/segmently-unit-economics

UC-1 — Should part of the users see the same plan with a trial?

Route: "trial or not", "add a free trial" — what the trial does to the economics (a trial OR another change first → UC-10; remove/lose/test → UC-13) (SKILL.md § Router) · Eval: P1 · MCP: uc-1, cases-uc-1 Contents: Turns · Calls · Answer skeleton · Boundary · Stops · Worked example

Turns

1. The trial — free or paid, and its length — skipped when the reader named it.
2. The trial product's share of buy-taps once both are shown — ONE question that states the plan's default in the same sentence: "What share of buy-taps do you expect <plan> with trial to take once both are shown? I keep <plan> at <its take as the CLI printed it> unless you give its own." Never a second question mark for the plan's take.

• Trial → paid q is not a question: run the reader's q if they gave one, else the book's 0.5 or the RevenueCat 2026 median for the trial's length (≤ 4 d 25.5 %, 5–9 d 37.4 %, 17–32 d 42.5 %, book entry BB-32), labelled "benchmark: RevenueCat 2026 median, a vendor's app-store peer set", and say the reader can replace it with their own.
• No trial take, no invented one. If the reader has no take for the trial product, do not supply one and never move points from one product's take to another (that is a take split, model arithmetic). Keep the plan's take as the CLI prints it, and either ask again (one question) or show a labelled sweep of the trial's take — "sweep points, not an estimate" — never one assumed value in the headline.

Calls

• The variant, by JSON edit: copy the plan to a new product (unique name, e.g. "Monthly with trial", new id), trial = { "kind": "free", "days": 7 }, onPaywall: true, takeOfTaps = the reader's number, trialConv = q; the plan's takeOfTaps = the reader's number (or unchanged). Put the new product NEXT TO the plan it copies, ahead of any onPaywall: false products — a minted link lists the paywall's products first, and ue link refuses one that does not round-trip (link_round_trip_failed). Completion and charges inherit the plan's unless the reader gives the trial's own.
• npx -y @segmently/cli ue evaluate <variant>.json --explain — one per q the reader wants to see and per sweep point, the base included; every sweep point is its own run and its own row (rule 9).
• The break-even q — SKILL.md § The bracket, on the trial product's trialConv (half a point; its window runs 2 % – 95 %): first the window's two ends, typed 0.01 and 1.0 — each comes back clamped, reported "typed 1 % → landed 2 %" and "typed 100 % → landed 95 %" — and they count toward its ≤ 8 runs; next the grid point nearest the value interpolated between the two LANDED ends' valuePerTap (valuePerTap is linear in trialConv; that value only picks the run and is never quoted), then its neighbours until two adjacent points bracket the base's valuePerTap (its steps 1–3; step 5 still applies if they don't).
• The break-even q is due in every UC-1 answer that prints a table, the swept-take branch included: bracket it on each printed take point (the window's ends first), or on one named point, saying which.
• npx -y @segmently/cli ue link <variant>.json --label '<name>' — every row and both bracket points.

Answer skeleton

1. The table: value per tap and CAC per payer, today and each variant, with the absolute difference beside both values; every row its own run. Caution: a free trial delays the first charge, and CAC per payer rises when trial takers do not convert.
2. The break-even q (SKILL.md § The bracket, steps 3–4 — at a landed end, "clears at every point I tried" or "never clears"): the first point whose own run reaches the base's value per tap, beside the last point that did not — on the swept-take branch, per printed take point or on one named point (say which).
3. readiness.line, whole, from EVERY variant you print: the trial-verdict day count moves with the trial product's take — never carry one variant's line across the other rows, and never call it invariant without evaluating the rows you assert it for (rule 9).
4. The links: privacy.sentence first when privacy.due, each link under the ue link call that minted it.
5. The verdict, then the boundary: walk.opener, the base's walk.lines, then the lines of § Boundary.
6. The one question, while the trial product's take is still missing: Turn 2's (on the ask-again path only).

Boundary

Assembled (SKILL.md § The boundary) from the rendered walk: walk.opener, the base's walk.lines, then the variant's own inputs, each a line of its own:

• both takes — assumed: the reader's belief until the mix is counted;
• the trial product (price, cadence, its trial) — measured: the reader's answer;
• the trial product's charges and completion — assumed: inherited from <plan> (the book's) unless the reader gave its own; quote the variant run's own walk.lines entries for them, word for word;
• trialConv — the reader's (measured), else the book's 0.5 or the labelled RevenueCat median (assumed); swept points and both bracket points — assumed: a labelled sweep, at their landed values;
• not testable soon: trial-converter retention — unknown until observed (cases/uc-04-revenuecat.md); the trial verdict itself needs the readiness.line days.

Stops

• No trial take from the reader: no invented one — ask again, or the labelled sweep (Turns).
• The takes pass 100 %: the CLI warns takes_normalized — ask again, do not rescale.
• A clamped end is quoted at its landed value from then on, in every later sentence about it.
• To test this change: cases/uc-13-test-an-offer-change.md, with the variant file as --variant.

Worked example

Base: the seed scenario, named before its evaluate — label "seed at the book's volume" and measuredOn "benchmark book, nothing measured" set in seed.json, so its evaluate prints no warning (Monthly $19.99 take 25.5 %, 2.86 charges; Annual $119.99 take 14.9 %; value per tap $32.46, CAC per payer $70.72).

