Document 10 of 14 · docs/RISKS.md

Risks, and what the plan does about each

Risk How likely What it would cost What is in the plan for it
Rendering service unavailable or changed (Cloudflare Browser Run is young and was renamed once already in 2026) medium paid orders stall Orders stay paid and retry; the success page and email both handle "taking longer"; the Chromium-in-Vercel fallback is documented with the current package versions (ARCHITECTURE.md 4.2); the weekly Action renders one PDF so a breakage is found before a customer finds it.
Fonts fail to load in the render, so PDFs ship with fallback faces medium at first, low after wrong-looking files, refunds The runtime sets data-ready only after document.fonts.ready; the renderer waits for it; pdfcheck asserts every font is embedded on every free file and on the test order.
The preview and the PDF disagree (a browser quirk, a print-only CSS rule) low by construction the one thing the site promises not to do One HTML file, one runtime, one stylesheet; the sweep compares the sheet's computed layout in the customizer with the rendered page's text; the paper check catches the rest.
Search traffic takes months and the first eight weeks look like nothing certain morale, and the temptation to change things early The eight-week rule is written down twice; Phase 2 is thirty templates and Pinterest regardless of impressions; the calendar month pages give the earliest measurable signal because of Q4 seasonality.
Pinterest never catches medium the second traffic lever is missing Google-first design; Pinterest is additive. Rich Pins and a pin per template cost one script.
Free version cannibalises paid medium revenue below expectation The paid product is the personalization, not the design; business and event templates are useless un-personalized; the email list is the second monetization path. Reviewed at the eight-week read; gating the free download is one flag.
Piracy of the templates (save the HTML, remove the proof) certain at some level small Accepted. The audience pays for the finished file and the fit guard, not for secrecy. No DRM effort beyond the proof overlay and the print stylesheet.
Supabase free-plan pause (does not apply on Pro, kept as a note) or a paused project after a quiet month low on Pro checkout errors Keep-warm query in the weekly Action anyway; Pro org already.
A regulatory template slips in (a lease "because customers asked") medium over years legal exposure, the Studio Standard broken Section 5 exclusion, in the skill, in the catalog, and in this file. The answer is a link to a lawyer, not a prettier lease.
Vercel Hobby used commercially by accident certain if not caught terms violation SETUP.md and the cost table both say Pro the day live keys go in.
Lovable keeps syncing after the merge medium if forgotten broken builds, confusion First line of SETUP.md section 2 and the handoff.
The design drifts as thirty templates get built by different sessions high without a gate a catalogue that looks like thirty sites DESIGN.md section 6, the four meta.json drafts as reference, the skill, and the paper check. Two faces per template, colours only through --t-*.
AI copy tells creep into descriptions high without a gate trust, and the owner has had to re-request it many times Human-voice pass is in the skill, in COPY.md, and every description is rewritten by a person from a draft.
The owner's time: thirty templates is thirty paper checks certain throughput One template per session through the skill; the paper check is five minutes; the calendar and the certificate family share sources so they come faster.

What to read at eight weeks (the first honest look)

Decide then, not before: which families to deepen, whether the free download gets gated, whether bundles ship, whether the all-access pass is worth accounts.

All documents