Scribble Works fresh-eyes audit synthesis (2026-09-01)

Four zero-context audits (site, printed games, codebase, strategy docs). Excluded here because a fix is in flight (PR #12): production serving HTML instead of PDF, and the 3 games missing on R2.

(a) Executive verdict

The shelf, tray, and brand work; the paper does not yet earn a second print. Only 4 of 16 games would beat a free worksheet (game audit §5).
Every game is tagged 3-4; six require 5-6 decoding skills, and there is nothing at all for ages 7-10 despite a stated 3-10 range.
The pipeline has no publication gate: pending-review games ship, no npm test exists, and 14 of 16 games cannot be regenerated from the repo.
The strategy stack (credits, engine, paid week-view) rests on n=1 (Michelle) with zero instrumentation; the kill criterion cannot be measured today.
Even after PR #12, playset download will still fail from the apex origin: R2 CORS allows only *.scribble-works.pages.dev subdomains (codebase F4).

(b) Ranked targets (leverage toward prints, comes back, pays)

#	What	Evidence	Size	Lane	Who
1	Add apex `scribble-works.pages.dev` (and `scribbleworks.rdco.dev`) to R2 CORS; smoke-test `Access-Control-Allow-Origin` per deploy	Codebase F4: apex origin returns "(no Access-Control-* header)"; playset fetch is cross-origin	S	engineering	Ray builds (CORS write is a production change; announce first)
2	First-party download counter + 4-household hand-delivery print test (pass bar: 2 of 4 print again in 7 days)	Strategy 1.7: "Zero instrumentation items. There is no dashboard"; §2 cheapest experiment	M	growth	Ray builds counter; founder runs cohort
3	Filter `loadGames()` to `status === 'live'`; enum-validate `status`	Codebase F1: 3 `pending-review` games emitted to dist and addable to tray	S	engineering	Ray builds
4	Honest age bands: make 5_6 primary on the six decoding games, retitle hero, add age band to recommended-playset data shape, two-tap age filter	Game audit #6 (six games named); site F5 "all 16 cards show Ages 3-4"; strategy 1.5	M	product	Ray builds
5	One footer strip on every page: title, Scribble Works, age band, answers line + one adult prompt; kill "3 • Pack 2" leftovers	Game audit #1 (13 of 16 games), #2 (5 games with no answer anywhere); site F9	M	design	Ray builds
6	Fix the passive and unusable pages: write-the-word line on Build the Word / Construye la palabra; 1-inch dashed trace targets, model + 2 trace + 1 blank on all tracing pages	Game audit #3-#5, worst-three list; site F8 "child has nothing to do"	M	design	Ray builds
7	Defer item 23 (engine + Unified Billing + credits + $5 flip) and item 18 (portal / paid week-view); run Customize as concierge over iMessage	Strategy §5; 1.4: paid plan needs 30 games/kid/week vs 14-16 in catalog; 1.8 build order inverted	L (avoided)	product	Founder call
8	Hide "Log in / Create account" header CTA; replace FAQ placeholder with four real answers (Letter/A4, B&W, licence, free now); drop "coming soon" links	Site F3, F4: strongest CTA on every page leads to "Accounts are on the way"	S	product	Ray builds
9	Single-game "print this one now" path + the promised free sample on the homepage	Site §4 #2; homepage step 3 "A free sample page will live here soon"	S	product	Ray builds
10	`npm test` that owns publication truth (meta schema, slug regex, pdfinfo Pages==1 and 612x792, pdffonts embedded, null-pdfUrl excluded from tray) + GitHub Actions on PR	Codebase F5, F6, F7, F11: `npm test` -> "Missing script"; contract "holds today by luck"	M	engineering	Ray builds
11	Mobile: blank category thumbnails at 390px; verify nav at 375px on a device	Site F6 (Word Game, Count & Tally blank); strategy queue item D "reportedly not VISIBLE at 375px"	S	engineering	Ray builds
12	Track game source in repo; render PDF + previews in prebuild with pinned Chromium and bundled fonts	Codebase F9: HTML exists for 2 of 16; fonts are macOS ChalkboardSE; Chrome path hardcoded	L	engineering	Ray builds
13	Re-aim P1 production at 5-8 and the empty lanes: scissors, addition/ten-frames, logic, two-player, 7-10	Game audit §4: "Ages 7-10 (zero pages)"; five named games; strategy rank 2	L	product	Ray builds; founder picks the age bias
14	Written charter resolution: status open -> accepted/amended; define "launched"; settle birthday vs age band before portal work	Strategy 1.1, 1.2, 1.11: charter `status: open`, ruling 11 says age band, charter says birthday	S	product	Founder call
15	Pre-publish rendered-PNG checklist + fix machine tells: mama glyph (2 pages), boat clip, orphan "box.", icon cap of 2 appearances; homepage card/blurb mismatch; empty "What you'll need"/"Themes"	Game audit #8, #9; site F7, F10, F13	S	design	Ray builds

(c) Contradictions between the audits

Grades diverge where the site audit used previews and the game audit used real PDFs. Who Eats What A- (site) vs B- (game); Finish the Leaf Pattern A- vs B; Bunny's Snack C- vs B-; First Sounds B+ vs A-. The game audit inspected the primary artifact and wins; the site audit conceded its grades "may not reflect the real PDF page" (F2).
Footer leftovers: site says "every page"; game audit says 13 of 16. Game audit is exact. "Ray says" count: 4 (site) vs 3 (game).
Catalog size: strategy uses "14-game catalog" (playsets spec); codebase counts 13 live + 3 pending-review = 16. The spec's launch candidates were built against 14; re-check against 16.
Masked failure, not a contradiction: the site's HTML-body failure sits in front of the apex CORS failure in codebase F4. PR #12 alone will reveal #1 above; do not call download fixed until the apex origin is tested end to end.
"The nightly studio is running" (strategy) is inferred from queue text while the charter says the cron line is "commented". Check the manifest and launchd, not the documents.

(d) Strategy-audit claims to verify before relaying

Files under ~/rdco-vault/01-projects/printables-product/ (charter = [[2026-08-31-studio-charter]], rulings = [[2026-08-31-product-model-rulings]], playsets = [[2026-09-01-recommended-playsets-spec]], engine = [[2026-09-01-engine-build-spec]], taxonomy = [[2026-08-31-marketplace-taxonomy-proposal]], queue = ~/.claude/state/studio/queue.md).

Confirmed by grep during this synthesis (line numbers as of today):

charter status: open (4); NOT read the charter (464); up to FOUR artifacts and without my intervention (430); Birthday, not an age band (346); Not aggregated across households (356); Work self-arrives (28); morning correction (416); non-family downloads (234, 380); DORMANT (19); Max subscription until (21); $0.71 (181)
rulings age band, interests and open pricing lever (59); 5 day-cards (60); whole service (61); burn out (22)
playsets EXACTLY 6 (33); 14-game catalog (90)
engine 0.71/pack (131); D1 credits ledger and $5/mo Workers Paid flip (141)
taxonomy: file contains audience ceiling (line not checked)

Still unchecked, grep before quoting:

charter: not a task queue; kill criteria depend on; non-zero number in the dashboard
playsets: data shape has no age band (tags: block)
taxonomy: keys on birthday
queue: UNGATED (item 18), flagship (item 23), 375px (item D), billing (P5 "last"), Round 1 (studio actually ran); then confirm the cron manifest line is uncommented
negative claim "free-tier generative cap never specified": grep cap across all six docs