Reddix is building an implementation of RAFCF, and editing RAFCF at the same time. Both are better with organizations that actually move money through agents. A design partner does three things in order: sees its own exposure, gates one real tier on one real surface, and puts its objections into Draft 0.3 — in public, with a disposition and a reason, like every other comment.
The same three constituencies the editorial board is drawn from, because the questions are the same ones.
Agents that already create, change, cancel or settle a financial obligation — or will this year. You need to know what they are putting in motion before someone else asks you.
You will be asked to form a view on agent-initiated spend against a control framework that does not exist yet. Better to be one of the people who wrote what is testable.
Governed volume, tier distribution and held-action counts are the exposure. If the categories are wrong, they are wrong on your book — and the time to say so is now.
Each one is a decision point. Stopping after the first is a normal outcome, not a failed one.
Shadow mode, read-only. Nothing touches your payment path. At the end you have your own governed volume — the total value your agents put in motion — your own tier distribution, and a count of the actions that would have been held for approval. That report is yours whether or not you go further.
Pick a single routed surface and a single tier, and let Reddix hold what sits at or above it until a second person approves. A hold is a durable state, not a held-open connection, so nothing waits on a socket. You are running the controls before general availability, and you keep the evidence chain either way.
By this point you have run the thing and found the places the standard does not fit your rails. Those go into the comment log with your name on them, and into the disposition list that ships with Draft 0.3. Partners see the draft before it publishes; they do not get a private answer instead of a public one.
The stages are the program, not a sales sequence. Stage one is available on its own — ask for the exposure report and stop there, and nothing else is expected of you.
This is the part that makes it a partnership rather than a discount.
Read-only access to one surface, one named person who can answer questions about how your approvals actually work today, and your objections in writing — published with attribution, answered in public. We are asking for the thing only you have: the rails, surfaces and control environments you work with, and the places Draft 0.2 does not survive contact with them.
Naming you is optional and separate. Comments on the standard are attributed because the standards process requires it. Being publicly identified as a Reddix design partner is a different question, and we will not raise it in the same conversation.
Stated plainly, because the failure mode of a partner program is a reader who assumed more than was offered.
It is not a universal undo. Reddix gates actions before they commit and recovers value where a mechanism exists. Where none exists, the system degrades to prepare-and-instruct, then to a documented claim — and it says so in the record rather than reporting a success.
Partnership does not buy a disposition. Reddix sells an implementation of RAFCF and edits RAFCF, which is a conflict of interest we would rather name than manage quietly. A partner's comment gets the same treatment as a stranger's: published, answered with a reason, and declined where we think it is wrong.
Tell us what your agents touch and which surface you would start on. We will come back with what stage one would look like on it, or with the reason we think you should wait.
No mail client? Write to ask@reddix.ai — tell us your name, organization and role, the surface you would start on, and how approvals work there today.