Catch the docs up to a week in which the read path moved twice

[mspinola]

Four documents, all stale in the same direction: they describe a price path that stopped
being true when ADR-0007 step 4 landed, the bars were imported, and the boot guard shipped.

docs/data-path.md

Carried a sentence written two days ago that is now wrong: "until the futures bars are
imported into the marketdata store, a local checkout renders positioning without prices."
They were imported on 2026-08-09.

Rather than delete the section, it now records what closed and what did not. The gap is
closed; the split is not. The bars exist in both stores and nothing reconciles them, so
every cotdata price run leaves this store behind until the import is re-run, which the boot
guard reports as stale.

New "Refusing early" section documents the guard: the deliberate asymmetry (nothing at
all refuses, a subset warns) and why get_bars was not changed. The next reader will
reasonably ask why the loudness lives in a separate call, and the answer belongs next to
the thing it explains.

Also: half of the dotted edge that section describes is now built and half is not. The
startup check shipped; the UI badge did not, so the history-depth leak is still silent
inside a rendered chart even though it is loud at boot. Said so, rather than letting "the
fix for the leaks above" imply both.

docs/data_architecture.md

Called Databento the primary price provider with a Yahoo fallback, describing a deployment
that has not existed for some time, and still named pardo_quant_framework.

was	now
Databento: source of truth for daily OHLC	Norgate via marketdata (Windows-only producer); Databento listed as dormant, unowned, read by nothing
Yahoo: fallback for ICE softs	the live source for the equity/ETF domain
pardo_quant_framework (3 places)	npf, with a note on why the naming matters
"degrades gracefully to Yahoo"	vendor resolution is per symbol, never blended (ADR-0006), and what a miss does depends on the caller

It also gains a scope note pointing at data-path.md, which is verified against the source
rather than written from the design, so a reader finding the two in disagreement knows
which wins.

README.md

An empty marketdata store no longer boots, so "populate both stores" stopped being advice
and became a precondition. Added the error a fresh clone will see, the reason bars cannot
be fetched on macOS or Linux, and the import command for a box that still has them under
$COTDATA_STORE/prices/.

Not in this repo: the workspace CLAUDE.md

Corrected in place (it is not in any repo, but it is what every session here reads first):

• /damage was still listed as a live crowdmon consumer, four months after that page was
  deleted in Remove the /damage page now that crowdmon is deprecated #22 and The last crowdmon references leave the repo with the page #25. The crowdmon section's open question is now answered in one
  direction: its two launchd jobs publish a panel nothing reads.
• cotdata.get_prices was still called the live API.
• The marketdata store was described as equity-only and deliberately never refreshed. That
  one needed splitting rather than replacing: the equities half IS still deliberately
  frozen, because npf's pinned snapshot compares updated_at, and the futures half is new
  and read live.

ruff clean, 138 tests pass.

🤖 Generated with Claude Code