ADR-0007 — Data-source economics: multi-chain tokens, free macro, a price guard — and what we refused

Status: Accepted

Context

The Phase-5 backlog proposed four data-source expansions: audit the real Graph gateway price before mainnet, decide subscription-vs-per-query economics, add new sources (macro, earnings transcripts, forward EPS estimates), and extend token coverage beyond Uniswap-v3-on-Ethereum. Before writing code we researched, for each candidate, the three numbers that actually decide it: per-query cost, commercial redistribution rights, and per-query (ideally x402) availability.

The research produced one organising finding: jim's core invariant — "every number traces to a public-domain primary source" — is not just a compliance stance, it is what makes the economics work. The candidates split cleanly, and the split is identical whether you sort by cost, by redistribution rights, or by invariant-fit:

• Public-domain / on-chain (EDGAR, US-gov macro, on-chain DEX data): free or ~$0.0002/query, freely redistributable, cite-able to a primary source.
• Proprietary (earnings transcripts, I/B/E/S-style EPS estimates, equity index levels): $149/mo → $1M+/yr, redistribution of derived works prohibited without an enterprise license, and not public domain — so they break the invariant and the unit economics at a $0.25–$0.50 sale price.

Decision

Build the public-domain / on-chain lane and refuse the proprietary one. Add a deterministic price guard so dynamic x402 pricing can't overcharge us.

• Multi-chain EVM tokens (thegraph.py). All Uniswap-v3 deployments share one GraphQL schema, so one query + one parser now serves Ethereum, Base, Arbitrum, and Polygon via a ChainSpec registry. The identifier carries the chain (WETH, WETH:base, 0x…:arbitrum); the cache key is chain-qualified; citations name the chain. Settlement is unchanged (still x402 on Base). On-chain data is public and The Graph charges a service fee, not a data license, so redistribution is clean. Aerodrome (a Solidly fork with a different schema) is a registry entry deferred to a follow-up parser.
• Free macro source (macro.py, product macro). Fed funds (NY Fed EFFR), CPI (BLS), and 2y/10y Treasury yields (U.S. Treasury) + a derived 2s10s spread — every figure cited to a US-government primary source (public domain, redistributable). Deliberately not FRED: FRED's API ToS forbids caching/redistribution, while the underlying agency data is public domain (17 U.S.C. §105), so we go straight to the agencies. Free data → pure-margin product, like fundamentals. Best-effort: a down upstream drops its reading, never fails the run.
• Dynamic-price guard (client.py PriceCapExceeded, threaded via base.py procure). The x402 price is set by the seller in the 402 header and is not pre-published; the unpaid pre-flight now reads it and refuses before settling if it exceeds the per-query budget. graph_probe.py became the pre-mainnet audit: it decodes the live price and PASS/FAILs it against the same ceiling.

Refused (with reasons), not merely deferred:

• Earnings transcripts & forward-EPS estimates — proprietary; every vetted vendor prohibits redistribution of derived works without a $75k–$1M+/yr enterprise license; none are on x402. At $0.50/query the cheapest option (FMP $149/mo) needs ~300 sales/month just to cover the subscription, before the redistribution license it still wouldn't grant. They break the invariant. (Narrow public-domain exception noted but not built: 8-K Exhibit 99.2 transcripts on EDGAR, ~20–30% coverage.)
• Equity index levels (S&P 500, sector benchmarks) — proprietary; no public-domain path exists. Macro uses gov-statistic context instead.
• Solana — not on The Graph; the redistributable providers (CoinGecko/Gecko­ Terminal) need an Enterprise license, the cheap ones (Birdeye, Jupiter, DefiLlama) prohibit redistribution, and none sell over x402.

The economics (the subscription-vs-per-query question)

Per-query (x402) wins at jim's scale by a wide margin; a flat subscription only wins at sustained high volume. The break-even is subscription_fee ÷ per_query_cost:

So: per-query/x402 is correct for jim today. The token product's real data cost is ~100× below the $0.01 the config historically assumed and far under the $0.10 ceiling — margin on token research is effectively 100%. The guard, not a subscription, is the thing the buy side actually needed.

Consequences

Positive

• Token coverage goes from one chain to four with no new payment surface and clean redistribution; the parser and settlement path are unchanged.
• A genuinely new free, redistributable, public-domain source (macro) — pure margin, fully inside the invariant.
• The mainnet buy leg is de-risked: a dynamic over-price is refused, not paid, and graph_probe audits it before cutover.
• We avoided sinking time into proprietary integrations that would have been legally unusable for a redistribute-derived-insight product.

Negative / trade-offs

• The community Uniswap-v3 subgraphs (Base/Arbitrum/Polygon) are "not official Uniswap Labs deployments"; Base's has thin indexer signal. Live mainnet use should verify subgraph health (the mock makes this moot on testnet).
• The real x402 price is still confirmed only by a live probe; our estimate (~$0.0002) is triangulated, not a contract — hence the guard.
• Macro hits three different gov endpoints with bespoke parsing (JSON + Atom XML); each is best-effort, so a feed change degrades coverage rather than crashing.
• No earnings-call or forward-estimate coverage — a real product gap we accept in exchange for keeping the invariant and the margins intact.

ELI5 / what I learned

I went shopping for new data. Some of it — government numbers, on-chain prices — is free or nearly free, and I'm allowed to republish what I compute from it. The rest — earnings-call transcripts, Wall-Street profit forecasts, the S&P 500 number — is expensive and comes with a "you may not pass this on" sticker, so I'd be paying a fortune to build something I'm not even allowed to sell. So I bought the free, shareable stuff (more chains of token data, plus Fed/inflation/Treasury context) and walked past the rest. I also learned the on-chain shop doesn't show its price until you're at the till, so I taught the wallet to check the price first and refuse if it's too high. The lesson: the rule "only use sources I can freely cite and resell" isn't a constraint on the business — it's the part that makes the business make money.