docs(rfc): Codex declines container decomposition — the partition was 46% overstated

[pjm17971]

pjm asked for the Codex adversarial pass before acting on
#656. It returned do not build it
at high confidence, and it's right.

I re-verified every decisive claim against the source before accepting. All hold
— and my own first check was the faulty one: a backwards-searching regex
reported "not deprecated" for all five props. Codex's line citations were exact.

The defect

cursor, cursorSequence, cursorTime, crosshairSnap, cursorFormat are
already @deprecated in favour of mounted cursor components
(ChartContainer.tsx:432, 461, 576, 590, 675). §4 called only onRegionSelect
and regionSelectModifier deprecated. There are seven, not two — so half the
"shared-x effects" group is migration residue from the interaction wave, not
evidence for <Domain>. snap was also mis-filed as mapping when it gates
annotation-to-annotation guide snapping (annotations.tsx:374-388).

claimed to <Domain>	24 of 38
defensible	13 (18 counting contested)
overstatement	11 props — 46% of the claim

Every error ran the same direction: making the cluster look larger.

Two claims withdrawn

• "Fixes Friction feat: add LiveView for composable live views #24 by construction" is false. Suppression is global
  because every row runs a query over all annotations (Layers.tsx:519).
  The fix is row-scoping it — available today, no decomposition needed.
• "Annotations are layers" wasn't a proposal. They're already children of
  <Layers> (annotations.tsx:30). What's actually unhomed is their cross-row
  coordination, which the RFC never located. The recursion was asserted, not
  closed.

What Codex added

The third option §10.1 missed: keep ChartContainer as a single-domain
facade and add an outer ChartGroup only for the uncommon multi-domain case —
or decompose internally via BoxContext/DomainContext with no public tree
break at all. Also flagged that useChartFrame() exposes one xScale/xKind/
bands (useChartFrame.ts:154-191), which multiple domains would leave
undefined — I'd treated that as a footnote.

Outcome

• RFC marked DECLINED; §1–§11 left unedited as the record, review + author
  response appended per the streaming.md convention.
• [PND-XAXISOWN] is unblocked — the reviewer's explicit finding is that
  xAxis="auto" is not wasted work. (Consumer evidence gathered en route:
  Tidal passes showAxis at 2 of 2 sites, estela at 5 of 14.)
• Friction feat: add LiveView for composable live views #24 now carries its located code fix instead of the doc sentence
  it proposed.
• PLAN.md's "structural work must precede consumers" rule survives, with the
  guard this cost us: urgency is a reason to review harder, not less. That
  RFC reached "decide in weeks" while its own §10 was unresolved and its
  evidence was wrong.

Docs only.

🤖 Generated with Claude Code