Work log — 2026-03-25

What shipped

Two major themes: the learning feedback loop and the homepage overhaul. Both represent the product crossing a maturity threshold — from “functional tool” to “product that can be marketed.”

Skip tracking and the learning feedback loop. Implemented the complete article impression tracking system: frontend fires impression events when articles load (#462), the user_learn handler computes skipped articles by diffing impressions against engagements (#463), and a database migration adds the impression event type to the CHECK constraint plus a unique partial index for deduplication (#464). This closes the biggest gap in PRODUCT.md’s “learns you over time” promise — the system now observes the 90% of articles users see but don’t engage with, not just the 10% they explicitly vote on.

Homepage overhaul driven by Playwright review. Started with a cold honest assessment of the homepage via Playwright — wrote a detailed review to /tmp/eclectis-playwright-review.txt scoring it 7/10. The single biggest gap: no product screenshot. Executed 10 homepage issues in a single session: added a sample briefing preview card (#465, the #1 conversion fix), podcast as 4th output (#466), elevated learning loop messaging (#467), fixed scroll-reveal animations hiding content for reduced-motion users and crawlers (#468), replaced the generic hero headline with “All your sources. One briefing.” (#469), reframed free tier pricing to lead with value (#470), added persona-driven scenarios with competitor monitoring callout (#472, #473), and added footer links (#474). The remaining homepage issues (#471 social proof, #472 personas) were either shipped or moved to backlog as content decisions.

Production bug fix and hardening. Fixed a 500 error on briefing pages caused by isomorphic-dompurify using JSDOM which fails in Vercel’s serverless runtime. Swapped to sanitize-html (pure JS), then caught and fixed a regression where inline styles were being stripped (#475). Also fixed blog prose-invert making text unreadable in light mode (#476), added a branded 404 page (#477), fixed the error boundary hardlinking to /articles (#479), and moved a redundant import out of a loop (#478).

Briefing and podcast refinements. Changed the briefing article window from a fixed 48-hour lookback to “since last briefing” with a 24-hour minimum floor. This prevents repeated articles across consecutive briefings while ensuring regeneration always has content. Applied the same logic to podcast generation. Also improved podcast script prompts to name-drop specific authors, publications, and article titles.

Scout pipeline cleared. Ran two full scout passes (error handling, security, dead code, performance, UX, features) plus a targeted follow-up on the day’s changes. All findings were executed or triaged to backlog. The codebase is clean — dead code scan found zero unused exports, security scan found no critical issues.

Articles page UX. Separated the filter/sort controls into two labeled groups (#344) — filters (source, status, score) on the left, sort on the right, with clear “Filter” and “Sort” labels.

Completed

• #344 — Confusing sort/filters (separated into labeled groups)
• #457 — Raindrop/Readwise push handlers crash on network errors
• #458 — Confirmation dialogs for passkey delete and integration remove
• #459 — Passkey challenge cleanup error logging
• #460 — Track article skips (decomposed → #462-464)
• #461 — Full data export with error tracking
• #462 — Track article impressions on /articles page
• #463 — Incorporate skip signals into user_learn prompt
• #464 — Impression event type and dedup index
• #465 — Sample briefing preview on homepage
• #466 — Podcast as 4th output on homepage
• #467 — Learning loop messaging elevated
• #468 — Scroll-reveal animation fix for reduced-motion/crawlers
• #469 — Hero headline replaced
• #470 — Free tier pricing reframed
• #472 — Persona-driven scenarios added
• #473 — Competitor monitoring use case named
• #474 — Footer links (blog, privacy, terms)
• #475 — Fix sanitize-html stripping inline styles
• #476 — Fix blog prose-invert in light mode
• #477 — Custom branded 404 page
• #478 — Move urlparse import to module level
• #479 — Fix error boundary hardlink to /articles

Also closed: #454 (Stripe env vars verified), #455 (false alarm), #376 (won’t fix — premise wrong), #77 (won’t fix — not reproducible)

Briefing/podcast fixes (no issue numbers): replaced isomorphic-dompurify with sanitize-html, changed briefing lookback to “since last briefing” with 24h min, improved podcast script name-dropping, matched podcast article window to briefing cutoff.

Carry-over

• #471 — Homepage social proof (backlog — needs real users/testimonials)
• #64 — Email inbox scanning for newsletters (backlog/discussion)
• Next.js 16 canary Turbopack build crash persists locally. Vercel builds work fine.

Risks

• Impression tracking volume — monitor engagement_events table growth. Unique partial index prevents per-article duplicates, but each page visit creates ~20 rows. First week of data will reveal the actual growth rate.
• sanitize-html inline style permissiveness — "*": ["style"] allows all inline styles. sanitize-html’s built-in CSS sanitization prevents expression() and other XSS vectors, but worth monitoring for edge cases in user-submitted content (though we don’t have any — all HTML is Claude-generated).

Flags and watch-outs

• Homepage footer links to /privacy and /terms which don’t exist yet — they’ll 404 until placeholder pages are created.
• The hero headline change (“All your sources. One briefing. Gets smarter every day.”) is a significant positioning shift. Watch for user/visitor feedback.
• Briefing “since last briefing” cutoff: if the scheduler misses a day, the next briefing will have 48+ hours of articles. The 24h minimum floor prevents empty briefings on manual regen, but a missed schedule creates a larger-than-normal briefing.