The public record
Agent Traffic Index
Which agents are reading the commerce web, at what rate, and what they ask for. Aggregated from gateway traffic, published to a floor.
The k-anonymity floor is 25 independent stores. Nothing is published until the floor is met.
Counting independent stores…
This page exists before the data does deliberately. The floor is easier to keep when it was published before anyone knew whether it would be inconvenient, and it means the commitment can be checked against the date it appeared rather than taken on trust.
Rules
What we will publish, and what we will not
Every one of these is decided now, while nothing is at stake in deciding it.
Shares and percentiles only
When this publishes it will carry shares of a mix and distribution percentiles. There is no absolute network total, because publishing a total alongside all of its parts is strictly worse than publishing either one.
Monthly, and that is a privacy parameter
Cadence sets the resolution available to a differencing attack: the finer the period, the smaller the change that can be attributed to one store joining or leaving. It is not a product knob and it will not be tightened because a chart would look livelier.
Pre-declared cuts, no slicer
A fixed list of cuts, published in advance. No query builder, no interactive filter, no "explore the data" surface — an attacker with a slicer can compose an arbitrary number of queries against the same population, and no per-query floor survives that.
k counts independent stores
25 distinct, non-affiliated, non-first-party stores per cell. trymumm.com posts events into the same log; counting our own store toward an independence floor would make the floor a fiction, so first-party stores are excluded from the numerator, the denominator, and k.
One store can never be the cell
No single store may contribute more than 20% of a published cell. 25 stores where one of them is 95% of the traffic is that one store with extra steps, and k alone does not catch it.
Opting out removes you from k, too
A per-store opt-out excludes the store from the numerator, the denominator, and the k count — not just from anything that names it. A store that opted out cannot be part of the reason another cell cleared the floor.
The wall
Whose data this is, and whose it is not
The Agent Traffic Index aggregates merchant-side gateway traffic: which requester asked, what it asked for, what it received. The subject of every number is a site, an agent, or a request — never a person.
Data whose subject is a shopper does not appear here, at any aggregation level, in any cut, at any price. "A buying agent could not find your shipping cost in 62% of attempts" describes a website and may be published. "Buyers with a $400 budget want this product" describes people and does not cross, aggregated or not.
There is also no price data of any kind in this index, and there will not be. We benchmark agent readability, traffic, and conversion; never price, and never discount depth.
Methodology · The readability index · Scanner and crawler disclosure