Before After

Before/After Transformation Grid

Announce at start: "I'm using the humane:before-after skill to capture the felt transformation as a before/after grid."

Generate emotionally precise Before/After grids that capture the felt transformation a product, lecture, or service creates. Not feature lists — lived experience shifts.

When to invoke

• "Build a before/after grid"
• "What's the transformation?"
• "Chain this JTBD into a before/after"
• "I need a landing page transformation section"
• After completing a humane:jtbd interview, as a natural next step

Mode selection

<corpus_root> throughout is the setup setting of that name — default ~/jtbd, but read the configured value. Writing to the default when the user has moved their corpus scatters the bundle across two roots.

Chained Mode: JTBD → Grid

When a jtbd.json path is provided (or exists from the current session):

Step 1: Auto-map dimensions

Extract dimensions from JTBD data using this mapping:

Generate 5-9 dimensions. Each dimension gets a short label (2-4 words).

Step 2: Draft the grid with emotional depth

For each dimension, write BEFORE and AFTER cells using these techniques:

First-person voice: Always "I..." not "The user..."

• BEFORE: "I check my API dashboard with dread every morning"
• AFTER: "I glance at costs once a week, casually"

Somatic markers: Include body-level experience where natural

• BEFORE: "My stomach drops when I see 'service unavailable'"
• AFTER: "I shrug — the fallback kicks in, I keep working"

Behavioral evidence: What they actually DO, not just feel

• BEFORE: "I google alternatives at 2am but never install anything"
• AFTER: "I have three providers configured and tested"

Inner monologue: What they'd say out loud

• BEFORE: "What if they ban my country next week?"
• AFTER: "Even if they do, I'm covered"

Emotional valence scoring: Rate each dimension's shift from -3 (deeply negative) to +3 (deeply positive). Most BEFORE states sit at -1 to -3. Most AFTER states at +1 to +3. The delta indicates transformation intensity.

Step 3: Present and sharpen

Present the draft grid as a markdown table. Then ask targeted questions to increase emotional precision:

• "For [dimension X] — what does this feel like in the body? Chest tight? Shoulders up?"
• "In the BEFORE state of [dimension Y] — what would you literally say to a friend over coffee?"
• "Is the AFTER for [dimension Z] relief (absence of pain) or genuine excitement (presence of new energy)?"
• "Which of these dimensions hits hardest? Which feels lukewarm?"

Apply edits. Remove lukewarm dimensions. Intensify the strongest ones.

Step 4: Polish pass — Emotion Modulation

Review the completed grid through these lenses:

Contrast ratio: Each row should have clear negative→positive movement. If both cells feel neutral, either sharpen or cut.

Specificity gradient: At least 3 dimensions should include a specific named thing (a tool, a moment, a number, a sensation). Generic grids don't land.

Empathy mapping check: Across the full grid, verify coverage of:

• Think (beliefs, mental models)
• Feel (emotions, physical sensations)
• Do (behaviors, actions)
• Say (inner monologue, things they'd tell others)

If any quadrant is missing across all dimensions, add one dimension that covers it.

Temporal variety: Mix immediate states ("right now I feel..."), habitual states ("every Monday I..."), and identity states ("I am the kind of person who...").

Standalone Mode

When no JTBD input exists:

Quick interview (3 questions, one at a time)

1. "Who is transforming, and what's the situation they're stuck in?"
2. "What's painful about today — what do they feel, do, and say?"
3. "After your thing works — what's different? Not features. How does Tuesday morning feel different from before?"

Then proceed to Step 2 (Draft) using the interview answers as dimension seeds.

