Cross-project synthesis — 2026-03-01

Projects covered

• Paulos — Skopos briefing pipeline (5 issues shipped)
• Authexis — Search infrastructure and scoring architecture (12+ issues shipped)
• Polymathic-h — Newsletter automation, image optimization, social sharing (5 issues shipped)

Key accomplishments

Paulos completed a focused sprint on the Skopos morning briefing, adding three new data sources: PostHog product metrics (DAU, events, sessions with week-over-week deltas), a Fastmail IMAP inbox scanner (unread/flagged counts, recent subjects), and an ICS calendar feed scanner (today’s events with all-day/timed handling). Each renders as a colored card in the HTML email. The grind pipeline handled two parallel agents successfully, though cherry-pick conflicts required manual integration when both agents touched the same files.

Authexis replaced Google CSE with Serper.dev for article discovery, solving the international content problem — Roland’s workspace now searches across 7 countries and tripled his article library from 86 to 278. A latent batch-scoring bug surfaced at scale (Claude can’t reliably produce 229-element JSON arrays), fixed by switching to individual Haiku calls per result. Also shipped article sorting, telemetry tracking, briefing email improvements, and admin visibility controls.

Polymathic-h had the most dramatic wins on the infrastructure side. Deleted 953 MB of orphaned WordPress images, moved remaining images into Hugo’s asset pipeline for automatic WebP conversion (public images went from ~1.8 GB to 82 MB), and created a shared resolve-image.html partial replacing duplicated 28-line blocks across four templates. Also built a newsletter send automation pipeline via pre-commit hooks + Brevo API, and hardened OG/JSON-LD structured data after testing against LinkedIn and Google validators.

Cross-cutting themes

Building the morning pane of glass. Paulos’s Skopos is becoming a genuine operational dashboard — calendar, inbox, product metrics, and project status all in one email. Authexis’s briefing email improvements and Polymathic-h’s newsletter automation are parallel efforts toward the same goal: information that arrives on schedule without manual intervention.

Scale revealing architectural weaknesses. Authexis’s batch scoring collapsed when Roland’s multi-country search returned 232 results. Paulos’s grind pipeline showed its merge-conflict pain point when two agents modified the same 6 files. Polymathic-h discovered that Hugo was silently publishing 2,098 raw images from a WordPress artifact directory. In each case, what worked at small scale broke at the next level up, and the fix was a cleaner architecture (individual scoring, manual integration, proper asset pipeline).

Infrastructure that pays off in multiples. The resolve-image.html partial in Polymathic-h, the per-workspace search_countries in Authexis, and the scanner/card pattern in Paulos’s Skopos — all are investments in composable infrastructure that make the next feature cheaper. The image pipeline alone cut page weight by 95%.

Carry-over

• Authexis: Kelly Pentecost’s scan returning 0 results; tags field format inconsistency; temp debug scripts in web/
• Polymathic-h: GH-38 responsive srcset/lazy loading (prepped, ready to grind); share buttons on posts; share CTA in newsletter; inline signup CTA copy
• Paulos: Nothing carrying over — all issues shipped clean

Risks

• Polymathic-h image pipeline deploy: ~2,400 file changes. First Cloudflare Pages build will process 300+ images from scratch — could be slow or hit memory limits.
• Authexis Serper API costs: Individual Haiku scoring means ~232 API calls per scan for multi-country workspaces. Pennies today, but could scale.
• Paulos calendar RRULE gap: ICS scanner doesn’t expand recurring events. Most calendar providers materialize near-future recurrences, but it’s an unverified assumption.
• Serper API key security: In Railway env vars and local .env — confirm gitignore coverage.