Thirty days · March 19 — April 18, 2026
Publication pipeline held steady at one-point-four projects a day. Membership moved sharply positive. Two thirds of shipping effort left the distribution OS behind and poured into the Intelligence app, where Atlas V2, Visual DNA, and Instagram Discovery all went live in this window without a single public announcement.
Editorial velocity held at 1.4 new projects per day. The pipeline is not linear — it batches. One day in March saw 18 projects queued at once; most days see two or three. This matches the rhythm of sourcing a cluster, processing, then releasing in waves.
8.3 events a day, every day, for thirty days — no single zero-publish day. The reliability work earlier this month (Pinterest migration to Zapier, IG scheduler fix, task state hygiene) is what made that possible.
The distribution OS processed 709 tasks in 30 days. Throughput ran 23.7 new tasks per day against 5.3 completed. The review queue is where human bandwidth lives, and the 278-task standing backlog describes that constraint.
The 33% skip rate is a feature. It is how the OS filters duplicates and off-edition content. The skip lane keeps the review queue from drowning.
The 39% review backlog is the bottleneck. Every task there needs a human decision. 278 standing is roughly ten days of unreviewed content at current throughput — worth a review sprint, or a policy that auto-approves tasks meeting simple criteria.
5% failure rate is acceptable. Most failures traced to Pinterest 401 token events or Instagram rate-limit throttles. Both are mitigated now — Pinterest via Zapier, IG via today’s scheduler refactor.
The Submissions tier is the story of the month. Going from ten to one-hundred-twenty-four paid submissions in thirty days is a twelve-fold step-change that maps directly to the Submit pricing and flow work finished in late March. A new revenue line that did not meaningfully exist before the window is now the fastest-moving one.
Twelve-fold growth. The pricing work shipped in late March is paying back inside thirty days. The fastest single revenue lever we have right now.
of total members (283 / 3,713) logged in during the last 30 days
Most readers never authenticate. Newsletter and public pages are carrying the relationship. This is a constraint on any feature locked behind login — either accept that the member-gated surface serves a narrow audience, or invest in pulling more members into the logged-in state with a member-only drip.
One-hundred-fourteen commits across two repos. The Intelligence app received twice the shipping effort of the distribution OS. That asymmetry is deliberate — the OS has matured into maintained infrastructure while Intelligence is in build-out phase.
React dashboard rebuild — 13 pages, 25 JSON endpoints. The admin surface is no longer a Flask render chain.
Email campaign launch. First send in production, round-robin sender to respect Resend 100/day cap.
Intelligence tagger + CLAUDE.md refactor. Two productionized features plus a doc-structure rethink.
Intelligence AI search prototype — search scaffold over the full archive.
Three long-standing bugs retired — Instagram landscape cards, Pinterest API hardening, email DB out of iCloud.
Reel caption handle verification from the handle database. Eliminates wrong-@ mentions in video output.
Email event tracking in the OS dashboard. Open / click data lives in the OS now, not just Resend.
Reliability stabilization. Webflow retry, Pillow arch fix, heartbeat wrapper. The daily cron is now genuinely boring.
Pinterest migration Make.com → Zapier. Ongoing monthly token refresh retired. Carousel multi-schedule refactor and Atlas hamburger fix also shipped today.
Neon palette on white — distinct visual language for /intelligence/*.
Image search, neighbors widget, poetic-tag cleanup — atomic features that compose into Visual DNA.
3-view toggle — Projects / Images / Galaxy. The interaction model for the map.
Image-level constellation data — 1,349 images across 20 clusters. The data layer underneath Visual DNA.
Visual DNA launch — purpose-built image exploration surface.
Instagram Discovery page at /intelligence/discovery. Taste scanner now has a user-visible UI.
2,593 Instagram posts scraped across 127 accounts via Playwright. The dataset underneath Discovery.
CDN resize for map thumbnails (≈100× smaller). Map now fast at 1,000 projects.
Atlas V2 launch — full-bleed Mapbox with Thisispaper nav. The flagship /intelligence/* page.
Atlas hamburger menu fix — hashed CSS class mismatch diagnosed and corrected live on the flagship page.
Five readings. Taken together they describe a studio shifting from “building the rails” to “building on the rails.”
Ten to one-hundred-twenty-four paid submissions in thirty days is not growth, it’s a step-change. The pricing work from late March paid back in three weeks. If even half of the trajectory holds, this line will outpace monthly Thisispaper+ revenue within the next cycle. Worth doubling the funnel: a per-article “submit similar” CTA in the article footer, a dedicated landing page, an email nudge on the newsletter archive.
Seven-point-six percent of members logged in during the window. That constrains every feature we put behind authentication. Either accept that login-gated surfaces serve a narrow audience (and build accordingly), or invest in pulling members into the logged-in state — email-first invites to Arena channels, downloadable PDFs that require login, a weekly member-only email.
Three-quarters of the month’s commits went to /intelligence/*. Visual DNA, Instagram Discovery, and Atlas V2 all shipped publicly without a public announcement. The gap between shipping velocity and external storytelling is wide. One piece — “What we built in April” — would close it.
Eight-point-three events per day for thirty days without a single zero-publish day is a plateau — sustainable, predictable, reliable. The reliability work earlier in the month (Pinterest migration, IG scheduler, task state hygiene) is what made it possible. The next gain here is not volume but mix: shifting some Pinterest throughput to Arena or LinkedIn, where we aren’t publishing yet.
The newsletter is the primary relationship with thirty-four-hundred free members. The Mon/Thu pattern works — when honored. It slipped twice this month, including a same-day double-send on April 14. Protecting the Mon/Thu lock on a calendar is the cheapest reliability work we could do in the next cycle.
Ranked by leverage. The first two compound; the rest stabilize.