# [Benchmark Title]

**Layer:** L3 Benchmark
**Parent Spec:** [Spec name + L2-XXX hash]
**Type:** [I-benchmark (fixed test set) / P-benchmark (parametric)]
**Task-type tags:** [select 1+: design / sci-sim / reconstruct / mismatch]
**Stake to register:** max(1 PWM, ceil($1 / TWAP_30d))

---

## Description

[1-2 paragraph description of what this benchmark tests. What scientific problem does it represent?]

---

## Example Data

### Input
[What input looks like. Include 1-2 sample images or numerical descriptions. Shape, dtype, value range.]

### Output (Ground Truth)
[What the ground-truth output looks like.]

### Reference Result
[Best known score on this benchmark, if any. PSNR / SSIM / recovery ratio / etc.]

### Downloadable Example
[URL or GCS path to a small example .npy bundle that users can download to preview.]

---

## Evaluation

### Dataset
[For I-benchmark: list the N fixed test scenes by URL or GCS path.]
[For P-benchmark: provide the parametric Ω + epsilon_fn formula.]

### Metric
[Specific metric formula. e.g., Q_int = (PSNR_dB - 25) × 10, clamped to [0, 100].]

### Pass Threshold
[Q_int score required to be accepted as a verified solution.]

---

## Real-World Relevance

[Why this benchmark matters for actual scientific use. Concrete example downstream applications.]

---

## Royalty Distribution

[Royalty share for benchmark builder is 15% of each cert against this benchmark. Confirm the address.]

---

## Standards Met

- [ ] Test data specified (URLs or parametric formula)
- [ ] Metric defined as an exact formula
- [ ] Pass threshold given
- [ ] Example data downloadable
- [ ] Real-world relevance explained

---

## AI Authoring Notes

Notes for ChatGPT / Claude / other AI tools:

> "Fill in this benchmark template under spec **{parent spec name}**. My use case is **{describe}**. Choose a canonical test dataset and cite its source. Define the metric as an exact formula. Set a pass threshold that reflects current SOTA. Provide a downloadable example. Follow the section structure above exactly. Return the result as a complete MD file."