Why some subnets print and others do not

Every earlier lesson in this course pointed at one piece of the answer — the chain's rules, a contest's own code, the pull of money behind a judge. This last lesson puts all of it against a live measurement of the network at once, and asks the question the whole course was built to answer: when a contest earns well, what does that actually tell you?

Read against this course's own tracked cohort of subnets at block 8,871,732, earnings are not spread evenly at all: the three highest-earning contests alone took in 70.2% @ 8871732 of everything the network paid out, and the top six took in 81.6% @ 8871732. A number that lopsided invites a story — that the leaders must simply be doing better work — and the rest of this lesson exists to test whether that's actually the story the measurements support.

That story is only one of three separate explanations for a number like that, and they don't all say the same thing about a subnet's quality. The first is mechanism: a well-run contest attracts real work, validators score it well, and the reward follows. The second is stake: judgment on this network is not one-participant-one-vote, and whoever backs a validator with the most stake decides more of what counts as good, whether or not the work actually is. The third is price: a subnet's whole slice of a day's emission is set by how much people want to hold its own token that day — a statement about market appetite, not a vote on the work. [source] [source] [source]

Mechanism is the one of the three this course's own measurements were never going to see directly. The chain can watch whether validators agree with each other and pay out accordingly, but it has no way to check whether the scoring code producing that agreement is measuring anything real — a contest built around a bad test can reach the same smooth agreement, and get paid exactly as cleanly, as one built around a good one. [source] Agreement is cheap to fake, too: a validator that skips judging entirely and copies everyone else's already-published numbers has historically earned better standing than the honest judges it copied, precisely because the network's own rule rewards agreement. From outside, a subnet earning well because its work is good and a subnet earning well because its judges are copying each other can look identical. [source]

Stake, unlike mechanism, leaves a trace this snapshot can actually show. A validator's influence scales with the stake behind it, and the biggest backers tend to spread themselves across many contests at once rather than commit to one — which is exactly how a large position keeps growing. [source] [source] One measurable piece of that is a subnet's root proportion: how much of its own judging power comes from money staked into the root network at large, rather than staked specifically into that subnet's own token. In this snapshot, every subnet whose emission rank and holder rank disagree sharply enough for this cohort to flag it for rank disagreement — InfiniteQuant among them, at roughly seventeen percent on this measure (17.3% @ 8871732, flagged true @ 8871732) — carries a higher root proportion than any of this snapshot's highest earners, whose own figures sit in a tight band around fifteen percent (14.8% @ 8871732). One contest reaches roughly sixty-four percent on this same measure (63.9% @ 8871732) — well over four times the leaders' own share. None of this says the flagged contests are judged badly. It says who is doing the judging looks different there, and that difference tracks the very rank mismatch the flag exists to catch.

Price leaves the clearest trace of the three, because it's the one variable this network's own formula for a subnet's slice of emission is actually built from. lium.io, this snapshot's own top earner, took in 31.8% @ 8871732 of the network's payout on the back of an alpha price up roughly twenty-two percent over the trailing month (22.18304707820676 @ 8871732). iota, in second place with 26.2% @ 8871732, got there despite its own price actually falling that same month (-1.425902984618844 @ 8871732) and the week before it (-1.021158438567247 @ 8871732) — a reminder that a contest's share tracks a slow-moving average of price, not the latest tick, so a short slide doesn't immediately show up as a smaller slice. [source]

Apex is the same subnet the previous lesson opened up, and its own numbers in this snapshot are worth sitting with. Its slice of the day's emission here is a measured 0.0% @ 8871732 — literally this cohort's lowest tier, C @ 8871732 — even though it still carries real judges and just over three thousand holders (3358 @ 8871732), and a root proportion of roughly fourteen and a half percent (14.5% @ 8871732) that sits in almost the same narrow band as this snapshot's own top earner, lium.io, at roughly fifteen percent (14.6% @ 8871732). That closeness rules out the easiest explanation: whatever is holding Apex at zero here, it is not an unusually root-heavy validator mix the way it is for this cohort's rank disagreement contests above. Apex's own alpha price was also down over the trailing month (-8.90614614103279 @ 8871732) — consistent with a price explanation, though consistent with is not the same as proven by.

Unverified By the network's own design there's no floor at exactly zero for emission share by default — a subnet with a low but positive price should still take home some small nonzero slice. [source] A subnet switched off from receiving emission entirely is the one case where a genuine zero, not just a small share, is the designed outcome. [source] Whether that's what's happening to Apex, or whether this is a rounding artifact of the data provider's own reporting, is not something this snapshot lets me tell apart — so take this paragraph as this author's own flagged uncertainty, not a documented fact.

The one thing this course does independently know about Apex's own mechanism didn't come from this snapshot at all — it came from reading the code. Its validator doesn't score anything itself; it asks a separate service outside this course's own pinned copy of the repository for the numbers, and when that service can't be reached, its own fallback is to submit an average of whatever every other judge already published, weighted by how much money backs each one — the same shortcut an earlier lesson described as a way to cheat, built in here as a safety net instead. [source] [source] None of that proves Apex's own zero here either. A subnet can run exactly this kind of fallback and still earn well, if its price holds up; a contest can run an honest, carefully-designed mechanism and still earn nothing, if nobody is buying its token that month. This snapshot's own measurements — concentration, root proportion, price movement, rank disagreement — cannot tell mechanism, stake, and price apart for Apex, and they cannot tell them apart for any other single contest in this cohort either, top earners included. A contest at the top of this ranking might be there because the work is genuinely good, because a few large stakers back it, or because its token is simply in fashion this month, and today's numbers alone do not say which.

What these measurements do support is narrower, and still useful: real statements about how concentrated emission is, how many holders and judges a subnet carries, how its token's price has actually moved, and whether its rank by earnings and its rank by popularity agree or pull apart. What they do not support is a verdict on whether any one contest's mechanism is good — that question, this course has shown again and again, is exactly the one the chain itself was never built to answer on its own. Across the whole network this snapshot measures, 3554.8694508093054 @ 8871732 TAO gets minted and handed out this way, every single day, to whichever mix of mechanism, stake, and price happens to be moving that day's price averages. Reading the code tells you what a contest is supposed to reward. Reading a live snapshot tells you what it is actually paying for right now. Neither one, alone, tells you why.