The takes are the READER's answers, never computed or supplied by the skill. Two example answers a reader might give (the numbers below are theirs in this example — a reader who has no trial take gets a question or a labelled sweep, never one of these numbers): (a) "with both shown, Monthly keeps 15.3 % of buy-taps and Monthly with trial takes 13.2 %"; (b) "Monthly keeps its 25.5 % (as the CLI prints it) and the trial adds 6 %". Each variant: new product "Monthly with trial" (same price and cadence, trial = {kind:"free", days:7}, takeOfTaps and trialConv as stated), the Monthly's takeOfTaps as stated.

npx -y @segmently/cli ue evaluate seed.json --explain
npx -y @segmently/cli ue evaluate trial-a-q55.json --explain

Break-even q (SKILL.md § The bracket, on trialConv, one evaluate run per point, the control's 0.5-point grid). The window's ends first: trialConv 1.0 comes back products[1].trialConv was clamped to 95.0% and 0.01 comes back clamped to 2.0% — reported "typed 100 % → landed 95 %" and "typed 1 % → landed 2 %". (a) lands at $26.78 at 2 % and $33.80 at 95 %, so the next run is the grid point interpolated between those landed ends — 0.775, which reaches today ($32.47, $0.01 above today's $32.46) — then its neighbour 0.77, which does not ($32.44, $0.02 below): four runs, and (a)'s break-even is 0.775. (b) already reaches today at the landed 2 % ($32.53): it clears at every point I tried, down to 2 %, the lowest the calculator takes — in (b) the trial's takers are all new buyers by the reader's own answer.

The readiness of every variant — it is not the base's to carry, and the trial-verdict day count moves with the trial product's take. Each variant's readiness.line — printed whole:

• (a) paywall mix decidable after 6 days · trial verdict after 54 days · annual renewals due at month 12, observable from month 13
• (b) paywall mix decidable after 6 days · trial verdict after 107 days · annual renewals due at month 12, observable from month 13

The (a) q 0.55 variant gets the measuredOn note "reader-stated takes, this session; trial conversion swept"; its ue link result says privacy.due true (carries prices, takes, trialConv), so the privacy sentence comes first, on its own line. privacy.sentence — printed whole:

This link carries your numbers in plain text — browser history, referrers and analytics can see it. I can give you the JSON file instead.

npx -y @segmently/cli ue link trial-a-q55.json --label 'trial a q55'

https://www.segmently.ai/unit-economics?v=2&label=trial%20a%20q55&cps=1.5&p1=30.0&p2=50.0&p3=35.0&vol=45000:budget_month&measured=reader-stated%20takes%2C%20this%20session%3B%20trial%20conversion%20swept&p=Monthly%20subscription~19.99~1month~none~15.30~100.0~2.86~&p=Monthly%20with%20trial~19.99~1month~7d-free~13.20~100.0~2.86~55.0&p=Annual%20subscription~119.99~1year~none~14.90~100.0~1.00~&off=Lite%20monthly~4.99~1month~none~~100.0~2.86~&off=One-time%20add-on~29~once~none~~100.0~1.00~&ref=skill

Every other row and both (a) bracket points are minted the same way.

Verdict with boundary: whether the trial pays depends on how many of its takers would have bought the plan anyway — that is the reader's belief (answer a vs b), not something the book or the CLI knows; the test that settles it needs the trial verdict's 54–107 days.

The boundary: walk.opener and the seed's walk.lines — printed whole, word for word — then the variant's own inputs (the trial product's charges and completion from the (a) q 0.55 run's own walk.lines, word for word):

These figures are this scenario's arithmetic on the inputs above.

• cost per start $1.50, p1 30 %, p2 50 % and p3 35 % — assumed: the book
• Monthly subscription: $19.99, monthly, no trial — assumed: the book's seed paywall
• Monthly subscription's take 25.5 % — assumed: the CLI's init split
• Monthly subscription's charges 2.86 — assumed: the book
• Monthly subscription's completion 100 % — assumed: the book
• Annual subscription: $119.99, yearly, no trial — assumed: the book's seed paywall
• Annual subscription's take 14.9 % — assumed: the CLI's init split
• Annual subscription's charges 1 (its cadence cap) — assumed: the book
• Annual subscription's completion 100 % — assumed: the book
• $45,000 a month — assumed: the book's volume
• the book's first renewal behind payback (monthly 60 %, from words.payback) — assumed: the book
• (a) Monthly subscription keeps 15.3 % and Monthly with trial takes 13.2 % of buy-taps; (b) Monthly subscription keeps its 25.5 % and the trial adds 6 % — assumed: the reader's belief until the mix is counted
• Monthly with trial: $19.99, monthly, a 7-day free trial — measured: the reader's answer
• Monthly with trial's charges 2.86 — assumed: the book
• Monthly with trial's completion 100 % — assumed: the book
• trial → paid 30 %, 55 % and 80 %, and the (a) bracket points 77 % and 77.5 % — assumed: a labelled sweep (the book's is 50 %)
• trial-converter retention — not testable soon: unknown until observed (RevenueCat); the trial verdict itself needs 54 days in (a) and 107 days in (b) (readiness.line)