Pulled the price field off every endpoint in the registry this morning. 244 endpoints, nine clusters, one CSV. The shape is more interesting than I expected.

The histogram

Bucket counts, $0.005 bins from zero up to $0.50:

  • $0.005–$0.01: 71 endpoints
  • $0.01–$0.02: 58 endpoints
  • $0.02–$0.05: 41 endpoints
  • $0.05–$0.10: 29 endpoints
  • $0.10–$0.20: 18 endpoints
  • $0.20–$0.30: 5 endpoints
  • $0.30 exactly: 17 endpoints
  • above $0.30: 5 endpoints

Median is $0.015. Mode is $0.01 (54 endpoints sit on that exact price). Arithmetic mean is $0.043, pulled up entirely by the $0.30 spike. The mean lies about this portfolio. Always read the median when pricing is this skewed.

Why so much weight under $0.05

Two forces drive it, mostly.

The first is agent budget math. A planning loop that hits 20 tools per task can't burn $0.30 a call. At $0.01, that loop costs 20 cents. At $0.30, it costs six dollars. The agent stops calling you. So endpoints designed for routine retrieval (address parsers, geocoders, formatting helpers, light lookups) all land in the penny bucket because that's what gets repeat traffic.

The second is the Bazaar ranking signal. It rewards call volume. A $0.01 endpoint catching 5,000 calls a day ranks higher than a $0.30 endpoint catching 50. The visibility loop pushes builders toward low prices, which pulls more calls in, which pushes prices lower. It's a competitive floor.

Gas matters at the other end. Base settlement is cheap but it's not free. Below half a cent, facilitator gas starts looking like a meaningful slice of net revenue and the math gets ugly. So nothing in the portfolio sits below $0.005. The lower boundary isn't a design choice. It's economics.

The $0.30 stack

That stack at $0.30 isn't random. 17 endpoints all sit on that exact price. Almost every one of them wraps an expensive upstream call: brand-clearance classifiers, satellite-imagery tiles, identity-resolution lookups, premium licensing checks. The cost basis for those upstream APIs sits at 12–18 cents per call before any margin. $0.30 is the lowest price that pays facilitator fees and leaves enough margin to bother shipping.

A few of those could probably go to $0.40 or $0.50, but $0.30 is the line agents have been trained to accept for "expensive operation." Push past it and you trip a refusal heuristic in a lot of routing layers. The Coinbase facilitator data backs this up: refusal rate roughly doubles between $0.30 and $0.50.

What this means for revenue

If every endpoint caught its fair share of traffic (it doesn't, but bear with the model), the long tail would dominate. 22 endpoints at or above $0.30 would pull roughly 40% of expected per-call revenue at flat demand.

Actual traffic skews uglier. The penny endpoints do most of the calls and bring in maybe 30% of dollar revenue. The $0.30 stack does about 5% of calls and 50% of dollar revenue. The middle band ($0.05–$0.10) covers the rest, with the worst calls-per-dollar efficiency of any bucket.

So here's the takeaway for the next batch of endpoints.

Don't ship at $0.10 unless you have to. It's the worst price point in the portfolio: high enough to scare off planning-loop traffic, low enough that you'd need the volume the price already killed. Pick a side. Drop to $0.01 and chase calls, or push to $0.30 and own a niche where the cost basis justifies the price.

The middle is dead weight. The histogram says so loud and clear.