Introduction
I share one invite code for Muse, a personal AI agent. The code is YAN3V3. If someone joins Muse and enters the code in Settings within 48 hours, both sides receive 1 billion Muse tokens. The code has 30 uses in total.
As of November 19, 2026, 3 of those uses are confirmed redeemed and 27 remain. I run the distribution like a small operation: the code sits in 9 distribution pools (rotation, queue, and fill-up types), and I also publish articles — like this one — that carry the code at the end. An AI agent helps me maintain the listings.
This article is about the part nobody warns you about: I can count the redemptions, but I mostly cannot tell where they came from.
The one redemption I can attribute
The first redemption came through a pool called muse300. That pool emails the code's owner when a code is handed out to a new user. The notification arrived, and the owner (me) later confirmed on the invite screen that the token balance had actually increased — handed out, then redeemed. Two data points, one chain, fully attributed.
That is the only fully attributed redemption I have. It is also the reason I know the difference between the two events that every other channel blurs together: a code being taken and a code being used.
What the pools cannot tell me
Most distribution pools expose no per-code redemption data at all. Three concrete examples:
- One major pool states on its own page, in plain text, that after a code is taken, the pool cannot tell whether it was actually used — the number next to a code stays whatever the owner originally posted. So its listing tells me my code is present, and nothing else.
- Another board tracks how many of a code's 30 uses have been "unlocked" by its queue. At my last weekly check, my code still showed 0 of 30 unlocked there — while my invite screen already said 3 redemptions. The board's counter and reality are measuring different things.
- The remaining pools show pool-level totals only (how many codes are listed, how many draws happened). My code is in there somewhere. That is the entire signal.
So for 2 of my 3 redemptions, the pool channel is a black box: possible, unprovable, and impossible to rule out.
What the articles can and cannot tell me
Articles at least report views. Measured on the author dashboards on November 19, 2026:
- My first Qiita article (published November 17) had 432 views cumulative.
- Three Qiita articles published on the morning of November 19 totaled 199 views within a few hours.
- My dev.to articles stayed under 25 views each.
Here is the tempting story: the 2 unattributed redemptions appeared on the same morning that five of my articles went live across Qiita, Zenn, and dev.to. Timing fits the articles.
Here is why I don't believe my own tempting story: view counts include crawlers and my own verification checks; Qiita's 199 views and dev.to's under-25 are not the same audience; and the pools were running the whole time, silently. "Same morning" is a correlation with at least two live alternative explanations. The honest record in my ledger says: 2 redemptions, source unknown.
Running an operation on partial measurements
Since I cannot measure the thing I actually care about (which channel redeems), I measure everything around it and keep the operation honest:
- Listings are re-checked weekly. For each of the 9 pools I record what is observable: pool size, my code's presence, any counter the site exposes, and any trace of reaction. A listing that disappears gets re-posted once under its original conditions (free, code only).
- One pool decays with time, so it is automated. Its ranking weight depends on how recently the listing was updated (highest within 6 hours). A scheduled job re-saves the listing every 6 hours to hold that weight. The job only reports failures; silence means it ran.
- Declared numbers are never padded. The listing asks for remaining uses, and the job declares the true number from my invite screen — 27 now, not a rounder 29 or 30. A wrong declared count would be a small lie that compounds every 6 hours.
- Each channel keeps a "traces" column in the ledger. muse300 has a trace (1 handout email → 1 confirmed redemption). Every other channel's column says the same thing: no observable trace yet. Maintenance effort goes to channels with traces first.
What I decided anyway
I keep both channels running. The pools cost almost nothing to maintain once automated, and one of them produced the only proven redemption. The articles are measurable in views if not in redemptions, they compound (432 cumulative views on the first one and still counting), and they document the project for readers who may never use a code at all.
What I stopped doing is pretending the dashboard is the operation. The ground truth for this whole setup is a single number on my invite screen — 3 used, 27 left — and everything else is a proxy with a label on it saying how much I trust it. Running the operation well means keeping those labels accurate, not making the proxies look better than they are.
Invite code
Muse has an invite code system. My code is YAN3V3. If you join Muse and enter the code in Settings within 48 hours, we both receive 1 billion Muse tokens.
- Join: https://muse.ai/join
- App: Settings → Redeem token
- Web: Settings → General → Usage → Use invite code
- Code page: https://muse.ai/s/muse-invite-code-yan3v3-english-xuf6xz9w2fxkgi/DVsk9awGJU_n6SJYZDXaNZxCPGoHiCRuQxqE3pAX2kY
This is a personal invite code share, not an official Muse announcement.

