# Examples Directory (Few-Shot Calibration)

This directory is optimized for LLM few-shot learning and Stage 5 critique calibration.

## How an LLM Must Use These Examples
- Compare every candidate output against both success and failure examples.
- Use failures as **hard anti-pattern detectors**.
- If a candidate resembles failure traits, reject and regenerate.
- Map design decisions to `STANDARDS.md` sections explicitly.

## Few-Shot Example Schema (Required)
Every example in this directory should include:
1. **Outcome** (Success or Failure)
2. **Rules Demonstrated** (specific `STANDARDS.md` sections)
3. **Workflow Checkpoint** (where issue was caught or validated)
4. **Why It Passed or Failed** (specific, non-generic explanation)
5. **Correction Pattern** (what to change next)
6. **Transfer Rule** (what future runs should reuse)

## Available Cases
- `example-01-success-minimalist.md` — **Positive**. Swiss Minimalist, textless-plate execution.
- `example-02-failure-slop.md` — **Negative**. Slop overload and recovery loop.
- `example-03-end-to-end.md` — **Positive**. Full 7-stage execution trace.
- `example-04-success-vintage-screenprint.md` — **Positive**. Analog texture and trope suppression.
- `example-05-failure-typography.md` — **Negative**. Typography distortion and hierarchy breakdown.
- `example-06-success-theater-conceptual.md` — **Positive**. Theater conceptual metaphor with disciplined composition.

Template for new examples: `examples/FEW_SHOT_TEMPLATE.md`.

## Linear Read Order for LLMs
1. `example-01-success-minimalist.md`
2. `example-02-failure-slop.md`
3. `example-05-failure-typography.md`
4. `example-03-end-to-end.md`
5. `example-04-success-vintage-screenprint.md`
6. `example-06-success-theater-conceptual.md`

## Adding New Examples
To add a new example:
1. Copy an existing file.
2. Use naming: `example-XX-<success|failure>-<theme>.md`.
3. Use the required schema above.
4. Add a rule mapping table and explicit workflow checkpoint.
5. Add the file to the Available Cases and Linear Read Order.
