UC-13 — Test a change to the offer

Route: "is it worth changing the price / a trial / the paywall mix / a downsell / an upsell", "how much can conversion drop if we remove the trial", "a paid trial instead of a free one", "the trial as a downsell", "test this change" (SKILL.md § Router) · Eval: P13 P14 P15 P16 · MCP: none (CLI 1.5.0) Contents: Turns · Calls · Answer skeleton · Boundary · Stops · Worked example

Turns

Fix the base — the project file .ue/<slug>.json; what the reader did not give stays the book's, labelled book. A variant never overwrites the base: it is a --lever spec or a variant file (--variant).
Name the change as a --lever spec (SKILL.md § The verbs). A downsell or upsell with no target named: the CLI proposes one (edit.proposed, walked "assumed: the CLI's proposal") — say so; the reader may correct the target, its price or its trial in the spec. A variant file the reader built → --variant <file>.
Run the lever with no belief — the row's sentences.structural: the break-even, the pool a day, what the horizon can read; with the base link, the variant link and the variant's sentences.diff.
Ask once for the reaction the lever needs, in the reader's words: a trial beside a plan — "where will trial starters come from — new buyers, or from <plan> (how many of 10)?"; a price, or a trial removed or made paid — "<plan> takes <its take in the walk line> of buy-taps today; what share do you expect after the change?" → take:<that share>, the reader's own share, never a part of today's take you work out; a downsell or an upsell — the acceptance. "I don't know" → the break-even is the answer, with the test that reads it; stop there.
With a belief → --belief <spec>=<name>:<value>: the gain, the margin, the test's n per arm and days; a card only when the reader asks for one (--card; field 5 prints the card block as cases/uc-11-hypothesis-card.md does).
Budget and horizon: a budget the reader states goes into volume first (cases/uc-10-which-experiment-first.md's rule); several changes → --together (2 or 3 arms): sentences.together, and recommendation.single when they do not all read within the horizon.
After the test → ue stat read … --lever <spec> → holds / fails / not_enough; --rebase on holds only. While it runs → once a day, ue stat watch --project-file .ue/<slug>.json --lever <spec> --a <n>/<x> --b <n>/<x> --guardrail <step>=… --today <step>=<n>/<x>,<n>/<x> --belief <name>:<value>[,…] (cases/uc-08-test-plan.md), the card's own beliefs as its --belief took them; a step the change moves prints "crash check only" — pass its last full day in --today (never its cumulative counts for a read). A downsell, upsell or upgrade reads --b alone in both (the control shows no such offer).

The ledger: a card writes a planned experiment (card.experiment.id); when the test goes live, ue exp start <project> <id>; every read is recorded and printed against the plan (context.progress.sentence); a test stopped without a re-base → ue exp close <project> <id> --reason "<why>". "What is planned, running or tested?" → ue exp list <project>.

Calls

npx -y @segmently/cli ue rank .ue/<slug>.json --lever <spec> [--belief <spec>=<name>:<value>[,…]] [--horizon <N>] — ONE run per turn; --variant <file> in place of --lever; --together with 2 or 3 arms; --card .ue/<slug>/card-<n>.json with exactly one arm.
npx -y @segmently/cli ue stat read [--a <n>/<x>] --b <n>/<x> --project-file .ue/<slug>.json --lever <spec> [--arm <spec>=<n>/<x>[,<control n>/<x>]] [--horizon <N>] [--rebase] — the control part only when the arm's decisive metric differs from --lever's; its n equals --a's.
npx -y @segmently/cli ue exp list|start|close .ue/<slug>.json … — the ledger (SKILL.md § The verbs).
Links: each structural row's links.base and links.variant, minted by the rank run like ue link (label, measured, round trip, privacy) — no ue link call for them.

Answer skeleton

sentences.ledger, when the run printed one — above everything else of the run.
sentences.structural, whole (SKILL.md § Blocks the CLI renders).
The privacy sentence when a link's privacy.due — then the base link, then each variant link, every one under a line naming the ue rank run that minted it; under each variant link its sentences.diff line(s).
sentences.together, when the run has several arms.
sentences.warnings, each line on its own line (SKILL.md § Blocks the CLI renders: a rank run that prints no table).
boundary, whole (SKILL.md § The boundary, rendered).
The one question of the turn — Turn 4's reaction, or none.