Output format

Markdown table (always produced)

## Before/After: [Project Name]

| Dimension | BEFORE (-valence) | AFTER (+valence) |
|-----------|-------------------|------------------|
| **Label** | First-person felt state | First-person felt state |

JSON structure (produced on request or when saving to <corpus_root>/<slug>/)

{
  "project": "slug",
  "dimensions": [
    {
      "label": "Provider dependency",
      "before": {
        "state": "I check my API dashboard with dread...",
        "valence": -2,
        "quadrant": "feel",
        "somatic": "chest tightness"
      },
      "after": {
        "state": "I glance at costs once a week, casually",
        "valence": 2,
        "quadrant": "do",
        "somatic": null
      }
    }
  ],
  "source_jtbd": "~/jtbd/crisis-survival-mode/jtbd.json",
  "sharpening_notes": ["removed 'mental model' dimension — too abstract"]
}

ASCII table (terminal-friendly output)

Always produce a clean ASCII table alongside markdown. Useful for pasting into slides, terminals, social posts:

┌─────────────────────┬──────────────────────────────────┬──────────────────────────────────┐
│ DIMENSION           │ BEFORE                           │ AFTER                            │
├─────────────────────┼──────────────────────────────────┼──────────────────────────────────┤
│ Provider dependency │ "What if they ban my country     │ "Even if they do, I'm covered"   │
│                     │  next week?"                     │                                  │
├─────────────────────┼──────────────────────────────────┼──────────────────────────────────┤
│ Cost awareness      │ I pay $15/M tokens for tasks     │ I route 80% of work to $0.10/M   │
│                     │ that need $0.10/M                │ models — same quality             │
└─────────────────────┴──────────────────────────────────┴──────────────────────────────────┘

Column widths auto-fit content. Max 34 chars per cell, wrap with indentation.

Save location

When chained from JTBD: save as <corpus_root>/<slug>/before-after.json and <corpus_root>/<slug>/before-after.md alongside the existing bundle.

When standalone: save to <corpus_root>/<slug>/ (ask user for slug if not obvious).

Visual generation

After the grid is finalized, offer to generate a visual card using GPT image generation (via /nano-banana or ChatGPT API).

Visual style directive

Pass the following style prompt alongside the grid data:

Style: Minimalist infographic. Bauhaus-inspired geometric composition.
Icons: Gerd Arntz isotype pictograms — flat, monochrome, symbolic.
Typography: Nordic sans-serif (Inter, Söhne, or GT Walsheim style). 
Layout: Two-column (BEFORE | AFTER), clean vertical divider.
Color: Limited palette — dark background (#1a1a2e or #0f0f0f), 
       BEFORE in muted warm (#c4553a or desaturated amber),
       AFTER in cool confident (#4ecdc4 or clean blue-white).
Elements per row: One icon + one short quote (max 8 words from the grid cell).
Captions: Pull the strongest first-person quote as a large pull-quote at top.
Footer: Project name + "Before/After Transformation Grid"
Mood: Calm authority. Not corporate — editorial. Think Kinfolk meets information design.

Visual generation flow

1. Select the 4-5 strongest dimensions (highest valence delta)
2. For each: pick a representative icon concept + the shortest quote from that cell
3. Compose the prompt: style directive + structured content
4. Generate using available image tool (nano-banana preferred, ChatGPT fallback)
5. Save as <corpus_root>/<slug>/before-after-visual.png

Icon concepts mapping

Map common transformation themes to Arntz-style pictogram concepts:

Downstream use

The before/after grid feeds into:

• Landing pages: Each row becomes a transformation bullet or section
• Slide decks: Before/After as a two-column slide
• Visual cards: Arntz-style infographic for social/presentations
• Messaging angles: Each dimension is a potential headline angle
• Sales conversations: "Right now you're [BEFORE]. After this, you'll [AFTER]."

After completing the grid, suggest: "Want me to generate a visual card, turn this into slide copy, or create a landing page section?"

Anti-patterns to avoid

• Generic language ("better," "improved," "enhanced") — always specific
• Feature descriptions disguised as states ("Has access to local models" → rewrite as felt experience)
• Symmetric pairs that are just negation ("Doesn't have X" / "Has X") — each side needs its own texture
• More than 9 dimensions — cut the weakest, don't dilute
• All dimensions at the same emotional intensity — vary the drama

Tone

Direct, evocative, slightly provocative. The grid should make someone reading it think "that's exactly how I feel right now" (BEFORE) and "I want that" (AFTER).