/segmently-unit-economics

UC-6 — Explain a scenario that does not clear

Route: pastes a segmently.ai/unit-economics link; "why doesn't it clear", "we lose money — where is the gap?" (SKILL.md § Router) · Eval: P6 · MCP: uc-6

Turns

• None required: the pasted link (or the project) is the base. An input the reader states is filled in before the runs; nothing else is asked before the answer.

Calls

• npx -y @segmently/cli ue parse "<link>" → npx -y @segmently/cli ue evaluate "<link>" --explain — the link itself while the reader states no change, so its walk.lines name "your link". No project: the variants go under .ue/<slug>/ (a slug naming the reader's link), each a copy of ue parse's scenario with ONE change.
• npx -y @segmently/cli ue evaluate <variant>.json --explain — each located cause (each product alone on the paywall; with and without each mechanic) and each bracket point, its own run (rule 9).
• The bracket per swept product knob (SKILL.md § The bracket): paymentsCounted (0.01) and completionRate (half a point) — always, even with one product, nothing deducted and no mechanic; a value worked out by hand is only where it starts, never a cell.
• npx -y @segmently/cli ue link <file> --label '<name>' — the reader's link re-minted from ue parse's scenario with its label and a measuredOn set: the source the reader named; none named: "reader-stated, source not given" — also when every value on the link equals the book's ("benchmark book, nothing measured", SKILL.md § Files, is for a scenario the skill built, never for a link the reader pasted) — and both bracket points of every sweep.

Answer skeleton

1. The gap: the two CLI values side by side (requiredValuePerTap, valuePerTap) and their absolute difference.
2. Where it sits: by product (each alone on the paywall), by deduction (the waterfall rows), by mechanic (with and without) — each from its own run. When all three come back empty (one product, nothing deducted, no mechanic), say only that; never conclude the gap is in the chain alone — the product's own knobs move value per tap too, and the sweeps are their runs.
3. The link's words.breakEvens, printed whole — the chain rows' today and break-even cells are its values. A rate listed in breakEvens.unreachable is not ranked at all — its line states it in words.
4. Rank fixes by distance to break-even, a table with exactly these columns: lever | today | break-even | distance | measurable / belief.
   • today and break-even are the two CLI values, side by side, in their own units; a swept knob's row names both bracket points, its break-even cell the first clearing point at its landed value.
   • distance is the ABSOLUTE difference in the lever's own unit — the difference of the two values exactly as printed in the today and break-even cells: "cps $1.50 → ≤ $0.90, $0.60 lower"; "p3 35.0 % → ≥ 58.3 %, 23.3 points higher". A "% relative", "Δ %" or "×" column is a ratio you computed: never print one, and never rank by one.
   • Group the rows by unit — $ (cps) first, then points (p1, p2, p3 together), then each swept knob in its own unit (charges, then completion points) — and inside each group order by absolute distance, smallest first, whatever order breakEvens.chain prints them in; never call one lever "closest" or "the easiest" across units — each distance stands in its own unit.
   • measurable / belief is a required column, one word per row — a lever the reader can instrument and measure today (cps, paymentsCounted, completionRate, the chain rates their analytics already counts) vs one that is a belief until tested (a take, a chain step they do not instrument). A row with no verdict in that column is an unfinished table.
5. The links: privacy.sentence first when privacy.due; the reader's link re-minted and both bracket points of every sweep, each under the ue link call that minted it. The evaluate's own measured_missing, cured by that re-mint, is reported once: "measured_missing on the evaluate's link → re-minted with measured=<note>".
6. The verdict, then the boundary: walk.opener, the link's own walk.lines, then § Boundary's lines.

Boundary

Assembled (SKILL.md § The boundary) from the rendered walk: walk.opener, the link's own walk.lines (ue evaluate "<link>" --explain; a pasted link carries no source note unless measured= is on it), then:

• each sweep a rank row names, a line of its own at the landed values of both bracket runs: "the <knob> sweep points <last miss> and <first clearing> — assumed: a labelled sweep";
• not testable soon: each belief row of the rank table, until the reader instruments it.

Stops

• A rate in breakEvens.unreachable is not ranked; its break-even above 100 % is never a target.
• A swept endpoint that comes back clamped is reported typed → landed on its row (rule 9).