A watch (Turn 7) prints watch.lines whole, then watch.boundary; a read, read.lines whole, then context.boundary; a re-base adds its context.rebase block (SKILL.md § Blocks the CLI renders).

Boundary

Rendered: the rank run's boundary whole — its structural lines name each reaction, proposal and break-even; a read's context.boundary whole. Nothing after it about a row or the verdict.

Stops

"never clears" / "clears at every point" at the landed end are the answer — never a break-even of your own, never "≈".
A refusal: its fix with your values, run again (SKILL.md § The verbs); a value only the reader has → ask.
No belief and no benchmark → no gain: the row's point is its break-even, and no recommendation names it.
Over MCP the structural levers, --variant, --together and ue exp do not exist (CLI 1.5.0): say so and stop.

Worked example

The owner's live case (2026-09-28): Monthly $29 and Annual $99 on the paywall, the book's chain and volume (ue init live --product "Monthly~29~1month~none" --product "Annual~99~1year~none"). The change: a 7-day free-trial Annual beside the paid one.

Turn 3 — the lever with no belief

npx -y @segmently/cli ue rank .ue/live.json --lever trial-add:Annual=free~7

The run's sentences.structural, sentences.diff and boundary — printed whole:

trial_vs_no_trial (Annual 7d trial beside Annual (a 7-day free trial), taking 18.73 % of buy-taps): pays while the share of trial starters taken from Annual stays ≤ 37 %, at trial → paid 37.4 % → 37.5 % (benchmark: RevenueCat 2026 median for a 7-day trial (BB-32)) (buy-taps, 51.81 a day).

No belief given: within 30 days this traffic can tell payers per buy-tap 6.07 points away from its break-even 40.5 %.

Guardrail — Annual's take may fall to 7.97 % at the break-even.

Guardrail — Annual 7d trial's charges counted: counted at one charge — it cannot fall — a calendar read (RevenueCat → paymentsCounted), not a two-arm test.

Assumed: the calculator's proposal for a product joining the paywall — Annual 7d trial's take 18.73 %.

The variant wins if payers per buy-tap holds ≥ 40.5 % (the calculator's break-even) at the planned look.

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

ue rank .ue/live.json --lever trial-add:Annual=free~7 — links.base:

https://www.segmently.ai/unit-economics?v=2&label=live&cps=1.5&p1=30.0&p2=50.0&p3=35.0&vol=45000:budget_month&measured=the%20owner's%20own%20paywall%2C%202026-09-28&p=Monthly~29~1month~none~25.50~100.0~2.86~&p=Annual~99~1year~none~14.90~100.0~1.00~&ref=skill

ue rank .ue/live.json --lever trial-add:Annual=free~7 — links.variant:

https://www.segmently.ai/unit-economics?v=2&label=Annual%207d%20trial%20beside%20(break-even)&cps=1.5&p1=30.0&p2=50.0&p3=35.0&vol=45000:budget_month&measured=the%20owner's%20own%20paywall%2C%202026-09-28%3B%20the%20calculator's%20break-even&p=Monthly~29~1month~none~25.50~100.0~2.86~&p=Annual~99~1year~none~7.97~100.0~1.00~&p=Annual%207d%20trial~99~1year~7d-free~18.73~100.0~1.00~37.5&ref=skill

Annual 7d trial beside (break-even) vs the base: Annual 7d trial beside Annual (a 7-day free trial); Annual's take 14.9 % → 7.97 %; Annual 7d trial takes 18.73 %. Value per tap $35.99 vs $35.90 (+$0.09); profit per start $0.39 vs $0.38 (+$0.01); payers per buy-tap 40.49 % vs 40.4 % (+0.09 points); payback month 3 vs 3 (no change).

