The tier line

Founder, 2026-09-02 18:28 ET: "not sure I have a good dividing line for free vs free authenticated vs paid authenticated plans." Every number is from the charter or the rulings, or is marked as an assumption. The four decisions at the end gate any build.

1. One principle, three tiers

The line follows the gateway. Anything pre-rendered or produced by code costs us nothing per use and is free to everyone. Anything that spends an AI-gateway call sits behind a verified sign-in and a hard cap, because the sign-in is what a cap can be keyed to. Anything generated on demand to complete a plan sits behind payment, and credits are that cap. This is rulings 25 and 26 in one sentence, sitting on ruling 20's corrected cost basis. Ruling 20 as corrected supplies only that basis (code = zero marginal cost, gateway = metered); the founder explicitly held the paywall mapping open as "a different discussion", so the tier line below is this plan's proposal, not his ruling. The planner is the deliberate exception: it spends a gateway call and stays open because the founder ruled it the tire-kick moment, so it gets a browser-keyed budget instead of a sign-in.

"Absent" means visible with a sign-in button on it, never hidden. A stranger sees the whole product and pays nothing to print.

2. Why sign in without paying

The free account has to be worth an email address on its own, or the paid tier has no top of funnel. Five things that cost us no gateway call:

• Saved kids. Up to eight, age band from birthday, interests, focus topics. The planner stops asking "how old" every visit.
• The rolling two-week plan. One game per kid per day, one kid on screen at a time, filled from the shelf by age, interest and focus with the spec's variety rules. Print the fortnight as one playset, parent page first.
• History. Done ticks, carry-forward, saved playsets, customized pages: "find last week's thing without rebuilding it."
• Feedback with replies. Only verified accounts can send (ruling 25), so only they see "received" become "fixed" with a before-and-after.
• The Customize demo budget. Three a day, twenty for the life of the account. The founder's lean (Customize as the severely restricted demo) written as a number.

Deliberately not on the list: nudges (none in v2, MA-D8; push email is its own decision) and UGC theme credit (variants are anonymous per ruling 21, so sign-in buys a toggle, not a byline).

3. The paid unit

Monthly subscription with credits, priced from the charter's contribution table. One credit = one on-demand generation: a blank slot filled, or a planned day customized with words and art. The charter costs a credit at $0.71 (a 7-page pack on Sonnet 5, 8 model calls, carried as a ceiling). A single page should land well under that; the charter's own arithmetic gives about $0.10 per page ($0.71 / 7), unmeasured, so the math below is worst case.

Recommendation: launch on the $9 / 7-credit row. It is the only row that satisfies the charter's standing rule (full-burn contribution must not fall as price rises), it needs the fewest subscribers to cover the scarce input, and seven credits maps to the product: one made-to-order day a week, the shelf covering the rest. The $7 / 6 row is the fallback if the 30-day read says price is the objection rather than value; it needs 55% more subscribers for the same cover.

Over the cap: generation stops, nothing else does. The slot reads "This month's seven are used; the shelf fills the rest until ." No auto-refill and no top-up pack at launch (Decision 4): credits exist to cap runaway generation, and a top-up makes the cap a suggestion before one month of burn data exists. Past-due keeps the calendar readable and drops the household to free caps (spec §2.6). Cancel deletes nothing.

Annual: not at launch. It locks a price we have not validated against unmeasured retention. Revisit at the first price review.

No per-kid charge (MA-D6, RM-D8): one household price, eight kids, the ninth routes to the classroom waitlist.

4. Guardrails per tier

• Anonymous. Turnstile on the planner. Per-browser (IP-hash) daily cap and an hourly burst (live: 10/hour on words, 6/hour on art). Global monthly ceilings ([[B-security-abuse]]: 1,000 word calls, 300 image generations, planner default 1,400) as the last brake and total spend cap. Known gap: the ceilings are KV read-modify-write and bypassable by parallel requests; the atomic-counter fix is queued and must land before the paid flag flips, because a credits ledger with the same race is a billing bug.
• Signed-in free. Caps keyed to the household, not the IP (feat/accounts-v2-limits): 3 + 3 per day, lifetime 20. Disposable-email deny-list; the sign-in limiter must fail closed ([[2026-09-02-audit-synthesis]] row 4, still open). Per-endpoint kill switches and the global "generative off." Charge the daily gate per generation, not per request ([[B-security-abuse]] L3: one slot currently buys up to four image calls).
• Paid. The D1 credits ledger is the cap: reserve, enqueue, settle, per the charter's tier-2 design. The daily 3 + 3 stays as a burst brake so a compromised account cannot burn a month in an hour. Stripe holds the card; we hold nothing.
• All tiers. The child's name never reaches the model (ruling 22). The UGC row stores no name and no account. The kid cap of 8 (RM-D8) applies regardless of plan.

