/segmently-unit-economics

UC-10 — Which experiment first: gain and cost to learn

Route: "which experiment first", "which metric to move first", "what to optimise" — a ranking of every lever, no change named or two to choose between ("a trial, or the paywall change first?") (SKILL.md § Router) · Eval: P10 · MCP: uc-10, cases-uc-10 Contents: Turns · Calls · Answer skeleton · Boundary · Stops · Worked example

Turns

• Nothing is asked before the table: the base is .ue/<slug>.json (none → cases/uc-00-launch-card.md's interview), the daily budget sits in volume, the horizons default to 30 and 90 days. A budget in the reader's message goes into volume ({ "amount": <budget>, "unit": "budget_day" }) with a measuredOn naming the budget as theirs ("your budget $<budget> a day; the rest from the book") before the run (SKILL.md § Files) — never the book's volume.
• A lift the reader states goes in as --override <lever>=<lift> (their "+25 % on the paywall" → --override p3=0.25); none stated → the CLI's default on every lever, labelled "assumed lift". Never propose a lift.
• After the table: ONE question, naming only the lever it asks about. "What lift do you expect on <lever>, as +X %?" The reader's answer to that question: a lift stated → a second ue rank run with --override <lever>=<lift>, printed as the skeleton; the default kept → no new run — record it, and restate no row; a restated row carries the table run's boundary whole under it.

Calls

• npx -y @segmently/cli ue rank .ue/<slug>.json [--override <lever>=<lift>] [--horizon <N>[,<M>]] — ONE run, JSON.
• Links: recommendation.links of that run (the two lifted variants; they carry the base's label and measuredOn — set both in the base before the run). No ue link call here.

Answer skeleton

1. sentences.ledger, when the run printed one — above the table
2. sentences.traffic
3. every sentences.table line, a blank line, every sentences.belowTable line as its own - bullet, a blank line, sentences.pick (SKILL.md § Blocks the CLI renders, the rank table row) — the belowTable lines are sentences.ranking, sentences.population, each sentences.realism, sentences.ceiling, sentences.blocked, sentences.price, then every sentences.warnings line
4. the privacy sentence (privacy.sentence — due for recommendation.links whatever recommendation.privacy says), then recommendation.links, each with the lever it lifts; under a structural pick's link, its row's sentences.diff line(s), whole
5. boundary — whole (SKILL.md § The boundary, rendered)
6. the one question

Boundary

Rendered: print boundary whole, after the pick and the links. Nothing about a row or the pick after it.

Stops

• recommendation.first null → sentences.pick says no lever is learnable within the horizon at this budget; no link.
• A lever with lift 0 (sentences.ceiling / sentences.blocked) is never attempted with a variant of your own; renewals is a calendar read with no sample size.
• price is ranked as a threshold on each plan's take (the price: rows); a revenue-per-user price test is cases/uc-08-test-plan.md's means test (σ from the reader's data).
• A structural row the reader wants to pursue → cases/uc-13-test-an-offer-change.md (its break-even, the reaction question, the test).

Worked example

Base: the seed scenario, named before the run as a real answer names its base — label "seed at the book's volume" and measuredOn "benchmark book, nothing measured" set in seed.json, so both recommended links carry them — 986.84 funnel starts a day (base.startsPerDay) at $45,000 a month, from the run below — every lever at the CLI's default +10 % relative ("assumed lift"), horizons 30 and 90 days. ONE run, in JSON:

npx -y @segmently/cli ue rank seed.json

This run's sentences.ledger is absent — no card has opened an entry yet, so nothing prints above the table.

sentences.traffic, every sentences.table line, every sentences.belowTable line (sentences.ranking, sentences.population, each sentences.realism, sentences.ceiling, sentences.blocked, sentences.price, then every sentences.warnings line) and sentences.pick — printed whole, in this order:

986.84 funnel starts a day (base.startsPerDay) at $45,000 a month

• Ranked by gain per day of testing: p1 and p2 are learnable inside 30 days in this run's realistic column.
• The population a test can use depends on the lever's place in the funnel — 986.84 funnel starts a day for p1, 296.05 funnel starts past step 1 a day for p2, 148.03 funnel starts reaching the paywall a day for p3, 51.81 buy-taps a day for bought, 51.81 buy-taps a day for mix, 20.93 buy-taps that pick a paid product a day for close, 0 trial starters a day for trialConv and 20.93 payers a day for renewals.
• Within 30 days you can only see a ≥ 11.6% change on p3.
• Within 30 days you can only see a ≥ 17.4% change on bought.
• Within 30 days you can only see a ≥ 36.3% change on mix.
• close: at ceiling — no lift to test (at its ceiling: payment completion is already 100.0% and the calculator's own control stops there).
• trialConv: no product on the paywall has a trial.
• Price is ranked below as a threshold on each plan's take (the price: rows); a revenue-per-user price test still needs the variance of revenue per user from your own data.
• 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

p1 is the pick (recommendation.first), p2 the runner-up (recommendation.runnerUp).

None of the 12 structural rows is realistic, so none can out-rank p1 or p2 here — a structural row only reaches a plan of its own with a stated belief, cases/uc-13-test-an-offer-change.md; this run prints no sentences.diff.

base: profitPerStart 0.20, roas 1.14; recommendation.links — the two lifted variants, round-trip checked.

Both recommended links carry the base's label= and measured= — set in the base before the run, so no label_missing / measured_missing is left to handle. Here recommendation.privacy says due false for both — the lift is ue rank's default, not the reader's number — but each carries a lifted rate off the book (p1 33 % in first), so ue check asks for the sentence; it is always allowed. 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.

recommendation.links.first of npx -y @segmently/cli ue rank seed.json (p1 lifted): https://www.segmently.ai/unit-economics?v=2&label=seed%20at%20the%20book's%20volume&cps=1.5&p1=33.0&p2=50.0&p3=35.0&vol=45000:budget_month&measured=benchmark%20book%2C%20nothing%20measured&p=Monthly%20subscription~19.99~1month~none~25.50~100.0~2.86~&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

recommendation.links.runnerUp of the same run (p2 lifted): https://www.segmently.ai/unit-economics?v=2&label=seed%20at%20the%20book's%20volume&cps=1.5&p1=30.0&p2=55.0&p3=35.0&vol=45000:budget_month&measured=benchmark%20book%2C%20nothing%20measured&p=Monthly%20subscription~19.99~1month~none~25.50~100.0~2.86~&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

boundary — printed whole (the price sentence is already under the table):

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
• 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.
• the 12 structural rows of this table — their targets, prices and trials: assumed: the CLI's proposals (price ± the run's lift, a 7-day free trial, a $29 add-on, the annual downsells)
• 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.