The run also warns, each line of sentences.warnings whole:

rank bought: +10.00% was asked; the calculator's own control puts it on its grid and the lifted scenario carries +10.15% — every figure of this row uses the landed lift
rank renewals: +10.00% was asked; the calculator's own control puts it on its grid and the lifted scenario carries +10.14% — every figure of this row uses the landed lift

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: $29.00, monthly, no trial — measured: your project file (price, cadence, trial)
Monthly's take 25.5 % — assumed: the CLI's init split
Monthly's charges 2.86 — assumed: the book
Monthly's completion 100 % — assumed: the book
Annual: $99.00, yearly, no trial — measured: your project file (price, cadence, trial)
Annual's take 14.9 % — assumed: the CLI's init split
Annual's charges 1 (its cadence cap) — assumed: the book
Annual'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
the +10 % on p1, p2, p3, bought, mix and renewals — assumed: ue rank's default ("assumed lift"), at each lever's landed lift in the table.
trial-add:Annual=free~7: Annual 7d trial's take 18.73 % — assumed: the calculator's proposal for a product joining the paywall
trial-add:Annual=free~7: trial_vs_no_trial (Annual 7d trial beside Annual (a 7-day free trial), taking 18.73 % of buy-taps): pays while the share of trial starters taken from Annual stays ≤ 37 %, at trial → paid 37.4 % → 37.5 % (benchmark: RevenueCat 2026 median for a 7-day trial (BB-32)) (buy-taps, 51.81 a day).
the default horizons, 30 and 90 days (ue rank's own) — assumed
the 50/50 split of the test this table plans — assumed: the test design
p1 is decidable in 7.63 days and p2 in 10.57 days at this budget.
not testable soon, in this 30-day run: p3 40.24 days, bought 88.05 days and mix 360.21 days — past 30 days at this budget, each at its own days; renewals — a calendar read.

Turn 4 — the reaction, asked once

Where will trial starters come from — new buyers, or from Annual (how many of 10)?

The reader: "About 10 of every 100 buy-taps would pick it, and half of those bought Annual anyway."

Turn 5 — with the belief

npx -y @segmently/cli ue rank .ue/live.json --lever trial-add:Annual=free~7 --belief 'trial-add:Annual=free~7=take:0.1,from:Annual:0.5'

The run's sentences.structural, sentences.diff and boundary — printed whole:

trial_vs_no_trial (Annual 7d trial beside Annual (a 7-day free trial), taking 10 % of buy-taps): pays while trial → paid stays ≥ 50 %, at 50 % from Annual of the trial's takers (your belief) (buy-taps, 51.81 a day).

trial → paid: benchmark 37.4 % → landed 37.5 % on the calculator's grid.

At trial → paid 37.4 % (benchmark: RevenueCat 2026 median for a 7-day trial (BB-32)), it loses $1,949.06 a month; the test needs 12,057 per arm, 475.44 days.

Guardrail — Annual's take may fall to 9.9 % at the break-even.

Guardrail — Annual 7d trial's charges counted: counted at one charge — it cannot fall — a calendar read (RevenueCat → paymentsCounted), not a two-arm test.

The variant wins if payers per buy-tap holds ≥ 40.4 % (the calculator's break-even) at the planned look.

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

ue rank .ue/live.json --lever trial-add:Annual=free~7 --belief 'trial-add:Annual=free~7=take:0.1,from:Annual:0.5' — links.base:

ue rank .ue/live.json --lever trial-add:Annual=free~7 --belief 'trial-add:Annual=free~7=take:0.1,from:Annual:0.5' — links.variant:

https://www.segmently.ai/unit-economics?v=2&label=Annual%207d%20trial%20beside%20(benchmark)&cps=1.5&p1=30.0&p2=50.0&p3=35.0&vol=45000:budget_month&measured=the%20owner's%20own%20paywall%2C%202026-09-28%3B%20a%20benchmark&p=Monthly~29~1month~none~25.50~100.0~2.86~&p=Annual~99~1year~none~9.90~100.0~1.00~&p=Annual%207d%20trial~99~1year~7d-free~10.00~100.0~1.00~37.5&ref=skill

Annual 7d trial beside (benchmark) vs the base: Annual 7d trial beside Annual (a 7-day free trial); Annual's take 14.9 % → 9.9 %; Annual 7d trial takes 10 %. Value per tap $34.66 vs $35.90 (−$1.24); profit per start $0.32 vs $0.38 (−$0.06); payers per buy-tap 39.15 % vs 40.4 % (−1.25 points); payback month 4 vs 3.