5. Conversion moments and the copy rule

One rule everywhere: state the fact, name the free path first, one button, no timer. No countdown, no fake scarcity, never address the child, never "your kid will miss out." This is a kids' product and the parent is tired.

1. Customize on a planned day, over the daily three. "Today's three customizations are used. Tomorrow's are free. Paid households get seven a month, any day." Print it as-is first, See the paid plan second.
2. A blank slot the shelf cannot fill. "Nothing on the shelf fits 4 years old and mermaids yet. Paid households can have one made; it takes a few minutes and Ray checks it first." Pick from the shelf first, Have one made second.
3. The twentieth lifetime customization. "That was your twentieth free customization. Everything else stays free. The paid plan is $9 a month for seven made-to-order pages." One button to billing, one to close.
4. A second through eighth kid. No wall. The ninth asks "Setting this up for a class?" and offers the waitlist, not a price.
5. Anonymous, third planner use in a day (number is an assumption). "Sign in and Ray remembers who this is for." A sign-in ask, never a price.

Until Stripe exists (founder-owned account, spec §4), every "see the paid plan" button lands on a one-line intent card: "Coming soon. Tell us you want it."

6. Thirty days of measurement before touching price

Events carry tier, a household hash where one exists, and a first-visit flag ([[D-execution-vs-vision]]'s missing dimension). None carry a child field.

1. Anonymous-to-signed-in rate. signin_completed with source (which wall or link) over download with first_visit=true.
2. Free Customize burn. customize_ok per household per day and cap_hit with cap=daily|lifetime. The share of households that ever hit 3/day and the days-to-twenty distribution decide whether the demo budget starves or leaks.
3. Plan activation and week-2 return. plan_filled and slot_done. Households with a slot_done in the second week of their plan is the retention number that justifies the free plan being free.
4. Shelf shortfall rate. slot_shortfall per plan_filled, with age band and the interest tags that missed. Paid demand and the catalog gap in one number.
5. Paywall intent. paywall_seen with moment=1..3 and paywall_intent (the coming-soon tap). Intent per seen, by moment, is the price signal we have before a card exists.

Plus synthesis decision 3's per-call cost column on the UGC row, so $0.71 becomes a measured number and the allotment can rise without breaking the standing rule.

7. Decisions for the founder

Silence proceeds on the default; billing is behind a flag, so none blocks this week.

1. Is the shelf-filled two-week plan free-authenticated? This reverses ruling 12 ("The plan IS a screen, paid only... Free accounts see a single 'this week's playset' card instead of the calendar"). It does not reverse MA-D5: the founder delegated MA-D5 to this plan on 2026-09-02 18:28 ET, so answering it here is the delegation, not an override. Ruling 12 is the only founder ruling being reversed. Ray default: yes, free. Curation costs code, not gateway, and the paywall moments above presuppose a free plan with blank days. Cost of being wrong: giving away what the charter calls the product, and reversal is a takeaway from live households, worse than never giving it.
2. Launch price row. Ray default: $9 / 7 credits. Cost of being wrong: too high shows up as low paywall_intent at moment 3 before any card exists, reversible; too low means raising a price on charged households, which churns them.
3. Free Customize budget: 3/day plus 20 lifetime, or 3/day with a monthly reset? Ray default: lifetime 20. It creates moment 3. Cost of being wrong: too tight and parents never feel the magic; too loose and the gateway bill comes from households that will never pay.
4. No top-up pack and no annual plan at launch? Ray default: yes to both absences until 30 days of burn data. Cost of being wrong: a few heavy users churn at the wall in month one, and the cap held. The reverse error makes the cap a suggestion before we know the burn.

Not legal advice; sign-in and kid-profile surfaces inherit the relationship model's counsel flags (RM-D0), and nothing with a kid profile reaches production before counsel answers.

Related

[[2026-08-31-studio-charter]] §4 · [[2026-08-31-product-model-rulings]] (rulings 12, 14, 20-28) · [[2026-09-02-members-area-spec]] §2.4, §2.6, MA-D5, MA-D6 · [[2026-09-02-account-relationship-model-decision]] §6 · [[2026-09-02-audit-synthesis]] · audit-2026-09-02/B-security-abuse.md (M6, L3) · audit-2026-09-02/D-execution-vs-vision.md

Ruling 31 (18:44 ET) - Decision 1 resolved

The founder drew the line at curation, not at the gateway call: signed-in free gets the rolling two-week plan and the shelf and fills it by hand; paid gets playsets pre-filled from the child profile and focus topics, customized, and delivered by email (SMS later) on schedule, so a paying parent need not return to the site. Decision 1's default ("shelf-filled plan free") is withdrawn; ruling 12 stands in this form. Decisions 2, 3 and 4 are unchanged and still open.