"Correction" is now "Ideal response", and the box invites the actual answer

[vijay-tylertech]

The label was wrong and the label is what gets read. The field has always wanted
the answer the agent SHOULD have given -- the old help text even said so -- but
"Correction" reads as a place for notes about the answer, so that is how it got
used. That matters concretely: Eval Review re-asks the question after the knowledge
files change and scores the new Foundry response against this text, and notes cannot
be compared to an answer. A box holding "this is wrong" scores meaninglessly and the
check cannot say whether the fix landed.

The recovered 2026-08-28 workspace review is the proof of intent: 2,300 characters of
formatted reply, not a critique. That is the shape the pipeline is built around.

• Renamed across the tool vocabulary -- UI labels, help text, prompts, the CSV
  header, the download filename (ideal-responses-<stamp>.csv), and the docs in
  CLAUDE.md, DAILY-WORKFLOW.md, contributor-prompting-guide.md, transcripts/README
  and ONBOARDING. Domain uses of the word are untouched: "corrections to an
  upstream source" in the Knowledge-* corpora and the templates means something
  else entirely.
• "Answer given" is now "Foundry response", in the form and in the CSV header.
• NEW "Copy Foundry response" button beside the field. Most of a bad answer is
  usually fine, and retyping the correct paragraphs to fix one sentence is the
  friction that made this box get used for notes. It copies silently into an empty
  box and warns before overwriting text already there.
• The fetch scaffolding "Review - verdict: - should have said:" is no
  longer written -- it put an annotation prompt where the answer belongs. The 67
  transcripts already carrying it render blank and shed it on first save, so no
  bulk rewrite of review data. reviewtext.PLACEHOLDERS stays, and legacy and new
  blocks are verified to behave identically through has_feedback/body_feedback.
• An ideal response IDENTICAL to the given answer no longer counts as a change.
  Otherwise pressing Copy and then deciding nothing was wrong left the button
  reading "Changes suggested", which would put the transcript into Eval Review to
  test a fix that does not exist.
• doc_popover now renders blank-line-separated paragraphs. about is escaped, so a
  long entry collapsed into one unreadable block.

Two notes on wording. The intro banner is the dictated text with one change: "the
response you expect Foundry to have provided" became "the response Foundry was
expected to provide", per the standing rule against "you"-pointed copy. And it says
Summary section rather than tab, because that is what it is -- same page, below
the exchanges.

Verified in Chrome: labels correct and "Answer given"/"Correction" absent; the legacy
placeholder rendering blank; Copy copying 1,178 chars byte-identical to the response,
silent when empty and warning on refill with the text preserved on cancel; the mark
label walking empty -> unedited copy -> edited -> back to verbatim -> cleared as
No/No/Changes/No/No; and both mark labels setting review_status=reviewed with saveDoc
stubbed so nothing was written. All six scripts parse; contrast green.

Claude-Session: https://claude.ai/code/session_01VVNYqGza2dFdeWFX7qXnwb