feat(notebooks): require feed tile summary when leading markdown is noisy

[leonardthethird]

Summary

Adds a conditional validation rule to the notebook create/edit form: if the first ~250 characters of the notebook's markdown contain formatting tokens that the feed tile cannot cleanly render as plain text, feed_tile_summary becomes a required field, with a red inline explanation under the field telling the author why.

Motivation

Notebook feed tiles fall back to getMarkdownSummary(markdown, ...) when feed_tile_summary is empty. That helper runs strip-markdown with keep: ["link"], then walks tokens with a regex that preserves inline code as well. The result: notebooks whose openings include links, inline / fenced code, raw HTML, images, or tables render with visible syntax on feed cards (e.g. [click here](https://...), backticks, <iframe ...>). Authors then have to discover the Feed Tile Summary field on their own to clean it up after the fact.

This shifts the discovery to creation time. The author is told, at submit, exactly why the summary is required and what to do.

Implementation

• front_end/src/app/(main)/questions/components/notebook_form.tsx
  • New FEED_TILE_PREVIEW_CHAR_LIMIT = 250 constant (conservative upper bound; actual tile budget is width/height-derived at render time).
  • New NOISY_MARKDOWN_PATTERNS list of regexes: images, links, inline code, fenced code, HTML tags, table rows. Plain emphasis (**bold**, *italic*) is intentionally excluded since strip-markdown cleans those up on the tile.
  • createNotebookSchema now wraps its z.object({...}) in a .superRefine() that adds a ZodIssueCode.custom issue with path: ["feed_tile_summary"] when the markdown preview is noisy and the summary is empty / whitespace-only. The existing Textarea already renders field errors as red text via FormError.
• front_end/messages/en.json
  • New i18n key feedTileSummaryRequiredForFormattedContent with the explainer text shown under the field.

Server-side validation was intentionally skipped — this is a UX rule, not a data integrity rule, and the existing notebook write serializer has no equivalent cross-field validation pattern.

Test plan

• Create a notebook with a clean plain-text opening — form submits without requiring Feed Tile Summary.
• Create a notebook whose first paragraph contains a markdown link ([text](url)) — submit blocked, red explainer text appears under Feed Tile Summary.
• Repeat with: inline code `foo`, fenced code block, raw <iframe> tag, image syntax ![](...), and a markdown table — each should trigger the requirement.
• Fill in Feed Tile Summary in each case — form submits successfully.
• Edit an existing notebook with noisy leading markdown but no summary — form correctly requires the summary on save.
• Confirm field-level error renders red (matching existing form_field styling).

🤖 Generated with Claude Code

Summary by CodeRabbit

• Improvements

  • Notebook form now requires a short plain-text feed tile summary when the notebook’s visible start contains formatted content (e.g., tables or raw HTML) that wouldn’t render cleanly on a feed; the form will prompt and validate that summary.

• Localization

  • Added an English message clarifying the feed tile summary requirement for formatted content.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Adds detection of "noisy" notebook markdown previews (raw HTML or table-like rows) and a Zod cross-field validation that requires a non-empty feed_tile_summary when the truncated preview matches those patterns; also adds an English i18n message key for the validation error.

Changes

Feed Tile Summary Validation for Formatted Content

Layer / File(s)	Summary
Noisy tile preview detection front_end/src/app/(main)/questions/components/notebook_form.tsx	Adds import of getMarkdownSummary, preview sizing constants, raw-source and rendered-preview regex patterns, and hasNoisyTilePreview which renders a truncated markdown preview and checks for raw HTML tags or table-like pipe rows.
Feed tile summary cross-field validation front_end/src/app/(main)/questions/components/notebook_form.tsx	Adds a Zod superRefine to createNotebookSchema that trims feed_tile_summary and, if empty when the preview is noisy, reports a Zod issue on feed_tile_summary using the new translation key.
English localization entry front_end/messages/en.json	Inserts feedTileSummaryRequiredForFormattedContent near feedTileSummaryPlaceholder with guidance that a short plain-text summary is required when the visible start contains a table or raw HTML.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• Metaculus/metaculus#4113: Earlier PR that added the feed_tile_summary field and UI wiring used by this cross-field validation.

Suggested reviewers

• hlbmtc

Poem

A rabbit hops through markdown streams,
Sees pipes and tags that break the dreams,
"Give me plain words, short and bright," it pleads,
So feed tiles show the clearest leads. 🐇✨

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes the main change: adding a requirement for feed tile summaries when leading markdown contains formatting that doesn't render cleanly.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/notebook-feed-tile-summary-validation

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Cleanup: Preview Environment Removed (Stale)

This preview environment has been deleted because the PR was marked as Stale (no activity detected).

Resource	Status
🌐 Preview App	Deleted
🗄️ PostgreSQL Branch	Deleted
⚡ Redis Database	Deleted
🔧 GitHub Deployments	Removed
📦 Docker Image	Retained (auto-cleanup via GHCR policies)

Push a new commit to this PR to recreate the preview environment.

