Bidding above the floor

The problem with a purely fixed price

The original rate card was fixed, published and unchangeable. That solved one real problem and created another.

It solved being priced out. In a search auction the clearing price is the price. When a better-funded competitor enters your keyword, the floor rises under you and eventually you cannot buy at all. That is the arithmetic that pushes one-time-licence desktop apps out of paid search entirely.

But a fixed price cannot clear a market. If three advertisers want creator-tool inventory and there is only enough for two, a fixed price has no way to decide, and it systematically underpays the apps supplying that inventory. Supply then does not grow, because supplying more earns no more. A price that never responds to demand is not a protection, it is a cap on the whole network.

Two layers

Layer	What it does	Who sets it
Floor	The published rate card. Any approved advertiser can always buy at it. It never rises.	GHAN, published in code
Bid	Optional. Buys priority in contested inventory, above the floor.	The market

You do not have to participate in the second layer. Leave the bid unset and you buy at floor price, forever, at the price on the pricing page.

How the auction works

Second price. You set max_bid_cents — the most you are willing to pay for a cleared install in that category. If you win, you pay one cent more than the next-highest competing bid, never your own maximum.

That single property is what removes the objection to auctions. Under second-price clearing, bidding your honest value is the best strategy available to you. There is no gain from watching competitors, no gain from re-bidding, and no spiral, because the thing you are competing against is somebody else's honest value rather than their nerve.

Per category, per slot type. There is no single global auction. Contention is resolved inside a category band for a given lifecycle slot, because a first_success impression in a creator tool is a different asset from an idle impression in a utility.

On cleared installs, not clicks. This is the difference that matters most. A bid is a price for an install that is still in use 48 hours later. If the install does not survive, you pay nothing regardless of what you bid. You cannot lose money to a high bid on traffic that never converts, which is the normal way an auction hurts you.

A worked example

Creator tools, floor EUR 24. Three advertisers want the same contested impression:

Advertiser	Max bid	Outcome
A	EUR 41	Wins. Pays EUR 33.01
B	EUR 33	Loses this impression
C	unset (floor buyer)	Buys from the reserved tranche at EUR 24

A bid 41 and paid 33.01 — one cent above B. A's honest maximum cost them nothing extra. C never entered the auction and still bought at 24.

The serving app earns 70 percent of 33.01, not of 24. That is the market doing its job: the app that supplied a contested placement captured the contention.

The three protections that stay

Bidding is added on top of the guarantees, not instead of them.

1. The floor never rises. It is a published number in packages/shared/ratecard.ts, changed only as a versioned change announced in the changelog before it takes effect. No amount of bidding moves it. This is the whole of what "never enter an auction" was protecting, and it is intact.

2. Thirty percent is reserved. In every category, 30 percent of inventory is sold only at floor price and is never auctioned. It is allocated round-robin among approved floor-price advertisers. A bidder with unlimited money wins priority in the contested 70 percent and cannot touch the reserved 30.

3. Pair caps still bind. Net flow between any two apps stays capped at EUR 500 per month until both have behaviour scores, and clearing above EUR 500 per day per pair still moves to day-7 engagement. A bid cannot buy its way out of a fraud control.

Why this is not the thing that broke search advertising

The objection to auctions in the original manifesto was real but imprecisely aimed. Stated properly, there are four things wrong with the search auction for a small desktop app, and bidding is only implicated in one of them:

The actual problem	Present on GHAN?
The clearing price is the only price — no floor to fall back to	No. The floor is guaranteed and permanent
You bid per click, and pay whether or not anything happens	No. You bid per cleared install
First-price dynamics reward nerve over value	No. Second price
A funded competitor can take 100% of your category	No. 30% reserved at floor

Price discovery was never the problem. Price discovery with no floor, on the wrong unit, with the wrong clearing rule, was.

Setting a bid

create_campaign {
  app_id: "app_...",
  categories: ["creator_tools"],
  daily_budget_cents: 20000,
  max_bid_cents: 4100          // optional; omit to buy at floor
}

max_bid_cents must be at or above the category floor. Omit it and the campaign is a floor-price campaign. Change it at any time; it takes effect on the next placement fetch.

get_stats reports avg_clearing_price_cents alongside the floor, so you can always see what contention actually cost you — and see when it cost you nothing, which for most categories most of the time is the expected answer.

For the apps serving placements

You earn 70 percent of the price that cleared, not 70 percent of the floor — or 84 percent of it in value if you take credits.

This is the half of the argument that makes the market worth having. Under a purely fixed price, an app with genuinely valuable inventory — high retention, high-intent moments, a category people want — earned exactly the same as an app with poor inventory. There was no mechanism by which being good at this paid more, and therefore no mechanism by which the network grew the supply it most needed.