Selling the reviews you already wrote

If you have used Google Maps for a decade you have probably written a few hundred reviews. Google's terms are explicit that the licence you granted is non-exclusive and that the content stays yours [1]. 4me2 takes that literally: pull your reviews out through Takeout, sign each one with Ed25519 under a did:web identity you control, host them yourself, and charge AI agents per request using x402.

The pipeline works. Export to signed credential to a running server takes one command each, and the paid endpoints return HTTP 402 until payment clears. Three things broke on the way there, and the market numbers we checked afterwards changed how we think it should be sold.

A DID document no resolver could read

The whole provenance claim rests on a third party being able to fetch /.well-known/did.json, recover your public key, and check the signature. Ours published this:

"publicKeyMultibase": "zgk3M3ZZYNPCl4r81RIbk9_yTv8xEs2kCc4YmmOKNIME"

The leading z declares base58btc under the multibase spec. The value after it is base64url — note the underscore, which base58 does not contain — and the Ed25519 multicodec prefix is missing entirely. Every signature verified perfectly in our own tests, because our own code encoded and decoded it the same wrong way. Any spec-compliant resolver would have failed.

We swapped the verification method to JsonWebKey2020 with a publicKeyJwk, which is the same JWK shape the signing path already used, and added a test that decodes the published key and compares it byte-for-byte against the source. That test would have caught the original bug on day one. The previous test asserted only that the field was a string.

Three prices, one payload

4me2 sells three tiers per review: display at $0.001, summarize at $0.005, and train at $0.05. In the first implementation all three endpoints returned identical bytes with a different string in a tier field. A buyer paying attention would have paid a tenth of a cent for training-grade data forever.

Now each tier hands back a different artifact. display returns the review text and rating, with no signed credential attached. summarize adds the structured fields an inference pipeline wants: geo, place identifiers, sub-ratings, and the structured answers. train returns the full signed JWT-VC with an explicit training grant naming the author's DID as licensor. The price now maps to an artifact and a stated licence rather than to a label.

The mainnet facilitator

The x402 quickstart points you at the public facilitator at x402.org/facilitator, and every example uses Base Sepolia. Flip NETWORK to Base mainnet (eip155:8453) while leaving that facilitator in place and the server starts, prints its startup banner, reports itself healthy, and then throws on the first paid request:

RouteConfigurationError: x402 Route Configuration Errors:
  - Route "GET /api/reviews/<id>/display": Facilitator does not
    support scheme "exact" on network "eip155:8453"

Verified against @x402/core 2.18. The public facilitator settles testnets. Mainnet settlement needs one that supports it, such as Coinbase's CDP endpoint, which requires API credentials. The process looks healthy right up until money is involved.

The worse version costs real money. Leave the default testnet setting in production and everything works. Agents pay, the server returns data, the logs look clean, and the payments settle in Base Sepolia USDC, which is worth nothing. The author would have earned zero and had no signal that anything was wrong. The node now refuses to start on a testnet unless ALLOW_TESTNET=true is set explicitly, and refuses mainnet paired with a facilitator that cannot settle it. Both errors name the exact variable that fixes them.

What the data is actually worth

Platforms monetise your reviews, so cut yourself in. We went looking for the numbers behind that before building any further.

Raw point-of-interest data has been commoditised. Foursquare open-sourced a POI dataset of over 100 million places, and the Overture Maps Foundation — backed by Meta, Microsoft, Amazon, TomTom and Esri — publishes more than 75 million places for free [2]. Coordinates and business names are not the scarce input.

On the payments side, Coinbase reports 169 million x402 payments and 100,000 sellers in the protocol's first year. Independent measurement disagrees sharply: CoinDesk, citing Artemis data from March 2026, put real daily volume near 131,000 transactions and about $28,000, averaging roughly $0.20 per payment, with about half the activity flagged as gamed [3]. The rails are being built out faster than the demand for them.

Run 4me2's own numbers against that. An author with 100 reviews who licenses the entire corpus at the training tier earns about five dollars, once. As individual passive income it does not work, and saying otherwise would be the kind of claim the SEC charged Delphia over in 2024 when it overstated how client data fed its models [4].

What does have a market is the part that is scarce. Overture reports that ChatGPT surfaces roughly 1.2% of local businesses and that 83% of restaurants are absent from AI answers [2]. The gap is fresh, attributed, human opinion about places, which is exactly what a decade of your reviews contains and what a free POI dump does not. Yelp has signed paid licensing deals with OpenAI and Perplexity for local data [5]. ProRata runs a recurring revenue share with more than 500 publishers rather than a flat buyout [6]. Cloudflare bought the licensing marketplace Human Native in January 2026 and folded it into Pay Per Crawl [7].

Where this goes

The individual sovereign node stays, because some people want it and the cryptography is the interesting part. It is not the growth path. A single signed corpus sitting on one person's domain is a rounding error to any buyer, and the friction to get there today is real: a domain, a deployed server, a wallet, and a private key you must not lose.

The version with a market is aggregate. Pool verified, attributed, freshly-exported reviews across many authors, licence the corpus through rails that already exist — Coinbase's x402 Bazaar catalogues any endpoint automatically on its first settled payment, and RSL and Pay Per Crawl cover the crawler side — then share revenue back to the people who wrote the reviews. A co-op with cryptographic receipts, in other words.

The code is Apache 2.0 and the pipeline is real today: github.com/tymrtn/4me2. Managed nodes, which remove the server and wallet setup, are what the waitlist at x4me2.com is for.