---

Automated cleanup by weekly maintenance

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@front_end/src/app/`(main)/questions/components/notebook_form.tsx:
- Around line 272-276: Remove the temporary debug console logging calls in the
notebook_form.tsx onValid handler (the console.log("[FEED_TILE_DEBUG] onValid ->
submitting", ...) lines and the similar debug log at 279-281); delete these
console statements so no console.debug/console.log remains in the onValid
submission path (keep all functional logic unchanged).
- Around line 93-101: Temporary diagnostic console logging inside the
superRefine validation (the console.log call logging "[FEED_TILE_DEBUG]
superRefine fired" and fields like data.markdown, summary, noisy, willAddIssue)
must be removed before merge; delete the console.log statement (or replace it
with a dev-only debug logger behind an environment/feature flag) so no markdown
snippets or internal validation state are emitted to production consoles.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 517260d1-a100-4dc3-8da1-60115213c737

📥 Commits

Reviewing files that changed from the base of the PR and between 8b97652 and c9d44aa.

📒 Files selected for processing (1)

• front_end/src/app/(main)/questions/components/notebook_form.tsx

[SylvainChevalier]

@leonardthethird should also trigger on ordered lists and bullet lists?

Latex seems to render fine, though:

[leonardthethird]

@claude make it also trigger on lists per @SylvainChevalier 's comment. Any other edge cases worth capturing while we're at it?

[claude]

Claude finished @leonardthethird's task in 4m 20s —— View job

---

Done — extended detection to lists, blockquotes, and fenced code

Pushed as c64bf87.

What changed

strip-markdown removes structural markers before render, so lists/blockquotes/code fences are invisible in the rendered preview — I have to detect them against the raw source. Split detection into two passes:

• NOISY_RENDERED_PATTERNS (unchanged): raw HTML, table pipes — checked in the post-strip preview output.
• NOISY_RAW_SOURCE_PATTERNS (new): bullet list, ordered list, blockquote, fenced code — checked against the first TILE_VISIBLE_SOURCE_CHARS = 600 chars of source. The {0,3} leading-space allowance matches CommonMark's tolerance for block constructs. Tightened source window so we don't pester authors about a list buried 1500 chars into a long plain-text intro that won't appear on the tile anyway.

Edge cases I considered and the calls I made

Case	Call	Reason
Bullet/ordered lists	Trigger	Sylvain's ask; markers stripped → run-on prose
Blockquotes	Trigger	Same failure mode as lists
Fenced code blocks	Trigger	Loses monospace; code as prose is unreadable
Headings (#)	Skip	strip-markdown leaves clean text; reads OK on tile
Inline code (`)	Skip	strip-markdown removes backticks; reads OK
Emphasis (**, *)	Skip	strip-markdown removes; already excluded in original PR
Images	Skip	Removed entirely by strip-markdown
LaTeX ($...$)	Skip	Sylvain confirmed it re-renders correctly
Horizontal rule (* * * style)	Accepted false-positive	Would match the bullet-list regex, but HRs at the start of a notebook are vanishingly rare
Links ([txt](url))	Skip	Already an intentional skip in the existing impl — tile's MarkdownEditor re-renders them as real <a> tags

i18n

Updated feedTileSummaryRequiredForFormattedContent to mention "a list, blockquote, code block, table, or raw HTML" so the explainer text matches what now triggers it.

Test plan additions for the PR description

• Notebook starting with - item 1\n- item 2 → blocked, requires summary
• Notebook starting with 1. step one\n2. step two → blocked
• Notebook starting with > quoted thought → blocked
• Notebook starting with ```js\ncode\n``` → blocked
• Notebook starting with a LaTeX expression $E = mc^2$ → not blocked (Sylvain's case)
• Notebook with a clean plain-text intro and a list at character ~1500 → not blocked (outside the tile-visible window)
  · branch: feat/notebook-feed-tile-summary-validation

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity. It will not be closed automatically, but please consider updating it or closing it if it is no longer relevant.

[SylvainChevalier]

@claude resolve the conflicts.

[claude]

Claude finished @SylvainChevalier's task in 5m 50s —— View job

---

Conflicts resolved on feat/notebook-feed-tile-summary-validation

• Fetched latest origin/main state and inspected conflict scope
• Verified notebook_form.tsx on main is identical to the merge base — no changes needed
• Merged en.json: pulled in every new key added on main, kept feedTileSummaryRequiredForFormattedContent placed right after feedTileSummaryPlaceholder
• Validated resulting en.json parses (jq . front_end/messages/en.json) — 2359 keys
• Pushed as 3be035d

What conflicted, and why the resolution is safe

The only real conflict was in front_end/messages/en.json. Since the branch diverged in May, main added a batch of new i18n keys (Labor Hub, Midterms Hub, list/grid layout, project tags, histogram helpers, etc.) and fixed one string (settingsChangeEmailAddressSuccess: "new address" → "old address"). Our branch only added a single trailing key. Resolution: base off main and re-insert feedTileSummaryRequiredForFormattedContent inline with feedTileSummaryPlaceholder, matching the original PR's grouping intent. All of main's changes are preserved verbatim.

notebook_form.tsx shows zero drift between the merge base and current main, so this branch's version is already the correct merged state — no edit was needed. The GitHub PR "conflict" flag on that file was a spurious ancestor-graph artifact from the branch's staleness rather than an actual textual conflict.

Note on approach

The runner is on a shallow clone and git fetch / git merge aren't permitted in this environment, so I did the 3-way merge manually against origin/main and pushed the result as a single fast-forward commit rather than a merge commit. The tree contents are identical to what git merge origin/main would have produced; you can git merge origin/main yourself once you pull if you'd prefer the merge topology.
· branch: feat/notebook-feed-tile-summary-validation

[ncarazon]

Sorry, I approved wrong PR. This one is not reviewed yet