# Prompt 9 — Recognise a Vibe-Charting Session in Progress

> Adapted from Chapter 13 (Yuki worked example, 1:47 a.m. dashboard polish). A diagnostic entry: the prompt sequence that produces vibe charting, and the disciplined alternative that ships the same intent without the failure mode.
>
> **Verified against**: Power BI Copilot, June 2026 surface (any style-generation prompt cycle)

## The pattern this entry is about

Unlike the other entries in this library, this one is *diagnostic*: it shows both the failure mode and the disciplined alternative side-by-side. Chapter 13's five diagnostics identify vibe-charting sessions; this entry shows what the same session looks like when it is not vibe-charted.

## The failure — prompt roulette

- **Crystallize (round 1)**: *"Make it look better."*
- **Assemble**: Copilot returns rounded corners, a soft blue palette, and a card layout.
- **Refine (round 2)**: *"Warmer."*
- **Assemble**: Palette shifts warmer.
- **Refine (round 3)**: *"Less corporate."*
- **Assemble**: Softer typography, more whitespace.
- **Refine (round 4)**: *"More like the Stripe dashboard but not exactly."*
- **Assemble**: Final version; shipped as-is.

**Ship decision (the failure)**: Shipped in under a minute after the fourth render, on the strength of *it feels right*. All five diagnostics fire — ambiguity burden held entirely by the analyst; no oracle; acceptance rate near 1.0 with zero verification prompts and sub-thirty-second decision; four rerolls with no repair; over-reliance signals all present.

The WCAG 4.5:1 contrast standard was an available oracle. Yuki did not consult it. The final dashboard ships at 3.1:1 contrast — inaccessible to a nontrivial fraction of the audience — and nobody catches it because nobody checks.

## The disciplined alternative — same intent, criteria-anchored

- **Crystallize**: *"Rebuild the dashboard's visual style so it satisfies three specific criteria: (1) all body text meets the WCAG 4.5:1 contrast threshold, verified by the Coblis simulator or an equivalent; (2) one accent colour used once per Chapter 5's ink budget, on the primary KPI; (3) the primary KPI is visually dominant per Chapter 12's blurred-thumbnail test, meaning at desktop and mobile widths the primary is the first thing the eye lands on."*
- **Scope**: Copilot restates the three criteria and names one it inferred (the accent will be applied to the primary KPI card and to nothing else). Confirm.
- **Assemble**: Copilot returns a version that meets each criterion.
- **Refine**: Check each criterion explicitly — run the contrast simulator on all body text, count the accent-colour uses, run the blurred-thumbnail test at desktop and mobile widths.

**Ship decision (the disciplined version)**: Ship with a three-line defensibility bundle: the criteria the redesign was built to, the check that passed each one, and the accessibility standard cited. Roughly the same total time as the reroll sequence; the acceptance is now auditable rather than felt.

## Diagnostic quick-check

Before shipping any AI-generated visual style, ask:

> **Could I turn my last prompt into a stated pass/fail criterion?**

If the answer is *"I would have to think about what correct even means here,"* the prompt is a reroll and the acceptance is a feeling. Every subsequent verification step becomes optional in practice, and the ship decision is a vibe.

See the [companion to Chapter 13](/ch/13/) for the full five-diagnostics + four-failure-mode treatment.

---

**When to use this prompt**: Before opening a style-generation prompt cycle, or when catching yourself mid-cycle. The distinction is not the tool or the surface — it is the presence or absence of a stated criterion the output can be checked against.

**Dataset**: [yuki-session](/datasets/yuki-session/) — the annotated four-prompt session with the failing-diagnostic scorecard.