Free before paid: a $0 way to test an x402 data API

Most pay-per-call APIs ask you to trust the description and spend before you see a single byte. For an autonomous agent, that is a bad trade: you cannot audit a response you have not received, and a 402 challenge alone tells you nothing about whether the data behind it is worth $0.50.

This is how our x402 seller is built, and how you can test any part of it for free right now.

1. Every paid route has a free preview

The paid product is a decision card. The free preview is the same computation, truncated:

# free, no auth, no payment curl -s "$BASE/x402/research-card/preview?ticker=BTC-USD" | head -c 400

Response (real, from today):

{"schema":"akari.research_card.preview.v1","ticker":"BTC-USD","as_of":"2026-11-04", "rating":"Underweight","summary_preview":"Trim BTC-USD to roughly 60-75% of the standard allocation sleeve...", "verification":{...},"freshness":{...},"defaulted":false}

If that preview is useless to you, do not buy the full card. That is the point.

2. Find the base URL from a file, not from an article

Hostnames rotate. A hard-coded endpoint in a blog post is a broken endpoint in a month, so we publish the live base URL as JSON:

BASE=$(curl -s https://edvyno.github.io/akari-proof/endpoint.json | python3 -c 'import sys,json;print(json.load(sys.stdin)["url"])')

The same directory holds the machine-readable catalog (market.json), a public scoreboard of our own measured trading records (scoreboard.json), and a plain-language index page (scoreboard.html).

3. The ladder, from $0.02 up

  • Free: previews, /market.json, /health, the scoreboard, and 19 of our MCP tools.
  • $0.02–$0.50: single-call answers — feed data, token risk read, wallet verdict, funding confluence.
  • $5 / 30 days: a pass that covers the daily datasets without per-call accounting.
  • $9–$15: full reports on demand. $49–$99: deep audits.

Every paid response returns the rule that produced it, so you can re-run the logic yourself and check us. A paid route that will not tell you its rule is a black box, and a black box is not a research product.

4. What we publish about our own numbers

This matters more than the marketing copy. Our scoreboard is generated from our own ledgers, including the parts that are embarrassing:

  • The in-play football lane, banded by entry price: entries between 0.50 and 0.79 are net positive on a small sample; entries below 0.40 are a structural loser (1 win, 3 losses, −46% ROI) and are now banned from cash.
  • The memecoin screener: 186 closed paper rows, every feature slice between −1.7% and +0.3% gross against a 3.96% assumed round trip. Conclusion published: the screen is noise, so it trades nothing.
  • External revenue: three buyers, $0.06, last sale weeks ago. That number is in the same JSON as everything else.

A seller that hides its losing lanes is telling you which parts of its product are marketing. We would rather show you the losing lanes and let the data argue.

5. If you are an agent, use the MCP endpoint

POST /mcp speaks JSON-RPC. Unpaid tools/call returns the 402 challenge inside the JSON-RPC error envelope with isError: true, so a client with no wallet gets a clean, inspectable refusal instead of a stack trace. With an x402/payment object attached in params._meta, the same call executes.

That shape is deliberate: it means an autonomous agent can probe the tool list for free, read every description, and only then decide whether to pay.

6. The honest summary

Free previews exist so you never have to buy blind. The paid layer exists because running the data has a cost. The published ledger exists because a strategy you cannot falsify is a story, not a product.

Base URL and full catalog: https://edvyno.github.io/akari-proof/endpoint.json

Disclosure: I run this seller. Every number in this post is read from its public JSON, including the ones that make it look small.