ADR-0007: fully implemented — the databento deviation is closed

[mspinola]

Follow-up to #22, which recorded databento as the one part of the Decision that had not been implemented. It has been: marketdata#15 ported the provider, cotdata#108 removed it. Both merged.

cotdata 0.5.0 has no bar API, no price producer of any kind, no vendor mapping in its registry, and no optional data dependency.

Doc only — 407 tests pass, and docs/adr/ is excluded from the prose guard.

The deviation section becomes history, not a deletion

The gap was real for a few hours, and an earlier reader would have been told to plan around it. So the section stays and records what it cost while it lasted — and what closing it bought:

cotdata's store is now COT-only, not just its public API. store.write_prices/read_prices, config.prices_dir() as a live path, the [databento] extra, the cotdata-prices entry point and the whole producer-half machinery went with the provider; the registry lost its vendor columns and resolve_source. Two packages can no longer both write futures bars — the coupling the contract-specs amendment argued against under "one vendor integration, one home".

One finding promoted from anecdote to expectation

The port turned up three rules that were correct in a single-domain package and wrong in a two-domain one:

1. the store's tier axis — one frame per symbol cannot hold backadj and unadj
2. the manifest's vendor key
3. databento's symbol-root default, which would have handed every equity a GLBX root and let resolve_source route SPY to a vendor that cannot serve it

A single-domain package encodes assumptions its author never had to name, and each surfaces only when a second domain exists to contradict it. Worth expecting on the next extract in this style, rather than meeting three times as a surprise. That now sits beside the existing finding from #22 (a call-site census is not a coverage census).

Open questions

Two of the three original ones are resolved; store-root convergence is what remains, and it is no longer only cosmetic — a launcher now has two roots to get right, and one was already found half-wired.

The Linux-server question is simpler than it was (a databento-sourced server now reads the same store as everyone else) but it carries a concrete action: that box's run-prices.sh still calls cotdata-prices, which no longer exists. It has to be repointed at marketdata-update --ingest-databento / --build-databento whichever way the question is settled.

The livebook/bin/daily.sh finding is marked fixed (livebook#11).

---

Generated by Claude Code