridor, what is being taken, by what method, and how has the pattern moved quarter by quarter?
- The data it took
- Six quarters of Verisk CargoNet analyses, Q1 2025 to Q2 2026; FBI and IC3 releases; state police and court filings; the trade press of record. 22 dated entries on the watch list.
- The model
- Losses and recorded events read side by side, quarter by quarter, with derived quarters marked as derived; the method, New Jersey and the consumer side read as separate tables; every watch entry dated, sourced and tagged.
- What the record held
- One corridor brief published so far: No. 1, The New Jersey corridor. Findings stay at pattern level: a method, a corridor, a commodity; never a person, never a client’s lane.
-
09
Analysis
Signed
A platform report, the method
What it showed A report is one dated page per platform: what each of the five shows, where the terms are silent, and which questions stay open.
The free report →How it was done
- The question
- Before a platform holds cards or a customer’s money, what does its own public record say, and what does it leave unsaid?
- The data it took
- One platform’s public pages: its affiliate terms, its refund and dispute language, its age and identity check, the processor named in the terms or at checkout, and an independent count of what is in its vault.
- The model
- Each of the five read against the platform’s own pages and dated; read with the chain record where the platform keeps one; every question the record cannot answer written down and held open until it is answered.
- What the record held
- No platform is named here. A question stays on the page as open until the record answers it; an open question is never written as an answer.
More of our work is private to the clients it was done for. Tell us the question: Pre-flight →
Before you send data. It stays inside your engagement: never in anything public, never in another client’s work. What to send, how, and what comes back: How to send your data →
Things and transactions, never people.