What it saves, what it costs

Every article in this library that says "the seed is floored" is making an accounting claim, and accounting claims should be written out as line items. This one does that.

Two warnings before the numbers. Every figure below is from the paper's own simulations, and every simulated return is a property of the stated crowd model, not a projection. And the paper is explicit on one point in particular: the seeding return is payment for cold-start risk-bearing, not a yield, and quoting it as an APY misquotes the paper.

What survives model choice is the floor rows and the identity-based claims. Those are mechanism properties. Everything else is a measurement of one simulated crowd.

The saving that is actually structural

Per-market capital moves from spent to parked

This is the whole economic argument, and it is one line:

A subsidy budget is consumed. A floored seed is a revolving float.

Compare the two balance sheets for a venue that wants ten thousand markets to exist.

Subsidy model (LMSR, CPMM, market-maker incentives): you allocate X per market, expect to lose some fraction of it to informed flow, and that fraction is gone. Ten thousand markets means ten thousand times the expected loss. The number scales with your catalogue and never comes back.

Float model (vested seed): you allocate S per leg, park it for the market's life, recover it in every branch, and redeploy. The number scales with your concurrent market count and your average market duration, not with your lifetime market count.

The arithmetic is small enough to be uncomfortable:

Against that, the real costs of the seed are three, and none of them is an expected loss to informed flow: fees, carry on locked capital, and resolution risk.

Small seeds are correct, not just cheap

The instinct is to seed heavily so markets look liquid. §9 measures what that does:

The seeder sits at its floor either way. So over-seeding crowds out exactly the participants you were trying to attract, and the policy that costs least is the policy the mechanism prefers. That is an unusual alignment and worth taking advantage of.

The seed is also your resolution bond

Not a separate line item. Vintage 0 doubles as the first rung of the escalation path in §12, which means the capital is doing two jobs. It also means it is slashable on adverse escalation rulings, which is what converts the direction premium from a bribe into a bond at risk.

Measured seeding outcomes

Creator seed, matched vintage 0, 20 seeds × 2,000 markets:

Read the floor rows, not the mean. The mean is crowd-model-dependent; the floor is the mechanism. And P6's unconditional form is exact: a market with zero subsequent flow returns the seed to the cent, minus fees — so in a market that attracts nothing at all, the seed returns −2% at a 2% fee.

The saving that depends on your current lock

Every minute your pools are closed is handle you do not take, at the peak-interest minute. Deleting the lock recovers it.

The study's number understates this badly for a real venue. Its lock arm closes only the final 5% of a market's life and loses 5.0% of total volume relative to unlocked. Real venues close for minutes or hours before an event, in a period where arrival rates are at their highest, so your recovery is your own number and you should compute it from your own closed-window traffic rather than from 5%.

Set against it: total volume in the study is $2,254 a market under the mechanism against classic's $2,845, about 21% lower, because rational late entry stops.

If you charge on handle, model both directions. Volume falls; the window you recover is real; whether they net positive depends on how long you currently close for and how much of your traffic sits inside that window.

The costs, priced

Carry, which is a real loss on slow markets

Capital is locked from entry to resolution, and this applies to the seed as well as to entrants.

At 4% and six months, the 2% hurdle is the same order as the entire observed late-tercile win-conditional yield (median 2.7%) and exceeds the median final-decile winner's post-fee return. On the seed side, carry turns the nominal floor into a real loss.

The paper's conclusion is unhedged: λ = 1 is a short-horizon mechanism. If your markets run for months, the floor is nominal only, and you should price accordingly.

The late price, which you are selling if you display it

0.174 final-phase Brier against the deployed lock baseline's 0.150. If your product surfaces a probability that anyone consumes as a probability — syndicated, cited, used by another product — you are degrading a thing you distribute. See the disqualification list.

The secondary layer, which is a build and a book

For long-enough markets this is transferable positions plus a cash-out affordance. For short recurring rounds it is a venue-side RFQ vault at a published spread, and that vault is a book you are running:

• It is a designated adverse-selection sink. It will be picked off precisely when information is most valuable.
• Its quote must not be a function of the pool ratio, or the manipulation the mechanism disarms returns through the vault at near-zero cost for a book-dominant party.
• So it must price from your own model over wash-excluded flow and off-market inputs.

The paper lists the vault's adverse-selection pricing as an open problem and calls it "the acknowledged cost of this step." Budget engineering and risk staff, not just a feature.

Resolution risk, now with sharper edges

The mechanism does not create resolution risk, but it changes its shape in ways that cost money to defend against (full article):

• Delay farming — a resolver holding a winning position profits from delay. Fixed by freezing the accumulator at a timestamp set at creation.
• Stall-into-void — a resolver on the losing side of a large vested payout profits from never resolving, and because accrued claims are public they know exactly what it saves them. Fixed by seed forfeiture plus permissionless fallback plus a bonded void declaration.
• The direction premium — the seed always pays more in the branch with more opposing volume, and the gap is publicly computable. Mitigated by displaying it live and making vintage 0 slashable.

None of these is free to implement, and all three are required before permissionless creation.

Your volume metric stops meaning anything

Not a cost in dollars, a cost in decision quality. A participant who dominates both books of a market can print wash volume at near-zero mechanism cost — the donation is (1−f)-scaled in their own share of the opposing book — so in exactly the whale-seeded regime §9 flags as the most likely drift, headline volume is manufacturable.

Consequence: track records must exclude self-vested flow (identifiable from the public log) and weight bonded, aged, distinct counterparties. If your growth reporting or your creator ranking runs on volume, both need rebuilding.

The one-page summary

The shape of the trade: you stop spending per-market capital and start spending on a layer above the pool. Whether that is a good deal depends entirely on how many markets you want to run — because the first cost scales with your catalogue and the second one does not.

Next: Choosing λ, market class by market class.