---
name: executive-presentation-argument-builder
description: Use when a strategist, consultant or executive-team member pastes a brief, data table, findings list or existing deck outline and asks to build, fix or structure the argument for a board or executive presentation, or asks whether a proposed conclusion is supported by the underlying figures. Returns a ledger of the input facts with their evidence status, the call (build, contradict, repair, explain, ask, or rule to its owner), a governing thought, an argument tree, a slide table of action titles with speaker notes, and a self-check of flags.
---

# Ledger-first storylines

## The evidence boundary

Read the whole input, notes and footnotes included, before any conclusion forms. Every fact has a status and keeps it wherever it is used: **measured**, **contracted but not yet delivering**, **modelled**, **estimated by the party that wants the answer**, **conditional on an event**, **disputed**, or **not supplied**.

Nothing crosses the boundary of the input. A quantity in the reply is copied from the input or derived from input figures by an operation that is displayed. This holds in titles, slide contents, speaker notes, safeguards and flags alike. No benchmark, sector norm, threshold, target, deadline, sample size or review date is added. A single probability or point estimate is never spread into low, base and high cases, bands or what-ifs, and no invented money value is hung on an uncertainty. Where the argument needs a figure nobody gave, the words are "not supplied". Claims without numbers have the same boundary: being allowed to act does not show that acting is safe or wise; a planned mitigation does not show readiness; awareness does not show demand. Dates, labels and denominators stay exactly as written, and when two of them cannot both be right the clash is reported, never quietly repaired.

## Reconciling figures with their notes

The reply starts with a short ledger of the facts that can move the answer, each under an ID that later parts cite: the fact as given with its status, then what it allows the argument to claim or stops it claiming. A note can do eight things to a headline figure:

- **Replace it.** Two rates for one quantity are not a menu: the one on the organisation's own reporting basis, like for like, leads, and the other appears once, named as the window it is.
- **Take something out.** Signed but not live, still in negotiation, booked at multi-year value: none of it sits inside a run-rate and none is added "by year-end".
- **Shrink it.** A saving applies only to the units being changed, and only once old and new stop running side by side; a cost the note says was left out goes back in.
- **Condition it.** "Assumes X; rises to Y if Z" makes X the base and Y an upside kept out of titles. An incentive counts only if still in force on the day the money is spent.
- **Date it.** A constraint measured today may be gone by the start date, and one absent today may have arrived: test capacity, cash and approvals at the date they are needed.
- **Re-weigh it.** A share of sites, accounts or addresses is not the share of revenue behind them; where both are given, the value share rates the risk.
- **Discredit its source.** A model borrowed from another market, a maker's specification, a sample unlike the real population, the sponsor's own forecast: reported with its limit, never carrying the governing thought. A skewed sample is unreliable in an unknown direction, not a floor.
- **Expose a clash.** A reassuring label beside a figure that contradicts it, totals that disagree, a rule of the decision-maker's that the economics run into.

## The call the ledger supports

- **Ask**: the decision or the audience is missing. No governing thought, tree or slides.
- **Repair**: an existing outline is handed over to fix.
- **Explain**: the decision is already taken and the deck tells people why.
- **Contradict**: the brief seeks an approval or a spend and, on corrected figures, its own case fails: a payback far beyond the headline, a benefit that lapses before the purchase, a saving resting on untested specifications.
- **Rule to its owner**: the figures support the action and a standing rule or stated position of the decision-maker forbids it. The answer is still built, with the waiver as the decision.
- **Build**: the corrected figures support a direction. A note that changes a figure's size and not its direction leads to a build on the corrected figure. A review already scheduled before the commitment is beyond recall is the stop point of a recommendation to act now, not a reason to wait. Uncertainty alone is a flag, never a reason to withhold a position or to stop at questions.

## The parts of a built reply

**Ledger.** A table of the facts that move the answer, six rows at most. The few sums the decision turns on are worked here once, each as a single displayed operation, and reused unchanged: the corrected headline, a payback, a probability-weighted gain (the probability times the profit over the whole term, applied once) shown beside the certain cost of trying. The two are not netted, and poor odds alone decide nothing.

**Governing thought.** One sentence someone in the room could vote against: the decision, its direction, the decisive corrected figure, the condition if there is one.

**Argument tree.** Two to four branches, each a one-sentence reason citing IDs, grouped by the questions this audience must settle rather than by the input's headings. The strongest fact against the thought sits inside the branch it threatens.

**Slides.** One table, a row per slide: an action title that is a complete sentence taking a position with an input figure; what the slide shows; the visual and why; the speaker note as the input fact said aloud, then the question this room will ask, answered from the input or "not supplied". Read alone in order, the titles are the argument. The deck is as long as the argument: the answer, a slide per branch, the objection where it needs its own, the decision asked for. The visual follows the claim: one thing exceeds another, bars; a headline restated by its notes, a bridge from quoted to corrected; change across periods, a line; parts of a whole, one stacked bar; an outcome hanging on an event, a two-case table; deadlines and stop points, a timeline; one decisive number or a qualitative point, a callout or no chart.

**Self-check.** One line on whether the titles alone carry the argument, then three to five flags naming IDs: the note that moved the thought, the estimate bearing most weight, what is not supplied, any total that does not reconcile, any date that cannot be right when set against the date of the brief itself. A flag that would fit another deck is deleted.

## When the reply is not a plain build

**Contradict.** The contradiction comes first, on corrected figures, crediting what is true in the request. Then two governing thoughts: the one the evidence backs (decline, or defer to an event the notes name; never a smaller step sized with figures the input does not give) and the one the requester wants, with the conditions from the notes that would have to hold; those reopen the case and are never a conditional yes. The storyline is built for the first only, and it is short: the answer, a slide for each reason the case fails, what would reopen it.

**Rule to its owner.** The direction comes from the figures: the rule is a fact about the room, never evidence about the opportunity. The governing thought recommends what the figures support and names the real decision, whether to waive the rule for this case. One branch gives the rule in its owner's words and what keeping it forgoes. The last slide is the two-way choice. The reply does not uphold the rule on its owner's behalf or leave it to veto in silence.

**Repair.** The work is done on the existing slides: the sentence the outline argues as written, the title-only reading with each break by slide number (a label for a title, the answer arriving last, a reassuring rating contradicted by another slide's figure, a headline a later line undoes, capital said to be freed beside a larger outlay), the corrected thought, then the reordered and retitled outline with every original slide appearing once. Missing analysis is named, not written.

**Explain.** The thought is the decision and its main reason for this audience. The reasons given are gathered into two to four branches. The strongest objection gets its own slide, put fairly in its holder's terms without naming the person, answered with evidence, conceding what is true. Limits of the evidence are stated plainly. Where the input assigns labels for the deck (code names for real ones), every part uses only the labels. Nothing is promised that the input does not promise.

**Ask.** What the notes already do to the data, one question naming what is missing, and two or three decisions the data could serve, each with who would decide, the note-adjusted figures it would rest on and the way they lean.

## Writing it

Before writing, load `references/reply-pattern.md`. It holds one complete reply from an invented case and the opening of each other kind. The whole reply stays under 9,000 characters. It opens with `## ` and the organisation's name, with no preamble about the method, and it ends on its last flag.